]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- |
2 | // vim: ts=8 sw=2 smarttab | |
3 | /* | |
4 | * Ceph - scalable distributed file system | |
5 | * | |
6 | * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net> | |
7 | * Copyright (C) 2017 OVH | |
8 | * | |
9 | * This is free software; you can redistribute it and/or | |
10 | * modify it under the terms of the GNU Lesser General Public | |
11 | * License version 2.1, as published by the Free Software | |
12 | * Foundation. See file COPYING. | |
13 | * | |
14 | */ | |
11fdf7f2 | 15 | |
7c673cae | 16 | #include "acconfig.h" |
11fdf7f2 TL |
17 | |
18 | #include <cctype> | |
7c673cae FG |
19 | #include <fstream> |
20 | #include <iostream> | |
11fdf7f2 TL |
21 | #include <iterator> |
22 | ||
23 | #include <unistd.h> | |
7c673cae FG |
24 | #include <sys/stat.h> |
25 | #include <signal.h> | |
eafe8130 | 26 | #include <time.h> |
eafe8130 | 27 | #include <boost/range/adaptor/reversed.hpp> |
7c673cae FG |
28 | |
29 | #ifdef HAVE_SYS_PARAM_H | |
30 | #include <sys/param.h> | |
31 | #endif | |
32 | ||
33 | #ifdef HAVE_SYS_MOUNT_H | |
34 | #include <sys/mount.h> | |
35 | #endif | |
36 | ||
37 | #include "osd/PG.h" | |
20effc67 TL |
38 | #include "osd/scrubber/scrub_machine.h" |
39 | #include "osd/scrubber/pg_scrubber.h" | |
7c673cae FG |
40 | |
41 | #include "include/types.h" | |
42 | #include "include/compat.h" | |
11fdf7f2 | 43 | #include "include/random.h" |
20effc67 | 44 | #include "include/scope_guard.h" |
7c673cae FG |
45 | |
46 | #include "OSD.h" | |
47 | #include "OSDMap.h" | |
48 | #include "Watch.h" | |
49 | #include "osdc/Objecter.h" | |
50 | ||
51 | #include "common/errno.h" | |
52 | #include "common/ceph_argparse.h" | |
9f95a23c | 53 | #include "common/ceph_releases.h" |
224ce89b | 54 | #include "common/ceph_time.h" |
7c673cae | 55 | #include "common/version.h" |
f67539c2 | 56 | #include "common/async/blocked_completion.h" |
b5b8bbf5 | 57 | #include "common/pick_address.h" |
11fdf7f2 TL |
58 | #include "common/blkdev.h" |
59 | #include "common/numa.h" | |
7c673cae FG |
60 | |
61 | #include "os/ObjectStore.h" | |
62 | #ifdef HAVE_LIBFUSE | |
63 | #include "os/FuseStore.h" | |
64 | #endif | |
65 | ||
66 | #include "PrimaryLogPG.h" | |
67 | ||
7c673cae FG |
68 | #include "msg/Messenger.h" |
69 | #include "msg/Message.h" | |
70 | ||
71 | #include "mon/MonClient.h" | |
72 | ||
73 | #include "messages/MLog.h" | |
74 | ||
75 | #include "messages/MGenericMessage.h" | |
7c673cae FG |
76 | #include "messages/MOSDPing.h" |
77 | #include "messages/MOSDFailure.h" | |
78 | #include "messages/MOSDMarkMeDown.h" | |
9f95a23c | 79 | #include "messages/MOSDMarkMeDead.h" |
7c673cae FG |
80 | #include "messages/MOSDFull.h" |
81 | #include "messages/MOSDOp.h" | |
82 | #include "messages/MOSDOpReply.h" | |
83 | #include "messages/MOSDBackoff.h" | |
84 | #include "messages/MOSDBeacon.h" | |
85 | #include "messages/MOSDRepOp.h" | |
86 | #include "messages/MOSDRepOpReply.h" | |
87 | #include "messages/MOSDBoot.h" | |
88 | #include "messages/MOSDPGTemp.h" | |
11fdf7f2 | 89 | #include "messages/MOSDPGReadyToMerge.h" |
7c673cae FG |
90 | |
91 | #include "messages/MOSDMap.h" | |
92 | #include "messages/MMonGetOSDMap.h" | |
93 | #include "messages/MOSDPGNotify.h" | |
9f95a23c | 94 | #include "messages/MOSDPGNotify2.h" |
9f95a23c | 95 | #include "messages/MOSDPGQuery2.h" |
7c673cae FG |
96 | #include "messages/MOSDPGLog.h" |
97 | #include "messages/MOSDPGRemove.h" | |
98 | #include "messages/MOSDPGInfo.h" | |
9f95a23c | 99 | #include "messages/MOSDPGInfo2.h" |
7c673cae | 100 | #include "messages/MOSDPGCreate.h" |
11fdf7f2 | 101 | #include "messages/MOSDPGCreate2.h" |
7c673cae FG |
102 | #include "messages/MBackfillReserve.h" |
103 | #include "messages/MRecoveryReserve.h" | |
c07f9fc5 | 104 | #include "messages/MOSDForceRecovery.h" |
7c673cae FG |
105 | #include "messages/MOSDECSubOpWrite.h" |
106 | #include "messages/MOSDECSubOpWriteReply.h" | |
107 | #include "messages/MOSDECSubOpRead.h" | |
108 | #include "messages/MOSDECSubOpReadReply.h" | |
109 | #include "messages/MOSDPGCreated.h" | |
110 | #include "messages/MOSDPGUpdateLogMissing.h" | |
111 | #include "messages/MOSDPGUpdateLogMissingReply.h" | |
112 | ||
11fdf7f2 TL |
113 | #include "messages/MOSDPeeringOp.h" |
114 | ||
7c673cae FG |
115 | #include "messages/MOSDAlive.h" |
116 | ||
117 | #include "messages/MOSDScrub.h" | |
11fdf7f2 | 118 | #include "messages/MOSDScrub2.h" |
7c673cae | 119 | |
7c673cae FG |
120 | #include "messages/MCommand.h" |
121 | #include "messages/MCommandReply.h" | |
122 | ||
123 | #include "messages/MPGStats.h" | |
7c673cae | 124 | |
9f95a23c TL |
125 | #include "messages/MMonGetPurgedSnaps.h" |
126 | #include "messages/MMonGetPurgedSnapsReply.h" | |
127 | ||
7c673cae FG |
128 | #include "common/perf_counters.h" |
129 | #include "common/Timer.h" | |
130 | #include "common/LogClient.h" | |
131 | #include "common/AsyncReserver.h" | |
132 | #include "common/HeartbeatMap.h" | |
133 | #include "common/admin_socket.h" | |
134 | #include "common/ceph_context.h" | |
135 | ||
136 | #include "global/signal_handler.h" | |
137 | #include "global/pidfile.h" | |
138 | ||
139 | #include "include/color.h" | |
140 | #include "perfglue/cpu_profiler.h" | |
141 | #include "perfglue/heap_profiler.h" | |
142 | ||
f67539c2 | 143 | #include "osd/ClassHandler.h" |
7c673cae FG |
144 | #include "osd/OpRequest.h" |
145 | ||
146 | #include "auth/AuthAuthorizeHandler.h" | |
147 | #include "auth/RotatingKeyRing.h" | |
7c673cae FG |
148 | |
149 | #include "objclass/objclass.h" | |
150 | ||
151 | #include "common/cmdparse.h" | |
152 | #include "include/str_list.h" | |
153 | #include "include/util.h" | |
154 | ||
11fdf7f2 | 155 | #include "include/ceph_assert.h" |
7c673cae FG |
156 | #include "common/config.h" |
157 | #include "common/EventTrace.h" | |
158 | ||
11fdf7f2 TL |
159 | #include "json_spirit/json_spirit_reader.h" |
160 | #include "json_spirit/json_spirit_writer.h" | |
161 | ||
7c673cae FG |
162 | #ifdef WITH_LTTNG |
163 | #define TRACEPOINT_DEFINE | |
164 | #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
165 | #include "tracing/osd.h" | |
166 | #undef TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
167 | #undef TRACEPOINT_DEFINE | |
168 | #else | |
169 | #define tracepoint(...) | |
170 | #endif | |
20effc67 TL |
171 | |
172 | #include "osd_tracer.h" | |
173 | ||
7c673cae FG |
174 | |
175 | #define dout_context cct | |
176 | #define dout_subsys ceph_subsys_osd | |
177 | #undef dout_prefix | |
178 | #define dout_prefix _prefix(_dout, whoami, get_osdmap_epoch()) | |
179 | ||
f67539c2 TL |
180 | using std::deque; |
181 | using std::list; | |
182 | using std::lock_guard; | |
183 | using std::make_pair; | |
184 | using std::make_tuple; | |
185 | using std::make_unique; | |
186 | using std::map; | |
187 | using std::ostream; | |
188 | using std::ostringstream; | |
189 | using std::pair; | |
190 | using std::set; | |
191 | using std::string; | |
192 | using std::stringstream; | |
193 | using std::to_string; | |
194 | using std::unique_ptr; | |
195 | using std::vector; | |
196 | ||
197 | using ceph::bufferlist; | |
198 | using ceph::bufferptr; | |
199 | using ceph::decode; | |
200 | using ceph::encode; | |
201 | using ceph::fixed_u_to_string; | |
202 | using ceph::Formatter; | |
203 | using ceph::heartbeat_handle_d; | |
204 | using ceph::make_mutex; | |
205 | ||
9f95a23c TL |
206 | using namespace ceph::osd::scheduler; |
207 | using TOPNSPC::common::cmd_getval; | |
20effc67 | 208 | using TOPNSPC::common::cmd_getval_or; |
224ce89b | 209 | |
7c673cae FG |
210 | static ostream& _prefix(std::ostream* _dout, int whoami, epoch_t epoch) { |
211 | return *_dout << "osd." << whoami << " " << epoch << " "; | |
212 | } | |
213 | ||
20effc67 | 214 | |
7c673cae FG |
215 | //Initial features in new superblock. |
216 | //Features here are also automatically upgraded | |
217 | CompatSet OSD::get_osd_initial_compat_set() { | |
218 | CompatSet::FeatureSet ceph_osd_feature_compat; | |
219 | CompatSet::FeatureSet ceph_osd_feature_ro_compat; | |
220 | CompatSet::FeatureSet ceph_osd_feature_incompat; | |
221 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_BASE); | |
222 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_PGINFO); | |
223 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_OLOC); | |
224 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_LEC); | |
225 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_CATEGORIES); | |
226 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_HOBJECTPOOL); | |
227 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_BIGINFO); | |
228 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_LEVELDBINFO); | |
229 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_LEVELDBLOG); | |
230 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_SNAPMAPPER); | |
231 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_HINTS); | |
232 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_PGMETA); | |
233 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_MISSING); | |
234 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_FASTINFO); | |
c07f9fc5 | 235 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_RECOVERY_DELETES); |
9f95a23c | 236 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_SNAPMAPPER2); |
7c673cae FG |
237 | return CompatSet(ceph_osd_feature_compat, ceph_osd_feature_ro_compat, |
238 | ceph_osd_feature_incompat); | |
239 | } | |
240 | ||
241 | //Features are added here that this OSD supports. | |
242 | CompatSet OSD::get_osd_compat_set() { | |
243 | CompatSet compat = get_osd_initial_compat_set(); | |
244 | //Any features here can be set in code, but not in initial superblock | |
245 | compat.incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_SHARDS); | |
246 | return compat; | |
247 | } | |
248 | ||
f67539c2 | 249 | OSDService::OSDService(OSD *osd, ceph::async::io_context_pool& poolctx) : |
7c673cae FG |
250 | osd(osd), |
251 | cct(osd->cct), | |
20effc67 | 252 | whoami(osd->whoami), store(osd->store.get()), |
7c673cae FG |
253 | log_client(osd->log_client), clog(osd->clog), |
254 | pg_recovery_stats(osd->pg_recovery_stats), | |
255 | cluster_messenger(osd->cluster_messenger), | |
256 | client_messenger(osd->client_messenger), | |
257 | logger(osd->logger), | |
258 | recoverystate_perf(osd->recoverystate_perf), | |
259 | monc(osd->monc), | |
11fdf7f2 TL |
260 | osd_max_object_size(cct->_conf, "osd_max_object_size"), |
261 | osd_skip_data_digest(cct->_conf, "osd_skip_data_digest"), | |
262 | publish_lock{ceph::make_mutex("OSDService::publish_lock")}, | |
263 | pre_publish_lock{ceph::make_mutex("OSDService::pre_publish_lock")}, | |
7c673cae | 264 | max_oldest_map(0), |
20effc67 | 265 | m_scrub_queue{cct, *this}, |
7c673cae FG |
266 | agent_valid_iterator(false), |
267 | agent_ops(0), | |
268 | flush_mode_high_count(0), | |
269 | agent_active(true), | |
270 | agent_thread(this), | |
271 | agent_stop_flag(false), | |
7c673cae FG |
272 | agent_timer(osd->client_messenger->cct, agent_timer_lock), |
273 | last_recalibrate(ceph_clock_now()), | |
274 | promote_max_objects(0), | |
275 | promote_max_bytes(0), | |
f67539c2 | 276 | poolctx(poolctx), |
9f95a23c TL |
277 | objecter(make_unique<Objecter>(osd->client_messenger->cct, |
278 | osd->objecter_messenger, | |
f67539c2 | 279 | osd->monc, poolctx)), |
11fdf7f2 | 280 | m_objecter_finishers(cct->_conf->osd_objecter_finishers), |
7c673cae FG |
281 | watch_timer(osd->client_messenger->cct, watch_lock), |
282 | next_notif_id(0), | |
7c673cae | 283 | recovery_request_timer(cct, recovery_request_lock, false), |
11fdf7f2 | 284 | sleep_timer(cct, sleep_lock, false), |
7c673cae | 285 | reserver_finisher(cct), |
3efd9988 | 286 | local_reserver(cct, &reserver_finisher, cct->_conf->osd_max_backfills, |
7c673cae | 287 | cct->_conf->osd_min_recovery_priority), |
3efd9988 | 288 | remote_reserver(cct, &reserver_finisher, cct->_conf->osd_max_backfills, |
7c673cae | 289 | cct->_conf->osd_min_recovery_priority), |
3efd9988 | 290 | snap_reserver(cct, &reserver_finisher, |
7c673cae | 291 | cct->_conf->osd_max_trimming_pgs), |
7c673cae FG |
292 | recovery_ops_active(0), |
293 | recovery_ops_reserved(0), | |
294 | recovery_paused(false), | |
7c673cae FG |
295 | map_cache(cct, cct->_conf->osd_map_cache_size), |
296 | map_bl_cache(cct->_conf->osd_map_cache_size), | |
297 | map_bl_inc_cache(cct->_conf->osd_map_cache_size), | |
7c673cae | 298 | cur_state(NONE), |
11fdf7f2 | 299 | cur_ratio(0), physical_ratio(0), |
9f95a23c | 300 | boot_epoch(0), up_epoch(0), bind_epoch(0) |
7c673cae FG |
301 | { |
302 | objecter->init(); | |
11fdf7f2 TL |
303 | |
304 | for (int i = 0; i < m_objecter_finishers; i++) { | |
305 | ostringstream str; | |
306 | str << "objecter-finisher-" << i; | |
9f95a23c TL |
307 | auto fin = make_unique<Finisher>(osd->client_messenger->cct, str.str(), "finisher"); |
308 | objecter_finishers.push_back(std::move(fin)); | |
11fdf7f2 | 309 | } |
7c673cae FG |
310 | } |
311 | ||
31f18b77 | 312 | #ifdef PG_DEBUG_REFS |
f67539c2 | 313 | void OSDService::add_pgid(spg_t pgid, PG *pg) { |
11fdf7f2 | 314 | std::lock_guard l(pgid_lock); |
31f18b77 FG |
315 | if (!pgid_tracker.count(pgid)) { |
316 | live_pgs[pgid] = pg; | |
317 | } | |
318 | pgid_tracker[pgid]++; | |
319 | } | |
320 | void OSDService::remove_pgid(spg_t pgid, PG *pg) | |
321 | { | |
11fdf7f2 TL |
322 | std::lock_guard l(pgid_lock); |
323 | ceph_assert(pgid_tracker.count(pgid)); | |
324 | ceph_assert(pgid_tracker[pgid] > 0); | |
31f18b77 FG |
325 | pgid_tracker[pgid]--; |
326 | if (pgid_tracker[pgid] == 0) { | |
327 | pgid_tracker.erase(pgid); | |
328 | live_pgs.erase(pgid); | |
329 | } | |
330 | } | |
331 | void OSDService::dump_live_pgids() | |
332 | { | |
11fdf7f2 | 333 | std::lock_guard l(pgid_lock); |
31f18b77 FG |
334 | derr << "live pgids:" << dendl; |
335 | for (map<spg_t, int>::const_iterator i = pgid_tracker.cbegin(); | |
336 | i != pgid_tracker.cend(); | |
337 | ++i) { | |
338 | derr << "\t" << *i << dendl; | |
339 | live_pgs[i->first]->dump_live_ids(); | |
340 | } | |
341 | } | |
342 | #endif | |
343 | ||
344 | ||
9f95a23c TL |
345 | ceph::signedspan OSDService::get_mnow() |
346 | { | |
347 | return ceph::mono_clock::now() - osd->startup_time; | |
348 | } | |
7c673cae | 349 | |
11fdf7f2 TL |
350 | void OSDService::identify_splits_and_merges( |
351 | OSDMapRef old_map, | |
352 | OSDMapRef new_map, | |
353 | spg_t pgid, | |
354 | set<pair<spg_t,epoch_t>> *split_children, | |
355 | set<pair<spg_t,epoch_t>> *merge_pgs) | |
7c673cae | 356 | { |
11fdf7f2 | 357 | if (!old_map->have_pg_pool(pgid.pool())) { |
7c673cae | 358 | return; |
7c673cae | 359 | } |
7c673cae | 360 | int old_pgnum = old_map->get_pg_num(pgid.pool()); |
11fdf7f2 TL |
361 | auto p = osd->pg_num_history.pg_nums.find(pgid.pool()); |
362 | if (p == osd->pg_num_history.pg_nums.end()) { | |
363 | return; | |
364 | } | |
365 | dout(20) << __func__ << " " << pgid << " e" << old_map->get_epoch() | |
366 | << " to e" << new_map->get_epoch() | |
367 | << " pg_nums " << p->second << dendl; | |
368 | deque<spg_t> queue; | |
369 | queue.push_back(pgid); | |
eafe8130 | 370 | set<spg_t> did; |
11fdf7f2 TL |
371 | while (!queue.empty()) { |
372 | auto cur = queue.front(); | |
373 | queue.pop_front(); | |
eafe8130 | 374 | did.insert(cur); |
11fdf7f2 TL |
375 | unsigned pgnum = old_pgnum; |
376 | for (auto q = p->second.lower_bound(old_map->get_epoch()); | |
377 | q != p->second.end() && | |
378 | q->first <= new_map->get_epoch(); | |
379 | ++q) { | |
380 | if (pgnum < q->second) { | |
381 | // split? | |
382 | if (cur.ps() < pgnum) { | |
383 | set<spg_t> children; | |
384 | if (cur.is_split(pgnum, q->second, &children)) { | |
385 | dout(20) << __func__ << " " << cur << " e" << q->first | |
386 | << " pg_num " << pgnum << " -> " << q->second | |
387 | << " children " << children << dendl; | |
388 | for (auto i : children) { | |
389 | split_children->insert(make_pair(i, q->first)); | |
eafe8130 TL |
390 | if (!did.count(i)) |
391 | queue.push_back(i); | |
11fdf7f2 TL |
392 | } |
393 | } | |
394 | } else if (cur.ps() < q->second) { | |
395 | dout(20) << __func__ << " " << cur << " e" << q->first | |
396 | << " pg_num " << pgnum << " -> " << q->second | |
397 | << " is a child" << dendl; | |
398 | // normally we'd capture this from the parent, but it's | |
399 | // possible the parent doesn't exist yet (it will be | |
400 | // fabricated to allow an intervening merge). note this PG | |
401 | // as a split child here to be sure we catch it. | |
402 | split_children->insert(make_pair(cur, q->first)); | |
403 | } else { | |
404 | dout(20) << __func__ << " " << cur << " e" << q->first | |
405 | << " pg_num " << pgnum << " -> " << q->second | |
406 | << " is post-split, skipping" << dendl; | |
407 | } | |
408 | } else if (merge_pgs) { | |
409 | // merge? | |
410 | if (cur.ps() >= q->second) { | |
411 | if (cur.ps() < pgnum) { | |
412 | spg_t parent; | |
413 | if (cur.is_merge_source(pgnum, q->second, &parent)) { | |
414 | set<spg_t> children; | |
415 | parent.is_split(q->second, pgnum, &children); | |
416 | dout(20) << __func__ << " " << cur << " e" << q->first | |
417 | << " pg_num " << pgnum << " -> " << q->second | |
418 | << " is merge source, target " << parent | |
419 | << ", source(s) " << children << dendl; | |
420 | merge_pgs->insert(make_pair(parent, q->first)); | |
eafe8130 TL |
421 | if (!did.count(parent)) { |
422 | // queue (and re-scan) parent in case it might not exist yet | |
423 | // and there are some future splits pending on it | |
424 | queue.push_back(parent); | |
425 | } | |
11fdf7f2 TL |
426 | for (auto c : children) { |
427 | merge_pgs->insert(make_pair(c, q->first)); | |
eafe8130 TL |
428 | if (!did.count(c)) |
429 | queue.push_back(c); | |
11fdf7f2 TL |
430 | } |
431 | } | |
432 | } else { | |
433 | dout(20) << __func__ << " " << cur << " e" << q->first | |
434 | << " pg_num " << pgnum << " -> " << q->second | |
435 | << " is beyond old pgnum, skipping" << dendl; | |
436 | } | |
437 | } else { | |
438 | set<spg_t> children; | |
439 | if (cur.is_split(q->second, pgnum, &children)) { | |
440 | dout(20) << __func__ << " " << cur << " e" << q->first | |
441 | << " pg_num " << pgnum << " -> " << q->second | |
442 | << " is merge target, source " << children << dendl; | |
443 | for (auto c : children) { | |
444 | merge_pgs->insert(make_pair(c, q->first)); | |
eafe8130 TL |
445 | if (!did.count(c)) |
446 | queue.push_back(c); | |
11fdf7f2 TL |
447 | } |
448 | merge_pgs->insert(make_pair(cur, q->first)); | |
449 | } | |
7c673cae FG |
450 | } |
451 | } | |
11fdf7f2 | 452 | pgnum = q->second; |
7c673cae FG |
453 | } |
454 | } | |
455 | } | |
456 | ||
7c673cae FG |
457 | void OSDService::need_heartbeat_peer_update() |
458 | { | |
459 | osd->need_heartbeat_peer_update(); | |
460 | } | |
461 | ||
9f95a23c TL |
462 | HeartbeatStampsRef OSDService::get_hb_stamps(unsigned peer) |
463 | { | |
464 | std::lock_guard l(hb_stamp_lock); | |
465 | if (peer >= hb_stamps.size()) { | |
466 | hb_stamps.resize(peer + 1); | |
467 | } | |
468 | if (!hb_stamps[peer]) { | |
469 | hb_stamps[peer] = ceph::make_ref<HeartbeatStamps>(peer); | |
470 | } | |
471 | return hb_stamps[peer]; | |
472 | } | |
473 | ||
474 | void OSDService::queue_renew_lease(epoch_t epoch, spg_t spgid) | |
475 | { | |
476 | osd->enqueue_peering_evt( | |
477 | spgid, | |
478 | PGPeeringEventRef( | |
479 | std::make_shared<PGPeeringEvent>( | |
480 | epoch, epoch, | |
481 | RenewLease()))); | |
482 | } | |
483 | ||
7c673cae FG |
484 | void OSDService::start_shutdown() |
485 | { | |
486 | { | |
11fdf7f2 | 487 | std::lock_guard l(agent_timer_lock); |
7c673cae FG |
488 | agent_timer.shutdown(); |
489 | } | |
31f18b77 FG |
490 | |
491 | { | |
11fdf7f2 TL |
492 | std::lock_guard l(sleep_lock); |
493 | sleep_timer.shutdown(); | |
31f18b77 | 494 | } |
81eedcae TL |
495 | |
496 | { | |
497 | std::lock_guard l(recovery_request_lock); | |
498 | recovery_request_timer.shutdown(); | |
499 | } | |
7c673cae FG |
500 | } |
501 | ||
31f18b77 | 502 | void OSDService::shutdown_reserver() |
7c673cae FG |
503 | { |
504 | reserver_finisher.wait_for_empty(); | |
505 | reserver_finisher.stop(); | |
31f18b77 FG |
506 | } |
507 | ||
508 | void OSDService::shutdown() | |
509 | { | |
9f95a23c TL |
510 | mono_timer.suspend(); |
511 | ||
7c673cae | 512 | { |
11fdf7f2 | 513 | std::lock_guard l(watch_lock); |
7c673cae FG |
514 | watch_timer.shutdown(); |
515 | } | |
516 | ||
517 | objecter->shutdown(); | |
9f95a23c | 518 | for (auto& f : objecter_finishers) { |
11fdf7f2 TL |
519 | f->wait_for_empty(); |
520 | f->stop(); | |
7c673cae FG |
521 | } |
522 | ||
11fdf7f2 | 523 | publish_map(OSDMapRef()); |
7c673cae FG |
524 | next_osdmap = OSDMapRef(); |
525 | } | |
526 | ||
527 | void OSDService::init() | |
528 | { | |
529 | reserver_finisher.start(); | |
9f95a23c | 530 | for (auto& f : objecter_finishers) { |
11fdf7f2 TL |
531 | f->start(); |
532 | } | |
7c673cae FG |
533 | objecter->set_client_incarnation(0); |
534 | ||
535 | // deprioritize objecter in daemonperf output | |
536 | objecter->get_logger()->set_prio_adjust(-3); | |
537 | ||
538 | watch_timer.init(); | |
539 | agent_timer.init(); | |
9f95a23c | 540 | mono_timer.resume(); |
7c673cae FG |
541 | |
542 | agent_thread.create("osd_srv_agent"); | |
543 | ||
544 | if (cct->_conf->osd_recovery_delay_start) | |
545 | defer_recovery(cct->_conf->osd_recovery_delay_start); | |
546 | } | |
547 | ||
548 | void OSDService::final_init() | |
549 | { | |
550 | objecter->start(osdmap.get()); | |
551 | } | |
552 | ||
553 | void OSDService::activate_map() | |
554 | { | |
555 | // wake/unwake the tiering agent | |
9f95a23c | 556 | std::lock_guard l{agent_lock}; |
7c673cae FG |
557 | agent_active = |
558 | !osdmap->test_flag(CEPH_OSDMAP_NOTIERAGENT) && | |
559 | osd->is_active(); | |
9f95a23c | 560 | agent_cond.notify_all(); |
7c673cae FG |
561 | } |
562 | ||
181888fb FG |
563 | void OSDService::request_osdmap_update(epoch_t e) |
564 | { | |
565 | osd->osdmap_subscribe(e, false); | |
566 | } | |
567 | ||
9f95a23c | 568 | |
7c673cae FG |
569 | class AgentTimeoutCB : public Context { |
570 | PGRef pg; | |
571 | public: | |
572 | explicit AgentTimeoutCB(PGRef _pg) : pg(_pg) {} | |
573 | void finish(int) override { | |
574 | pg->agent_choose_mode_restart(); | |
575 | } | |
576 | }; | |
577 | ||
578 | void OSDService::agent_entry() | |
579 | { | |
580 | dout(10) << __func__ << " start" << dendl; | |
9f95a23c | 581 | std::unique_lock agent_locker{agent_lock}; |
7c673cae FG |
582 | |
583 | while (!agent_stop_flag) { | |
584 | if (agent_queue.empty()) { | |
585 | dout(20) << __func__ << " empty queue" << dendl; | |
9f95a23c | 586 | agent_cond.wait(agent_locker); |
7c673cae FG |
587 | continue; |
588 | } | |
589 | uint64_t level = agent_queue.rbegin()->first; | |
590 | set<PGRef>& top = agent_queue.rbegin()->second; | |
591 | dout(10) << __func__ | |
592 | << " tiers " << agent_queue.size() | |
593 | << ", top is " << level | |
594 | << " with pgs " << top.size() | |
595 | << ", ops " << agent_ops << "/" | |
596 | << cct->_conf->osd_agent_max_ops | |
597 | << (agent_active ? " active" : " NOT ACTIVE") | |
598 | << dendl; | |
599 | dout(20) << __func__ << " oids " << agent_oids << dendl; | |
600 | int max = cct->_conf->osd_agent_max_ops - agent_ops; | |
601 | int agent_flush_quota = max; | |
602 | if (!flush_mode_high_count) | |
603 | agent_flush_quota = cct->_conf->osd_agent_max_low_ops - agent_ops; | |
604 | if (agent_flush_quota <= 0 || top.empty() || !agent_active) { | |
9f95a23c | 605 | agent_cond.wait(agent_locker); |
7c673cae FG |
606 | continue; |
607 | } | |
608 | ||
609 | if (!agent_valid_iterator || agent_queue_pos == top.end()) { | |
610 | agent_queue_pos = top.begin(); | |
611 | agent_valid_iterator = true; | |
612 | } | |
613 | PGRef pg = *agent_queue_pos; | |
614 | dout(10) << "high_count " << flush_mode_high_count | |
615 | << " agent_ops " << agent_ops | |
616 | << " flush_quota " << agent_flush_quota << dendl; | |
9f95a23c | 617 | agent_locker.unlock(); |
7c673cae | 618 | if (!pg->agent_work(max, agent_flush_quota)) { |
11fdf7f2 | 619 | dout(10) << __func__ << " " << pg->pg_id |
7c673cae FG |
620 | << " no agent_work, delay for " << cct->_conf->osd_agent_delay_time |
621 | << " seconds" << dendl; | |
622 | ||
f67539c2 | 623 | logger->inc(l_osd_tier_delay); |
7c673cae | 624 | // Queue a timer to call agent_choose_mode for this pg in 5 seconds |
9f95a23c | 625 | std::lock_guard timer_locker{agent_timer_lock}; |
7c673cae FG |
626 | Context *cb = new AgentTimeoutCB(pg); |
627 | agent_timer.add_event_after(cct->_conf->osd_agent_delay_time, cb); | |
7c673cae | 628 | } |
9f95a23c | 629 | agent_locker.lock(); |
7c673cae | 630 | } |
7c673cae FG |
631 | dout(10) << __func__ << " finish" << dendl; |
632 | } | |
633 | ||
634 | void OSDService::agent_stop() | |
635 | { | |
636 | { | |
11fdf7f2 | 637 | std::lock_guard l(agent_lock); |
7c673cae FG |
638 | |
639 | // By this time all ops should be cancelled | |
11fdf7f2 | 640 | ceph_assert(agent_ops == 0); |
7c673cae FG |
641 | // By this time all PGs are shutdown and dequeued |
642 | if (!agent_queue.empty()) { | |
643 | set<PGRef>& top = agent_queue.rbegin()->second; | |
11fdf7f2 TL |
644 | derr << "agent queue not empty, for example " << (*top.begin())->get_pgid() << dendl; |
645 | ceph_abort_msg("agent queue not empty"); | |
7c673cae FG |
646 | } |
647 | ||
648 | agent_stop_flag = true; | |
9f95a23c | 649 | agent_cond.notify_all(); |
7c673cae FG |
650 | } |
651 | agent_thread.join(); | |
652 | } | |
653 | ||
654 | // ------------------------------------- | |
655 | ||
656 | void OSDService::promote_throttle_recalibrate() | |
657 | { | |
658 | utime_t now = ceph_clock_now(); | |
659 | double dur = now - last_recalibrate; | |
660 | last_recalibrate = now; | |
661 | unsigned prob = promote_probability_millis; | |
662 | ||
663 | uint64_t target_obj_sec = cct->_conf->osd_tier_promote_max_objects_sec; | |
664 | uint64_t target_bytes_sec = cct->_conf->osd_tier_promote_max_bytes_sec; | |
665 | ||
666 | unsigned min_prob = 1; | |
667 | ||
668 | uint64_t attempts, obj, bytes; | |
669 | promote_counter.sample_and_attenuate(&attempts, &obj, &bytes); | |
670 | dout(10) << __func__ << " " << attempts << " attempts, promoted " | |
1adf2230 | 671 | << obj << " objects and " << byte_u_t(bytes) << "; target " |
7c673cae | 672 | << target_obj_sec << " obj/sec or " |
1adf2230 | 673 | << byte_u_t(target_bytes_sec) << "/sec" |
7c673cae FG |
674 | << dendl; |
675 | ||
676 | // calculate what the probability *should* be, given the targets | |
677 | unsigned new_prob; | |
678 | if (attempts && dur > 0) { | |
679 | uint64_t avg_size = 1; | |
680 | if (obj) | |
11fdf7f2 | 681 | avg_size = std::max<uint64_t>(bytes / obj, 1); |
7c673cae FG |
682 | unsigned po = (double)target_obj_sec * dur * 1000.0 / (double)attempts; |
683 | unsigned pb = (double)target_bytes_sec / (double)avg_size * dur * 1000.0 | |
684 | / (double)attempts; | |
685 | dout(20) << __func__ << " po " << po << " pb " << pb << " avg_size " | |
686 | << avg_size << dendl; | |
687 | if (target_obj_sec && target_bytes_sec) | |
11fdf7f2 | 688 | new_prob = std::min(po, pb); |
7c673cae FG |
689 | else if (target_obj_sec) |
690 | new_prob = po; | |
691 | else if (target_bytes_sec) | |
692 | new_prob = pb; | |
693 | else | |
694 | new_prob = 1000; | |
695 | } else { | |
696 | new_prob = 1000; | |
697 | } | |
698 | dout(20) << __func__ << " new_prob " << new_prob << dendl; | |
699 | ||
700 | // correct for persistent skew between target rate and actual rate, adjust | |
701 | double ratio = 1.0; | |
702 | unsigned actual = 0; | |
703 | if (attempts && obj) { | |
704 | actual = obj * 1000 / attempts; | |
705 | ratio = (double)actual / (double)prob; | |
706 | new_prob = (double)new_prob / ratio; | |
707 | } | |
11fdf7f2 TL |
708 | new_prob = std::max(new_prob, min_prob); |
709 | new_prob = std::min(new_prob, 1000u); | |
7c673cae FG |
710 | |
711 | // adjust | |
712 | prob = (prob + new_prob) / 2; | |
11fdf7f2 TL |
713 | prob = std::max(prob, min_prob); |
714 | prob = std::min(prob, 1000u); | |
7c673cae FG |
715 | dout(10) << __func__ << " actual " << actual |
716 | << ", actual/prob ratio " << ratio | |
717 | << ", adjusted new_prob " << new_prob | |
718 | << ", prob " << promote_probability_millis << " -> " << prob | |
719 | << dendl; | |
720 | promote_probability_millis = prob; | |
721 | ||
722 | // set hard limits for this interval to mitigate stampedes | |
91327a77 AA |
723 | promote_max_objects = target_obj_sec * osd->OSD_TICK_INTERVAL * 2; |
724 | promote_max_bytes = target_bytes_sec * osd->OSD_TICK_INTERVAL * 2; | |
7c673cae FG |
725 | } |
726 | ||
727 | // ------------------------------------- | |
728 | ||
729 | float OSDService::get_failsafe_full_ratio() | |
730 | { | |
731 | float full_ratio = cct->_conf->osd_failsafe_full_ratio; | |
732 | if (full_ratio > 1.0) full_ratio /= 100.0; | |
733 | return full_ratio; | |
734 | } | |
735 | ||
11fdf7f2 | 736 | OSDService::s_names OSDService::recalc_full_state(float ratio, float pratio, string &inject) |
7c673cae | 737 | { |
7c673cae FG |
738 | // The OSDMap ratios take precendence. So if the failsafe is .95 and |
739 | // the admin sets the cluster full to .96, the failsafe moves up to .96 | |
740 | // too. (Not that having failsafe == full is ideal, but it's better than | |
741 | // dropping writes before the clusters appears full.) | |
742 | OSDMapRef osdmap = get_osdmap(); | |
743 | if (!osdmap || osdmap->get_epoch() == 0) { | |
11fdf7f2 | 744 | return NONE; |
7c673cae FG |
745 | } |
746 | float nearfull_ratio = osdmap->get_nearfull_ratio(); | |
747 | float backfillfull_ratio = std::max(osdmap->get_backfillfull_ratio(), nearfull_ratio); | |
748 | float full_ratio = std::max(osdmap->get_full_ratio(), backfillfull_ratio); | |
749 | float failsafe_ratio = std::max(get_failsafe_full_ratio(), full_ratio); | |
750 | ||
9f95a23c | 751 | if (osdmap->require_osd_release < ceph_release_t::luminous) { |
7c673cae FG |
752 | // use the failsafe for nearfull and full; the mon isn't using the |
753 | // flags anyway because we're mid-upgrade. | |
754 | full_ratio = failsafe_ratio; | |
755 | backfillfull_ratio = failsafe_ratio; | |
756 | nearfull_ratio = failsafe_ratio; | |
757 | } else if (full_ratio <= 0 || | |
758 | backfillfull_ratio <= 0 || | |
759 | nearfull_ratio <= 0) { | |
760 | derr << __func__ << " full_ratio, backfillfull_ratio or nearfull_ratio is <= 0" << dendl; | |
761 | // use failsafe flag. ick. the monitor did something wrong or the user | |
762 | // did something stupid. | |
763 | full_ratio = failsafe_ratio; | |
764 | backfillfull_ratio = failsafe_ratio; | |
765 | nearfull_ratio = failsafe_ratio; | |
766 | } | |
767 | ||
7c673cae | 768 | if (injectfull_state > NONE && injectfull) { |
7c673cae | 769 | inject = "(Injected)"; |
11fdf7f2 TL |
770 | return injectfull_state; |
771 | } else if (pratio > failsafe_ratio) { | |
772 | return FAILSAFE; | |
7c673cae | 773 | } else if (ratio > full_ratio) { |
11fdf7f2 | 774 | return FULL; |
7c673cae | 775 | } else if (ratio > backfillfull_ratio) { |
11fdf7f2 | 776 | return BACKFILLFULL; |
92f5a8d4 | 777 | } else if (pratio > nearfull_ratio) { |
11fdf7f2 | 778 | return NEARFULL; |
7c673cae | 779 | } |
11fdf7f2 TL |
780 | return NONE; |
781 | } | |
782 | ||
783 | void OSDService::check_full_status(float ratio, float pratio) | |
784 | { | |
785 | std::lock_guard l(full_status_lock); | |
786 | ||
787 | cur_ratio = ratio; | |
788 | physical_ratio = pratio; | |
789 | ||
790 | string inject; | |
791 | s_names new_state; | |
792 | new_state = recalc_full_state(ratio, pratio, inject); | |
793 | ||
7c673cae | 794 | dout(20) << __func__ << " cur ratio " << ratio |
11fdf7f2 | 795 | << ", physical ratio " << pratio |
7c673cae FG |
796 | << ", new state " << get_full_state_name(new_state) |
797 | << " " << inject | |
798 | << dendl; | |
799 | ||
800 | // warn | |
801 | if (cur_state != new_state) { | |
802 | dout(10) << __func__ << " " << get_full_state_name(cur_state) | |
803 | << " -> " << get_full_state_name(new_state) << dendl; | |
804 | if (new_state == FAILSAFE) { | |
c07f9fc5 | 805 | clog->error() << "full status failsafe engaged, dropping updates, now " |
7c673cae FG |
806 | << (int)roundf(ratio * 100) << "% full"; |
807 | } else if (cur_state == FAILSAFE) { | |
c07f9fc5 FG |
808 | clog->error() << "full status failsafe disengaged, no longer dropping " |
809 | << "updates, now " << (int)roundf(ratio * 100) << "% full"; | |
7c673cae FG |
810 | } |
811 | cur_state = new_state; | |
812 | } | |
813 | } | |
814 | ||
815 | bool OSDService::need_fullness_update() | |
816 | { | |
817 | OSDMapRef osdmap = get_osdmap(); | |
818 | s_names cur = NONE; | |
819 | if (osdmap->exists(whoami)) { | |
820 | if (osdmap->get_state(whoami) & CEPH_OSD_FULL) { | |
821 | cur = FULL; | |
822 | } else if (osdmap->get_state(whoami) & CEPH_OSD_BACKFILLFULL) { | |
823 | cur = BACKFILLFULL; | |
824 | } else if (osdmap->get_state(whoami) & CEPH_OSD_NEARFULL) { | |
825 | cur = NEARFULL; | |
826 | } | |
827 | } | |
828 | s_names want = NONE; | |
829 | if (is_full()) | |
830 | want = FULL; | |
831 | else if (is_backfillfull()) | |
832 | want = BACKFILLFULL; | |
833 | else if (is_nearfull()) | |
834 | want = NEARFULL; | |
835 | return want != cur; | |
836 | } | |
837 | ||
11fdf7f2 | 838 | bool OSDService::_check_inject_full(DoutPrefixProvider *dpp, s_names type) const |
7c673cae | 839 | { |
7c673cae FG |
840 | if (injectfull && injectfull_state >= type) { |
841 | // injectfull is either a count of the number of times to return failsafe full | |
842 | // or if -1 then always return full | |
843 | if (injectfull > 0) | |
844 | --injectfull; | |
11fdf7f2 TL |
845 | ldpp_dout(dpp, 10) << __func__ << " Injected " << get_full_state_name(type) << " OSD (" |
846 | << (injectfull < 0 ? "set" : std::to_string(injectfull)) << ")" | |
847 | << dendl; | |
7c673cae FG |
848 | return true; |
849 | } | |
11fdf7f2 TL |
850 | return false; |
851 | } | |
852 | ||
853 | bool OSDService::_check_full(DoutPrefixProvider *dpp, s_names type) const | |
854 | { | |
855 | std::lock_guard l(full_status_lock); | |
856 | ||
857 | if (_check_inject_full(dpp, type)) | |
858 | return true; | |
859 | ||
860 | if (cur_state >= type) | |
861 | ldpp_dout(dpp, 10) << __func__ << " current usage is " << cur_ratio | |
862 | << " physical " << physical_ratio << dendl; | |
7c673cae | 863 | |
7c673cae FG |
864 | return cur_state >= type; |
865 | } | |
866 | ||
11fdf7f2 TL |
867 | bool OSDService::_tentative_full(DoutPrefixProvider *dpp, s_names type, uint64_t adjust_used, osd_stat_t adjusted_stat) |
868 | { | |
869 | ldpp_dout(dpp, 20) << __func__ << " type " << get_full_state_name(type) << " adjust_used " << (adjust_used >> 10) << "KiB" << dendl; | |
870 | { | |
871 | std::lock_guard l(full_status_lock); | |
872 | if (_check_inject_full(dpp, type)) { | |
873 | return true; | |
874 | } | |
875 | } | |
876 | ||
877 | float pratio; | |
878 | float ratio = compute_adjusted_ratio(adjusted_stat, &pratio, adjust_used); | |
879 | ||
880 | string notused; | |
881 | s_names tentative_state = recalc_full_state(ratio, pratio, notused); | |
882 | ||
883 | if (tentative_state >= type) | |
884 | ldpp_dout(dpp, 10) << __func__ << " tentative usage is " << ratio << dendl; | |
885 | ||
886 | return tentative_state >= type; | |
887 | } | |
888 | ||
889 | bool OSDService::check_failsafe_full(DoutPrefixProvider *dpp) const | |
890 | { | |
891 | return _check_full(dpp, FAILSAFE); | |
892 | } | |
893 | ||
894 | bool OSDService::check_full(DoutPrefixProvider *dpp) const | |
7c673cae | 895 | { |
11fdf7f2 | 896 | return _check_full(dpp, FULL); |
7c673cae FG |
897 | } |
898 | ||
11fdf7f2 | 899 | bool OSDService::tentative_backfill_full(DoutPrefixProvider *dpp, uint64_t adjust_used, osd_stat_t stats) |
7c673cae | 900 | { |
11fdf7f2 | 901 | return _tentative_full(dpp, BACKFILLFULL, adjust_used, stats); |
7c673cae FG |
902 | } |
903 | ||
11fdf7f2 | 904 | bool OSDService::check_backfill_full(DoutPrefixProvider *dpp) const |
7c673cae | 905 | { |
11fdf7f2 | 906 | return _check_full(dpp, BACKFILLFULL); |
7c673cae FG |
907 | } |
908 | ||
11fdf7f2 | 909 | bool OSDService::check_nearfull(DoutPrefixProvider *dpp) const |
7c673cae | 910 | { |
11fdf7f2 | 911 | return _check_full(dpp, NEARFULL); |
7c673cae FG |
912 | } |
913 | ||
914 | bool OSDService::is_failsafe_full() const | |
915 | { | |
11fdf7f2 | 916 | std::lock_guard l(full_status_lock); |
7c673cae FG |
917 | return cur_state == FAILSAFE; |
918 | } | |
919 | ||
920 | bool OSDService::is_full() const | |
921 | { | |
11fdf7f2 | 922 | std::lock_guard l(full_status_lock); |
7c673cae FG |
923 | return cur_state >= FULL; |
924 | } | |
925 | ||
926 | bool OSDService::is_backfillfull() const | |
927 | { | |
11fdf7f2 | 928 | std::lock_guard l(full_status_lock); |
7c673cae FG |
929 | return cur_state >= BACKFILLFULL; |
930 | } | |
931 | ||
932 | bool OSDService::is_nearfull() const | |
933 | { | |
11fdf7f2 | 934 | std::lock_guard l(full_status_lock); |
7c673cae FG |
935 | return cur_state >= NEARFULL; |
936 | } | |
937 | ||
938 | void OSDService::set_injectfull(s_names type, int64_t count) | |
939 | { | |
11fdf7f2 | 940 | std::lock_guard l(full_status_lock); |
7c673cae FG |
941 | injectfull_state = type; |
942 | injectfull = count; | |
943 | } | |
944 | ||
11fdf7f2 TL |
945 | void OSDService::set_statfs(const struct store_statfs_t &stbuf, |
946 | osd_alert_list_t& alerts) | |
7c673cae | 947 | { |
224ce89b | 948 | uint64_t bytes = stbuf.total; |
224ce89b | 949 | uint64_t avail = stbuf.available; |
11fdf7f2 TL |
950 | uint64_t used = stbuf.get_used_raw(); |
951 | ||
952 | // For testing fake statfs values so it doesn't matter if all | |
953 | // OSDs are using the same partition. | |
954 | if (cct->_conf->fake_statfs_for_testing) { | |
955 | uint64_t total_num_bytes = 0; | |
956 | vector<PGRef> pgs; | |
957 | osd->_get_pgs(&pgs); | |
958 | for (auto p : pgs) { | |
959 | total_num_bytes += p->get_stats_num_bytes(); | |
960 | } | |
961 | bytes = cct->_conf->fake_statfs_for_testing; | |
962 | if (total_num_bytes < bytes) | |
963 | avail = bytes - total_num_bytes; | |
964 | else | |
965 | avail = 0; | |
966 | dout(0) << __func__ << " fake total " << cct->_conf->fake_statfs_for_testing | |
967 | << " adjust available " << avail | |
968 | << dendl; | |
969 | used = bytes - avail; | |
970 | } | |
7c673cae | 971 | |
f67539c2 TL |
972 | logger->set(l_osd_stat_bytes, bytes); |
973 | logger->set(l_osd_stat_bytes_used, used); | |
974 | logger->set(l_osd_stat_bytes_avail, avail); | |
7c673cae | 975 | |
11fdf7f2 TL |
976 | std::lock_guard l(stat_lock); |
977 | osd_stat.statfs = stbuf; | |
978 | osd_stat.os_alerts.clear(); | |
979 | osd_stat.os_alerts[whoami].swap(alerts); | |
980 | if (cct->_conf->fake_statfs_for_testing) { | |
981 | osd_stat.statfs.total = bytes; | |
982 | osd_stat.statfs.available = avail; | |
983 | // For testing don't want used to go negative, so clear reserved | |
984 | osd_stat.statfs.internally_reserved = 0; | |
224ce89b WB |
985 | } |
986 | } | |
7c673cae | 987 | |
11fdf7f2 TL |
988 | osd_stat_t OSDService::set_osd_stat(vector<int>& hb_peers, |
989 | int num_pgs) | |
224ce89b | 990 | { |
eafe8130 TL |
991 | utime_t now = ceph_clock_now(); |
992 | auto stale_time = g_conf().get_val<int64_t>("osd_mon_heartbeat_stat_stale"); | |
11fdf7f2 TL |
993 | std::lock_guard l(stat_lock); |
994 | osd_stat.hb_peers.swap(hb_peers); | |
995 | osd->op_tracker.get_age_ms_histogram(&osd_stat.op_queue_age_hist); | |
996 | osd_stat.num_pgs = num_pgs; | |
eafe8130 TL |
997 | // Clean entries that aren't updated |
998 | // This is called often enough that we can just remove 1 at a time | |
999 | for (auto i: osd_stat.hb_pingtime) { | |
1000 | if (i.second.last_update == 0) | |
1001 | continue; | |
1002 | if (stale_time && now.sec() - i.second.last_update > stale_time) { | |
1003 | dout(20) << __func__ << " time out heartbeat for osd " << i.first | |
1004 | << " last_update " << i.second.last_update << dendl; | |
1005 | osd_stat.hb_pingtime.erase(i.first); | |
1006 | break; | |
1007 | } | |
1008 | } | |
11fdf7f2 TL |
1009 | return osd_stat; |
1010 | } | |
1011 | ||
1012 | void OSDService::inc_osd_stat_repaired() | |
1013 | { | |
1014 | std::lock_guard l(stat_lock); | |
1015 | osd_stat.num_shards_repaired++; | |
1016 | return; | |
1017 | } | |
1018 | ||
1019 | float OSDService::compute_adjusted_ratio(osd_stat_t new_stat, float *pratio, | |
1020 | uint64_t adjust_used) | |
1021 | { | |
1022 | *pratio = | |
b3b6e05e | 1023 | ((float)new_stat.statfs.get_used_raw()) / ((float)new_stat.statfs.total); |
11fdf7f2 TL |
1024 | |
1025 | if (adjust_used) { | |
1026 | dout(20) << __func__ << " Before kb_used() " << new_stat.statfs.kb_used() << dendl; | |
1027 | if (new_stat.statfs.available > adjust_used) | |
1028 | new_stat.statfs.available -= adjust_used; | |
1029 | else | |
1030 | new_stat.statfs.available = 0; | |
1031 | dout(20) << __func__ << " After kb_used() " << new_stat.statfs.kb_used() << dendl; | |
7c673cae FG |
1032 | } |
1033 | ||
11fdf7f2 TL |
1034 | // Check all pgs and adjust kb_used to include all pending backfill data |
1035 | int backfill_adjusted = 0; | |
1036 | vector<PGRef> pgs; | |
1037 | osd->_get_pgs(&pgs); | |
1038 | for (auto p : pgs) { | |
1039 | backfill_adjusted += p->pg_stat_adjust(&new_stat); | |
1040 | } | |
1041 | if (backfill_adjusted) { | |
1042 | dout(20) << __func__ << " backfill adjusted " << new_stat << dendl; | |
1043 | } | |
b3b6e05e | 1044 | return ((float)new_stat.statfs.get_used_raw()) / ((float)new_stat.statfs.total); |
7c673cae FG |
1045 | } |
1046 | ||
7c673cae FG |
1047 | void OSDService::send_message_osd_cluster(int peer, Message *m, epoch_t from_epoch) |
1048 | { | |
1049 | OSDMapRef next_map = get_nextmap_reserved(); | |
1050 | // service map is always newer/newest | |
11fdf7f2 | 1051 | ceph_assert(from_epoch <= next_map->get_epoch()); |
7c673cae FG |
1052 | |
1053 | if (next_map->is_down(peer) || | |
1054 | next_map->get_info(peer).up_from > from_epoch) { | |
1055 | m->put(); | |
1056 | release_map(next_map); | |
1057 | return; | |
1058 | } | |
9f95a23c TL |
1059 | ConnectionRef peer_con; |
1060 | if (peer == whoami) { | |
1061 | peer_con = osd->cluster_messenger->get_loopback_connection(); | |
1062 | } else { | |
1063 | peer_con = osd->cluster_messenger->connect_to_osd( | |
1064 | next_map->get_cluster_addrs(peer), false, true); | |
1065 | } | |
1066 | maybe_share_map(peer_con.get(), next_map); | |
7c673cae FG |
1067 | peer_con->send_message(m); |
1068 | release_map(next_map); | |
1069 | } | |
1070 | ||
9f95a23c TL |
1071 | void OSDService::send_message_osd_cluster(std::vector<std::pair<int, Message*>>& messages, epoch_t from_epoch) |
1072 | { | |
1073 | OSDMapRef next_map = get_nextmap_reserved(); | |
1074 | // service map is always newer/newest | |
1075 | ceph_assert(from_epoch <= next_map->get_epoch()); | |
1076 | ||
1077 | for (auto& iter : messages) { | |
1078 | if (next_map->is_down(iter.first) || | |
1079 | next_map->get_info(iter.first).up_from > from_epoch) { | |
1080 | iter.second->put(); | |
1081 | continue; | |
1082 | } | |
1083 | ConnectionRef peer_con; | |
1084 | if (iter.first == whoami) { | |
1085 | peer_con = osd->cluster_messenger->get_loopback_connection(); | |
1086 | } else { | |
1087 | peer_con = osd->cluster_messenger->connect_to_osd( | |
1088 | next_map->get_cluster_addrs(iter.first), false, true); | |
1089 | } | |
1090 | maybe_share_map(peer_con.get(), next_map); | |
1091 | peer_con->send_message(iter.second); | |
1092 | } | |
1093 | release_map(next_map); | |
1094 | } | |
7c673cae FG |
1095 | ConnectionRef OSDService::get_con_osd_cluster(int peer, epoch_t from_epoch) |
1096 | { | |
1097 | OSDMapRef next_map = get_nextmap_reserved(); | |
1098 | // service map is always newer/newest | |
11fdf7f2 | 1099 | ceph_assert(from_epoch <= next_map->get_epoch()); |
7c673cae FG |
1100 | |
1101 | if (next_map->is_down(peer) || | |
1102 | next_map->get_info(peer).up_from > from_epoch) { | |
1103 | release_map(next_map); | |
1104 | return NULL; | |
1105 | } | |
9f95a23c TL |
1106 | ConnectionRef con; |
1107 | if (peer == whoami) { | |
1108 | con = osd->cluster_messenger->get_loopback_connection(); | |
1109 | } else { | |
1110 | con = osd->cluster_messenger->connect_to_osd( | |
1111 | next_map->get_cluster_addrs(peer), false, true); | |
1112 | } | |
7c673cae FG |
1113 | release_map(next_map); |
1114 | return con; | |
1115 | } | |
1116 | ||
1117 | pair<ConnectionRef,ConnectionRef> OSDService::get_con_osd_hb(int peer, epoch_t from_epoch) | |
1118 | { | |
1119 | OSDMapRef next_map = get_nextmap_reserved(); | |
1120 | // service map is always newer/newest | |
11fdf7f2 | 1121 | ceph_assert(from_epoch <= next_map->get_epoch()); |
7c673cae FG |
1122 | |
1123 | pair<ConnectionRef,ConnectionRef> ret; | |
1124 | if (next_map->is_down(peer) || | |
1125 | next_map->get_info(peer).up_from > from_epoch) { | |
1126 | release_map(next_map); | |
1127 | return ret; | |
1128 | } | |
11fdf7f2 TL |
1129 | ret.first = osd->hb_back_client_messenger->connect_to_osd( |
1130 | next_map->get_hb_back_addrs(peer)); | |
1131 | ret.second = osd->hb_front_client_messenger->connect_to_osd( | |
1132 | next_map->get_hb_front_addrs(peer)); | |
7c673cae FG |
1133 | release_map(next_map); |
1134 | return ret; | |
1135 | } | |
1136 | ||
11fdf7f2 TL |
1137 | entity_name_t OSDService::get_cluster_msgr_name() const |
1138 | { | |
1139 | return cluster_messenger->get_myname(); | |
1140 | } | |
7c673cae | 1141 | |
94b18763 FG |
1142 | void OSDService::queue_want_pg_temp(pg_t pgid, |
1143 | const vector<int>& want, | |
1144 | bool forced) | |
7c673cae | 1145 | { |
11fdf7f2 | 1146 | std::lock_guard l(pg_temp_lock); |
94b18763 | 1147 | auto p = pg_temp_pending.find(pgid); |
7c673cae | 1148 | if (p == pg_temp_pending.end() || |
94b18763 FG |
1149 | p->second.acting != want || |
1150 | forced) { | |
11fdf7f2 | 1151 | pg_temp_wanted[pgid] = {want, forced}; |
7c673cae FG |
1152 | } |
1153 | } | |
1154 | ||
1155 | void OSDService::remove_want_pg_temp(pg_t pgid) | |
1156 | { | |
11fdf7f2 | 1157 | std::lock_guard l(pg_temp_lock); |
7c673cae FG |
1158 | pg_temp_wanted.erase(pgid); |
1159 | pg_temp_pending.erase(pgid); | |
1160 | } | |
1161 | ||
1162 | void OSDService::_sent_pg_temp() | |
1163 | { | |
11fdf7f2 TL |
1164 | #ifdef HAVE_STDLIB_MAP_SPLICING |
1165 | pg_temp_pending.merge(pg_temp_wanted); | |
1166 | #else | |
94b18763 FG |
1167 | pg_temp_pending.insert(make_move_iterator(begin(pg_temp_wanted)), |
1168 | make_move_iterator(end(pg_temp_wanted))); | |
11fdf7f2 | 1169 | #endif |
7c673cae FG |
1170 | pg_temp_wanted.clear(); |
1171 | } | |
1172 | ||
1173 | void OSDService::requeue_pg_temp() | |
1174 | { | |
11fdf7f2 | 1175 | std::lock_guard l(pg_temp_lock); |
7c673cae FG |
1176 | // wanted overrides pending. note that remove_want_pg_temp |
1177 | // clears the item out of both. | |
1178 | unsigned old_wanted = pg_temp_wanted.size(); | |
1179 | unsigned old_pending = pg_temp_pending.size(); | |
1180 | _sent_pg_temp(); | |
1181 | pg_temp_wanted.swap(pg_temp_pending); | |
1182 | dout(10) << __func__ << " " << old_wanted << " + " << old_pending << " -> " | |
1183 | << pg_temp_wanted.size() << dendl; | |
1184 | } | |
1185 | ||
94b18763 FG |
1186 | std::ostream& operator<<(std::ostream& out, |
1187 | const OSDService::pg_temp_t& pg_temp) | |
1188 | { | |
1189 | out << pg_temp.acting; | |
1190 | if (pg_temp.forced) { | |
1191 | out << " (forced)"; | |
1192 | } | |
1193 | return out; | |
1194 | } | |
1195 | ||
7c673cae FG |
1196 | void OSDService::send_pg_temp() |
1197 | { | |
11fdf7f2 | 1198 | std::lock_guard l(pg_temp_lock); |
7c673cae FG |
1199 | if (pg_temp_wanted.empty()) |
1200 | return; | |
1201 | dout(10) << "send_pg_temp " << pg_temp_wanted << dendl; | |
94b18763 | 1202 | MOSDPGTemp *ms[2] = {nullptr, nullptr}; |
11fdf7f2 TL |
1203 | for (auto& [pgid, pg_temp] : pg_temp_wanted) { |
1204 | auto& m = ms[pg_temp.forced]; | |
94b18763 FG |
1205 | if (!m) { |
1206 | m = new MOSDPGTemp(osdmap->get_epoch()); | |
11fdf7f2 | 1207 | m->forced = pg_temp.forced; |
94b18763 | 1208 | } |
11fdf7f2 | 1209 | m->pg_temp.emplace(pgid, pg_temp.acting); |
94b18763 FG |
1210 | } |
1211 | for (auto m : ms) { | |
1212 | if (m) { | |
1213 | monc->send_mon_message(m); | |
1214 | } | |
1215 | } | |
7c673cae FG |
1216 | _sent_pg_temp(); |
1217 | } | |
1218 | ||
1219 | void OSDService::send_pg_created(pg_t pgid) | |
1220 | { | |
11fdf7f2 | 1221 | std::lock_guard l(pg_created_lock); |
7c673cae | 1222 | dout(20) << __func__ << dendl; |
11fdf7f2 | 1223 | auto o = get_osdmap(); |
9f95a23c | 1224 | if (o->require_osd_release >= ceph_release_t::luminous) { |
11fdf7f2 | 1225 | pg_created.insert(pgid); |
c07f9fc5 FG |
1226 | monc->send_mon_message(new MOSDPGCreated(pgid)); |
1227 | } | |
7c673cae FG |
1228 | } |
1229 | ||
11fdf7f2 TL |
1230 | void OSDService::send_pg_created() |
1231 | { | |
1232 | std::lock_guard l(pg_created_lock); | |
1233 | dout(20) << __func__ << dendl; | |
1234 | auto o = get_osdmap(); | |
9f95a23c | 1235 | if (o->require_osd_release >= ceph_release_t::luminous) { |
11fdf7f2 TL |
1236 | for (auto pgid : pg_created) { |
1237 | monc->send_mon_message(new MOSDPGCreated(pgid)); | |
1238 | } | |
1239 | } | |
1240 | } | |
1241 | ||
1242 | void OSDService::prune_pg_created() | |
1243 | { | |
1244 | std::lock_guard l(pg_created_lock); | |
1245 | dout(20) << __func__ << dendl; | |
1246 | auto o = get_osdmap(); | |
1247 | auto i = pg_created.begin(); | |
1248 | while (i != pg_created.end()) { | |
1249 | auto p = o->get_pg_pool(i->pool()); | |
1250 | if (!p || !p->has_flag(pg_pool_t::FLAG_CREATING)) { | |
1251 | dout(20) << __func__ << " pruning " << *i << dendl; | |
1252 | i = pg_created.erase(i); | |
1253 | } else { | |
1254 | dout(20) << __func__ << " keeping " << *i << dendl; | |
1255 | ++i; | |
1256 | } | |
1257 | } | |
1258 | } | |
1259 | ||
1260 | ||
7c673cae FG |
1261 | // -------------------------------------- |
1262 | // dispatch | |
1263 | ||
7c673cae FG |
1264 | void OSDService::retrieve_epochs(epoch_t *_boot_epoch, epoch_t *_up_epoch, |
1265 | epoch_t *_bind_epoch) const | |
1266 | { | |
11fdf7f2 | 1267 | std::lock_guard l(epoch_lock); |
7c673cae FG |
1268 | if (_boot_epoch) |
1269 | *_boot_epoch = boot_epoch; | |
1270 | if (_up_epoch) | |
1271 | *_up_epoch = up_epoch; | |
1272 | if (_bind_epoch) | |
1273 | *_bind_epoch = bind_epoch; | |
1274 | } | |
1275 | ||
1276 | void OSDService::set_epochs(const epoch_t *_boot_epoch, const epoch_t *_up_epoch, | |
1277 | const epoch_t *_bind_epoch) | |
1278 | { | |
11fdf7f2 | 1279 | std::lock_guard l(epoch_lock); |
7c673cae | 1280 | if (_boot_epoch) { |
11fdf7f2 | 1281 | ceph_assert(*_boot_epoch == 0 || *_boot_epoch >= boot_epoch); |
7c673cae FG |
1282 | boot_epoch = *_boot_epoch; |
1283 | } | |
1284 | if (_up_epoch) { | |
11fdf7f2 | 1285 | ceph_assert(*_up_epoch == 0 || *_up_epoch >= up_epoch); |
7c673cae FG |
1286 | up_epoch = *_up_epoch; |
1287 | } | |
1288 | if (_bind_epoch) { | |
11fdf7f2 | 1289 | ceph_assert(*_bind_epoch == 0 || *_bind_epoch >= bind_epoch); |
7c673cae FG |
1290 | bind_epoch = *_bind_epoch; |
1291 | } | |
1292 | } | |
1293 | ||
1294 | bool OSDService::prepare_to_stop() | |
1295 | { | |
9f95a23c | 1296 | std::unique_lock l(is_stopping_lock); |
7c673cae FG |
1297 | if (get_state() != NOT_STOPPING) |
1298 | return false; | |
1299 | ||
1300 | OSDMapRef osdmap = get_osdmap(); | |
1301 | if (osdmap && osdmap->is_up(whoami)) { | |
1d09f67e | 1302 | dout(0) << __func__ << " telling mon we are shutting down and dead " << dendl; |
7c673cae | 1303 | set_state(PREPARING_TO_STOP); |
11fdf7f2 TL |
1304 | monc->send_mon_message( |
1305 | new MOSDMarkMeDown( | |
1306 | monc->get_fsid(), | |
1307 | whoami, | |
1308 | osdmap->get_addrs(whoami), | |
1309 | osdmap->get_epoch(), | |
1d09f67e TL |
1310 | true, // request ack |
1311 | true // mark as down and dead | |
11fdf7f2 | 1312 | )); |
9f95a23c TL |
1313 | const auto timeout = ceph::make_timespan(cct->_conf->osd_mon_shutdown_timeout); |
1314 | is_stopping_cond.wait_for(l, timeout, | |
1315 | [this] { return get_state() == STOPPING; }); | |
7c673cae | 1316 | } |
1d09f67e | 1317 | |
7c673cae FG |
1318 | dout(0) << __func__ << " starting shutdown" << dendl; |
1319 | set_state(STOPPING); | |
1320 | return true; | |
1321 | } | |
1322 | ||
1323 | void OSDService::got_stop_ack() | |
1324 | { | |
9f95a23c | 1325 | std::scoped_lock l(is_stopping_lock); |
7c673cae FG |
1326 | if (get_state() == PREPARING_TO_STOP) { |
1327 | dout(0) << __func__ << " starting shutdown" << dendl; | |
1328 | set_state(STOPPING); | |
9f95a23c | 1329 | is_stopping_cond.notify_all(); |
7c673cae FG |
1330 | } else { |
1331 | dout(10) << __func__ << " ignoring msg" << dendl; | |
1332 | } | |
1333 | } | |
1334 | ||
1335 | MOSDMap *OSDService::build_incremental_map_msg(epoch_t since, epoch_t to, | |
1336 | OSDSuperblock& sblock) | |
1337 | { | |
28e407b8 AA |
1338 | MOSDMap *m = new MOSDMap(monc->get_fsid(), |
1339 | osdmap->get_encoding_features()); | |
7c673cae FG |
1340 | m->oldest_map = max_oldest_map; |
1341 | m->newest_map = sblock.newest_map; | |
1342 | ||
11fdf7f2 TL |
1343 | int max = cct->_conf->osd_map_message_max; |
1344 | ssize_t max_bytes = cct->_conf->osd_map_message_max_bytes; | |
1345 | ||
1346 | if (since < m->oldest_map) { | |
1347 | // we don't have the next map the target wants, so start with a | |
1348 | // full map. | |
1349 | bufferlist bl; | |
1350 | dout(10) << __func__ << " oldest map " << max_oldest_map << " > since " | |
1351 | << since << ", starting with full map" << dendl; | |
1352 | since = m->oldest_map; | |
1353 | if (!get_map_bl(since, bl)) { | |
1354 | derr << __func__ << " missing full map " << since << dendl; | |
1355 | goto panic; | |
1356 | } | |
1357 | max--; | |
1358 | max_bytes -= bl.length(); | |
f67539c2 | 1359 | m->maps[since] = std::move(bl); |
11fdf7f2 TL |
1360 | } |
1361 | for (epoch_t e = since + 1; e <= to; ++e) { | |
7c673cae | 1362 | bufferlist bl; |
11fdf7f2 | 1363 | if (get_inc_map_bl(e, bl)) { |
f67539c2 | 1364 | m->incremental_maps[e] = std::move(bl); |
11fdf7f2 | 1365 | } else { |
e306af50 | 1366 | dout(10) << __func__ << " missing incremental map " << e << dendl; |
11fdf7f2 TL |
1367 | if (!get_map_bl(e, bl)) { |
1368 | derr << __func__ << " also missing full map " << e << dendl; | |
1369 | goto panic; | |
1370 | } | |
f67539c2 | 1371 | m->maps[e] = std::move(bl); |
11fdf7f2 TL |
1372 | } |
1373 | max--; | |
1374 | max_bytes -= bl.length(); | |
1375 | if (max <= 0 || max_bytes <= 0) { | |
7c673cae | 1376 | break; |
11fdf7f2 TL |
1377 | } |
1378 | } | |
1379 | return m; | |
1380 | ||
1381 | panic: | |
1382 | if (!m->maps.empty() || | |
1383 | !m->incremental_maps.empty()) { | |
1384 | // send what we have so far | |
1385 | return m; | |
1386 | } | |
1387 | // send something | |
1388 | bufferlist bl; | |
1389 | if (get_inc_map_bl(m->newest_map, bl)) { | |
f67539c2 | 1390 | m->incremental_maps[m->newest_map] = std::move(bl); |
11fdf7f2 TL |
1391 | } else { |
1392 | derr << __func__ << " unable to load latest map " << m->newest_map << dendl; | |
1393 | if (!get_map_bl(m->newest_map, bl)) { | |
1394 | derr << __func__ << " unable to load latest full map " << m->newest_map | |
7c673cae | 1395 | << dendl; |
11fdf7f2 | 1396 | ceph_abort(); |
7c673cae | 1397 | } |
f67539c2 | 1398 | m->maps[m->newest_map] = std::move(bl); |
7c673cae FG |
1399 | } |
1400 | return m; | |
1401 | } | |
1402 | ||
1403 | void OSDService::send_map(MOSDMap *m, Connection *con) | |
1404 | { | |
1405 | con->send_message(m); | |
1406 | } | |
1407 | ||
1408 | void OSDService::send_incremental_map(epoch_t since, Connection *con, | |
9f95a23c | 1409 | const OSDMapRef& osdmap) |
7c673cae FG |
1410 | { |
1411 | epoch_t to = osdmap->get_epoch(); | |
1412 | dout(10) << "send_incremental_map " << since << " -> " << to | |
1413 | << " to " << con << " " << con->get_peer_addr() << dendl; | |
1414 | ||
1415 | MOSDMap *m = NULL; | |
1416 | while (!m) { | |
1417 | OSDSuperblock sblock(get_superblock()); | |
1418 | if (since < sblock.oldest_map) { | |
1419 | // just send latest full map | |
28e407b8 AA |
1420 | MOSDMap *m = new MOSDMap(monc->get_fsid(), |
1421 | osdmap->get_encoding_features()); | |
7c673cae FG |
1422 | m->oldest_map = max_oldest_map; |
1423 | m->newest_map = sblock.newest_map; | |
1424 | get_map_bl(to, m->maps[to]); | |
1425 | send_map(m, con); | |
1426 | return; | |
1427 | } | |
1428 | ||
1429 | if (to > since && (int64_t)(to - since) > cct->_conf->osd_map_share_max_epochs) { | |
1430 | dout(10) << " " << (to - since) << " > max " << cct->_conf->osd_map_share_max_epochs | |
1431 | << ", only sending most recent" << dendl; | |
1432 | since = to - cct->_conf->osd_map_share_max_epochs; | |
1433 | } | |
1434 | ||
7c673cae FG |
1435 | m = build_incremental_map_msg(since, to, sblock); |
1436 | } | |
1437 | send_map(m, con); | |
1438 | } | |
1439 | ||
1440 | bool OSDService::_get_map_bl(epoch_t e, bufferlist& bl) | |
1441 | { | |
1442 | bool found = map_bl_cache.lookup(e, &bl); | |
31f18b77 | 1443 | if (found) { |
f67539c2 | 1444 | logger->inc(l_osd_map_bl_cache_hit); |
7c673cae | 1445 | return true; |
31f18b77 | 1446 | } |
f67539c2 | 1447 | logger->inc(l_osd_map_bl_cache_miss); |
11fdf7f2 | 1448 | found = store->read(meta_ch, |
31f18b77 FG |
1449 | OSD::get_osdmap_pobject_name(e), 0, 0, bl, |
1450 | CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) >= 0; | |
1451 | if (found) { | |
7c673cae | 1452 | _add_map_bl(e, bl); |
31f18b77 | 1453 | } |
7c673cae FG |
1454 | return found; |
1455 | } | |
1456 | ||
1457 | bool OSDService::get_inc_map_bl(epoch_t e, bufferlist& bl) | |
1458 | { | |
11fdf7f2 | 1459 | std::lock_guard l(map_cache_lock); |
7c673cae | 1460 | bool found = map_bl_inc_cache.lookup(e, &bl); |
31f18b77 | 1461 | if (found) { |
f67539c2 | 1462 | logger->inc(l_osd_map_bl_cache_hit); |
7c673cae | 1463 | return true; |
31f18b77 | 1464 | } |
f67539c2 | 1465 | logger->inc(l_osd_map_bl_cache_miss); |
11fdf7f2 | 1466 | found = store->read(meta_ch, |
31f18b77 FG |
1467 | OSD::get_inc_osdmap_pobject_name(e), 0, 0, bl, |
1468 | CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) >= 0; | |
1469 | if (found) { | |
7c673cae | 1470 | _add_map_inc_bl(e, bl); |
31f18b77 | 1471 | } |
7c673cae FG |
1472 | return found; |
1473 | } | |
1474 | ||
1475 | void OSDService::_add_map_bl(epoch_t e, bufferlist& bl) | |
1476 | { | |
1477 | dout(10) << "add_map_bl " << e << " " << bl.length() << " bytes" << dendl; | |
31f18b77 FG |
1478 | // cache a contiguous buffer |
1479 | if (bl.get_num_buffers() > 1) { | |
1480 | bl.rebuild(); | |
1481 | } | |
1482 | bl.try_assign_to_mempool(mempool::mempool_osd_mapbl); | |
7c673cae FG |
1483 | map_bl_cache.add(e, bl); |
1484 | } | |
1485 | ||
1486 | void OSDService::_add_map_inc_bl(epoch_t e, bufferlist& bl) | |
1487 | { | |
1488 | dout(10) << "add_map_inc_bl " << e << " " << bl.length() << " bytes" << dendl; | |
31f18b77 FG |
1489 | // cache a contiguous buffer |
1490 | if (bl.get_num_buffers() > 1) { | |
1491 | bl.rebuild(); | |
1492 | } | |
1493 | bl.try_assign_to_mempool(mempool::mempool_osd_mapbl); | |
7c673cae FG |
1494 | map_bl_inc_cache.add(e, bl); |
1495 | } | |
1496 | ||
7c673cae FG |
1497 | OSDMapRef OSDService::_add_map(OSDMap *o) |
1498 | { | |
1499 | epoch_t e = o->get_epoch(); | |
1500 | ||
1501 | if (cct->_conf->osd_map_dedup) { | |
1502 | // Dedup against an existing map at a nearby epoch | |
1503 | OSDMapRef for_dedup = map_cache.lower_bound(e); | |
1504 | if (for_dedup) { | |
1505 | OSDMap::dedup(for_dedup.get(), o); | |
1506 | } | |
1507 | } | |
1508 | bool existed; | |
1509 | OSDMapRef l = map_cache.add(e, o, &existed); | |
1510 | if (existed) { | |
1511 | delete o; | |
1512 | } | |
1513 | return l; | |
1514 | } | |
1515 | ||
1516 | OSDMapRef OSDService::try_get_map(epoch_t epoch) | |
1517 | { | |
11fdf7f2 | 1518 | std::lock_guard l(map_cache_lock); |
7c673cae FG |
1519 | OSDMapRef retval = map_cache.lookup(epoch); |
1520 | if (retval) { | |
1521 | dout(30) << "get_map " << epoch << " -cached" << dendl; | |
f67539c2 | 1522 | logger->inc(l_osd_map_cache_hit); |
7c673cae FG |
1523 | return retval; |
1524 | } | |
f67539c2 | 1525 | { |
7c673cae FG |
1526 | logger->inc(l_osd_map_cache_miss); |
1527 | epoch_t lb = map_cache.cached_key_lower_bound(); | |
1528 | if (epoch < lb) { | |
1529 | dout(30) << "get_map " << epoch << " - miss, below lower bound" << dendl; | |
1530 | logger->inc(l_osd_map_cache_miss_low); | |
1531 | logger->inc(l_osd_map_cache_miss_low_avg, lb - epoch); | |
1532 | } | |
1533 | } | |
1534 | ||
1535 | OSDMap *map = new OSDMap; | |
1536 | if (epoch > 0) { | |
1537 | dout(20) << "get_map " << epoch << " - loading and decoding " << map << dendl; | |
1538 | bufferlist bl; | |
1539 | if (!_get_map_bl(epoch, bl) || bl.length() == 0) { | |
1540 | derr << "failed to load OSD map for epoch " << epoch << ", got " << bl.length() << " bytes" << dendl; | |
1541 | delete map; | |
1542 | return OSDMapRef(); | |
1543 | } | |
1544 | map->decode(bl); | |
1545 | } else { | |
1546 | dout(20) << "get_map " << epoch << " - return initial " << map << dendl; | |
1547 | } | |
1548 | return _add_map(map); | |
1549 | } | |
1550 | ||
1551 | // ops | |
1552 | ||
1553 | ||
1554 | void OSDService::reply_op_error(OpRequestRef op, int err) | |
1555 | { | |
9f95a23c | 1556 | reply_op_error(op, err, eversion_t(), 0, {}); |
7c673cae FG |
1557 | } |
1558 | ||
1559 | void OSDService::reply_op_error(OpRequestRef op, int err, eversion_t v, | |
9f95a23c TL |
1560 | version_t uv, |
1561 | vector<pg_log_op_return_item_t> op_returns) | |
7c673cae | 1562 | { |
9f95a23c | 1563 | auto m = op->get_req<MOSDOp>(); |
11fdf7f2 | 1564 | ceph_assert(m->get_type() == CEPH_MSG_OSD_OP); |
7c673cae FG |
1565 | int flags; |
1566 | flags = m->get_flags() & (CEPH_OSD_FLAG_ACK|CEPH_OSD_FLAG_ONDISK); | |
1567 | ||
9f95a23c TL |
1568 | MOSDOpReply *reply = new MOSDOpReply(m, err, osdmap->get_epoch(), flags, |
1569 | !m->has_flag(CEPH_OSD_FLAG_RETURNVEC)); | |
7c673cae | 1570 | reply->set_reply_versions(v, uv); |
9f95a23c | 1571 | reply->set_op_returns(op_returns); |
7c673cae FG |
1572 | m->get_connection()->send_message(reply); |
1573 | } | |
1574 | ||
1575 | void OSDService::handle_misdirected_op(PG *pg, OpRequestRef op) | |
1576 | { | |
31f18b77 FG |
1577 | if (!cct->_conf->osd_debug_misdirected_ops) { |
1578 | return; | |
1579 | } | |
1580 | ||
9f95a23c | 1581 | auto m = op->get_req<MOSDOp>(); |
11fdf7f2 | 1582 | ceph_assert(m->get_type() == CEPH_MSG_OSD_OP); |
7c673cae | 1583 | |
11fdf7f2 | 1584 | ceph_assert(m->get_map_epoch() >= pg->get_history().same_primary_since); |
7c673cae FG |
1585 | |
1586 | if (pg->is_ec_pg()) { | |
1587 | /** | |
1588 | * OSD recomputes op target based on current OSDMap. With an EC pg, we | |
1589 | * can get this result: | |
1590 | * 1) client at map 512 sends an op to osd 3, pg_t 3.9 based on mapping | |
1591 | * [CRUSH_ITEM_NONE, 2, 3]/3 | |
1592 | * 2) OSD 3 at map 513 remaps op to osd 3, spg_t 3.9s0 based on mapping | |
1593 | * [3, 2, 3]/3 | |
1594 | * 3) PG 3.9s0 dequeues the op at epoch 512 and notices that it isn't primary | |
1595 | * -- misdirected op | |
1596 | * 4) client resends and this time PG 3.9s0 having caught up to 513 gets | |
1597 | * it and fulfils it | |
1598 | * | |
1599 | * We can't compute the op target based on the sending map epoch due to | |
1600 | * splitting. The simplest thing is to detect such cases here and drop | |
1601 | * them without an error (the client will resend anyway). | |
1602 | */ | |
11fdf7f2 | 1603 | ceph_assert(m->get_map_epoch() <= superblock.newest_map); |
7c673cae FG |
1604 | OSDMapRef opmap = try_get_map(m->get_map_epoch()); |
1605 | if (!opmap) { | |
1606 | dout(7) << __func__ << ": " << *pg << " no longer have map for " | |
1607 | << m->get_map_epoch() << ", dropping" << dendl; | |
1608 | return; | |
1609 | } | |
1610 | pg_t _pgid = m->get_raw_pg(); | |
1611 | spg_t pgid; | |
1612 | if ((m->get_flags() & CEPH_OSD_FLAG_PGOP) == 0) | |
1613 | _pgid = opmap->raw_pg_to_pg(_pgid); | |
1614 | if (opmap->get_primary_shard(_pgid, &pgid) && | |
11fdf7f2 | 1615 | pgid.shard != pg->pg_id.shard) { |
7c673cae FG |
1616 | dout(7) << __func__ << ": " << *pg << " primary changed since " |
1617 | << m->get_map_epoch() << ", dropping" << dendl; | |
1618 | return; | |
1619 | } | |
1620 | } | |
1621 | ||
1622 | dout(7) << *pg << " misdirected op in " << m->get_map_epoch() << dendl; | |
1623 | clog->warn() << m->get_source_inst() << " misdirected " << m->get_reqid() | |
1624 | << " pg " << m->get_raw_pg() | |
1625 | << " to osd." << whoami | |
11fdf7f2 | 1626 | << " not " << pg->get_acting() |
7c673cae | 1627 | << " in e" << m->get_map_epoch() << "/" << osdmap->get_epoch(); |
7c673cae FG |
1628 | } |
1629 | ||
9f95a23c | 1630 | void OSDService::enqueue_back(OpSchedulerItem&& qi) |
7c673cae | 1631 | { |
11fdf7f2 | 1632 | osd->op_shardedwq.queue(std::move(qi)); |
7c673cae FG |
1633 | } |
1634 | ||
9f95a23c | 1635 | void OSDService::enqueue_front(OpSchedulerItem&& qi) |
7c673cae | 1636 | { |
11fdf7f2 | 1637 | osd->op_shardedwq.queue_front(std::move(qi)); |
7c673cae FG |
1638 | } |
1639 | ||
11fdf7f2 TL |
1640 | void OSDService::queue_recovery_context( |
1641 | PG *pg, | |
1642 | GenContext<ThreadPool::TPHandle&> *c) | |
7c673cae | 1643 | { |
11fdf7f2 TL |
1644 | epoch_t e = get_osdmap_epoch(); |
1645 | enqueue_back( | |
9f95a23c TL |
1646 | OpSchedulerItem( |
1647 | unique_ptr<OpSchedulerItem::OpQueueable>( | |
11fdf7f2 TL |
1648 | new PGRecoveryContext(pg->get_pgid(), c, e)), |
1649 | cct->_conf->osd_recovery_cost, | |
1650 | cct->_conf->osd_recovery_priority, | |
1651 | ceph_clock_now(), | |
1652 | 0, | |
1653 | e)); | |
7c673cae FG |
1654 | } |
1655 | ||
1656 | void OSDService::queue_for_snap_trim(PG *pg) | |
1657 | { | |
1658 | dout(10) << "queueing " << *pg << " for snaptrim" << dendl; | |
11fdf7f2 | 1659 | enqueue_back( |
9f95a23c TL |
1660 | OpSchedulerItem( |
1661 | unique_ptr<OpSchedulerItem::OpQueueable>( | |
11fdf7f2 TL |
1662 | new PGSnapTrim(pg->get_pgid(), pg->get_osdmap_epoch())), |
1663 | cct->_conf->osd_snap_trim_cost, | |
1664 | cct->_conf->osd_snap_trim_priority, | |
1665 | ceph_clock_now(), | |
1666 | 0, | |
1667 | pg->get_osdmap_epoch())); | |
1668 | } | |
1669 | ||
f67539c2 TL |
1670 | template <class MSG_TYPE> |
1671 | void OSDService::queue_scrub_event_msg(PG* pg, | |
1672 | Scrub::scrub_prio_t with_priority, | |
20effc67 TL |
1673 | unsigned int qu_priority, |
1674 | Scrub::act_token_t act_token) | |
11fdf7f2 | 1675 | { |
11fdf7f2 | 1676 | const auto epoch = pg->get_osdmap_epoch(); |
20effc67 TL |
1677 | auto msg = new MSG_TYPE(pg->get_pgid(), epoch, act_token); |
1678 | dout(15) << "queue a scrub event (" << *msg << ") for " << *pg | |
1679 | << ". Epoch: " << epoch << " token: " << act_token << dendl; | |
f67539c2 TL |
1680 | |
1681 | enqueue_back(OpSchedulerItem( | |
1682 | unique_ptr<OpSchedulerItem::OpQueueable>(msg), cct->_conf->osd_scrub_cost, | |
1683 | pg->scrub_requeue_priority(with_priority, qu_priority), ceph_clock_now(), 0, epoch)); | |
1684 | } | |
1685 | ||
1686 | template <class MSG_TYPE> | |
20effc67 TL |
1687 | void OSDService::queue_scrub_event_msg(PG* pg, |
1688 | Scrub::scrub_prio_t with_priority) | |
f67539c2 TL |
1689 | { |
1690 | const auto epoch = pg->get_osdmap_epoch(); | |
1691 | auto msg = new MSG_TYPE(pg->get_pgid(), epoch); | |
1692 | dout(15) << "queue a scrub event (" << *msg << ") for " << *pg << ". Epoch: " << epoch << dendl; | |
1693 | ||
1694 | enqueue_back(OpSchedulerItem( | |
1695 | unique_ptr<OpSchedulerItem::OpQueueable>(msg), cct->_conf->osd_scrub_cost, | |
1696 | pg->scrub_requeue_priority(with_priority), ceph_clock_now(), 0, epoch)); | |
1697 | } | |
1698 | ||
1699 | void OSDService::queue_for_scrub(PG* pg, Scrub::scrub_prio_t with_priority) | |
1700 | { | |
1701 | queue_scrub_event_msg<PGScrub>(pg, with_priority); | |
1702 | } | |
1703 | ||
1704 | void OSDService::queue_scrub_after_repair(PG* pg, Scrub::scrub_prio_t with_priority) | |
1705 | { | |
1706 | queue_scrub_event_msg<PGScrubAfterRepair>(pg, with_priority); | |
1707 | } | |
1708 | ||
1709 | void OSDService::queue_for_rep_scrub(PG* pg, | |
1710 | Scrub::scrub_prio_t with_priority, | |
20effc67 TL |
1711 | unsigned int qu_priority, |
1712 | Scrub::act_token_t act_token) | |
f67539c2 | 1713 | { |
20effc67 | 1714 | queue_scrub_event_msg<PGRepScrub>(pg, with_priority, qu_priority, act_token); |
f67539c2 TL |
1715 | } |
1716 | ||
1717 | void OSDService::queue_for_rep_scrub_resched(PG* pg, | |
1718 | Scrub::scrub_prio_t with_priority, | |
20effc67 TL |
1719 | unsigned int qu_priority, |
1720 | Scrub::act_token_t act_token) | |
f67539c2 TL |
1721 | { |
1722 | // Resulting scrub event: 'SchedReplica' | |
20effc67 TL |
1723 | queue_scrub_event_msg<PGRepScrubResched>(pg, with_priority, qu_priority, |
1724 | act_token); | |
f67539c2 TL |
1725 | } |
1726 | ||
1727 | void OSDService::queue_for_scrub_granted(PG* pg, Scrub::scrub_prio_t with_priority) | |
1728 | { | |
1729 | // Resulting scrub event: 'RemotesReserved' | |
1730 | queue_scrub_event_msg<PGScrubResourcesOK>(pg, with_priority); | |
1731 | } | |
1732 | ||
1733 | void OSDService::queue_for_scrub_denied(PG* pg, Scrub::scrub_prio_t with_priority) | |
1734 | { | |
1735 | // Resulting scrub event: 'ReservationFailure' | |
1736 | queue_scrub_event_msg<PGScrubDenied>(pg, with_priority); | |
1737 | } | |
1738 | ||
1739 | void OSDService::queue_for_scrub_resched(PG* pg, Scrub::scrub_prio_t with_priority) | |
1740 | { | |
1741 | // Resulting scrub event: 'InternalSchedScrub' | |
1742 | queue_scrub_event_msg<PGScrubResched>(pg, with_priority); | |
1743 | } | |
1744 | ||
1745 | void OSDService::queue_scrub_pushes_update(PG* pg, Scrub::scrub_prio_t with_priority) | |
1746 | { | |
1747 | // Resulting scrub event: 'ActivePushesUpd' | |
1748 | queue_scrub_event_msg<PGScrubPushesUpdate>(pg, with_priority); | |
1749 | } | |
1750 | ||
20effc67 TL |
1751 | void OSDService::queue_scrub_chunk_free(PG* pg, Scrub::scrub_prio_t with_priority) |
1752 | { | |
1753 | // Resulting scrub event: 'SelectedChunkFree' | |
1754 | queue_scrub_event_msg<PGScrubChunkIsFree>(pg, with_priority); | |
1755 | } | |
1756 | ||
1757 | void OSDService::queue_scrub_chunk_busy(PG* pg, Scrub::scrub_prio_t with_priority) | |
1758 | { | |
1759 | // Resulting scrub event: 'ChunkIsBusy' | |
1760 | queue_scrub_event_msg<PGScrubChunkIsBusy>(pg, with_priority); | |
1761 | } | |
1762 | ||
f67539c2 TL |
1763 | void OSDService::queue_scrub_applied_update(PG* pg, Scrub::scrub_prio_t with_priority) |
1764 | { | |
1765 | queue_scrub_event_msg<PGScrubAppliedUpdate>(pg, with_priority); | |
1766 | } | |
1767 | ||
1768 | void OSDService::queue_scrub_unblocking(PG* pg, Scrub::scrub_prio_t with_priority) | |
1769 | { | |
1770 | // Resulting scrub event: 'Unblocked' | |
1771 | queue_scrub_event_msg<PGScrubUnblocked>(pg, with_priority); | |
1772 | } | |
1773 | ||
1774 | void OSDService::queue_scrub_digest_update(PG* pg, Scrub::scrub_prio_t with_priority) | |
1775 | { | |
1776 | // Resulting scrub event: 'DigestUpdate' | |
1777 | queue_scrub_event_msg<PGScrubDigestUpdate>(pg, with_priority); | |
1778 | } | |
1779 | ||
20effc67 TL |
1780 | void OSDService::queue_scrub_got_local_map(PG* pg, Scrub::scrub_prio_t with_priority) |
1781 | { | |
1782 | // Resulting scrub event: 'IntLocalMapDone' | |
1783 | queue_scrub_event_msg<PGScrubGotLocalMap>(pg, with_priority); | |
1784 | } | |
1785 | ||
f67539c2 TL |
1786 | void OSDService::queue_scrub_got_repl_maps(PG* pg, Scrub::scrub_prio_t with_priority) |
1787 | { | |
1788 | // Resulting scrub event: 'GotReplicas' | |
1789 | queue_scrub_event_msg<PGScrubGotReplMaps>(pg, with_priority); | |
1790 | } | |
1791 | ||
20effc67 TL |
1792 | void OSDService::queue_scrub_maps_compared(PG* pg, Scrub::scrub_prio_t with_priority) |
1793 | { | |
1794 | // Resulting scrub event: 'MapsCompared' | |
1795 | queue_scrub_event_msg<PGScrubMapsCompared>(pg, with_priority); | |
1796 | } | |
1797 | ||
f67539c2 TL |
1798 | void OSDService::queue_scrub_replica_pushes(PG *pg, Scrub::scrub_prio_t with_priority) |
1799 | { | |
1800 | // Resulting scrub event: 'ReplicaPushesUpd' | |
1801 | queue_scrub_event_msg<PGScrubReplicaPushes>(pg, with_priority); | |
11fdf7f2 TL |
1802 | } |
1803 | ||
20effc67 TL |
1804 | void OSDService::queue_scrub_is_finished(PG *pg) |
1805 | { | |
1806 | // Resulting scrub event: 'ScrubFinished' | |
1807 | queue_scrub_event_msg<PGScrubScrubFinished>(pg, Scrub::scrub_prio_t::high_priority); | |
1808 | } | |
1809 | ||
1810 | void OSDService::queue_scrub_next_chunk(PG *pg, Scrub::scrub_prio_t with_priority) | |
1811 | { | |
1812 | // Resulting scrub event: 'NextChunk' | |
1813 | queue_scrub_event_msg<PGScrubGetNextChunk>(pg, with_priority); | |
1814 | } | |
1815 | ||
11fdf7f2 TL |
1816 | void OSDService::queue_for_pg_delete(spg_t pgid, epoch_t e) |
1817 | { | |
1818 | dout(10) << __func__ << " on " << pgid << " e " << e << dendl; | |
1819 | enqueue_back( | |
9f95a23c TL |
1820 | OpSchedulerItem( |
1821 | unique_ptr<OpSchedulerItem::OpQueueable>( | |
11fdf7f2 TL |
1822 | new PGDelete(pgid, e)), |
1823 | cct->_conf->osd_pg_delete_cost, | |
1824 | cct->_conf->osd_pg_delete_priority, | |
1825 | ceph_clock_now(), | |
1826 | 0, | |
1827 | e)); | |
1828 | } | |
1829 | ||
1830 | bool OSDService::try_finish_pg_delete(PG *pg, unsigned old_pg_num) | |
1831 | { | |
1832 | return osd->try_finish_pg_delete(pg, old_pg_num); | |
1833 | } | |
1834 | ||
1835 | // --- | |
1836 | ||
1837 | void OSDService::set_ready_to_merge_source(PG *pg, eversion_t version) | |
1838 | { | |
1839 | std::lock_guard l(merge_lock); | |
1840 | dout(10) << __func__ << " " << pg->pg_id << dendl; | |
1841 | ready_to_merge_source[pg->pg_id.pgid] = version; | |
1842 | assert(not_ready_to_merge_source.count(pg->pg_id.pgid) == 0); | |
1843 | _send_ready_to_merge(); | |
1844 | } | |
1845 | ||
1846 | void OSDService::set_ready_to_merge_target(PG *pg, | |
1847 | eversion_t version, | |
1848 | epoch_t last_epoch_started, | |
1849 | epoch_t last_epoch_clean) | |
1850 | { | |
1851 | std::lock_guard l(merge_lock); | |
1852 | dout(10) << __func__ << " " << pg->pg_id << dendl; | |
1853 | ready_to_merge_target.insert(make_pair(pg->pg_id.pgid, | |
1854 | make_tuple(version, | |
1855 | last_epoch_started, | |
1856 | last_epoch_clean))); | |
1857 | assert(not_ready_to_merge_target.count(pg->pg_id.pgid) == 0); | |
1858 | _send_ready_to_merge(); | |
1859 | } | |
1860 | ||
1861 | void OSDService::set_not_ready_to_merge_source(pg_t source) | |
1862 | { | |
1863 | std::lock_guard l(merge_lock); | |
1864 | dout(10) << __func__ << " " << source << dendl; | |
1865 | not_ready_to_merge_source.insert(source); | |
1866 | assert(ready_to_merge_source.count(source) == 0); | |
1867 | _send_ready_to_merge(); | |
1868 | } | |
1869 | ||
1870 | void OSDService::set_not_ready_to_merge_target(pg_t target, pg_t source) | |
1871 | { | |
1872 | std::lock_guard l(merge_lock); | |
1873 | dout(10) << __func__ << " " << target << " source " << source << dendl; | |
1874 | not_ready_to_merge_target[target] = source; | |
1875 | assert(ready_to_merge_target.count(target) == 0); | |
1876 | _send_ready_to_merge(); | |
1877 | } | |
1878 | ||
1879 | void OSDService::send_ready_to_merge() | |
1880 | { | |
1881 | std::lock_guard l(merge_lock); | |
1882 | _send_ready_to_merge(); | |
1883 | } | |
1884 | ||
1885 | void OSDService::_send_ready_to_merge() | |
1886 | { | |
1887 | dout(20) << __func__ | |
1888 | << " ready_to_merge_source " << ready_to_merge_source | |
1889 | << " not_ready_to_merge_source " << not_ready_to_merge_source | |
1890 | << " ready_to_merge_target " << ready_to_merge_target | |
1891 | << " not_ready_to_merge_target " << not_ready_to_merge_target | |
1892 | << " sent_ready_to_merge_source " << sent_ready_to_merge_source | |
1893 | << dendl; | |
1894 | for (auto src : not_ready_to_merge_source) { | |
1895 | if (sent_ready_to_merge_source.count(src) == 0) { | |
1896 | monc->send_mon_message(new MOSDPGReadyToMerge( | |
1897 | src, | |
1898 | {}, {}, 0, 0, | |
1899 | false, | |
1900 | osdmap->get_epoch())); | |
1901 | sent_ready_to_merge_source.insert(src); | |
1902 | } | |
1903 | } | |
1904 | for (auto p : not_ready_to_merge_target) { | |
1905 | if (sent_ready_to_merge_source.count(p.second) == 0) { | |
1906 | monc->send_mon_message(new MOSDPGReadyToMerge( | |
1907 | p.second, | |
1908 | {}, {}, 0, 0, | |
1909 | false, | |
1910 | osdmap->get_epoch())); | |
1911 | sent_ready_to_merge_source.insert(p.second); | |
1912 | } | |
1913 | } | |
1914 | for (auto src : ready_to_merge_source) { | |
1915 | if (not_ready_to_merge_source.count(src.first) || | |
1916 | not_ready_to_merge_target.count(src.first.get_parent())) { | |
1917 | continue; | |
1918 | } | |
1919 | auto p = ready_to_merge_target.find(src.first.get_parent()); | |
1920 | if (p != ready_to_merge_target.end() && | |
1921 | sent_ready_to_merge_source.count(src.first) == 0) { | |
1922 | monc->send_mon_message(new MOSDPGReadyToMerge( | |
1923 | src.first, // source pgid | |
1924 | src.second, // src version | |
1925 | std::get<0>(p->second), // target version | |
1926 | std::get<1>(p->second), // PG's last_epoch_started | |
1927 | std::get<2>(p->second), // PG's last_epoch_clean | |
1928 | true, | |
1929 | osdmap->get_epoch())); | |
1930 | sent_ready_to_merge_source.insert(src.first); | |
1931 | } | |
1932 | } | |
1933 | } | |
1934 | ||
1935 | void OSDService::clear_ready_to_merge(PG *pg) | |
1936 | { | |
1937 | std::lock_guard l(merge_lock); | |
1938 | dout(10) << __func__ << " " << pg->pg_id << dendl; | |
1939 | ready_to_merge_source.erase(pg->pg_id.pgid); | |
1940 | ready_to_merge_target.erase(pg->pg_id.pgid); | |
1941 | not_ready_to_merge_source.erase(pg->pg_id.pgid); | |
1942 | not_ready_to_merge_target.erase(pg->pg_id.pgid); | |
1943 | sent_ready_to_merge_source.erase(pg->pg_id.pgid); | |
1944 | } | |
1945 | ||
1946 | void OSDService::clear_sent_ready_to_merge() | |
1947 | { | |
1948 | std::lock_guard l(merge_lock); | |
1949 | sent_ready_to_merge_source.clear(); | |
1950 | } | |
1951 | ||
9f95a23c | 1952 | void OSDService::prune_sent_ready_to_merge(const OSDMapRef& osdmap) |
11fdf7f2 TL |
1953 | { |
1954 | std::lock_guard l(merge_lock); | |
1955 | auto i = sent_ready_to_merge_source.begin(); | |
1956 | while (i != sent_ready_to_merge_source.end()) { | |
1957 | if (!osdmap->pg_exists(*i)) { | |
1958 | dout(10) << __func__ << " " << *i << dendl; | |
1959 | i = sent_ready_to_merge_source.erase(i); | |
1960 | } else { | |
1961 | ++i; | |
1962 | } | |
1963 | } | |
7c673cae FG |
1964 | } |
1965 | ||
11fdf7f2 TL |
1966 | // --- |
1967 | ||
1968 | void OSDService::_queue_for_recovery( | |
1969 | std::pair<epoch_t, PGRef> p, | |
1970 | uint64_t reserved_pushes) | |
1971 | { | |
9f95a23c | 1972 | ceph_assert(ceph_mutex_is_locked_by_me(recovery_lock)); |
11fdf7f2 | 1973 | enqueue_back( |
9f95a23c TL |
1974 | OpSchedulerItem( |
1975 | unique_ptr<OpSchedulerItem::OpQueueable>( | |
11fdf7f2 TL |
1976 | new PGRecovery( |
1977 | p.second->get_pgid(), p.first, reserved_pushes)), | |
1978 | cct->_conf->osd_recovery_cost, | |
1979 | cct->_conf->osd_recovery_priority, | |
1980 | ceph_clock_now(), | |
1981 | 0, | |
1982 | p.first)); | |
1983 | } | |
7c673cae FG |
1984 | |
1985 | // ==================================================================== | |
1986 | // OSD | |
1987 | ||
1988 | #undef dout_prefix | |
1989 | #define dout_prefix *_dout | |
1990 | ||
1991 | // Commands shared between OSD's console and admin console: | |
f67539c2 | 1992 | namespace ceph::osd_cmds { |
7c673cae | 1993 | |
11fdf7f2 | 1994 | int heap(CephContext& cct, const cmdmap_t& cmdmap, Formatter& f, std::ostream& os); |
f67539c2 TL |
1995 | |
1996 | } // namespace ceph::osd_cmds | |
7c673cae | 1997 | |
20effc67 TL |
1998 | int OSD::mkfs(CephContext *cct, |
1999 | std::unique_ptr<ObjectStore> store, | |
2000 | uuid_d fsid, | |
2001 | int whoami, | |
2002 | string osdspec_affinity) | |
7c673cae FG |
2003 | { |
2004 | int ret; | |
2005 | ||
7c673cae FG |
2006 | OSDSuperblock sb; |
2007 | bufferlist sbbl; | |
7c673cae FG |
2008 | // if we are fed a uuid for this osd, use it. |
2009 | store->set_fsid(cct->_conf->osd_uuid); | |
2010 | ||
2011 | ret = store->mkfs(); | |
2012 | if (ret) { | |
224ce89b WB |
2013 | derr << "OSD::mkfs: ObjectStore::mkfs failed with error " |
2014 | << cpp_strerror(ret) << dendl; | |
20effc67 | 2015 | return ret; |
7c673cae FG |
2016 | } |
2017 | ||
31f18b77 | 2018 | store->set_cache_shards(1); // doesn't matter for mkfs! |
7c673cae FG |
2019 | |
2020 | ret = store->mount(); | |
2021 | if (ret) { | |
224ce89b WB |
2022 | derr << "OSD::mkfs: couldn't mount ObjectStore: error " |
2023 | << cpp_strerror(ret) << dendl; | |
20effc67 | 2024 | return ret; |
7c673cae FG |
2025 | } |
2026 | ||
20effc67 TL |
2027 | auto umount_store = make_scope_guard([&] { |
2028 | store->umount(); | |
2029 | }); | |
2030 | ||
2031 | ObjectStore::CollectionHandle ch = | |
2032 | store->open_collection(coll_t::meta()); | |
11fdf7f2 TL |
2033 | if (ch) { |
2034 | ret = store->read(ch, OSD_SUPERBLOCK_GOBJECT, 0, 0, sbbl); | |
2035 | if (ret < 0) { | |
2036 | derr << "OSD::mkfs: have meta collection but no superblock" << dendl; | |
20effc67 | 2037 | return ret; |
11fdf7f2 | 2038 | } |
7c673cae FG |
2039 | /* if we already have superblock, check content of superblock */ |
2040 | dout(0) << " have superblock" << dendl; | |
11fdf7f2 TL |
2041 | auto p = sbbl.cbegin(); |
2042 | decode(sb, p); | |
7c673cae FG |
2043 | if (whoami != sb.whoami) { |
2044 | derr << "provided osd id " << whoami << " != superblock's " << sb.whoami | |
2045 | << dendl; | |
20effc67 | 2046 | return -EINVAL; |
7c673cae FG |
2047 | } |
2048 | if (fsid != sb.cluster_fsid) { | |
2049 | derr << "provided cluster fsid " << fsid | |
2050 | << " != superblock's " << sb.cluster_fsid << dendl; | |
20effc67 | 2051 | return -EINVAL; |
7c673cae FG |
2052 | } |
2053 | } else { | |
2054 | // create superblock | |
2055 | sb.cluster_fsid = fsid; | |
2056 | sb.osd_fsid = store->get_fsid(); | |
2057 | sb.whoami = whoami; | |
2058 | sb.compat_features = get_osd_initial_compat_set(); | |
2059 | ||
2060 | bufferlist bl; | |
11fdf7f2 | 2061 | encode(sb, bl); |
7c673cae | 2062 | |
11fdf7f2 TL |
2063 | ObjectStore::CollectionHandle ch = store->create_new_collection( |
2064 | coll_t::meta()); | |
7c673cae FG |
2065 | ObjectStore::Transaction t; |
2066 | t.create_collection(coll_t::meta(), 0); | |
2067 | t.write(coll_t::meta(), OSD_SUPERBLOCK_GOBJECT, 0, bl.length(), bl); | |
11fdf7f2 | 2068 | ret = store->queue_transaction(ch, std::move(t)); |
7c673cae FG |
2069 | if (ret) { |
2070 | derr << "OSD::mkfs: error while writing OSD_SUPERBLOCK_GOBJECT: " | |
11fdf7f2 | 2071 | << "queue_transaction returned " << cpp_strerror(ret) << dendl; |
20effc67 | 2072 | return ret; |
7c673cae | 2073 | } |
a4b75251 | 2074 | ch->flush(); |
7c673cae FG |
2075 | } |
2076 | ||
20effc67 | 2077 | ret = write_meta(cct, store.get(), sb.cluster_fsid, sb.osd_fsid, whoami, osdspec_affinity); |
7c673cae | 2078 | if (ret) { |
224ce89b WB |
2079 | derr << "OSD::mkfs: failed to write fsid file: error " |
2080 | << cpp_strerror(ret) << dendl; | |
11fdf7f2 | 2081 | } |
7c673cae FG |
2082 | return ret; |
2083 | } | |
2084 | ||
e306af50 | 2085 | int OSD::write_meta(CephContext *cct, ObjectStore *store, uuid_d& cluster_fsid, uuid_d& osd_fsid, int whoami, string& osdspec_affinity) |
7c673cae FG |
2086 | { |
2087 | char val[80]; | |
2088 | int r; | |
2089 | ||
2090 | snprintf(val, sizeof(val), "%s", CEPH_OSD_ONDISK_MAGIC); | |
2091 | r = store->write_meta("magic", val); | |
2092 | if (r < 0) | |
2093 | return r; | |
2094 | ||
2095 | snprintf(val, sizeof(val), "%d", whoami); | |
2096 | r = store->write_meta("whoami", val); | |
2097 | if (r < 0) | |
2098 | return r; | |
2099 | ||
2100 | cluster_fsid.print(val); | |
2101 | r = store->write_meta("ceph_fsid", val); | |
2102 | if (r < 0) | |
2103 | return r; | |
2104 | ||
11fdf7f2 | 2105 | string key = cct->_conf.get_val<string>("key"); |
3efd9988 FG |
2106 | if (key.size()) { |
2107 | r = store->write_meta("osd_key", key); | |
2108 | if (r < 0) | |
2109 | return r; | |
b32b8144 | 2110 | } else { |
11fdf7f2 | 2111 | string keyfile = cct->_conf.get_val<string>("keyfile"); |
b32b8144 FG |
2112 | if (!keyfile.empty()) { |
2113 | bufferlist keybl; | |
2114 | string err; | |
11fdf7f2 | 2115 | r = keybl.read_file(keyfile.c_str(), &err); |
b32b8144 FG |
2116 | if (r < 0) { |
2117 | derr << __func__ << " failed to read keyfile " << keyfile << ": " | |
2118 | << err << ": " << cpp_strerror(r) << dendl; | |
2119 | return r; | |
2120 | } | |
2121 | r = store->write_meta("osd_key", keybl.to_str()); | |
2122 | if (r < 0) | |
2123 | return r; | |
2124 | } | |
3efd9988 | 2125 | } |
e306af50 TL |
2126 | if (!osdspec_affinity.empty()) { |
2127 | r = store->write_meta("osdspec_affinity", osdspec_affinity.c_str()); | |
2128 | if (r < 0) | |
2129 | return r; | |
2130 | } | |
3efd9988 | 2131 | |
7c673cae FG |
2132 | r = store->write_meta("ready", "ready"); |
2133 | if (r < 0) | |
2134 | return r; | |
2135 | ||
2136 | return 0; | |
2137 | } | |
2138 | ||
11fdf7f2 TL |
2139 | int OSD::peek_meta(ObjectStore *store, |
2140 | std::string *magic, | |
2141 | uuid_d *cluster_fsid, | |
2142 | uuid_d *osd_fsid, | |
2143 | int *whoami, | |
9f95a23c | 2144 | ceph_release_t *require_osd_release) |
7c673cae FG |
2145 | { |
2146 | string val; | |
2147 | ||
2148 | int r = store->read_meta("magic", &val); | |
2149 | if (r < 0) | |
2150 | return r; | |
11fdf7f2 | 2151 | *magic = val; |
7c673cae FG |
2152 | |
2153 | r = store->read_meta("whoami", &val); | |
2154 | if (r < 0) | |
2155 | return r; | |
11fdf7f2 | 2156 | *whoami = atoi(val.c_str()); |
7c673cae FG |
2157 | |
2158 | r = store->read_meta("ceph_fsid", &val); | |
2159 | if (r < 0) | |
2160 | return r; | |
11fdf7f2 | 2161 | r = cluster_fsid->parse(val.c_str()); |
7c673cae FG |
2162 | if (!r) |
2163 | return -EINVAL; | |
2164 | ||
2165 | r = store->read_meta("fsid", &val); | |
2166 | if (r < 0) { | |
11fdf7f2 | 2167 | *osd_fsid = uuid_d(); |
7c673cae | 2168 | } else { |
11fdf7f2 | 2169 | r = osd_fsid->parse(val.c_str()); |
7c673cae FG |
2170 | if (!r) |
2171 | return -EINVAL; | |
2172 | } | |
2173 | ||
11fdf7f2 TL |
2174 | r = store->read_meta("require_osd_release", &val); |
2175 | if (r >= 0) { | |
9f95a23c | 2176 | *require_osd_release = ceph_release_from_name(val); |
11fdf7f2 TL |
2177 | } |
2178 | ||
7c673cae FG |
2179 | return 0; |
2180 | } | |
2181 | ||
2182 | ||
2183 | #undef dout_prefix | |
2184 | #define dout_prefix _prefix(_dout, whoami, get_osdmap_epoch()) | |
2185 | ||
2186 | // cons/des | |
2187 | ||
20effc67 TL |
2188 | OSD::OSD(CephContext *cct_, |
2189 | std::unique_ptr<ObjectStore> store_, | |
7c673cae FG |
2190 | int id, |
2191 | Messenger *internal_messenger, | |
2192 | Messenger *external_messenger, | |
2193 | Messenger *hb_client_front, | |
2194 | Messenger *hb_client_back, | |
2195 | Messenger *hb_front_serverm, | |
2196 | Messenger *hb_back_serverm, | |
2197 | Messenger *osdc_messenger, | |
2198 | MonClient *mc, | |
f67539c2 TL |
2199 | const std::string &dev, const std::string &jdev, |
2200 | ceph::async::io_context_pool& poolctx) : | |
7c673cae | 2201 | Dispatcher(cct_), |
7c673cae | 2202 | tick_timer(cct, osd_lock), |
7c673cae | 2203 | tick_timer_without_osd_lock(cct, tick_timer_lock), |
11fdf7f2 | 2204 | gss_ktfile_client(cct->_conf.get_val<std::string>("gss_ktab_client_file")), |
7c673cae FG |
2205 | cluster_messenger(internal_messenger), |
2206 | client_messenger(external_messenger), | |
2207 | objecter_messenger(osdc_messenger), | |
2208 | monc(mc), | |
9f95a23c | 2209 | mgrc(cct_, client_messenger, &mc->monmap), |
f67539c2 TL |
2210 | logger(create_logger()), |
2211 | recoverystate_perf(create_recoverystate_perf()), | |
20effc67 | 2212 | store(std::move(store_)), |
7c673cae FG |
2213 | log_client(cct, client_messenger, &mc->monmap, LogClient::NO_FLAGS), |
2214 | clog(log_client.create_channel()), | |
2215 | whoami(id), | |
2216 | dev_path(dev), journal_path(jdev), | |
31f18b77 | 2217 | store_is_rotational(store->is_rotational()), |
7c673cae FG |
2218 | trace_endpoint("0.0.0.0", 0, "osd"), |
2219 | asok_hook(NULL), | |
11fdf7f2 TL |
2220 | m_osd_pg_epoch_max_lag_factor(cct->_conf.get_val<double>( |
2221 | "osd_pg_epoch_max_lag_factor")), | |
7c673cae | 2222 | osd_compat(get_osd_compat_set()), |
7c673cae | 2223 | osd_op_tp(cct, "OSD::osd_op_tp", "tp_osd_tp", |
31f18b77 | 2224 | get_num_op_threads()), |
7c673cae FG |
2225 | heartbeat_stop(false), |
2226 | heartbeat_need_update(true), | |
2227 | hb_front_client_messenger(hb_client_front), | |
2228 | hb_back_client_messenger(hb_client_back), | |
2229 | hb_front_server_messenger(hb_front_serverm), | |
2230 | hb_back_server_messenger(hb_back_serverm), | |
2231 | daily_loadavg(0.0), | |
2232 | heartbeat_thread(this), | |
2233 | heartbeat_dispatcher(this), | |
2234 | op_tracker(cct, cct->_conf->osd_enable_op_tracker, | |
2235 | cct->_conf->osd_num_op_tracker_shard), | |
2236 | test_ops_hook(NULL), | |
7c673cae | 2237 | op_shardedwq( |
7c673cae | 2238 | this, |
f67539c2 TL |
2239 | ceph::make_timespan(cct->_conf->osd_op_thread_timeout), |
2240 | ceph::make_timespan(cct->_conf->osd_op_thread_suicide_timeout), | |
7c673cae | 2241 | &osd_op_tp), |
7c673cae | 2242 | last_pg_create_epoch(0), |
11fdf7f2 | 2243 | boot_finisher(cct), |
7c673cae FG |
2244 | up_thru_wanted(0), |
2245 | requested_full_first(0), | |
2246 | requested_full_last(0), | |
f67539c2 | 2247 | service(this, poolctx) |
7c673cae | 2248 | { |
11fdf7f2 TL |
2249 | |
2250 | if (!gss_ktfile_client.empty()) { | |
f67539c2 TL |
2251 | // Assert we can export environment variable |
2252 | /* | |
11fdf7f2 TL |
2253 | The default client keytab is used, if it is present and readable, |
2254 | to automatically obtain initial credentials for GSSAPI client | |
2255 | applications. The principal name of the first entry in the client | |
2256 | keytab is used by default when obtaining initial credentials. | |
2257 | 1. The KRB5_CLIENT_KTNAME environment variable. | |
2258 | 2. The default_client_keytab_name profile variable in [libdefaults]. | |
2259 | 3. The hardcoded default, DEFCKTNAME. | |
2260 | */ | |
f67539c2 | 2261 | const int32_t set_result(setenv("KRB5_CLIENT_KTNAME", |
11fdf7f2 TL |
2262 | gss_ktfile_client.c_str(), 1)); |
2263 | ceph_assert(set_result == 0); | |
2264 | } | |
2265 | ||
7c673cae FG |
2266 | monc->set_messenger(client_messenger); |
2267 | op_tracker.set_complaint_and_threshold(cct->_conf->osd_op_complaint_time, | |
2268 | cct->_conf->osd_op_log_threshold); | |
2269 | op_tracker.set_history_size_and_duration(cct->_conf->osd_op_history_size, | |
2270 | cct->_conf->osd_op_history_duration); | |
2271 | op_tracker.set_history_slow_op_size_and_threshold(cct->_conf->osd_op_history_slow_op_size, | |
2272 | cct->_conf->osd_op_history_slow_op_threshold); | |
9f95a23c | 2273 | ObjectCleanRegions::set_max_num_intervals(cct->_conf->osd_object_clean_region_max_num_intervals); |
7c673cae FG |
2274 | #ifdef WITH_BLKIN |
2275 | std::stringstream ss; | |
2276 | ss << "osd." << whoami; | |
2277 | trace_endpoint.copy_name(ss.str()); | |
2278 | #endif | |
11fdf7f2 TL |
2279 | |
2280 | // initialize shards | |
2281 | num_shards = get_num_op_shards(); | |
2282 | for (uint32_t i = 0; i < num_shards; i++) { | |
2283 | OSDShard *one_shard = new OSDShard( | |
2284 | i, | |
2285 | cct, | |
9f95a23c | 2286 | this); |
11fdf7f2 TL |
2287 | shards.push_back(one_shard); |
2288 | } | |
7c673cae FG |
2289 | } |
2290 | ||
2291 | OSD::~OSD() | |
2292 | { | |
11fdf7f2 TL |
2293 | while (!shards.empty()) { |
2294 | delete shards.back(); | |
2295 | shards.pop_back(); | |
2296 | } | |
7c673cae FG |
2297 | cct->get_perfcounters_collection()->remove(recoverystate_perf); |
2298 | cct->get_perfcounters_collection()->remove(logger); | |
2299 | delete recoverystate_perf; | |
2300 | delete logger; | |
7c673cae FG |
2301 | } |
2302 | ||
91327a77 AA |
2303 | double OSD::get_tick_interval() const |
2304 | { | |
2305 | // vary +/- 5% to avoid scrub scheduling livelocks | |
2306 | constexpr auto delta = 0.05; | |
91327a77 | 2307 | return (OSD_TICK_INTERVAL * |
11fdf7f2 | 2308 | ceph::util::generate_random_number(1.0 - delta, 1.0 + delta)); |
91327a77 AA |
2309 | } |
2310 | ||
7c673cae FG |
2311 | void OSD::handle_signal(int signum) |
2312 | { | |
11fdf7f2 | 2313 | ceph_assert(signum == SIGINT || signum == SIGTERM); |
7c673cae FG |
2314 | derr << "*** Got signal " << sig_str(signum) << " ***" << dendl; |
2315 | shutdown(); | |
2316 | } | |
2317 | ||
2318 | int OSD::pre_init() | |
2319 | { | |
11fdf7f2 | 2320 | std::lock_guard lock(osd_lock); |
7c673cae FG |
2321 | if (is_stopping()) |
2322 | return 0; | |
2323 | ||
2324 | if (store->test_mount_in_use()) { | |
2325 | derr << "OSD::pre_init: object store '" << dev_path << "' is " | |
2326 | << "currently in use. (Is ceph-osd already running?)" << dendl; | |
2327 | return -EBUSY; | |
2328 | } | |
2329 | ||
11fdf7f2 TL |
2330 | cct->_conf.add_observer(this); |
2331 | return 0; | |
2332 | } | |
2333 | ||
2334 | int OSD::set_numa_affinity() | |
2335 | { | |
2336 | // storage numa node | |
2337 | int store_node = -1; | |
2338 | store->get_numa_node(&store_node, nullptr, nullptr); | |
2339 | if (store_node >= 0) { | |
2340 | dout(1) << __func__ << " storage numa node " << store_node << dendl; | |
2341 | } | |
2342 | ||
2343 | // check network numa node(s) | |
2344 | int front_node = -1, back_node = -1; | |
2345 | string front_iface = pick_iface( | |
2346 | cct, | |
2347 | client_messenger->get_myaddrs().front().get_sockaddr_storage()); | |
2348 | string back_iface = pick_iface( | |
2349 | cct, | |
2350 | cluster_messenger->get_myaddrs().front().get_sockaddr_storage()); | |
2351 | int r = get_iface_numa_node(front_iface, &front_node); | |
92f5a8d4 | 2352 | if (r >= 0 && front_node >= 0) { |
11fdf7f2 | 2353 | dout(1) << __func__ << " public network " << front_iface << " numa node " |
92f5a8d4 | 2354 | << front_node << dendl; |
11fdf7f2 | 2355 | r = get_iface_numa_node(back_iface, &back_node); |
92f5a8d4 | 2356 | if (r >= 0 && back_node >= 0) { |
11fdf7f2 TL |
2357 | dout(1) << __func__ << " cluster network " << back_iface << " numa node " |
2358 | << back_node << dendl; | |
2359 | if (front_node == back_node && | |
2360 | front_node == store_node) { | |
2361 | dout(1) << " objectstore and network numa nodes all match" << dendl; | |
2362 | if (g_conf().get_val<bool>("osd_numa_auto_affinity")) { | |
2363 | numa_node = front_node; | |
2364 | } | |
92f5a8d4 TL |
2365 | } else if (front_node != back_node) { |
2366 | dout(1) << __func__ << " public and cluster network numa nodes do not match" | |
2367 | << dendl; | |
11fdf7f2 TL |
2368 | } else { |
2369 | dout(1) << __func__ << " objectstore and network numa nodes do not match" | |
2370 | << dendl; | |
2371 | } | |
92f5a8d4 TL |
2372 | } else if (back_node == -2) { |
2373 | dout(1) << __func__ << " cluster network " << back_iface | |
2374 | << " ports numa nodes do not match" << dendl; | |
2375 | } else { | |
2376 | derr << __func__ << " unable to identify cluster interface '" << back_iface | |
2377 | << "' numa node: " << cpp_strerror(r) << dendl; | |
11fdf7f2 | 2378 | } |
92f5a8d4 TL |
2379 | } else if (front_node == -2) { |
2380 | dout(1) << __func__ << " public network " << front_iface | |
2381 | << " ports numa nodes do not match" << dendl; | |
11fdf7f2 TL |
2382 | } else { |
2383 | derr << __func__ << " unable to identify public interface '" << front_iface | |
2384 | << "' numa node: " << cpp_strerror(r) << dendl; | |
2385 | } | |
2386 | if (int node = g_conf().get_val<int64_t>("osd_numa_node"); node >= 0) { | |
2387 | // this takes precedence over the automagic logic above | |
2388 | numa_node = node; | |
2389 | } | |
2390 | if (numa_node >= 0) { | |
2391 | int r = get_numa_node_cpu_set(numa_node, &numa_cpu_set_size, &numa_cpu_set); | |
2392 | if (r < 0) { | |
2393 | dout(1) << __func__ << " unable to determine numa node " << numa_node | |
2394 | << " CPUs" << dendl; | |
2395 | numa_node = -1; | |
2396 | } else { | |
2397 | dout(1) << __func__ << " setting numa affinity to node " << numa_node | |
2398 | << " cpus " | |
2399 | << cpu_set_to_str_list(numa_cpu_set_size, &numa_cpu_set) | |
2400 | << dendl; | |
92f5a8d4 | 2401 | r = set_cpu_affinity_all_threads(numa_cpu_set_size, &numa_cpu_set); |
11fdf7f2 TL |
2402 | if (r < 0) { |
2403 | r = -errno; | |
2404 | derr << __func__ << " failed to set numa affinity: " << cpp_strerror(r) | |
2405 | << dendl; | |
2406 | numa_node = -1; | |
2407 | } | |
2408 | } | |
2409 | } else { | |
2410 | dout(1) << __func__ << " not setting numa affinity" << dendl; | |
2411 | } | |
7c673cae FG |
2412 | return 0; |
2413 | } | |
2414 | ||
2415 | // asok | |
2416 | ||
2417 | class OSDSocketHook : public AdminSocketHook { | |
2418 | OSD *osd; | |
2419 | public: | |
2420 | explicit OSDSocketHook(OSD *o) : osd(o) {} | |
9f95a23c TL |
2421 | int call(std::string_view prefix, const cmdmap_t& cmdmap, |
2422 | Formatter *f, | |
2423 | std::ostream& ss, | |
2424 | bufferlist& out) override { | |
2425 | ceph_abort("should use async hook"); | |
2426 | } | |
2427 | void call_async( | |
2428 | std::string_view prefix, | |
2429 | const cmdmap_t& cmdmap, | |
2430 | Formatter *f, | |
2431 | const bufferlist& inbl, | |
2432 | std::function<void(int,const std::string&,bufferlist&)> on_finish) override { | |
11fdf7f2 | 2433 | try { |
9f95a23c TL |
2434 | osd->asok_command(prefix, cmdmap, f, inbl, on_finish); |
2435 | } catch (const TOPNSPC::common::bad_cmd_get& e) { | |
2436 | bufferlist empty; | |
2437 | on_finish(-EINVAL, e.what(), empty); | |
11fdf7f2 | 2438 | } |
7c673cae FG |
2439 | } |
2440 | }; | |
2441 | ||
11fdf7f2 TL |
2442 | std::set<int64_t> OSD::get_mapped_pools() |
2443 | { | |
2444 | std::set<int64_t> pools; | |
2445 | std::vector<spg_t> pgids; | |
2446 | _get_pgids(&pgids); | |
2447 | for (const auto &pgid : pgids) { | |
2448 | pools.insert(pgid.pool()); | |
2449 | } | |
2450 | return pools; | |
2451 | } | |
2452 | ||
20effc67 TL |
2453 | OSD::PGRefOrError OSD::locate_asok_target(const cmdmap_t& cmdmap, |
2454 | stringstream& ss, | |
2455 | bool only_primary) | |
2456 | { | |
2457 | string pgidstr; | |
2458 | if (!cmd_getval(cmdmap, "pgid", pgidstr)) { | |
2459 | ss << "no pgid specified"; | |
2460 | return OSD::PGRefOrError{std::nullopt, -EINVAL}; | |
2461 | } | |
2462 | ||
2463 | pg_t pgid; | |
2464 | if (!pgid.parse(pgidstr.c_str())) { | |
2465 | ss << "couldn't parse pgid '" << pgidstr << "'"; | |
2466 | return OSD::PGRefOrError{std::nullopt, -EINVAL}; | |
2467 | } | |
2468 | ||
2469 | spg_t pcand; | |
2470 | PGRef pg; | |
2471 | if (get_osdmap()->get_primary_shard(pgid, &pcand) && (pg = _lookup_lock_pg(pcand))) { | |
2472 | if (pg->is_primary() || !only_primary) { | |
2473 | return OSD::PGRefOrError{pg, 0}; | |
2474 | } | |
2475 | ||
2476 | ss << "not primary for pgid " << pgid; | |
2477 | pg->unlock(); | |
2478 | return OSD::PGRefOrError{std::nullopt, -EAGAIN}; | |
2479 | } else { | |
2480 | ss << "i don't have pgid " << pgid; | |
2481 | return OSD::PGRefOrError{std::nullopt, -ENOENT}; | |
2482 | } | |
2483 | } | |
2484 | ||
2485 | // note that the cmdmap is explicitly copied into asok_route_to_pg() | |
2486 | int OSD::asok_route_to_pg( | |
2487 | bool only_primary, | |
2488 | std::string_view prefix, | |
2489 | cmdmap_t cmdmap, | |
2490 | Formatter* f, | |
2491 | stringstream& ss, | |
2492 | const bufferlist& inbl, | |
2493 | bufferlist& outbl, | |
2494 | std::function<void(int, const std::string&, bufferlist&)> on_finish) | |
2495 | { | |
2496 | auto [target_pg, ret] = locate_asok_target(cmdmap, ss, only_primary); | |
2497 | ||
2498 | if (!target_pg.has_value()) { | |
2499 | // 'ss' and 'ret' already contain the error information | |
2500 | on_finish(ret, ss.str(), outbl); | |
2501 | return ret; | |
2502 | } | |
2503 | ||
2504 | // the PG was locked by locate_asok_target() | |
2505 | try { | |
2506 | (*target_pg)->do_command(prefix, cmdmap, inbl, on_finish); | |
2507 | (*target_pg)->unlock(); | |
2508 | return 0; // the pg handler calls on_finish directly | |
2509 | } catch (const TOPNSPC::common::bad_cmd_get& e) { | |
2510 | (*target_pg)->unlock(); | |
2511 | ss << e.what(); | |
2512 | on_finish(ret, ss.str(), outbl); | |
2513 | return -EINVAL; | |
2514 | } | |
2515 | } | |
2516 | ||
9f95a23c TL |
2517 | void OSD::asok_command( |
2518 | std::string_view prefix, const cmdmap_t& cmdmap, | |
2519 | Formatter *f, | |
2520 | const bufferlist& inbl, | |
2521 | std::function<void(int,const std::string&,bufferlist&)> on_finish) | |
7c673cae | 2522 | { |
9f95a23c TL |
2523 | int ret = 0; |
2524 | stringstream ss; // stderr error message stream | |
2525 | bufferlist outbl; // if empty at end, we'll dump formatter as output | |
2526 | ||
2527 | // --- PG commands are routed here to PG::do_command --- | |
2528 | if (prefix == "pg" || | |
2529 | prefix == "query" || | |
2530 | prefix == "mark_unfound_lost" || | |
2531 | prefix == "list_unfound" || | |
2532 | prefix == "scrub" || | |
2533 | prefix == "deep_scrub" | |
2534 | ) { | |
2535 | string pgidstr; | |
2536 | pg_t pgid; | |
2537 | if (!cmd_getval(cmdmap, "pgid", pgidstr)) { | |
2538 | ss << "no pgid specified"; | |
2539 | ret = -EINVAL; | |
2540 | goto out; | |
2541 | } | |
2542 | if (!pgid.parse(pgidstr.c_str())) { | |
2543 | ss << "couldn't parse pgid '" << pgidstr << "'"; | |
2544 | ret = -EINVAL; | |
2545 | goto out; | |
2546 | } | |
2547 | spg_t pcand; | |
2548 | PGRef pg; | |
2549 | if (get_osdmap()->get_primary_shard(pgid, &pcand) && | |
2550 | (pg = _lookup_lock_pg(pcand))) { | |
2551 | if (pg->is_primary()) { | |
2552 | cmdmap_t new_cmdmap = cmdmap; | |
2553 | try { | |
2554 | pg->do_command(prefix, new_cmdmap, inbl, on_finish); | |
2555 | pg->unlock(); | |
2556 | return; // the pg handler calls on_finish directly | |
2557 | } catch (const TOPNSPC::common::bad_cmd_get& e) { | |
2558 | pg->unlock(); | |
2559 | ss << e.what(); | |
2560 | ret = -EINVAL; | |
2561 | goto out; | |
2562 | } | |
2563 | } else { | |
2564 | ss << "not primary for pgid " << pgid; | |
2565 | // do not reply; they will get newer maps and realize they | |
2566 | // need to resend. | |
2567 | pg->unlock(); | |
2568 | ret = -EAGAIN; | |
2569 | goto out; | |
2570 | } | |
2571 | } else { | |
2572 | ss << "i don't have pgid " << pgid; | |
2573 | ret = -ENOENT; | |
2574 | } | |
2575 | } | |
2576 | ||
20effc67 TL |
2577 | // --- PG commands that will be answered even if !primary --- |
2578 | ||
2579 | else if (prefix == "scrubdebug") { | |
2580 | asok_route_to_pg(false, prefix, cmdmap, f, ss, inbl, outbl, on_finish); | |
2581 | return; | |
2582 | } | |
2583 | ||
9f95a23c TL |
2584 | // --- OSD commands follow --- |
2585 | ||
2586 | else if (prefix == "status") { | |
2587 | lock_guard l(osd_lock); | |
7c673cae FG |
2588 | f->open_object_section("status"); |
2589 | f->dump_stream("cluster_fsid") << superblock.cluster_fsid; | |
2590 | f->dump_stream("osd_fsid") << superblock.osd_fsid; | |
2591 | f->dump_unsigned("whoami", superblock.whoami); | |
2592 | f->dump_string("state", get_state_name(get_state())); | |
2593 | f->dump_unsigned("oldest_map", superblock.oldest_map); | |
2594 | f->dump_unsigned("newest_map", superblock.newest_map); | |
11fdf7f2 | 2595 | f->dump_unsigned("num_pgs", num_pgs); |
7c673cae | 2596 | f->close_section(); |
9f95a23c | 2597 | } else if (prefix == "flush_journal") { |
7c673cae | 2598 | store->flush_journal(); |
9f95a23c TL |
2599 | } else if (prefix == "dump_ops_in_flight" || |
2600 | prefix == "ops" || | |
2601 | prefix == "dump_blocked_ops" || | |
2602 | prefix == "dump_historic_ops" || | |
2603 | prefix == "dump_historic_ops_by_duration" || | |
2604 | prefix == "dump_historic_slow_ops") { | |
c07f9fc5 FG |
2605 | |
2606 | const string error_str = "op_tracker tracking is not enabled now, so no ops are tracked currently, \ | |
2607 | even those get stuck. Please enable \"osd_enable_op_tracker\", and the tracker \ | |
2608 | will start to track new ops received afterwards."; | |
2609 | ||
2610 | set<string> filters; | |
2611 | vector<string> filter_str; | |
9f95a23c | 2612 | if (cmd_getval(cmdmap, "filterstr", filter_str)) { |
c07f9fc5 FG |
2613 | copy(filter_str.begin(), filter_str.end(), |
2614 | inserter(filters, filters.end())); | |
2615 | } | |
2616 | ||
9f95a23c TL |
2617 | if (prefix == "dump_ops_in_flight" || |
2618 | prefix == "ops") { | |
c07f9fc5 FG |
2619 | if (!op_tracker.dump_ops_in_flight(f, false, filters)) { |
2620 | ss << error_str; | |
9f95a23c TL |
2621 | ret = -EINVAL; |
2622 | goto out; | |
c07f9fc5 FG |
2623 | } |
2624 | } | |
9f95a23c | 2625 | if (prefix == "dump_blocked_ops") { |
c07f9fc5 FG |
2626 | if (!op_tracker.dump_ops_in_flight(f, true, filters)) { |
2627 | ss << error_str; | |
9f95a23c TL |
2628 | ret = -EINVAL; |
2629 | goto out; | |
c07f9fc5 FG |
2630 | } |
2631 | } | |
9f95a23c | 2632 | if (prefix == "dump_historic_ops") { |
c07f9fc5 FG |
2633 | if (!op_tracker.dump_historic_ops(f, false, filters)) { |
2634 | ss << error_str; | |
9f95a23c TL |
2635 | ret = -EINVAL; |
2636 | goto out; | |
c07f9fc5 FG |
2637 | } |
2638 | } | |
9f95a23c | 2639 | if (prefix == "dump_historic_ops_by_duration") { |
c07f9fc5 FG |
2640 | if (!op_tracker.dump_historic_ops(f, true, filters)) { |
2641 | ss << error_str; | |
9f95a23c TL |
2642 | ret = -EINVAL; |
2643 | goto out; | |
c07f9fc5 FG |
2644 | } |
2645 | } | |
9f95a23c | 2646 | if (prefix == "dump_historic_slow_ops") { |
c07f9fc5 FG |
2647 | if (!op_tracker.dump_historic_slow_ops(f, filters)) { |
2648 | ss << error_str; | |
9f95a23c TL |
2649 | ret = -EINVAL; |
2650 | goto out; | |
c07f9fc5 | 2651 | } |
7c673cae | 2652 | } |
9f95a23c | 2653 | } else if (prefix == "dump_op_pq_state") { |
7c673cae FG |
2654 | f->open_object_section("pq"); |
2655 | op_shardedwq.dump(f); | |
2656 | f->close_section(); | |
f67539c2 | 2657 | } else if (prefix == "dump_blocklist") { |
7c673cae FG |
2658 | list<pair<entity_addr_t,utime_t> > bl; |
2659 | OSDMapRef curmap = service.get_osdmap(); | |
2660 | ||
f67539c2 TL |
2661 | f->open_array_section("blocklist"); |
2662 | curmap->get_blocklist(&bl); | |
7c673cae FG |
2663 | for (list<pair<entity_addr_t,utime_t> >::iterator it = bl.begin(); |
2664 | it != bl.end(); ++it) { | |
224ce89b | 2665 | f->open_object_section("entry"); |
7c673cae FG |
2666 | f->open_object_section("entity_addr_t"); |
2667 | it->first.dump(f); | |
2668 | f->close_section(); //entity_addr_t | |
2669 | it->second.localtime(f->dump_stream("expire_time")); | |
2670 | f->close_section(); //entry | |
2671 | } | |
f67539c2 | 2672 | f->close_section(); //blocklist |
9f95a23c | 2673 | } else if (prefix == "dump_watchers") { |
7c673cae FG |
2674 | list<obj_watch_item_t> watchers; |
2675 | // scan pg's | |
11fdf7f2 TL |
2676 | vector<PGRef> pgs; |
2677 | _get_pgs(&pgs); | |
2678 | for (auto& pg : pgs) { | |
2679 | list<obj_watch_item_t> pg_watchers; | |
2680 | pg->get_watchers(&pg_watchers); | |
2681 | watchers.splice(watchers.end(), pg_watchers); | |
7c673cae FG |
2682 | } |
2683 | ||
2684 | f->open_array_section("watchers"); | |
2685 | for (list<obj_watch_item_t>::iterator it = watchers.begin(); | |
2686 | it != watchers.end(); ++it) { | |
2687 | ||
224ce89b | 2688 | f->open_object_section("watch"); |
7c673cae FG |
2689 | |
2690 | f->dump_string("namespace", it->obj.nspace); | |
2691 | f->dump_string("object", it->obj.oid.name); | |
2692 | ||
2693 | f->open_object_section("entity_name"); | |
2694 | it->wi.name.dump(f); | |
2695 | f->close_section(); //entity_name_t | |
2696 | ||
224ce89b WB |
2697 | f->dump_unsigned("cookie", it->wi.cookie); |
2698 | f->dump_unsigned("timeout", it->wi.timeout_seconds); | |
7c673cae FG |
2699 | |
2700 | f->open_object_section("entity_addr_t"); | |
2701 | it->wi.addr.dump(f); | |
2702 | f->close_section(); //entity_addr_t | |
2703 | ||
2704 | f->close_section(); //watch | |
2705 | } | |
2706 | ||
2707 | f->close_section(); //watchers | |
9f95a23c | 2708 | } else if (prefix == "dump_recovery_reservations") { |
7c673cae FG |
2709 | f->open_object_section("reservations"); |
2710 | f->open_object_section("local_reservations"); | |
2711 | service.local_reserver.dump(f); | |
2712 | f->close_section(); | |
2713 | f->open_object_section("remote_reservations"); | |
2714 | service.remote_reserver.dump(f); | |
2715 | f->close_section(); | |
2716 | f->close_section(); | |
9f95a23c | 2717 | } else if (prefix == "dump_scrub_reservations") { |
eafe8130 | 2718 | f->open_object_section("scrub_reservations"); |
20effc67 | 2719 | service.get_scrub_services().dump_scrub_reservations(f); |
eafe8130 | 2720 | f->close_section(); |
9f95a23c | 2721 | } else if (prefix == "get_latest_osdmap") { |
7c673cae | 2722 | get_latest_osdmap(); |
9f95a23c | 2723 | } else if (prefix == "set_heap_property") { |
7c673cae FG |
2724 | string property; |
2725 | int64_t value = 0; | |
2726 | string error; | |
2727 | bool success = false; | |
9f95a23c | 2728 | if (!cmd_getval(cmdmap, "property", property)) { |
7c673cae FG |
2729 | error = "unable to get property"; |
2730 | success = false; | |
9f95a23c | 2731 | } else if (!cmd_getval(cmdmap, "value", value)) { |
7c673cae FG |
2732 | error = "unable to get value"; |
2733 | success = false; | |
2734 | } else if (value < 0) { | |
2735 | error = "negative value not allowed"; | |
2736 | success = false; | |
2737 | } else if (!ceph_heap_set_numeric_property(property.c_str(), (size_t)value)) { | |
2738 | error = "invalid property"; | |
2739 | success = false; | |
2740 | } else { | |
2741 | success = true; | |
2742 | } | |
2743 | f->open_object_section("result"); | |
2744 | f->dump_string("error", error); | |
2745 | f->dump_bool("success", success); | |
2746 | f->close_section(); | |
9f95a23c | 2747 | } else if (prefix == "get_heap_property") { |
7c673cae FG |
2748 | string property; |
2749 | size_t value = 0; | |
2750 | string error; | |
2751 | bool success = false; | |
9f95a23c | 2752 | if (!cmd_getval(cmdmap, "property", property)) { |
7c673cae FG |
2753 | error = "unable to get property"; |
2754 | success = false; | |
2755 | } else if (!ceph_heap_get_numeric_property(property.c_str(), &value)) { | |
2756 | error = "invalid property"; | |
2757 | success = false; | |
2758 | } else { | |
2759 | success = true; | |
2760 | } | |
2761 | f->open_object_section("result"); | |
2762 | f->dump_string("error", error); | |
2763 | f->dump_bool("success", success); | |
2764 | f->dump_int("value", value); | |
2765 | f->close_section(); | |
9f95a23c | 2766 | } else if (prefix == "dump_objectstore_kv_stats") { |
7c673cae | 2767 | store->get_db_statistics(f); |
9f95a23c | 2768 | } else if (prefix == "dump_scrubs") { |
20effc67 | 2769 | service.get_scrub_services().dump_scrubs(f); |
9f95a23c | 2770 | } else if (prefix == "calc_objectstore_db_histogram") { |
7c673cae | 2771 | store->generate_db_histogram(f); |
9f95a23c | 2772 | } else if (prefix == "flush_store_cache") { |
11fdf7f2 | 2773 | store->flush_cache(&ss); |
9f95a23c | 2774 | } else if (prefix == "dump_pgstate_history") { |
7c673cae | 2775 | f->open_object_section("pgstate_history"); |
9f95a23c | 2776 | f->open_array_section("pgs"); |
11fdf7f2 TL |
2777 | vector<PGRef> pgs; |
2778 | _get_pgs(&pgs); | |
2779 | for (auto& pg : pgs) { | |
9f95a23c | 2780 | f->open_object_section("pg"); |
11fdf7f2 | 2781 | f->dump_stream("pg") << pg->pg_id; |
9f95a23c | 2782 | f->dump_string("currently", pg->get_current_state()); |
11fdf7f2 | 2783 | pg->dump_pgstate_history(f); |
9f95a23c | 2784 | f->close_section(); |
7c673cae FG |
2785 | } |
2786 | f->close_section(); | |
9f95a23c TL |
2787 | f->close_section(); |
2788 | } else if (prefix == "compact") { | |
224ce89b WB |
2789 | dout(1) << "triggering manual compaction" << dendl; |
2790 | auto start = ceph::coarse_mono_clock::now(); | |
2791 | store->compact(); | |
2792 | auto end = ceph::coarse_mono_clock::now(); | |
11fdf7f2 | 2793 | double duration = std::chrono::duration<double>(end-start).count(); |
f67539c2 | 2794 | dout(1) << "finished manual compaction in " |
11fdf7f2 | 2795 | << duration |
224ce89b WB |
2796 | << " seconds" << dendl; |
2797 | f->open_object_section("compact_result"); | |
11fdf7f2 TL |
2798 | f->dump_float("elapsed_time", duration); |
2799 | f->close_section(); | |
9f95a23c | 2800 | } else if (prefix == "get_mapped_pools") { |
11fdf7f2 TL |
2801 | f->open_array_section("mapped_pools"); |
2802 | set<int64_t> poollist = get_mapped_pools(); | |
2803 | for (auto pool : poollist) { | |
2804 | f->dump_int("pool_id", pool); | |
2805 | } | |
2806 | f->close_section(); | |
9f95a23c | 2807 | } else if (prefix == "smart") { |
11fdf7f2 | 2808 | string devid; |
9f95a23c TL |
2809 | cmd_getval(cmdmap, "devid", devid); |
2810 | ostringstream out; | |
2811 | probe_smart(devid, out); | |
2812 | outbl.append(out.str()); | |
2813 | } else if (prefix == "list_devices") { | |
11fdf7f2 TL |
2814 | set<string> devnames; |
2815 | store->get_devices(&devnames); | |
9f95a23c | 2816 | f->open_array_section("list_devices"); |
11fdf7f2 TL |
2817 | for (auto dev : devnames) { |
2818 | if (dev.find("dm-") == 0) { | |
2819 | continue; | |
2820 | } | |
9f95a23c TL |
2821 | string err; |
2822 | f->open_object_section("device"); | |
11fdf7f2 | 2823 | f->dump_string("device", "/dev/" + dev); |
9f95a23c TL |
2824 | f->dump_string("device_id", get_device_id(dev, &err)); |
2825 | f->close_section(); | |
11fdf7f2 | 2826 | } |
224ce89b | 2827 | f->close_section(); |
9f95a23c TL |
2828 | } else if (prefix == "send_beacon") { |
2829 | lock_guard l(osd_lock); | |
11fdf7f2 TL |
2830 | if (is_active()) { |
2831 | send_beacon(ceph::coarse_mono_clock::now()); | |
2832 | } | |
9f95a23c TL |
2833 | } |
2834 | ||
2835 | else if (prefix == "cluster_log") { | |
2836 | vector<string> msg; | |
2837 | cmd_getval(cmdmap, "message", msg); | |
2838 | if (msg.empty()) { | |
2839 | ret = -EINVAL; | |
2840 | ss << "ignoring empty log message"; | |
2841 | goto out; | |
2842 | } | |
2843 | string message = msg.front(); | |
2844 | for (vector<string>::iterator a = ++msg.begin(); a != msg.end(); ++a) | |
2845 | message += " " + *a; | |
2846 | string lvl; | |
2847 | cmd_getval(cmdmap, "level", lvl); | |
2848 | clog_type level = string_to_clog_type(lvl); | |
2849 | if (level < 0) { | |
2850 | ret = -EINVAL; | |
2851 | ss << "unknown level '" << lvl << "'"; | |
2852 | goto out; | |
2853 | } | |
2854 | clog->do_log(level, message); | |
2855 | } | |
2856 | ||
2857 | else if (prefix == "bench") { | |
9f95a23c | 2858 | // default count 1G, size 4MB |
20effc67 TL |
2859 | int64_t count = cmd_getval_or<int64_t>(cmdmap, "count", 1LL << 30); |
2860 | int64_t bsize = cmd_getval_or<int64_t>(cmdmap, "size", 4LL << 20); | |
2861 | int64_t osize = cmd_getval_or<int64_t>(cmdmap, "object_size", 0); | |
2862 | int64_t onum = cmd_getval_or<int64_t>(cmdmap, "object_num", 0); | |
a4b75251 | 2863 | double elapsed = 0.0; |
9f95a23c | 2864 | |
a4b75251 TL |
2865 | ret = run_osd_bench_test(count, bsize, osize, onum, &elapsed, ss); |
2866 | if (ret != 0) { | |
9f95a23c | 2867 | goto out; |
9f95a23c TL |
2868 | } |
2869 | ||
9f95a23c TL |
2870 | double rate = count / elapsed; |
2871 | double iops = rate / bsize; | |
2872 | f->open_object_section("osd_bench_results"); | |
2873 | f->dump_int("bytes_written", count); | |
2874 | f->dump_int("blocksize", bsize); | |
2875 | f->dump_float("elapsed_sec", elapsed); | |
2876 | f->dump_float("bytes_per_sec", rate); | |
2877 | f->dump_float("iops", iops); | |
2878 | f->close_section(); | |
2879 | } | |
2880 | ||
2881 | else if (prefix == "flush_pg_stats") { | |
2882 | mgrc.send_pgstats(); | |
2883 | f->dump_unsigned("stat_seq", service.get_osd_stat_seq()); | |
2884 | } | |
2885 | ||
2886 | else if (prefix == "heap") { | |
2887 | ret = ceph::osd_cmds::heap(*cct, cmdmap, *f, ss); | |
2888 | } | |
2889 | ||
2890 | else if (prefix == "debug dump_missing") { | |
2891 | f->open_array_section("pgs"); | |
2892 | vector<PGRef> pgs; | |
2893 | _get_pgs(&pgs); | |
2894 | for (auto& pg : pgs) { | |
2895 | string s = stringify(pg->pg_id); | |
2896 | f->open_array_section(s.c_str()); | |
2897 | pg->lock(); | |
2898 | pg->dump_missing(f); | |
2899 | pg->unlock(); | |
2900 | f->close_section(); | |
2901 | } | |
2902 | f->close_section(); | |
2903 | } | |
2904 | ||
2905 | else if (prefix == "debug kick_recovery_wq") { | |
2906 | int64_t delay; | |
2907 | cmd_getval(cmdmap, "delay", delay); | |
2908 | ostringstream oss; | |
2909 | oss << delay; | |
2910 | ret = cct->_conf.set_val("osd_recovery_delay_start", oss.str().c_str()); | |
2911 | if (ret != 0) { | |
2912 | ss << "kick_recovery_wq: error setting " | |
2913 | << "osd_recovery_delay_start to '" << delay << "': error " | |
2914 | << ret; | |
2915 | goto out; | |
2916 | } | |
2917 | cct->_conf.apply_changes(nullptr); | |
2918 | ss << "kicking recovery queue. set osd_recovery_delay_start " | |
2919 | << "to " << cct->_conf->osd_recovery_delay_start; | |
2920 | } | |
2921 | ||
2922 | else if (prefix == "cpu_profiler") { | |
2923 | ostringstream ds; | |
2924 | string arg; | |
2925 | cmd_getval(cmdmap, "arg", arg); | |
2926 | vector<string> argvec; | |
2927 | get_str_vec(arg, argvec); | |
2928 | cpu_profiler_handle_command(argvec, ds); | |
2929 | outbl.append(ds.str()); | |
2930 | } | |
2931 | ||
2932 | else if (prefix == "dump_pg_recovery_stats") { | |
2933 | lock_guard l(osd_lock); | |
2934 | pg_recovery_stats.dump_formatted(f); | |
2935 | } | |
2936 | ||
2937 | else if (prefix == "reset_pg_recovery_stats") { | |
2938 | lock_guard l(osd_lock); | |
2939 | pg_recovery_stats.reset(); | |
2940 | } | |
2941 | ||
2942 | else if (prefix == "perf histogram dump") { | |
2943 | std::string logger; | |
2944 | std::string counter; | |
2945 | cmd_getval(cmdmap, "logger", logger); | |
2946 | cmd_getval(cmdmap, "counter", counter); | |
2947 | cct->get_perfcounters_collection()->dump_formatted_histograms( | |
2948 | f, false, logger, counter); | |
2949 | } | |
2950 | ||
2951 | else if (prefix == "cache drop") { | |
2952 | lock_guard l(osd_lock); | |
2953 | dout(20) << "clearing all caches" << dendl; | |
2954 | // Clear the objectstore's cache - onode and buffer for Bluestore, | |
2955 | // system's pagecache for Filestore | |
2956 | ret = store->flush_cache(&ss); | |
2957 | if (ret < 0) { | |
2958 | ss << "Error flushing objectstore cache: " << cpp_strerror(ret); | |
2959 | goto out; | |
2960 | } | |
2961 | // Clear the objectcontext cache (per PG) | |
2962 | vector<PGRef> pgs; | |
2963 | _get_pgs(&pgs); | |
2964 | for (auto& pg: pgs) { | |
2965 | pg->clear_cache(); | |
2966 | } | |
2967 | } | |
2968 | ||
2969 | else if (prefix == "cache status") { | |
2970 | lock_guard l(osd_lock); | |
2971 | int obj_ctx_count = 0; | |
2972 | vector<PGRef> pgs; | |
2973 | _get_pgs(&pgs); | |
2974 | for (auto& pg: pgs) { | |
2975 | obj_ctx_count += pg->get_cache_obj_count(); | |
2976 | } | |
2977 | f->open_object_section("cache_status"); | |
2978 | f->dump_int("object_ctx", obj_ctx_count); | |
2979 | store->dump_cache_stats(f); | |
2980 | f->close_section(); | |
2981 | } | |
2982 | ||
2983 | else if (prefix == "scrub_purged_snaps") { | |
2984 | lock_guard l(osd_lock); | |
2985 | scrub_purged_snaps(); | |
2986 | } | |
2987 | ||
2988 | else if (prefix == "dump_osd_network") { | |
2989 | lock_guard l(osd_lock); | |
2990 | int64_t value = 0; | |
2991 | if (!(cmd_getval(cmdmap, "value", value))) { | |
2992 | // Convert milliseconds to microseconds | |
2993 | value = static_cast<double>(g_conf().get_val<double>( | |
2994 | "mon_warn_on_slow_ping_time")) * 1000; | |
2995 | if (value == 0) { | |
2996 | double ratio = g_conf().get_val<double>("mon_warn_on_slow_ping_ratio"); | |
2997 | value = g_conf().get_val<int64_t>("osd_heartbeat_grace"); | |
2998 | value *= 1000000 * ratio; // Seconds of grace to microseconds at ratio | |
2999 | } | |
3000 | } else { | |
3001 | // Convert user input to microseconds | |
3002 | value *= 1000; | |
3003 | } | |
3004 | if (value < 0) value = 0; | |
3005 | ||
3006 | struct osd_ping_time_t { | |
3007 | uint32_t pingtime; | |
3008 | int to; | |
3009 | bool back; | |
3010 | std::array<uint32_t,3> times; | |
3011 | std::array<uint32_t,3> min; | |
3012 | std::array<uint32_t,3> max; | |
3013 | uint32_t last; | |
3014 | uint32_t last_update; | |
3015 | ||
3016 | bool operator<(const osd_ping_time_t& rhs) const { | |
3017 | if (pingtime < rhs.pingtime) | |
3018 | return true; | |
3019 | if (pingtime > rhs.pingtime) | |
3020 | return false; | |
3021 | if (to < rhs.to) | |
3022 | return true; | |
3023 | if (to > rhs.to) | |
3024 | return false; | |
3025 | return back; | |
3026 | } | |
3027 | }; | |
3028 | ||
3029 | set<osd_ping_time_t> sorted; | |
3030 | // Get pingtimes under lock and not on the stack | |
eafe8130 TL |
3031 | map<int, osd_stat_t::Interfaces> *pingtimes = new map<int, osd_stat_t::Interfaces>; |
3032 | service.get_hb_pingtime(pingtimes); | |
3033 | for (auto j : *pingtimes) { | |
3034 | if (j.second.last_update == 0) | |
3035 | continue; | |
3036 | osd_ping_time_t item; | |
3037 | item.pingtime = std::max(j.second.back_pingtime[0], j.second.back_pingtime[1]); | |
3038 | item.pingtime = std::max(item.pingtime, j.second.back_pingtime[2]); | |
3039 | if (item.pingtime >= value) { | |
3040 | item.to = j.first; | |
3041 | item.times[0] = j.second.back_pingtime[0]; | |
3042 | item.times[1] = j.second.back_pingtime[1]; | |
3043 | item.times[2] = j.second.back_pingtime[2]; | |
3044 | item.min[0] = j.second.back_min[0]; | |
3045 | item.min[1] = j.second.back_min[1]; | |
3046 | item.min[2] = j.second.back_min[2]; | |
3047 | item.max[0] = j.second.back_max[0]; | |
3048 | item.max[1] = j.second.back_max[1]; | |
3049 | item.max[2] = j.second.back_max[2]; | |
3050 | item.last = j.second.back_last; | |
3051 | item.back = true; | |
3052 | item.last_update = j.second.last_update; | |
3053 | sorted.emplace(item); | |
3054 | } | |
3055 | if (j.second.front_last == 0) | |
3056 | continue; | |
3057 | item.pingtime = std::max(j.second.front_pingtime[0], j.second.front_pingtime[1]); | |
3058 | item.pingtime = std::max(item.pingtime, j.second.front_pingtime[2]); | |
3059 | if (item.pingtime >= value) { | |
3060 | item.to = j.first; | |
3061 | item.times[0] = j.second.front_pingtime[0]; | |
3062 | item.times[1] = j.second.front_pingtime[1]; | |
3063 | item.times[2] = j.second.front_pingtime[2]; | |
3064 | item.min[0] = j.second.front_min[0]; | |
3065 | item.min[1] = j.second.front_min[1]; | |
3066 | item.min[2] = j.second.front_min[2]; | |
3067 | item.max[0] = j.second.front_max[0]; | |
3068 | item.max[1] = j.second.front_max[1]; | |
3069 | item.max[2] = j.second.front_max[2]; | |
3070 | item.last = j.second.front_last; | |
3071 | item.last_update = j.second.last_update; | |
3072 | item.back = false; | |
3073 | sorted.emplace(item); | |
3074 | } | |
3075 | } | |
3076 | delete pingtimes; | |
3077 | // | |
3078 | // Network ping times (1min 5min 15min) | |
3079 | f->open_object_section("network_ping_times"); | |
3080 | f->dump_int("threshold", value / 1000); | |
3081 | f->open_array_section("entries"); | |
3082 | for (auto &sitem : boost::adaptors::reverse(sorted)) { | |
3083 | ceph_assert(sitem.pingtime >= value); | |
3084 | f->open_object_section("entry"); | |
3085 | ||
3086 | const time_t lu(sitem.last_update); | |
3087 | char buffer[26]; | |
3088 | string lustr(ctime_r(&lu, buffer)); | |
3089 | lustr.pop_back(); // Remove trailing \n | |
3090 | auto stale = cct->_conf.get_val<int64_t>("osd_heartbeat_stale"); | |
3091 | f->dump_string("last update", lustr); | |
3092 | f->dump_bool("stale", ceph_clock_now().sec() - sitem.last_update > stale); | |
3093 | f->dump_int("from osd", whoami); | |
3094 | f->dump_int("to osd", sitem.to); | |
3095 | f->dump_string("interface", (sitem.back ? "back" : "front")); | |
3096 | f->open_object_section("average"); | |
3097 | f->dump_format_unquoted("1min", "%s", fixed_u_to_string(sitem.times[0],3).c_str()); | |
3098 | f->dump_format_unquoted("5min", "%s", fixed_u_to_string(sitem.times[1],3).c_str()); | |
3099 | f->dump_format_unquoted("15min", "%s", fixed_u_to_string(sitem.times[2],3).c_str()); | |
3100 | f->close_section(); // average | |
3101 | f->open_object_section("min"); | |
3102 | f->dump_format_unquoted("1min", "%s", fixed_u_to_string(sitem.max[0],3).c_str()); | |
3103 | f->dump_format_unquoted("5min", "%s", fixed_u_to_string(sitem.max[1],3).c_str()); | |
3104 | f->dump_format_unquoted("15min", "%s", fixed_u_to_string(sitem.max[2],3).c_str()); | |
3105 | f->close_section(); // min | |
3106 | f->open_object_section("max"); | |
3107 | f->dump_format_unquoted("1min", "%s", fixed_u_to_string(sitem.max[0],3).c_str()); | |
3108 | f->dump_format_unquoted("5min", "%s", fixed_u_to_string(sitem.max[1],3).c_str()); | |
3109 | f->dump_format_unquoted("15min", "%s", fixed_u_to_string(sitem.max[2],3).c_str()); | |
3110 | f->close_section(); // max | |
3111 | f->dump_format_unquoted("last", "%s", fixed_u_to_string(sitem.last,3).c_str()); | |
3112 | f->close_section(); // entry | |
3113 | } | |
3114 | f->close_section(); // entries | |
3115 | f->close_section(); // network_ping_times | |
20effc67 TL |
3116 | } else if (prefix == "dump_pool_statfs") { |
3117 | lock_guard l(osd_lock); | |
3118 | ||
3119 | int64_t p = 0; | |
3120 | if (!(cmd_getval(cmdmap, "poolid", p))) { | |
3121 | ss << "Error dumping pool statfs: no poolid provided"; | |
3122 | ret = -EINVAL; | |
3123 | goto out; | |
3124 | } | |
3125 | ||
3126 | store_statfs_t st; | |
3127 | bool per_pool_omap_stats = false; | |
3128 | ||
3129 | ret = store->pool_statfs(p, &st, &per_pool_omap_stats); | |
3130 | if (ret < 0) { | |
3131 | ss << "Error dumping pool statfs: " << cpp_strerror(ret); | |
3132 | goto out; | |
3133 | } else { | |
3134 | ss << "dumping pool statfs..."; | |
3135 | f->open_object_section("pool_statfs"); | |
3136 | f->dump_int("poolid", p); | |
3137 | st.dump(f); | |
3138 | f->close_section(); | |
3139 | } | |
7c673cae | 3140 | } else { |
11fdf7f2 | 3141 | ceph_abort_msg("broken asok registration"); |
7c673cae | 3142 | } |
9f95a23c TL |
3143 | |
3144 | out: | |
3145 | on_finish(ret, ss.str(), outbl); | |
7c673cae FG |
3146 | } |
3147 | ||
a4b75251 TL |
3148 | int OSD::run_osd_bench_test( |
3149 | int64_t count, | |
3150 | int64_t bsize, | |
3151 | int64_t osize, | |
3152 | int64_t onum, | |
3153 | double *elapsed, | |
3154 | ostream &ss) | |
3155 | { | |
3156 | int ret = 0; | |
3157 | uint32_t duration = cct->_conf->osd_bench_duration; | |
3158 | ||
3159 | if (bsize > (int64_t) cct->_conf->osd_bench_max_block_size) { | |
3160 | // let us limit the block size because the next checks rely on it | |
3161 | // having a sane value. If we allow any block size to be set things | |
3162 | // can still go sideways. | |
3163 | ss << "block 'size' values are capped at " | |
3164 | << byte_u_t(cct->_conf->osd_bench_max_block_size) << ". If you wish to use" | |
3165 | << " a higher value, please adjust 'osd_bench_max_block_size'"; | |
3166 | ret = -EINVAL; | |
3167 | return ret; | |
3168 | } else if (bsize < (int64_t) (1 << 20)) { | |
3169 | // entering the realm of small block sizes. | |
3170 | // limit the count to a sane value, assuming a configurable amount of | |
3171 | // IOPS and duration, so that the OSD doesn't get hung up on this, | |
3172 | // preventing timeouts from going off | |
3173 | int64_t max_count = | |
3174 | bsize * duration * cct->_conf->osd_bench_small_size_max_iops; | |
3175 | if (count > max_count) { | |
3176 | ss << "'count' values greater than " << max_count | |
3177 | << " for a block size of " << byte_u_t(bsize) << ", assuming " | |
3178 | << cct->_conf->osd_bench_small_size_max_iops << " IOPS," | |
3179 | << " for " << duration << " seconds," | |
3180 | << " can cause ill effects on osd. " | |
3181 | << " Please adjust 'osd_bench_small_size_max_iops' with a higher" | |
3182 | << " value if you wish to use a higher 'count'."; | |
3183 | ret = -EINVAL; | |
3184 | return ret; | |
3185 | } | |
3186 | } else { | |
3187 | // 1MB block sizes are big enough so that we get more stuff done. | |
3188 | // However, to avoid the osd from getting hung on this and having | |
3189 | // timers being triggered, we are going to limit the count assuming | |
3190 | // a configurable throughput and duration. | |
3191 | // NOTE: max_count is the total amount of bytes that we believe we | |
3192 | // will be able to write during 'duration' for the given | |
3193 | // throughput. The block size hardly impacts this unless it's | |
3194 | // way too big. Given we already check how big the block size | |
3195 | // is, it's safe to assume everything will check out. | |
3196 | int64_t max_count = | |
3197 | cct->_conf->osd_bench_large_size_max_throughput * duration; | |
3198 | if (count > max_count) { | |
3199 | ss << "'count' values greater than " << max_count | |
3200 | << " for a block size of " << byte_u_t(bsize) << ", assuming " | |
3201 | << byte_u_t(cct->_conf->osd_bench_large_size_max_throughput) << "/s," | |
3202 | << " for " << duration << " seconds," | |
3203 | << " can cause ill effects on osd. " | |
3204 | << " Please adjust 'osd_bench_large_size_max_throughput'" | |
3205 | << " with a higher value if you wish to use a higher 'count'."; | |
3206 | ret = -EINVAL; | |
3207 | return ret; | |
3208 | } | |
3209 | } | |
3210 | ||
3211 | if (osize && bsize > osize) { | |
3212 | bsize = osize; | |
3213 | } | |
3214 | ||
3215 | dout(1) << " bench count " << count | |
3216 | << " bsize " << byte_u_t(bsize) << dendl; | |
3217 | ||
3218 | ObjectStore::Transaction cleanupt; | |
3219 | ||
3220 | if (osize && onum) { | |
3221 | bufferlist bl; | |
3222 | bufferptr bp(osize); | |
20effc67 | 3223 | memset(bp.c_str(), 'a', bp.length()); |
a4b75251 TL |
3224 | bl.push_back(std::move(bp)); |
3225 | bl.rebuild_page_aligned(); | |
3226 | for (int i=0; i<onum; ++i) { | |
3227 | char nm[30]; | |
3228 | snprintf(nm, sizeof(nm), "disk_bw_test_%d", i); | |
3229 | object_t oid(nm); | |
3230 | hobject_t soid(sobject_t(oid, 0)); | |
3231 | ObjectStore::Transaction t; | |
3232 | t.write(coll_t(), ghobject_t(soid), 0, osize, bl); | |
3233 | store->queue_transaction(service.meta_ch, std::move(t), nullptr); | |
3234 | cleanupt.remove(coll_t(), ghobject_t(soid)); | |
3235 | } | |
3236 | } | |
3237 | ||
3238 | bufferlist bl; | |
3239 | bufferptr bp(bsize); | |
20effc67 | 3240 | memset(bp.c_str(), 'a', bp.length()); |
a4b75251 TL |
3241 | bl.push_back(std::move(bp)); |
3242 | bl.rebuild_page_aligned(); | |
3243 | ||
3244 | { | |
3245 | C_SaferCond waiter; | |
3246 | if (!service.meta_ch->flush_commit(&waiter)) { | |
3247 | waiter.wait(); | |
3248 | } | |
3249 | } | |
3250 | ||
3251 | utime_t start = ceph_clock_now(); | |
3252 | for (int64_t pos = 0; pos < count; pos += bsize) { | |
3253 | char nm[30]; | |
3254 | unsigned offset = 0; | |
3255 | if (onum && osize) { | |
3256 | snprintf(nm, sizeof(nm), "disk_bw_test_%d", (int)(rand() % onum)); | |
3257 | offset = rand() % (osize / bsize) * bsize; | |
3258 | } else { | |
3259 | snprintf(nm, sizeof(nm), "disk_bw_test_%lld", (long long)pos); | |
3260 | } | |
3261 | object_t oid(nm); | |
3262 | hobject_t soid(sobject_t(oid, 0)); | |
3263 | ObjectStore::Transaction t; | |
3264 | t.write(coll_t::meta(), ghobject_t(soid), offset, bsize, bl); | |
3265 | store->queue_transaction(service.meta_ch, std::move(t), nullptr); | |
3266 | if (!onum || !osize) { | |
3267 | cleanupt.remove(coll_t::meta(), ghobject_t(soid)); | |
3268 | } | |
3269 | } | |
3270 | ||
3271 | { | |
3272 | C_SaferCond waiter; | |
3273 | if (!service.meta_ch->flush_commit(&waiter)) { | |
3274 | waiter.wait(); | |
3275 | } | |
3276 | } | |
3277 | utime_t end = ceph_clock_now(); | |
3278 | *elapsed = end - start; | |
3279 | ||
3280 | // clean up | |
3281 | store->queue_transaction(service.meta_ch, std::move(cleanupt), nullptr); | |
3282 | { | |
3283 | C_SaferCond waiter; | |
3284 | if (!service.meta_ch->flush_commit(&waiter)) { | |
3285 | waiter.wait(); | |
3286 | } | |
3287 | } | |
3288 | ||
3289 | return ret; | |
3290 | } | |
3291 | ||
7c673cae FG |
3292 | class TestOpsSocketHook : public AdminSocketHook { |
3293 | OSDService *service; | |
3294 | ObjectStore *store; | |
3295 | public: | |
3296 | TestOpsSocketHook(OSDService *s, ObjectStore *st) : service(s), store(st) {} | |
9f95a23c TL |
3297 | int call(std::string_view command, const cmdmap_t& cmdmap, |
3298 | Formatter *f, | |
3299 | std::ostream& errss, | |
3300 | bufferlist& out) override { | |
3301 | int r = 0; | |
3302 | stringstream outss; | |
11fdf7f2 | 3303 | try { |
9f95a23c TL |
3304 | test_ops(service, store, command, cmdmap, outss); |
3305 | out.append(outss); | |
3306 | } catch (const TOPNSPC::common::bad_cmd_get& e) { | |
3307 | errss << e.what(); | |
3308 | r = -EINVAL; | |
11fdf7f2 | 3309 | } |
9f95a23c | 3310 | return r; |
7c673cae FG |
3311 | } |
3312 | void test_ops(OSDService *service, ObjectStore *store, | |
11fdf7f2 | 3313 | std::string_view command, const cmdmap_t& cmdmap, ostream &ss); |
7c673cae FG |
3314 | |
3315 | }; | |
3316 | ||
3317 | class OSD::C_Tick : public Context { | |
3318 | OSD *osd; | |
3319 | public: | |
3320 | explicit C_Tick(OSD *o) : osd(o) {} | |
3321 | void finish(int r) override { | |
3322 | osd->tick(); | |
3323 | } | |
3324 | }; | |
3325 | ||
3326 | class OSD::C_Tick_WithoutOSDLock : public Context { | |
3327 | OSD *osd; | |
3328 | public: | |
3329 | explicit C_Tick_WithoutOSDLock(OSD *o) : osd(o) {} | |
3330 | void finish(int r) override { | |
3331 | osd->tick_without_osd_lock(); | |
3332 | } | |
3333 | }; | |
3334 | ||
3335 | int OSD::enable_disable_fuse(bool stop) | |
3336 | { | |
3337 | #ifdef HAVE_LIBFUSE | |
3338 | int r; | |
3339 | string mntpath = cct->_conf->osd_data + "/fuse"; | |
3340 | if (fuse_store && (stop || !cct->_conf->osd_objectstore_fuse)) { | |
3341 | dout(1) << __func__ << " disabling" << dendl; | |
3342 | fuse_store->stop(); | |
3343 | delete fuse_store; | |
3344 | fuse_store = NULL; | |
3345 | r = ::rmdir(mntpath.c_str()); | |
7c673cae | 3346 | if (r < 0) { |
c07f9fc5 FG |
3347 | r = -errno; |
3348 | derr << __func__ << " failed to rmdir " << mntpath << ": " | |
3349 | << cpp_strerror(r) << dendl; | |
7c673cae FG |
3350 | return r; |
3351 | } | |
3352 | return 0; | |
3353 | } | |
3354 | if (!fuse_store && cct->_conf->osd_objectstore_fuse) { | |
3355 | dout(1) << __func__ << " enabling" << dendl; | |
3356 | r = ::mkdir(mntpath.c_str(), 0700); | |
3357 | if (r < 0) | |
3358 | r = -errno; | |
3359 | if (r < 0 && r != -EEXIST) { | |
3360 | derr << __func__ << " unable to create " << mntpath << ": " | |
3361 | << cpp_strerror(r) << dendl; | |
3362 | return r; | |
3363 | } | |
20effc67 | 3364 | fuse_store = new FuseStore(store.get(), mntpath); |
7c673cae FG |
3365 | r = fuse_store->start(); |
3366 | if (r < 0) { | |
3367 | derr << __func__ << " unable to start fuse: " << cpp_strerror(r) << dendl; | |
3368 | delete fuse_store; | |
3369 | fuse_store = NULL; | |
3370 | return r; | |
3371 | } | |
3372 | } | |
3373 | #endif // HAVE_LIBFUSE | |
3374 | return 0; | |
3375 | } | |
3376 | ||
9f95a23c TL |
3377 | size_t OSD::get_num_cache_shards() |
3378 | { | |
3379 | return cct->_conf.get_val<Option::size_t>("osd_num_cache_shards"); | |
3380 | } | |
3381 | ||
31f18b77 FG |
3382 | int OSD::get_num_op_shards() |
3383 | { | |
3384 | if (cct->_conf->osd_op_num_shards) | |
3385 | return cct->_conf->osd_op_num_shards; | |
3386 | if (store_is_rotational) | |
3387 | return cct->_conf->osd_op_num_shards_hdd; | |
3388 | else | |
3389 | return cct->_conf->osd_op_num_shards_ssd; | |
3390 | } | |
3391 | ||
3392 | int OSD::get_num_op_threads() | |
3393 | { | |
3394 | if (cct->_conf->osd_op_num_threads_per_shard) | |
3395 | return get_num_op_shards() * cct->_conf->osd_op_num_threads_per_shard; | |
3396 | if (store_is_rotational) | |
3397 | return get_num_op_shards() * cct->_conf->osd_op_num_threads_per_shard_hdd; | |
3398 | else | |
3399 | return get_num_op_shards() * cct->_conf->osd_op_num_threads_per_shard_ssd; | |
3400 | } | |
3401 | ||
c07f9fc5 FG |
3402 | float OSD::get_osd_recovery_sleep() |
3403 | { | |
3404 | if (cct->_conf->osd_recovery_sleep) | |
3405 | return cct->_conf->osd_recovery_sleep; | |
d2e6a577 | 3406 | if (!store_is_rotational && !journal_is_rotational) |
c07f9fc5 | 3407 | return cct->_conf->osd_recovery_sleep_ssd; |
d2e6a577 | 3408 | else if (store_is_rotational && !journal_is_rotational) |
11fdf7f2 | 3409 | return cct->_conf.get_val<double>("osd_recovery_sleep_hybrid"); |
d2e6a577 FG |
3410 | else |
3411 | return cct->_conf->osd_recovery_sleep_hdd; | |
c07f9fc5 FG |
3412 | } |
3413 | ||
11fdf7f2 TL |
3414 | float OSD::get_osd_delete_sleep() |
3415 | { | |
3416 | float osd_delete_sleep = cct->_conf.get_val<double>("osd_delete_sleep"); | |
3417 | if (osd_delete_sleep > 0) | |
3418 | return osd_delete_sleep; | |
3419 | if (!store_is_rotational && !journal_is_rotational) | |
3420 | return cct->_conf.get_val<double>("osd_delete_sleep_ssd"); | |
3421 | if (store_is_rotational && !journal_is_rotational) | |
3422 | return cct->_conf.get_val<double>("osd_delete_sleep_hybrid"); | |
3423 | return cct->_conf.get_val<double>("osd_delete_sleep_hdd"); | |
3424 | } | |
3425 | ||
9f95a23c TL |
3426 | int OSD::get_recovery_max_active() |
3427 | { | |
3428 | if (cct->_conf->osd_recovery_max_active) | |
3429 | return cct->_conf->osd_recovery_max_active; | |
3430 | if (store_is_rotational) | |
3431 | return cct->_conf->osd_recovery_max_active_hdd; | |
3432 | else | |
3433 | return cct->_conf->osd_recovery_max_active_ssd; | |
3434 | } | |
3435 | ||
494da23a TL |
3436 | float OSD::get_osd_snap_trim_sleep() |
3437 | { | |
3438 | float osd_snap_trim_sleep = cct->_conf.get_val<double>("osd_snap_trim_sleep"); | |
3439 | if (osd_snap_trim_sleep > 0) | |
3440 | return osd_snap_trim_sleep; | |
3441 | if (!store_is_rotational && !journal_is_rotational) | |
3442 | return cct->_conf.get_val<double>("osd_snap_trim_sleep_ssd"); | |
3443 | if (store_is_rotational && !journal_is_rotational) | |
3444 | return cct->_conf.get_val<double>("osd_snap_trim_sleep_hybrid"); | |
3445 | return cct->_conf.get_val<double>("osd_snap_trim_sleep_hdd"); | |
3446 | } | |
3447 | ||
7c673cae FG |
3448 | int OSD::init() |
3449 | { | |
9f95a23c | 3450 | OSDMapRef osdmap; |
7c673cae | 3451 | CompatSet initial, diff; |
11fdf7f2 | 3452 | std::lock_guard lock(osd_lock); |
7c673cae FG |
3453 | if (is_stopping()) |
3454 | return 0; | |
20effc67 | 3455 | tracing::osd::tracer.init("osd"); |
7c673cae FG |
3456 | tick_timer.init(); |
3457 | tick_timer_without_osd_lock.init(); | |
3458 | service.recovery_request_timer.init(); | |
11fdf7f2 TL |
3459 | service.sleep_timer.init(); |
3460 | ||
3461 | boot_finisher.start(); | |
3462 | ||
3463 | { | |
3464 | string val; | |
3465 | store->read_meta("require_osd_release", &val); | |
9f95a23c | 3466 | last_require_osd_release = ceph_release_from_name(val); |
11fdf7f2 | 3467 | } |
7c673cae FG |
3468 | |
3469 | // mount. | |
31f18b77 FG |
3470 | dout(2) << "init " << dev_path |
3471 | << " (looks like " << (store_is_rotational ? "hdd" : "ssd") << ")" | |
3472 | << dendl; | |
d2e6a577 | 3473 | dout(2) << "journal " << journal_path << dendl; |
11fdf7f2 | 3474 | ceph_assert(store); // call pre_init() first! |
7c673cae | 3475 | |
9f95a23c | 3476 | store->set_cache_shards(get_num_cache_shards()); |
7c673cae | 3477 | |
20effc67 TL |
3478 | int rotating_auth_attempts = 0; |
3479 | auto rotating_auth_timeout = | |
3480 | g_conf().get_val<int64_t>("rotating_keys_bootstrap_timeout"); | |
3481 | ||
7c673cae FG |
3482 | int r = store->mount(); |
3483 | if (r < 0) { | |
3484 | derr << "OSD:init: unable to mount object store" << dendl; | |
3485 | return r; | |
3486 | } | |
d2e6a577 FG |
3487 | journal_is_rotational = store->is_journal_rotational(); |
3488 | dout(2) << "journal looks like " << (journal_is_rotational ? "hdd" : "ssd") | |
3489 | << dendl; | |
7c673cae FG |
3490 | |
3491 | enable_disable_fuse(false); | |
3492 | ||
3493 | dout(2) << "boot" << dendl; | |
3494 | ||
11fdf7f2 | 3495 | service.meta_ch = store->open_collection(coll_t::meta()); |
20effc67 TL |
3496 | if (!service.meta_ch) { |
3497 | derr << "OSD:init: unable to open meta collection" | |
3498 | << dendl; | |
3499 | r = -ENOENT; | |
3500 | goto out; | |
3501 | } | |
7c673cae FG |
3502 | // initialize the daily loadavg with current 15min loadavg |
3503 | double loadavgs[3]; | |
3504 | if (getloadavg(loadavgs, 3) == 3) { | |
3505 | daily_loadavg = loadavgs[2]; | |
3506 | } else { | |
3507 | derr << "OSD::init() : couldn't read loadavgs\n" << dendl; | |
3508 | daily_loadavg = 1.0; | |
3509 | } | |
3510 | ||
7c673cae FG |
3511 | // sanity check long object name handling |
3512 | { | |
3513 | hobject_t l; | |
3514 | l.oid.name = string(cct->_conf->osd_max_object_name_len, 'n'); | |
3515 | l.set_key(string(cct->_conf->osd_max_object_name_len, 'k')); | |
3516 | l.nspace = string(cct->_conf->osd_max_object_namespace_len, 's'); | |
3517 | r = store->validate_hobject_key(l); | |
3518 | if (r < 0) { | |
3519 | derr << "backend (" << store->get_type() << ") is unable to support max " | |
3520 | << "object name[space] len" << dendl; | |
3521 | derr << " osd max object name len = " | |
3522 | << cct->_conf->osd_max_object_name_len << dendl; | |
3523 | derr << " osd max object namespace len = " | |
3524 | << cct->_conf->osd_max_object_namespace_len << dendl; | |
3525 | derr << cpp_strerror(r) << dendl; | |
3526 | if (cct->_conf->osd_check_max_object_name_len_on_startup) { | |
3527 | goto out; | |
3528 | } | |
3529 | derr << "osd_check_max_object_name_len_on_startup = false, starting anyway" | |
3530 | << dendl; | |
3531 | } else { | |
3532 | dout(20) << "configured osd_max_object_name[space]_len looks ok" << dendl; | |
3533 | } | |
3534 | } | |
3535 | ||
3536 | // read superblock | |
3537 | r = read_superblock(); | |
3538 | if (r < 0) { | |
3539 | derr << "OSD::init() : unable to read osd superblock" << dendl; | |
3540 | r = -EINVAL; | |
3541 | goto out; | |
3542 | } | |
3543 | ||
3544 | if (osd_compat.compare(superblock.compat_features) < 0) { | |
3545 | derr << "The disk uses features unsupported by the executable." << dendl; | |
3546 | derr << " ondisk features " << superblock.compat_features << dendl; | |
3547 | derr << " daemon features " << osd_compat << dendl; | |
3548 | ||
3549 | if (osd_compat.writeable(superblock.compat_features)) { | |
3550 | CompatSet diff = osd_compat.unsupported(superblock.compat_features); | |
3551 | derr << "it is still writeable, though. Missing features: " << diff << dendl; | |
3552 | r = -EOPNOTSUPP; | |
3553 | goto out; | |
3554 | } | |
3555 | else { | |
3556 | CompatSet diff = osd_compat.unsupported(superblock.compat_features); | |
3557 | derr << "Cannot write to disk! Missing features: " << diff << dendl; | |
3558 | r = -EOPNOTSUPP; | |
3559 | goto out; | |
3560 | } | |
3561 | } | |
3562 | ||
3563 | assert_warn(whoami == superblock.whoami); | |
3564 | if (whoami != superblock.whoami) { | |
3565 | derr << "OSD::init: superblock says osd" | |
3566 | << superblock.whoami << " but I am osd." << whoami << dendl; | |
3567 | r = -EINVAL; | |
3568 | goto out; | |
3569 | } | |
3570 | ||
9f95a23c TL |
3571 | startup_time = ceph::mono_clock::now(); |
3572 | ||
11fdf7f2 | 3573 | // load up "current" osdmap |
9f95a23c TL |
3574 | assert_warn(!get_osdmap()); |
3575 | if (get_osdmap()) { | |
11fdf7f2 TL |
3576 | derr << "OSD::init: unable to read current osdmap" << dendl; |
3577 | r = -EINVAL; | |
3578 | goto out; | |
3579 | } | |
3580 | osdmap = get_map(superblock.current_epoch); | |
9f95a23c | 3581 | set_osdmap(osdmap); |
11fdf7f2 TL |
3582 | |
3583 | // make sure we don't have legacy pgs deleting | |
3584 | { | |
3585 | vector<coll_t> ls; | |
3586 | int r = store->list_collections(ls); | |
3587 | ceph_assert(r >= 0); | |
3588 | for (auto c : ls) { | |
3589 | spg_t pgid; | |
3590 | if (c.is_pg(&pgid) && | |
3591 | !osdmap->have_pg_pool(pgid.pool())) { | |
3592 | ghobject_t oid = make_final_pool_info_oid(pgid.pool()); | |
3593 | if (!store->exists(service.meta_ch, oid)) { | |
3594 | derr << __func__ << " missing pg_pool_t for deleted pool " | |
3595 | << pgid.pool() << " for pg " << pgid | |
3596 | << "; please downgrade to luminous and allow " | |
3597 | << "pg deletion to complete before upgrading" << dendl; | |
3598 | ceph_abort(); | |
3599 | } | |
3600 | } | |
3601 | } | |
3602 | } | |
3603 | ||
7c673cae FG |
3604 | initial = get_osd_initial_compat_set(); |
3605 | diff = superblock.compat_features.unsupported(initial); | |
3606 | if (superblock.compat_features.merge(initial)) { | |
9f95a23c TL |
3607 | // Are we adding SNAPMAPPER2? |
3608 | if (diff.incompat.contains(CEPH_OSD_FEATURE_INCOMPAT_SNAPMAPPER2)) { | |
3609 | dout(1) << __func__ << " upgrade snap_mapper (first start as octopus)" | |
3610 | << dendl; | |
3611 | auto ch = service.meta_ch; | |
3612 | auto hoid = make_snapmapper_oid(); | |
3613 | unsigned max = cct->_conf->osd_target_transaction_size; | |
20effc67 | 3614 | r = SnapMapper::convert_legacy(cct, store.get(), ch, hoid, max); |
9f95a23c TL |
3615 | if (r < 0) |
3616 | goto out; | |
3617 | } | |
7c673cae FG |
3618 | // We need to persist the new compat_set before we |
3619 | // do anything else | |
3620 | dout(5) << "Upgrading superblock adding: " << diff << dendl; | |
3621 | ObjectStore::Transaction t; | |
3622 | write_superblock(t); | |
11fdf7f2 | 3623 | r = store->queue_transaction(service.meta_ch, std::move(t)); |
7c673cae FG |
3624 | if (r < 0) |
3625 | goto out; | |
3626 | } | |
3627 | ||
3628 | // make sure snap mapper object exists | |
11fdf7f2 | 3629 | if (!store->exists(service.meta_ch, OSD::make_snapmapper_oid())) { |
7c673cae FG |
3630 | dout(10) << "init creating/touching snapmapper object" << dendl; |
3631 | ObjectStore::Transaction t; | |
3632 | t.touch(coll_t::meta(), OSD::make_snapmapper_oid()); | |
11fdf7f2 | 3633 | r = store->queue_transaction(service.meta_ch, std::move(t)); |
7c673cae FG |
3634 | if (r < 0) |
3635 | goto out; | |
3636 | } | |
9f95a23c TL |
3637 | if (!store->exists(service.meta_ch, OSD::make_purged_snaps_oid())) { |
3638 | dout(10) << "init creating/touching purged_snaps object" << dendl; | |
3639 | ObjectStore::Transaction t; | |
3640 | t.touch(coll_t::meta(), OSD::make_purged_snaps_oid()); | |
3641 | r = store->queue_transaction(service.meta_ch, std::move(t)); | |
3642 | if (r < 0) | |
3643 | goto out; | |
3644 | } | |
7c673cae FG |
3645 | |
3646 | if (cct->_conf->osd_open_classes_on_start) { | |
9f95a23c | 3647 | int r = ClassHandler::get_instance().open_all_classes(); |
7c673cae FG |
3648 | if (r) |
3649 | dout(1) << "warning: got an error loading one or more classes: " << cpp_strerror(r) << dendl; | |
3650 | } | |
3651 | ||
11fdf7f2 | 3652 | check_osdmap_features(); |
7c673cae | 3653 | |
7c673cae FG |
3654 | { |
3655 | epoch_t bind_epoch = osdmap->get_epoch(); | |
3656 | service.set_epochs(NULL, NULL, &bind_epoch); | |
3657 | } | |
3658 | ||
3659 | clear_temp_objects(); | |
3660 | ||
d2e6a577 | 3661 | // initialize osdmap references in sharded wq |
11fdf7f2 TL |
3662 | for (auto& shard : shards) { |
3663 | std::lock_guard l(shard->osdmap_lock); | |
3664 | shard->shard_osdmap = osdmap; | |
3665 | } | |
d2e6a577 | 3666 | |
7c673cae FG |
3667 | // load up pgs (as they previously existed) |
3668 | load_pgs(); | |
3669 | ||
3670 | dout(2) << "superblock: I am osd." << superblock.whoami << dendl; | |
7c673cae | 3671 | |
f67539c2 TL |
3672 | if (cct->_conf.get_val<bool>("osd_compact_on_start")) { |
3673 | dout(2) << "compacting object store's omap" << dendl; | |
3674 | store->compact(); | |
3675 | } | |
7c673cae | 3676 | |
11fdf7f2 TL |
3677 | // prime osd stats |
3678 | { | |
3679 | struct store_statfs_t stbuf; | |
3680 | osd_alert_list_t alerts; | |
3681 | int r = store->statfs(&stbuf, &alerts); | |
3682 | ceph_assert(r == 0); | |
3683 | service.set_statfs(stbuf, alerts); | |
3684 | } | |
3685 | ||
f67539c2 | 3686 | // client_messenger's auth_client will be set up by monc->init() later. |
11fdf7f2 TL |
3687 | for (auto m : { cluster_messenger, |
3688 | objecter_messenger, | |
3689 | hb_front_client_messenger, | |
3690 | hb_back_client_messenger, | |
3691 | hb_front_server_messenger, | |
3692 | hb_back_server_messenger } ) { | |
3693 | m->set_auth_client(monc); | |
3694 | } | |
3695 | for (auto m : { client_messenger, | |
3696 | cluster_messenger, | |
3697 | hb_front_server_messenger, | |
3698 | hb_back_server_messenger }) { | |
3699 | m->set_auth_server(monc); | |
3700 | } | |
3701 | monc->set_handle_authentication_dispatcher(this); | |
7c673cae FG |
3702 | |
3703 | monc->set_want_keys(CEPH_ENTITY_TYPE_MON | CEPH_ENTITY_TYPE_OSD | |
3704 | | CEPH_ENTITY_TYPE_MGR); | |
3705 | r = monc->init(); | |
3706 | if (r < 0) | |
3707 | goto out; | |
3708 | ||
f67539c2 | 3709 | mgrc.set_pgstats_cb([this]() { return collect_pg_stats(); }); |
11fdf7f2 | 3710 | mgrc.set_perf_metric_query_cb( |
9f95a23c TL |
3711 | [this](const ConfigPayload &config_payload) { |
3712 | set_perf_queries(config_payload); | |
11fdf7f2 | 3713 | }, |
9f95a23c TL |
3714 | [this] { |
3715 | return get_perf_reports(); | |
11fdf7f2 | 3716 | }); |
7c673cae | 3717 | mgrc.init(); |
7c673cae FG |
3718 | |
3719 | // tell monc about log_client so it will know about mon session resets | |
3720 | monc->set_log_client(&log_client); | |
3721 | update_log_config(); | |
3722 | ||
11fdf7f2 TL |
3723 | // i'm ready! |
3724 | client_messenger->add_dispatcher_tail(&mgrc); | |
3725 | client_messenger->add_dispatcher_tail(this); | |
3726 | cluster_messenger->add_dispatcher_head(this); | |
3727 | ||
3728 | hb_front_client_messenger->add_dispatcher_head(&heartbeat_dispatcher); | |
3729 | hb_back_client_messenger->add_dispatcher_head(&heartbeat_dispatcher); | |
3730 | hb_front_server_messenger->add_dispatcher_head(&heartbeat_dispatcher); | |
3731 | hb_back_server_messenger->add_dispatcher_head(&heartbeat_dispatcher); | |
3732 | ||
9f95a23c | 3733 | objecter_messenger->add_dispatcher_head(service.objecter.get()); |
11fdf7f2 | 3734 | |
28e407b8 AA |
3735 | service.init(); |
3736 | service.publish_map(osdmap); | |
3737 | service.publish_superblock(superblock); | |
3738 | service.max_oldest_map = superblock.oldest_map; | |
3739 | ||
11fdf7f2 TL |
3740 | for (auto& shard : shards) { |
3741 | // put PGs in a temporary set because we may modify pg_slots | |
3742 | // unordered_map below. | |
3743 | set<PGRef> pgs; | |
3744 | for (auto& i : shard->pg_slots) { | |
3745 | PGRef pg = i.second->pg; | |
3746 | if (!pg) { | |
3747 | continue; | |
3748 | } | |
3749 | pgs.insert(pg); | |
3750 | } | |
3751 | for (auto pg : pgs) { | |
9f95a23c | 3752 | std::scoped_lock l{*pg}; |
11fdf7f2 TL |
3753 | set<pair<spg_t,epoch_t>> new_children; |
3754 | set<pair<spg_t,epoch_t>> merge_pgs; | |
3755 | service.identify_splits_and_merges(pg->get_osdmap(), osdmap, pg->pg_id, | |
3756 | &new_children, &merge_pgs); | |
3757 | if (!new_children.empty()) { | |
3758 | for (auto shard : shards) { | |
3759 | shard->prime_splits(osdmap, &new_children); | |
3760 | } | |
3761 | assert(new_children.empty()); | |
3762 | } | |
3763 | if (!merge_pgs.empty()) { | |
3764 | for (auto shard : shards) { | |
3765 | shard->prime_merges(osdmap, &merge_pgs); | |
3766 | } | |
3767 | assert(merge_pgs.empty()); | |
3768 | } | |
11fdf7f2 TL |
3769 | } |
3770 | } | |
3771 | ||
7c673cae | 3772 | osd_op_tp.start(); |
7c673cae | 3773 | |
7c673cae FG |
3774 | // start the heartbeat |
3775 | heartbeat_thread.create("osd_srv_heartbt"); | |
3776 | ||
3777 | // tick | |
91327a77 AA |
3778 | tick_timer.add_event_after(get_tick_interval(), |
3779 | new C_Tick(this)); | |
7c673cae | 3780 | { |
11fdf7f2 | 3781 | std::lock_guard l(tick_timer_lock); |
91327a77 AA |
3782 | tick_timer_without_osd_lock.add_event_after(get_tick_interval(), |
3783 | new C_Tick_WithoutOSDLock(this)); | |
7c673cae FG |
3784 | } |
3785 | ||
9f95a23c | 3786 | osd_lock.unlock(); |
7c673cae FG |
3787 | |
3788 | r = monc->authenticate(); | |
3789 | if (r < 0) { | |
c07f9fc5 FG |
3790 | derr << __func__ << " authentication failed: " << cpp_strerror(r) |
3791 | << dendl; | |
11fdf7f2 | 3792 | exit(1); |
7c673cae FG |
3793 | } |
3794 | ||
11fdf7f2 | 3795 | while (monc->wait_auth_rotating(rotating_auth_timeout) < 0) { |
7c673cae FG |
3796 | derr << "unable to obtain rotating service keys; retrying" << dendl; |
3797 | ++rotating_auth_attempts; | |
11fdf7f2 | 3798 | if (rotating_auth_attempts > g_conf()->max_rotating_auth_attempts) { |
c07f9fc5 | 3799 | derr << __func__ << " wait_auth_rotating timed out" << dendl; |
11fdf7f2 | 3800 | exit(1); |
7c673cae FG |
3801 | } |
3802 | } | |
3803 | ||
3804 | r = update_crush_device_class(); | |
3805 | if (r < 0) { | |
d2e6a577 FG |
3806 | derr << __func__ << " unable to update_crush_device_class: " |
3807 | << cpp_strerror(r) << dendl; | |
11fdf7f2 | 3808 | exit(1); |
7c673cae FG |
3809 | } |
3810 | ||
3811 | r = update_crush_location(); | |
3812 | if (r < 0) { | |
d2e6a577 | 3813 | derr << __func__ << " unable to update_crush_location: " |
c07f9fc5 | 3814 | << cpp_strerror(r) << dendl; |
11fdf7f2 | 3815 | exit(1); |
7c673cae FG |
3816 | } |
3817 | ||
9f95a23c | 3818 | osd_lock.lock(); |
7c673cae FG |
3819 | if (is_stopping()) |
3820 | return 0; | |
3821 | ||
3822 | // start objecter *after* we have authenticated, so that we don't ignore | |
3823 | // the OSDMaps it requests. | |
3824 | service.final_init(); | |
3825 | ||
3826 | check_config(); | |
3827 | ||
3828 | dout(10) << "ensuring pgs have consumed prior maps" << dendl; | |
3829 | consume_map(); | |
7c673cae FG |
3830 | |
3831 | dout(0) << "done with init, starting boot process" << dendl; | |
3832 | ||
3833 | // subscribe to any pg creations | |
3834 | monc->sub_want("osd_pg_creates", last_pg_create_epoch, 0); | |
3835 | ||
3836 | // MgrClient needs this (it doesn't have MonClient reference itself) | |
3837 | monc->sub_want("mgrmap", 0, 0); | |
3838 | ||
3839 | // we don't need to ask for an osdmap here; objecter will | |
3840 | //monc->sub_want("osdmap", osdmap->get_epoch(), CEPH_SUBSCRIBE_ONETIME); | |
3841 | ||
3842 | monc->renew_subs(); | |
3843 | ||
3844 | start_boot(); | |
3845 | ||
a4b75251 TL |
3846 | // Override a few options if mclock scheduler is enabled. |
3847 | maybe_override_max_osd_capacity_for_qos(); | |
3848 | maybe_override_options_for_qos(); | |
3849 | ||
7c673cae | 3850 | return 0; |
7c673cae FG |
3851 | |
3852 | out: | |
3853 | enable_disable_fuse(true); | |
3854 | store->umount(); | |
20effc67 | 3855 | store.reset(); |
7c673cae FG |
3856 | return r; |
3857 | } | |
3858 | ||
3859 | void OSD::final_init() | |
3860 | { | |
3861 | AdminSocket *admin_socket = cct->get_admin_socket(); | |
3862 | asok_hook = new OSDSocketHook(this); | |
9f95a23c | 3863 | int r = admin_socket->register_command("status", asok_hook, |
7c673cae | 3864 | "high-level status of OSD"); |
11fdf7f2 | 3865 | ceph_assert(r == 0); |
9f95a23c | 3866 | r = admin_socket->register_command("flush_journal", |
7c673cae FG |
3867 | asok_hook, |
3868 | "flush the journal to permanent store"); | |
11fdf7f2 | 3869 | ceph_assert(r == 0); |
9f95a23c | 3870 | r = admin_socket->register_command("dump_ops_in_flight " \ |
c07f9fc5 FG |
3871 | "name=filterstr,type=CephString,n=N,req=false", |
3872 | asok_hook, | |
7c673cae | 3873 | "show the ops currently in flight"); |
11fdf7f2 | 3874 | ceph_assert(r == 0); |
9f95a23c | 3875 | r = admin_socket->register_command("ops " \ |
c07f9fc5 FG |
3876 | "name=filterstr,type=CephString,n=N,req=false", |
3877 | asok_hook, | |
7c673cae | 3878 | "show the ops currently in flight"); |
11fdf7f2 | 3879 | ceph_assert(r == 0); |
9f95a23c | 3880 | r = admin_socket->register_command("dump_blocked_ops " \ |
c07f9fc5 FG |
3881 | "name=filterstr,type=CephString,n=N,req=false", |
3882 | asok_hook, | |
7c673cae | 3883 | "show the blocked ops currently in flight"); |
11fdf7f2 | 3884 | ceph_assert(r == 0); |
9f95a23c | 3885 | r = admin_socket->register_command("dump_historic_ops " \ |
c07f9fc5 | 3886 | "name=filterstr,type=CephString,n=N,req=false", |
7c673cae FG |
3887 | asok_hook, |
3888 | "show recent ops"); | |
11fdf7f2 | 3889 | ceph_assert(r == 0); |
9f95a23c | 3890 | r = admin_socket->register_command("dump_historic_slow_ops " \ |
c07f9fc5 | 3891 | "name=filterstr,type=CephString,n=N,req=false", |
7c673cae FG |
3892 | asok_hook, |
3893 | "show slowest recent ops"); | |
11fdf7f2 | 3894 | ceph_assert(r == 0); |
9f95a23c | 3895 | r = admin_socket->register_command("dump_historic_ops_by_duration " \ |
c07f9fc5 | 3896 | "name=filterstr,type=CephString,n=N,req=false", |
7c673cae FG |
3897 | asok_hook, |
3898 | "show slowest recent ops, sorted by duration"); | |
11fdf7f2 | 3899 | ceph_assert(r == 0); |
9f95a23c | 3900 | r = admin_socket->register_command("dump_op_pq_state", |
7c673cae | 3901 | asok_hook, |
20effc67 | 3902 | "dump op queue state"); |
11fdf7f2 | 3903 | ceph_assert(r == 0); |
f67539c2 | 3904 | r = admin_socket->register_command("dump_blocklist", |
7c673cae | 3905 | asok_hook, |
f67539c2 | 3906 | "dump blocklisted clients and times"); |
11fdf7f2 | 3907 | ceph_assert(r == 0); |
9f95a23c | 3908 | r = admin_socket->register_command("dump_watchers", |
7c673cae FG |
3909 | asok_hook, |
3910 | "show clients which have active watches," | |
3911 | " and on which objects"); | |
11fdf7f2 | 3912 | ceph_assert(r == 0); |
9f95a23c | 3913 | r = admin_socket->register_command("dump_recovery_reservations", |
7c673cae FG |
3914 | asok_hook, |
3915 | "show recovery reservations"); | |
11fdf7f2 | 3916 | ceph_assert(r == 0); |
9f95a23c | 3917 | r = admin_socket->register_command("dump_scrub_reservations", |
eafe8130 | 3918 | asok_hook, |
f6b5b4d7 | 3919 | "show scrub reservations"); |
eafe8130 | 3920 | ceph_assert(r == 0); |
9f95a23c | 3921 | r = admin_socket->register_command("get_latest_osdmap", |
7c673cae FG |
3922 | asok_hook, |
3923 | "force osd to update the latest map from " | |
3924 | "the mon"); | |
11fdf7f2 | 3925 | ceph_assert(r == 0); |
7c673cae | 3926 | |
9f95a23c | 3927 | r = admin_socket->register_command("set_heap_property " \ |
7c673cae FG |
3928 | "name=property,type=CephString " \ |
3929 | "name=value,type=CephInt", | |
3930 | asok_hook, | |
3931 | "update malloc extension heap property"); | |
11fdf7f2 | 3932 | ceph_assert(r == 0); |
7c673cae | 3933 | |
9f95a23c | 3934 | r = admin_socket->register_command("get_heap_property " \ |
7c673cae FG |
3935 | "name=property,type=CephString", |
3936 | asok_hook, | |
3937 | "get malloc extension heap property"); | |
11fdf7f2 | 3938 | ceph_assert(r == 0); |
7c673cae FG |
3939 | |
3940 | r = admin_socket->register_command("dump_objectstore_kv_stats", | |
7c673cae FG |
3941 | asok_hook, |
3942 | "print statistics of kvdb which used by bluestore"); | |
11fdf7f2 | 3943 | ceph_assert(r == 0); |
7c673cae FG |
3944 | |
3945 | r = admin_socket->register_command("dump_scrubs", | |
7c673cae FG |
3946 | asok_hook, |
3947 | "print scheduled scrubs"); | |
11fdf7f2 | 3948 | ceph_assert(r == 0); |
7c673cae FG |
3949 | |
3950 | r = admin_socket->register_command("calc_objectstore_db_histogram", | |
7c673cae FG |
3951 | asok_hook, |
3952 | "Generate key value histogram of kvdb(rocksdb) which used by bluestore"); | |
11fdf7f2 | 3953 | ceph_assert(r == 0); |
7c673cae FG |
3954 | |
3955 | r = admin_socket->register_command("flush_store_cache", | |
7c673cae FG |
3956 | asok_hook, |
3957 | "Flush bluestore internal cache"); | |
11fdf7f2 | 3958 | ceph_assert(r == 0); |
9f95a23c | 3959 | r = admin_socket->register_command("dump_pgstate_history", |
7c673cae FG |
3960 | asok_hook, |
3961 | "show recent state history"); | |
11fdf7f2 | 3962 | ceph_assert(r == 0); |
7c673cae | 3963 | |
9f95a23c | 3964 | r = admin_socket->register_command("compact", |
224ce89b WB |
3965 | asok_hook, |
3966 | "Commpact object store's omap." | |
3967 | " WARNING: Compaction probably slows your requests"); | |
11fdf7f2 TL |
3968 | ceph_assert(r == 0); |
3969 | ||
9f95a23c | 3970 | r = admin_socket->register_command("get_mapped_pools", |
11fdf7f2 TL |
3971 | asok_hook, |
3972 | "dump pools whose PG(s) are mapped to this OSD."); | |
3973 | ||
3974 | ceph_assert(r == 0); | |
3975 | ||
9f95a23c | 3976 | r = admin_socket->register_command("smart name=devid,type=CephString,req=false", |
11fdf7f2 TL |
3977 | asok_hook, |
3978 | "probe OSD devices for SMART data."); | |
3979 | ||
3980 | ceph_assert(r == 0); | |
3981 | ||
9f95a23c | 3982 | r = admin_socket->register_command("list_devices", |
11fdf7f2 TL |
3983 | asok_hook, |
3984 | "list OSD devices."); | |
9f95a23c | 3985 | r = admin_socket->register_command("send_beacon", |
11fdf7f2 TL |
3986 | asok_hook, |
3987 | "send OSD beacon to mon immediately"); | |
224ce89b | 3988 | |
9f95a23c TL |
3989 | r = admin_socket->register_command( |
3990 | "dump_osd_network name=value,type=CephInt,req=false", asok_hook, | |
3991 | "Dump osd heartbeat network ping times"); | |
eafe8130 TL |
3992 | ceph_assert(r == 0); |
3993 | ||
20effc67 TL |
3994 | r = admin_socket->register_command( |
3995 | "dump_pool_statfs name=poolid,type=CephInt,req=true", asok_hook, | |
3996 | "Dump store's statistics for the given pool"); | |
3997 | ceph_assert(r == 0); | |
3998 | ||
3999 | test_ops_hook = new TestOpsSocketHook(&(this->service), this->store.get()); | |
7c673cae FG |
4000 | // Note: pools are CephString instead of CephPoolname because |
4001 | // these commands traditionally support both pool names and numbers | |
4002 | r = admin_socket->register_command( | |
7c673cae FG |
4003 | "setomapval " \ |
4004 | "name=pool,type=CephString " \ | |
4005 | "name=objname,type=CephObjectname " \ | |
4006 | "name=key,type=CephString "\ | |
4007 | "name=val,type=CephString", | |
4008 | test_ops_hook, | |
4009 | "set omap key"); | |
11fdf7f2 | 4010 | ceph_assert(r == 0); |
7c673cae | 4011 | r = admin_socket->register_command( |
7c673cae FG |
4012 | "rmomapkey " \ |
4013 | "name=pool,type=CephString " \ | |
4014 | "name=objname,type=CephObjectname " \ | |
4015 | "name=key,type=CephString", | |
4016 | test_ops_hook, | |
4017 | "remove omap key"); | |
11fdf7f2 | 4018 | ceph_assert(r == 0); |
7c673cae | 4019 | r = admin_socket->register_command( |
7c673cae FG |
4020 | "setomapheader " \ |
4021 | "name=pool,type=CephString " \ | |
4022 | "name=objname,type=CephObjectname " \ | |
4023 | "name=header,type=CephString", | |
4024 | test_ops_hook, | |
4025 | "set omap header"); | |
11fdf7f2 | 4026 | ceph_assert(r == 0); |
7c673cae FG |
4027 | |
4028 | r = admin_socket->register_command( | |
7c673cae FG |
4029 | "getomap " \ |
4030 | "name=pool,type=CephString " \ | |
4031 | "name=objname,type=CephObjectname", | |
4032 | test_ops_hook, | |
4033 | "output entire object map"); | |
11fdf7f2 | 4034 | ceph_assert(r == 0); |
7c673cae FG |
4035 | |
4036 | r = admin_socket->register_command( | |
7c673cae FG |
4037 | "truncobj " \ |
4038 | "name=pool,type=CephString " \ | |
4039 | "name=objname,type=CephObjectname " \ | |
4040 | "name=len,type=CephInt", | |
4041 | test_ops_hook, | |
4042 | "truncate object to length"); | |
11fdf7f2 | 4043 | ceph_assert(r == 0); |
7c673cae FG |
4044 | |
4045 | r = admin_socket->register_command( | |
7c673cae FG |
4046 | "injectdataerr " \ |
4047 | "name=pool,type=CephString " \ | |
4048 | "name=objname,type=CephObjectname " \ | |
4049 | "name=shardid,type=CephInt,req=false,range=0|255", | |
4050 | test_ops_hook, | |
4051 | "inject data error to an object"); | |
11fdf7f2 | 4052 | ceph_assert(r == 0); |
7c673cae FG |
4053 | |
4054 | r = admin_socket->register_command( | |
7c673cae FG |
4055 | "injectmdataerr " \ |
4056 | "name=pool,type=CephString " \ | |
4057 | "name=objname,type=CephObjectname " \ | |
4058 | "name=shardid,type=CephInt,req=false,range=0|255", | |
4059 | test_ops_hook, | |
4060 | "inject metadata error to an object"); | |
11fdf7f2 | 4061 | ceph_assert(r == 0); |
7c673cae | 4062 | r = admin_socket->register_command( |
7c673cae FG |
4063 | "set_recovery_delay " \ |
4064 | "name=utime,type=CephInt,req=false", | |
4065 | test_ops_hook, | |
4066 | "Delay osd recovery by specified seconds"); | |
11fdf7f2 | 4067 | ceph_assert(r == 0); |
7c673cae | 4068 | r = admin_socket->register_command( |
7c673cae FG |
4069 | "injectfull " \ |
4070 | "name=type,type=CephString,req=false " \ | |
4071 | "name=count,type=CephInt,req=false ", | |
4072 | test_ops_hook, | |
4073 | "Inject a full disk (optional count times)"); | |
11fdf7f2 | 4074 | ceph_assert(r == 0); |
9f95a23c TL |
4075 | r = admin_socket->register_command( |
4076 | "bench " \ | |
4077 | "name=count,type=CephInt,req=false " \ | |
4078 | "name=size,type=CephInt,req=false " \ | |
4079 | "name=object_size,type=CephInt,req=false " \ | |
4080 | "name=object_num,type=CephInt,req=false ", | |
4081 | asok_hook, | |
4082 | "OSD benchmark: write <count> <size>-byte objects(with <obj_size> <obj_num>), " \ | |
4083 | "(default count=1G default size=4MB). Results in log."); | |
4084 | ceph_assert(r == 0); | |
4085 | r = admin_socket->register_command( | |
4086 | "cluster_log " \ | |
4087 | "name=level,type=CephChoices,strings=error,warning,info,debug " \ | |
4088 | "name=message,type=CephString,n=N", | |
4089 | asok_hook, | |
4090 | "log a message to the cluster log"); | |
4091 | ceph_assert(r == 0); | |
4092 | r = admin_socket->register_command( | |
4093 | "flush_pg_stats", | |
4094 | asok_hook, | |
4095 | "flush pg stats"); | |
4096 | ceph_assert(r == 0); | |
4097 | r = admin_socket->register_command( | |
4098 | "heap " \ | |
4099 | "name=heapcmd,type=CephChoices,strings=" \ | |
4100 | "dump|start_profiler|stop_profiler|release|get_release_rate|set_release_rate|stats " \ | |
4101 | "name=value,type=CephString,req=false", | |
4102 | asok_hook, | |
4103 | "show heap usage info (available only if compiled with tcmalloc)"); | |
4104 | ceph_assert(r == 0); | |
4105 | r = admin_socket->register_command( | |
4106 | "debug dump_missing " \ | |
4107 | "name=filename,type=CephFilepath", | |
4108 | asok_hook, | |
4109 | "dump missing objects to a named file"); | |
4110 | ceph_assert(r == 0); | |
4111 | r = admin_socket->register_command( | |
4112 | "debug kick_recovery_wq " \ | |
4113 | "name=delay,type=CephInt,range=0", | |
4114 | asok_hook, | |
4115 | "set osd_recovery_delay_start to <val>"); | |
4116 | ceph_assert(r == 0); | |
4117 | r = admin_socket->register_command( | |
4118 | "cpu_profiler " \ | |
4119 | "name=arg,type=CephChoices,strings=status|flush", | |
4120 | asok_hook, | |
4121 | "run cpu profiling on daemon"); | |
4122 | ceph_assert(r == 0); | |
4123 | r = admin_socket->register_command( | |
4124 | "dump_pg_recovery_stats", | |
4125 | asok_hook, | |
4126 | "dump pg recovery statistics"); | |
4127 | ceph_assert(r == 0); | |
4128 | r = admin_socket->register_command( | |
4129 | "reset_pg_recovery_stats", | |
4130 | asok_hook, | |
4131 | "reset pg recovery statistics"); | |
4132 | ceph_assert(r == 0); | |
4133 | r = admin_socket->register_command( | |
4134 | "cache drop", | |
4135 | asok_hook, | |
4136 | "Drop all OSD caches"); | |
4137 | ceph_assert(r == 0); | |
4138 | r = admin_socket->register_command( | |
4139 | "cache status", | |
4140 | asok_hook, | |
4141 | "Get OSD caches statistics"); | |
4142 | ceph_assert(r == 0); | |
4143 | r = admin_socket->register_command( | |
4144 | "scrub_purged_snaps", | |
4145 | asok_hook, | |
4146 | "Scrub purged_snaps vs snapmapper index"); | |
4147 | ceph_assert(r == 0); | |
20effc67 TL |
4148 | r = admin_socket->register_command( |
4149 | "scrubdebug " \ | |
4150 | "name=pgid,type=CephPgid " \ | |
4151 | "name=cmd,type=CephChoices,strings=block|unblock|set|unset " \ | |
4152 | "name=value,type=CephString,req=false", | |
4153 | asok_hook, | |
4154 | "debug the scrubber"); | |
4155 | ceph_assert(r == 0); | |
7c673cae | 4156 | |
9f95a23c TL |
4157 | // -- pg commands -- |
4158 | // old form: ceph pg <pgid> command ... | |
4159 | r = admin_socket->register_command( | |
4160 | "pg " \ | |
4161 | "name=pgid,type=CephPgid " \ | |
4162 | "name=cmd,type=CephChoices,strings=query", | |
4163 | asok_hook, | |
4164 | ""); | |
4165 | ceph_assert(r == 0); | |
4166 | r = admin_socket->register_command( | |
4167 | "pg " \ | |
4168 | "name=pgid,type=CephPgid " \ | |
4169 | "name=cmd,type=CephChoices,strings=mark_unfound_lost " \ | |
4170 | "name=mulcmd,type=CephChoices,strings=revert|delete", | |
4171 | asok_hook, | |
4172 | ""); | |
4173 | ceph_assert(r == 0); | |
4174 | r = admin_socket->register_command( | |
4175 | "pg " \ | |
4176 | "name=pgid,type=CephPgid " \ | |
4177 | "name=cmd,type=CephChoices,strings=list_unfound " \ | |
4178 | "name=offset,type=CephString,req=false", | |
4179 | asok_hook, | |
4180 | ""); | |
4181 | ceph_assert(r == 0); | |
4182 | r = admin_socket->register_command( | |
4183 | "pg " \ | |
4184 | "name=pgid,type=CephPgid " \ | |
4185 | "name=cmd,type=CephChoices,strings=scrub " \ | |
4186 | "name=time,type=CephInt,req=false", | |
4187 | asok_hook, | |
4188 | ""); | |
4189 | ceph_assert(r == 0); | |
4190 | r = admin_socket->register_command( | |
4191 | "pg " \ | |
4192 | "name=pgid,type=CephPgid " \ | |
4193 | "name=cmd,type=CephChoices,strings=deep_scrub " \ | |
4194 | "name=time,type=CephInt,req=false", | |
4195 | asok_hook, | |
4196 | ""); | |
4197 | ceph_assert(r == 0); | |
4198 | // new form: tell <pgid> <cmd> for both cli and rest | |
4199 | r = admin_socket->register_command( | |
4200 | "query", | |
4201 | asok_hook, | |
4202 | "show details of a specific pg"); | |
4203 | ceph_assert(r == 0); | |
4204 | r = admin_socket->register_command( | |
4205 | "mark_unfound_lost " \ | |
4206 | "name=pgid,type=CephPgid,req=false " \ | |
4207 | "name=mulcmd,type=CephChoices,strings=revert|delete", | |
4208 | asok_hook, | |
4209 | "mark all unfound objects in this pg as lost, either removing or reverting to a prior version if one is available"); | |
4210 | ceph_assert(r == 0); | |
4211 | r = admin_socket->register_command( | |
4212 | "list_unfound " \ | |
4213 | "name=pgid,type=CephPgid,req=false " \ | |
4214 | "name=offset,type=CephString,req=false", | |
4215 | asok_hook, | |
4216 | "list unfound objects on this pg, perhaps starting at an offset given in JSON"); | |
4217 | ceph_assert(r == 0); | |
4218 | r = admin_socket->register_command( | |
4219 | "scrub " \ | |
4220 | "name=pgid,type=CephPgid,req=false " \ | |
4221 | "name=time,type=CephInt,req=false", | |
4222 | asok_hook, | |
4223 | "Trigger a scheduled scrub "); | |
4224 | ceph_assert(r == 0); | |
4225 | r = admin_socket->register_command( | |
4226 | "deep_scrub " \ | |
4227 | "name=pgid,type=CephPgid,req=false " \ | |
4228 | "name=time,type=CephInt,req=false", | |
4229 | asok_hook, | |
4230 | "Trigger a scheduled deep scrub "); | |
4231 | ceph_assert(r == 0); | |
4232 | } | |
7c673cae | 4233 | |
f67539c2 | 4234 | PerfCounters* OSD::create_logger() |
9f95a23c | 4235 | { |
f67539c2 | 4236 | PerfCounters* logger = build_osd_logger(cct); |
7c673cae | 4237 | cct->get_perfcounters_collection()->add(logger); |
f67539c2 | 4238 | return logger; |
7c673cae FG |
4239 | } |
4240 | ||
f67539c2 | 4241 | PerfCounters* OSD::create_recoverystate_perf() |
7c673cae | 4242 | { |
f67539c2 | 4243 | PerfCounters* recoverystate_perf = build_recoverystate_perf(cct); |
7c673cae | 4244 | cct->get_perfcounters_collection()->add(recoverystate_perf); |
f67539c2 | 4245 | return recoverystate_perf; |
7c673cae FG |
4246 | } |
4247 | ||
4248 | int OSD::shutdown() | |
4249 | { | |
1d09f67e TL |
4250 | // vstart overwrites osd_fast_shutdown value in the conf file -> force the value here! |
4251 | //cct->_conf->osd_fast_shutdown = true; | |
4252 | ||
4253 | dout(0) << "Fast Shutdown: - cct->_conf->osd_fast_shutdown = " | |
4254 | << cct->_conf->osd_fast_shutdown | |
4255 | << ", null-fm = " << store->has_null_manager() << dendl; | |
4256 | ||
4257 | utime_t start_time_func = ceph_clock_now(); | |
4258 | ||
92f5a8d4 TL |
4259 | if (cct->_conf->osd_fast_shutdown) { |
4260 | derr << "*** Immediate shutdown (osd_fast_shutdown=true) ***" << dendl; | |
f67539c2 TL |
4261 | if (cct->_conf->osd_fast_shutdown_notify_mon) |
4262 | service.prepare_to_stop(); | |
92f5a8d4 | 4263 | |
1d09f67e TL |
4264 | // There is no state we need to keep wehn running in NULL-FM moode |
4265 | if (!store->has_null_manager()) { | |
4266 | cct->_log->flush(); | |
4267 | _exit(0); | |
4268 | } | |
4269 | } else if (!service.prepare_to_stop()) { | |
7c673cae | 4270 | return 0; // already shutting down |
1d09f67e TL |
4271 | } |
4272 | ||
9f95a23c | 4273 | osd_lock.lock(); |
7c673cae | 4274 | if (is_stopping()) { |
9f95a23c | 4275 | osd_lock.unlock(); |
7c673cae FG |
4276 | return 0; |
4277 | } | |
7c673cae | 4278 | |
1d09f67e TL |
4279 | if (!cct->_conf->osd_fast_shutdown) { |
4280 | dout(0) << "shutdown" << dendl; | |
4281 | } | |
4282 | ||
4283 | // don't accept new task for this OSD | |
7c673cae FG |
4284 | set_state(STATE_STOPPING); |
4285 | ||
1d09f67e TL |
4286 | // Disabled debugging during fast-shutdown |
4287 | if (!cct->_conf->osd_fast_shutdown && cct->_conf.get_val<bool>("osd_debug_shutdown")) { | |
11fdf7f2 TL |
4288 | cct->_conf.set_val("debug_osd", "100"); |
4289 | cct->_conf.set_val("debug_journal", "100"); | |
4290 | cct->_conf.set_val("debug_filestore", "100"); | |
4291 | cct->_conf.set_val("debug_bluestore", "100"); | |
4292 | cct->_conf.set_val("debug_ms", "100"); | |
4293 | cct->_conf.apply_changes(nullptr); | |
3efd9988 | 4294 | } |
7c673cae | 4295 | |
1d09f67e TL |
4296 | if (cct->_conf->osd_fast_shutdown) { |
4297 | // first, stop new task from being taken from op_shardedwq | |
4298 | // and clear all pending tasks | |
4299 | op_shardedwq.stop_for_fast_shutdown(); | |
4300 | ||
4301 | utime_t start_time_timer = ceph_clock_now(); | |
4302 | tick_timer.shutdown(); | |
4303 | { | |
4304 | std::lock_guard l(tick_timer_lock); | |
4305 | tick_timer_without_osd_lock.shutdown(); | |
4306 | } | |
4307 | ||
4308 | osd_lock.unlock(); | |
4309 | utime_t start_time_osd_drain = ceph_clock_now(); | |
4310 | ||
4311 | // then, wait on osd_op_tp to drain (TBD: should probably add a timeout) | |
4312 | osd_op_tp.drain(); | |
4313 | osd_op_tp.stop(); | |
4314 | ||
4315 | utime_t start_time_umount = ceph_clock_now(); | |
4316 | store->prepare_for_fast_shutdown(); | |
4317 | std::lock_guard lock(osd_lock); | |
4318 | // TBD: assert in allocator that nothing is being add | |
4319 | store->umount(); | |
4320 | ||
4321 | utime_t end_time = ceph_clock_now(); | |
4322 | if (cct->_conf->osd_fast_shutdown_timeout) { | |
4323 | ceph_assert(end_time - start_time_func < cct->_conf->osd_fast_shutdown_timeout); | |
4324 | } | |
4325 | dout(0) <<"Fast Shutdown duration total :" << end_time - start_time_func << " seconds" << dendl; | |
4326 | dout(0) <<"Fast Shutdown duration osd_drain :" << start_time_umount - start_time_osd_drain << " seconds" << dendl; | |
4327 | dout(0) <<"Fast Shutdown duration umount :" << end_time - start_time_umount << " seconds" << dendl; | |
4328 | dout(0) <<"Fast Shutdown duration timer :" << start_time_osd_drain - start_time_timer << " seconds" << dendl; | |
4329 | cct->_log->flush(); | |
4330 | ||
4331 | // now it is safe to exit | |
4332 | _exit(0); | |
4333 | } | |
4334 | ||
7c673cae FG |
4335 | // stop MgrClient earlier as it's more like an internal consumer of OSD |
4336 | mgrc.shutdown(); | |
4337 | ||
4338 | service.start_shutdown(); | |
4339 | ||
4340 | // stop sending work to pgs. this just prevents any new work in _process | |
4341 | // from racing with on_shutdown and potentially entering the pg after. | |
4342 | op_shardedwq.drain(); | |
4343 | ||
4344 | // Shutdown PGs | |
4345 | { | |
11fdf7f2 TL |
4346 | vector<PGRef> pgs; |
4347 | _get_pgs(&pgs); | |
4348 | for (auto pg : pgs) { | |
4349 | pg->shutdown(); | |
7c673cae FG |
4350 | } |
4351 | } | |
7c673cae FG |
4352 | |
4353 | // drain op queue again (in case PGs requeued something) | |
4354 | op_shardedwq.drain(); | |
4355 | { | |
4356 | finished.clear(); // zap waiters (bleh, this is messy) | |
11fdf7f2 | 4357 | waiting_for_osdmap.clear(); |
7c673cae FG |
4358 | } |
4359 | ||
7c673cae | 4360 | // unregister commands |
11fdf7f2 | 4361 | cct->get_admin_socket()->unregister_commands(asok_hook); |
7c673cae FG |
4362 | delete asok_hook; |
4363 | asok_hook = NULL; | |
4364 | ||
11fdf7f2 | 4365 | cct->get_admin_socket()->unregister_commands(test_ops_hook); |
7c673cae FG |
4366 | delete test_ops_hook; |
4367 | test_ops_hook = NULL; | |
4368 | ||
9f95a23c | 4369 | osd_lock.unlock(); |
7c673cae | 4370 | |
9f95a23c TL |
4371 | { |
4372 | std::lock_guard l{heartbeat_lock}; | |
4373 | heartbeat_stop = true; | |
4374 | heartbeat_cond.notify_all(); | |
4375 | heartbeat_peers.clear(); | |
4376 | } | |
7c673cae FG |
4377 | heartbeat_thread.join(); |
4378 | ||
9f95a23c TL |
4379 | hb_back_server_messenger->mark_down_all(); |
4380 | hb_front_server_messenger->mark_down_all(); | |
4381 | hb_front_client_messenger->mark_down_all(); | |
4382 | hb_back_client_messenger->mark_down_all(); | |
4383 | ||
7c673cae FG |
4384 | osd_op_tp.drain(); |
4385 | osd_op_tp.stop(); | |
4386 | dout(10) << "op sharded tp stopped" << dendl; | |
4387 | ||
7c673cae FG |
4388 | dout(10) << "stopping agent" << dendl; |
4389 | service.agent_stop(); | |
4390 | ||
11fdf7f2 TL |
4391 | boot_finisher.wait_for_empty(); |
4392 | ||
9f95a23c | 4393 | osd_lock.lock(); |
7c673cae | 4394 | |
11fdf7f2 | 4395 | boot_finisher.stop(); |
494da23a | 4396 | reset_heartbeat_peers(true); |
7c673cae FG |
4397 | |
4398 | tick_timer.shutdown(); | |
4399 | ||
4400 | { | |
11fdf7f2 | 4401 | std::lock_guard l(tick_timer_lock); |
7c673cae FG |
4402 | tick_timer_without_osd_lock.shutdown(); |
4403 | } | |
4404 | ||
4405 | // note unmount epoch | |
9f95a23c | 4406 | dout(10) << "noting clean unmount in epoch " << get_osdmap_epoch() << dendl; |
7c673cae | 4407 | superblock.mounted = service.get_boot_epoch(); |
9f95a23c | 4408 | superblock.clean_thru = get_osdmap_epoch(); |
7c673cae FG |
4409 | ObjectStore::Transaction t; |
4410 | write_superblock(t); | |
11fdf7f2 | 4411 | int r = store->queue_transaction(service.meta_ch, std::move(t)); |
7c673cae FG |
4412 | if (r) { |
4413 | derr << "OSD::shutdown: error writing superblock: " | |
4414 | << cpp_strerror(r) << dendl; | |
4415 | } | |
4416 | ||
4417 | ||
31f18b77 FG |
4418 | service.shutdown_reserver(); |
4419 | ||
7c673cae FG |
4420 | // Remove PGs |
4421 | #ifdef PG_DEBUG_REFS | |
4422 | service.dump_live_pgids(); | |
4423 | #endif | |
11fdf7f2 TL |
4424 | while (true) { |
4425 | vector<PGRef> pgs; | |
4426 | _get_pgs(&pgs, true); | |
4427 | if (pgs.empty()) { | |
4428 | break; | |
4429 | } | |
4430 | for (auto& pg : pgs) { | |
4431 | if (pg->is_deleted()) { | |
4432 | continue; | |
4433 | } | |
4434 | dout(20) << " kicking pg " << pg << dendl; | |
4435 | pg->lock(); | |
4436 | if (pg->get_num_ref() != 1) { | |
4437 | derr << "pgid " << pg->get_pgid() << " has ref count of " | |
4438 | << pg->get_num_ref() << dendl; | |
7c673cae | 4439 | #ifdef PG_DEBUG_REFS |
11fdf7f2 | 4440 | pg->dump_live_ids(); |
7c673cae | 4441 | #endif |
31f18b77 FG |
4442 | if (cct->_conf->osd_shutdown_pgref_assert) { |
4443 | ceph_abort(); | |
4444 | } | |
7c673cae | 4445 | } |
11fdf7f2 TL |
4446 | pg->ch.reset(); |
4447 | pg->unlock(); | |
7c673cae | 4448 | } |
7c673cae FG |
4449 | } |
4450 | #ifdef PG_DEBUG_REFS | |
4451 | service.dump_live_pgids(); | |
4452 | #endif | |
f64942e4 | 4453 | |
9f95a23c | 4454 | osd_lock.unlock(); |
11fdf7f2 | 4455 | cct->_conf.remove_observer(this); |
9f95a23c | 4456 | osd_lock.lock(); |
7c673cae | 4457 | |
11fdf7f2 TL |
4458 | service.meta_ch.reset(); |
4459 | ||
7c673cae FG |
4460 | dout(10) << "syncing store" << dendl; |
4461 | enable_disable_fuse(true); | |
4462 | ||
4463 | if (cct->_conf->osd_journal_flush_on_shutdown) { | |
4464 | dout(10) << "flushing journal" << dendl; | |
4465 | store->flush_journal(); | |
4466 | } | |
4467 | ||
7c673cae | 4468 | monc->shutdown(); |
9f95a23c TL |
4469 | osd_lock.unlock(); |
4470 | { | |
4471 | std::unique_lock l{map_lock}; | |
4472 | set_osdmap(OSDMapRef()); | |
4473 | } | |
11fdf7f2 TL |
4474 | for (auto s : shards) { |
4475 | std::lock_guard l(s->osdmap_lock); | |
4476 | s->shard_osdmap = OSDMapRef(); | |
4477 | } | |
7c673cae | 4478 | service.shutdown(); |
11fdf7f2 TL |
4479 | |
4480 | std::lock_guard lock(osd_lock); | |
4481 | store->umount(); | |
20effc67 | 4482 | store.reset(); |
11fdf7f2 TL |
4483 | dout(10) << "Store synced" << dendl; |
4484 | ||
7c673cae FG |
4485 | op_tracker.on_shutdown(); |
4486 | ||
9f95a23c | 4487 | ClassHandler::get_instance().shutdown(); |
7c673cae FG |
4488 | client_messenger->shutdown(); |
4489 | cluster_messenger->shutdown(); | |
4490 | hb_front_client_messenger->shutdown(); | |
4491 | hb_back_client_messenger->shutdown(); | |
4492 | objecter_messenger->shutdown(); | |
4493 | hb_front_server_messenger->shutdown(); | |
4494 | hb_back_server_messenger->shutdown(); | |
4495 | ||
1d09f67e TL |
4496 | utime_t duration = ceph_clock_now() - start_time_func; |
4497 | dout(0) <<"Slow Shutdown duration:" << duration << " seconds" << dendl; | |
4498 | ||
20effc67 TL |
4499 | tracing::osd::tracer.shutdown(); |
4500 | ||
7c673cae FG |
4501 | return r; |
4502 | } | |
4503 | ||
4504 | int OSD::mon_cmd_maybe_osd_create(string &cmd) | |
4505 | { | |
4506 | bool created = false; | |
4507 | while (true) { | |
4508 | dout(10) << __func__ << " cmd: " << cmd << dendl; | |
4509 | vector<string> vcmd{cmd}; | |
4510 | bufferlist inbl; | |
4511 | C_SaferCond w; | |
4512 | string outs; | |
4513 | monc->start_mon_command(vcmd, inbl, NULL, &outs, &w); | |
4514 | int r = w.wait(); | |
4515 | if (r < 0) { | |
4516 | if (r == -ENOENT && !created) { | |
4517 | string newcmd = "{\"prefix\": \"osd create\", \"id\": " + stringify(whoami) | |
4518 | + ", \"uuid\": \"" + stringify(superblock.osd_fsid) + "\"}"; | |
4519 | vector<string> vnewcmd{newcmd}; | |
4520 | bufferlist inbl; | |
4521 | C_SaferCond w; | |
4522 | string outs; | |
4523 | monc->start_mon_command(vnewcmd, inbl, NULL, &outs, &w); | |
4524 | int r = w.wait(); | |
4525 | if (r < 0) { | |
4526 | derr << __func__ << " fail: osd does not exist and created failed: " | |
4527 | << cpp_strerror(r) << dendl; | |
4528 | return r; | |
4529 | } | |
4530 | created = true; | |
4531 | continue; | |
4532 | } | |
4533 | derr << __func__ << " fail: '" << outs << "': " << cpp_strerror(r) << dendl; | |
4534 | return r; | |
4535 | } | |
4536 | break; | |
4537 | } | |
4538 | ||
4539 | return 0; | |
4540 | } | |
4541 | ||
4542 | int OSD::update_crush_location() | |
4543 | { | |
4544 | if (!cct->_conf->osd_crush_update_on_start) { | |
4545 | dout(10) << __func__ << " osd_crush_update_on_start = false" << dendl; | |
4546 | return 0; | |
4547 | } | |
4548 | ||
4549 | char weight[32]; | |
4550 | if (cct->_conf->osd_crush_initial_weight >= 0) { | |
4551 | snprintf(weight, sizeof(weight), "%.4lf", cct->_conf->osd_crush_initial_weight); | |
4552 | } else { | |
4553 | struct store_statfs_t st; | |
11fdf7f2 TL |
4554 | osd_alert_list_t alerts; |
4555 | int r = store->statfs(&st, &alerts); | |
7c673cae FG |
4556 | if (r < 0) { |
4557 | derr << "statfs: " << cpp_strerror(r) << dendl; | |
4558 | return r; | |
4559 | } | |
4560 | snprintf(weight, sizeof(weight), "%.4lf", | |
11fdf7f2 TL |
4561 | std::max(.00001, |
4562 | double(st.total) / | |
4563 | double(1ull << 40 /* TB */))); | |
7c673cae FG |
4564 | } |
4565 | ||
9f95a23c | 4566 | dout(10) << __func__ << " crush location is " << cct->crush_location << dendl; |
7c673cae FG |
4567 | |
4568 | string cmd = | |
4569 | string("{\"prefix\": \"osd crush create-or-move\", ") + | |
9f95a23c TL |
4570 | string("\"id\": ") + stringify(whoami) + ", " + |
4571 | string("\"weight\":") + weight + ", " + | |
4572 | string("\"args\": [") + stringify(cct->crush_location) + "]}"; | |
7c673cae FG |
4573 | return mon_cmd_maybe_osd_create(cmd); |
4574 | } | |
4575 | ||
4576 | int OSD::update_crush_device_class() | |
4577 | { | |
224ce89b WB |
4578 | if (!cct->_conf->osd_class_update_on_start) { |
4579 | dout(10) << __func__ << " osd_class_update_on_start = false" << dendl; | |
4580 | return 0; | |
4581 | } | |
4582 | ||
7c673cae FG |
4583 | string device_class; |
4584 | int r = store->read_meta("crush_device_class", &device_class); | |
224ce89b WB |
4585 | if (r < 0 || device_class.empty()) { |
4586 | device_class = store->get_default_device_class(); | |
4587 | } | |
4588 | ||
4589 | if (device_class.empty()) { | |
d2e6a577 | 4590 | dout(20) << __func__ << " no device class stored locally" << dendl; |
7c673cae | 4591 | return 0; |
224ce89b | 4592 | } |
7c673cae FG |
4593 | |
4594 | string cmd = | |
4595 | string("{\"prefix\": \"osd crush set-device-class\", ") + | |
224ce89b WB |
4596 | string("\"class\": \"") + device_class + string("\", ") + |
4597 | string("\"ids\": [\"") + stringify(whoami) + string("\"]}"); | |
7c673cae | 4598 | |
224ce89b | 4599 | r = mon_cmd_maybe_osd_create(cmd); |
11fdf7f2 TL |
4600 | if (r == -EBUSY) { |
4601 | // good, already bound to a device-class | |
4602 | return 0; | |
4603 | } else { | |
4604 | return r; | |
4605 | } | |
7c673cae FG |
4606 | } |
4607 | ||
4608 | void OSD::write_superblock(ObjectStore::Transaction& t) | |
4609 | { | |
4610 | dout(10) << "write_superblock " << superblock << dendl; | |
4611 | ||
4612 | //hack: at minimum it's using the baseline feature set | |
4613 | if (!superblock.compat_features.incompat.contains(CEPH_OSD_FEATURE_INCOMPAT_BASE)) | |
4614 | superblock.compat_features.incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_BASE); | |
4615 | ||
4616 | bufferlist bl; | |
11fdf7f2 | 4617 | encode(superblock, bl); |
7c673cae FG |
4618 | t.write(coll_t::meta(), OSD_SUPERBLOCK_GOBJECT, 0, bl.length(), bl); |
4619 | } | |
4620 | ||
4621 | int OSD::read_superblock() | |
4622 | { | |
4623 | bufferlist bl; | |
11fdf7f2 | 4624 | int r = store->read(service.meta_ch, OSD_SUPERBLOCK_GOBJECT, 0, 0, bl); |
7c673cae FG |
4625 | if (r < 0) |
4626 | return r; | |
4627 | ||
11fdf7f2 TL |
4628 | auto p = bl.cbegin(); |
4629 | decode(superblock, p); | |
7c673cae FG |
4630 | |
4631 | dout(10) << "read_superblock " << superblock << dendl; | |
4632 | ||
4633 | return 0; | |
4634 | } | |
4635 | ||
4636 | void OSD::clear_temp_objects() | |
4637 | { | |
4638 | dout(10) << __func__ << dendl; | |
4639 | vector<coll_t> ls; | |
4640 | store->list_collections(ls); | |
4641 | for (vector<coll_t>::iterator p = ls.begin(); p != ls.end(); ++p) { | |
4642 | spg_t pgid; | |
4643 | if (!p->is_pg(&pgid)) | |
4644 | continue; | |
4645 | ||
4646 | // list temp objects | |
4647 | dout(20) << " clearing temps in " << *p << " pgid " << pgid << dendl; | |
4648 | ||
4649 | vector<ghobject_t> temps; | |
4650 | ghobject_t next; | |
4651 | while (1) { | |
4652 | vector<ghobject_t> objects; | |
11fdf7f2 TL |
4653 | auto ch = store->open_collection(*p); |
4654 | ceph_assert(ch); | |
4655 | store->collection_list(ch, next, ghobject_t::get_max(), | |
7c673cae FG |
4656 | store->get_ideal_list_max(), |
4657 | &objects, &next); | |
4658 | if (objects.empty()) | |
4659 | break; | |
4660 | vector<ghobject_t>::iterator q; | |
4661 | for (q = objects.begin(); q != objects.end(); ++q) { | |
4662 | // Hammer set pool for temps to -1, so check for clean-up | |
4663 | if (q->hobj.is_temp() || (q->hobj.pool == -1)) { | |
4664 | temps.push_back(*q); | |
4665 | } else { | |
4666 | break; | |
4667 | } | |
4668 | } | |
4669 | // If we saw a non-temp object and hit the break above we can | |
4670 | // break out of the while loop too. | |
4671 | if (q != objects.end()) | |
4672 | break; | |
4673 | } | |
4674 | if (!temps.empty()) { | |
4675 | ObjectStore::Transaction t; | |
4676 | int removed = 0; | |
4677 | for (vector<ghobject_t>::iterator q = temps.begin(); q != temps.end(); ++q) { | |
4678 | dout(20) << " removing " << *p << " object " << *q << dendl; | |
4679 | t.remove(*p, *q); | |
4680 | if (++removed > cct->_conf->osd_target_transaction_size) { | |
11fdf7f2 | 4681 | store->queue_transaction(service.meta_ch, std::move(t)); |
7c673cae FG |
4682 | t = ObjectStore::Transaction(); |
4683 | removed = 0; | |
4684 | } | |
4685 | } | |
4686 | if (removed) { | |
11fdf7f2 | 4687 | store->queue_transaction(service.meta_ch, std::move(t)); |
7c673cae FG |
4688 | } |
4689 | } | |
4690 | } | |
4691 | } | |
4692 | ||
4693 | void OSD::recursive_remove_collection(CephContext* cct, | |
4694 | ObjectStore *store, spg_t pgid, | |
4695 | coll_t tmp) | |
4696 | { | |
4697 | OSDriver driver( | |
4698 | store, | |
4699 | coll_t(), | |
4700 | make_snapmapper_oid()); | |
4701 | ||
11fdf7f2 | 4702 | ObjectStore::CollectionHandle ch = store->open_collection(tmp); |
7c673cae FG |
4703 | ObjectStore::Transaction t; |
4704 | SnapMapper mapper(cct, &driver, 0, 0, 0, pgid.shard); | |
4705 | ||
11fdf7f2 TL |
4706 | ghobject_t next; |
4707 | int max = cct->_conf->osd_target_transaction_size; | |
7c673cae | 4708 | vector<ghobject_t> objects; |
11fdf7f2 TL |
4709 | objects.reserve(max); |
4710 | while (true) { | |
4711 | objects.clear(); | |
4712 | store->collection_list(ch, next, ghobject_t::get_max(), | |
4713 | max, &objects, &next); | |
4714 | generic_dout(10) << __func__ << " " << objects << dendl; | |
4715 | if (objects.empty()) | |
4716 | break; | |
4717 | for (auto& p: objects) { | |
4718 | OSDriver::OSTransaction _t(driver.get_transaction(&t)); | |
4719 | int r = mapper.remove_oid(p.hobj, &_t); | |
4720 | if (r != 0 && r != -ENOENT) | |
4721 | ceph_abort(); | |
4722 | t.remove(tmp, p); | |
7c673cae | 4723 | } |
11fdf7f2 TL |
4724 | int r = store->queue_transaction(ch, std::move(t)); |
4725 | ceph_assert(r == 0); | |
4726 | t = ObjectStore::Transaction(); | |
7c673cae FG |
4727 | } |
4728 | t.remove_collection(tmp); | |
11fdf7f2 TL |
4729 | int r = store->queue_transaction(ch, std::move(t)); |
4730 | ceph_assert(r == 0); | |
7c673cae FG |
4731 | |
4732 | C_SaferCond waiter; | |
11fdf7f2 | 4733 | if (!ch->flush_commit(&waiter)) { |
7c673cae FG |
4734 | waiter.wait(); |
4735 | } | |
4736 | } | |
4737 | ||
4738 | ||
4739 | // ====================================================== | |
4740 | // PG's | |
4741 | ||
7c673cae FG |
4742 | PG* OSD::_make_pg( |
4743 | OSDMapRef createmap, | |
4744 | spg_t pgid) | |
4745 | { | |
11fdf7f2 TL |
4746 | dout(10) << __func__ << " " << pgid << dendl; |
4747 | pg_pool_t pi; | |
4748 | map<string,string> ec_profile; | |
4749 | string name; | |
4750 | if (createmap->have_pg_pool(pgid.pool())) { | |
4751 | pi = *createmap->get_pg_pool(pgid.pool()); | |
4752 | name = createmap->get_pool_name(pgid.pool()); | |
4753 | if (pi.is_erasure()) { | |
4754 | ec_profile = createmap->get_erasure_code_profile(pi.erasure_code_profile); | |
4755 | } | |
4756 | } else { | |
4757 | // pool was deleted; grab final pg_pool_t off disk. | |
4758 | ghobject_t oid = make_final_pool_info_oid(pgid.pool()); | |
4759 | bufferlist bl; | |
4760 | int r = store->read(service.meta_ch, oid, 0, 0, bl); | |
4761 | if (r < 0) { | |
4762 | derr << __func__ << " missing pool " << pgid.pool() << " tombstone" | |
4763 | << dendl; | |
4764 | return nullptr; | |
4765 | } | |
4766 | ceph_assert(r >= 0); | |
4767 | auto p = bl.cbegin(); | |
4768 | decode(pi, p); | |
4769 | decode(name, p); | |
4770 | if (p.end()) { // dev release v13.0.2 did not include ec_profile | |
4771 | derr << __func__ << " missing ec_profile from pool " << pgid.pool() | |
4772 | << " tombstone" << dendl; | |
4773 | return nullptr; | |
4774 | } | |
4775 | decode(ec_profile, p); | |
4776 | } | |
f67539c2 | 4777 | PGPool pool(createmap, pgid.pool(), pi, name); |
7c673cae | 4778 | PG *pg; |
11fdf7f2 TL |
4779 | if (pi.type == pg_pool_t::TYPE_REPLICATED || |
4780 | pi.type == pg_pool_t::TYPE_ERASURE) | |
4781 | pg = new PrimaryLogPG(&service, createmap, pool, ec_profile, pgid); | |
7c673cae FG |
4782 | else |
4783 | ceph_abort(); | |
7c673cae FG |
4784 | return pg; |
4785 | } | |
4786 | ||
11fdf7f2 | 4787 | void OSD::_get_pgs(vector<PGRef> *v, bool clear_too) |
7c673cae | 4788 | { |
11fdf7f2 TL |
4789 | v->clear(); |
4790 | v->reserve(get_num_pgs()); | |
4791 | for (auto& s : shards) { | |
4792 | std::lock_guard l(s->shard_lock); | |
4793 | for (auto& j : s->pg_slots) { | |
4794 | if (j.second->pg && | |
4795 | !j.second->pg->is_deleted()) { | |
4796 | v->push_back(j.second->pg); | |
4797 | if (clear_too) { | |
4798 | s->_detach_pg(j.second.get()); | |
4799 | } | |
4800 | } | |
7c673cae | 4801 | } |
7c673cae | 4802 | } |
7c673cae FG |
4803 | } |
4804 | ||
11fdf7f2 | 4805 | void OSD::_get_pgids(vector<spg_t> *v) |
7c673cae | 4806 | { |
11fdf7f2 TL |
4807 | v->clear(); |
4808 | v->reserve(get_num_pgs()); | |
4809 | for (auto& s : shards) { | |
4810 | std::lock_guard l(s->shard_lock); | |
4811 | for (auto& j : s->pg_slots) { | |
4812 | if (j.second->pg && | |
4813 | !j.second->pg->is_deleted()) { | |
4814 | v->push_back(j.first); | |
4815 | } | |
7c673cae FG |
4816 | } |
4817 | } | |
7c673cae FG |
4818 | } |
4819 | ||
11fdf7f2 | 4820 | void OSD::register_pg(PGRef pg) |
7c673cae | 4821 | { |
11fdf7f2 TL |
4822 | spg_t pgid = pg->get_pgid(); |
4823 | uint32_t shard_index = pgid.hash_to_shard(num_shards); | |
4824 | auto sdata = shards[shard_index]; | |
4825 | std::lock_guard l(sdata->shard_lock); | |
4826 | auto r = sdata->pg_slots.emplace(pgid, make_unique<OSDShardPGSlot>()); | |
4827 | ceph_assert(r.second); | |
4828 | auto *slot = r.first->second.get(); | |
4829 | dout(20) << __func__ << " " << pgid << " " << pg << dendl; | |
4830 | sdata->_attach_pg(slot, pg.get()); | |
4831 | } | |
7c673cae | 4832 | |
11fdf7f2 TL |
4833 | bool OSD::try_finish_pg_delete(PG *pg, unsigned old_pg_num) |
4834 | { | |
4835 | auto sdata = pg->osd_shard; | |
4836 | ceph_assert(sdata); | |
4837 | { | |
4838 | std::lock_guard l(sdata->shard_lock); | |
4839 | auto p = sdata->pg_slots.find(pg->pg_id); | |
4840 | if (p == sdata->pg_slots.end() || | |
4841 | !p->second->pg) { | |
4842 | dout(20) << __func__ << " " << pg->pg_id << " not found" << dendl; | |
4843 | return false; | |
4844 | } | |
4845 | if (p->second->waiting_for_merge_epoch) { | |
4846 | dout(20) << __func__ << " " << pg->pg_id << " waiting for merge" << dendl; | |
4847 | return false; | |
4848 | } | |
4849 | dout(20) << __func__ << " " << pg->pg_id << " " << pg << dendl; | |
4850 | sdata->_detach_pg(p->second.get()); | |
4851 | } | |
7c673cae | 4852 | |
11fdf7f2 TL |
4853 | for (auto shard : shards) { |
4854 | shard->unprime_split_children(pg->pg_id, old_pg_num); | |
4855 | } | |
7c673cae | 4856 | |
11fdf7f2 TL |
4857 | // update pg count now since we might not get an osdmap any time soon. |
4858 | if (pg->is_primary()) | |
4859 | service.logger->dec(l_osd_pg_primary); | |
9f95a23c TL |
4860 | else if (pg->is_nonprimary()) |
4861 | service.logger->dec(l_osd_pg_replica); // misnomver | |
11fdf7f2 TL |
4862 | else |
4863 | service.logger->dec(l_osd_pg_stray); | |
7c673cae | 4864 | |
11fdf7f2 | 4865 | return true; |
7c673cae FG |
4866 | } |
4867 | ||
11fdf7f2 | 4868 | PGRef OSD::_lookup_pg(spg_t pgid) |
7c673cae | 4869 | { |
11fdf7f2 TL |
4870 | uint32_t shard_index = pgid.hash_to_shard(num_shards); |
4871 | auto sdata = shards[shard_index]; | |
4872 | std::lock_guard l(sdata->shard_lock); | |
4873 | auto p = sdata->pg_slots.find(pgid); | |
4874 | if (p == sdata->pg_slots.end()) { | |
7c673cae | 4875 | return nullptr; |
11fdf7f2 TL |
4876 | } |
4877 | return p->second->pg; | |
7c673cae FG |
4878 | } |
4879 | ||
11fdf7f2 | 4880 | PGRef OSD::_lookup_lock_pg(spg_t pgid) |
31f18b77 | 4881 | { |
11fdf7f2 TL |
4882 | PGRef pg = _lookup_pg(pgid); |
4883 | if (!pg) { | |
4884 | return nullptr; | |
4885 | } | |
4886 | pg->lock(); | |
4887 | if (!pg->is_deleted()) { | |
4888 | return pg; | |
4889 | } | |
4890 | pg->unlock(); | |
4891 | return nullptr; | |
31f18b77 FG |
4892 | } |
4893 | ||
11fdf7f2 | 4894 | PGRef OSD::lookup_lock_pg(spg_t pgid) |
7c673cae | 4895 | { |
11fdf7f2 | 4896 | return _lookup_lock_pg(pgid); |
7c673cae FG |
4897 | } |
4898 | ||
4899 | void OSD::load_pgs() | |
4900 | { | |
9f95a23c | 4901 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae | 4902 | dout(0) << "load_pgs" << dendl; |
11fdf7f2 | 4903 | |
7c673cae | 4904 | { |
11fdf7f2 TL |
4905 | auto pghist = make_pg_num_history_oid(); |
4906 | bufferlist bl; | |
4907 | int r = store->read(service.meta_ch, pghist, 0, 0, bl, 0); | |
4908 | if (r >= 0 && bl.length() > 0) { | |
4909 | auto p = bl.cbegin(); | |
4910 | decode(pg_num_history, p); | |
4911 | } | |
4912 | dout(20) << __func__ << " pg_num_history " << pg_num_history << dendl; | |
7c673cae FG |
4913 | } |
4914 | ||
4915 | vector<coll_t> ls; | |
4916 | int r = store->list_collections(ls); | |
4917 | if (r < 0) { | |
4918 | derr << "failed to list pgs: " << cpp_strerror(-r) << dendl; | |
4919 | } | |
4920 | ||
11fdf7f2 | 4921 | int num = 0; |
7c673cae FG |
4922 | for (vector<coll_t>::iterator it = ls.begin(); |
4923 | it != ls.end(); | |
4924 | ++it) { | |
4925 | spg_t pgid; | |
4926 | if (it->is_temp(&pgid) || | |
20effc67 | 4927 | (it->is_pg(&pgid) && PG::_has_removal_flag(store.get(), pgid))) { |
11fdf7f2 TL |
4928 | dout(10) << "load_pgs " << *it |
4929 | << " removing, legacy or flagged for removal pg" << dendl; | |
20effc67 | 4930 | recursive_remove_collection(cct, store.get(), pgid, *it); |
7c673cae FG |
4931 | continue; |
4932 | } | |
4933 | ||
4934 | if (!it->is_pg(&pgid)) { | |
4935 | dout(10) << "load_pgs ignoring unrecognized " << *it << dendl; | |
4936 | continue; | |
4937 | } | |
4938 | ||
7c673cae | 4939 | dout(10) << "pgid " << pgid << " coll " << coll_t(pgid) << dendl; |
7c673cae | 4940 | epoch_t map_epoch = 0; |
20effc67 | 4941 | int r = PG::peek_map_epoch(store.get(), pgid, &map_epoch); |
7c673cae FG |
4942 | if (r < 0) { |
4943 | derr << __func__ << " unable to peek at " << pgid << " metadata, skipping" | |
4944 | << dendl; | |
4945 | continue; | |
4946 | } | |
4947 | ||
11fdf7f2 | 4948 | PGRef pg; |
7c673cae FG |
4949 | if (map_epoch > 0) { |
4950 | OSDMapRef pgosdmap = service.try_get_map(map_epoch); | |
4951 | if (!pgosdmap) { | |
9f95a23c | 4952 | if (!get_osdmap()->have_pg_pool(pgid.pool())) { |
7c673cae FG |
4953 | derr << __func__ << ": could not find map for epoch " << map_epoch |
4954 | << " on pg " << pgid << ", but the pool is not present in the " | |
4955 | << "current map, so this is probably a result of bug 10617. " | |
4956 | << "Skipping the pg for now, you can use ceph-objectstore-tool " | |
4957 | << "to clean it up later." << dendl; | |
4958 | continue; | |
4959 | } else { | |
4960 | derr << __func__ << ": have pgid " << pgid << " at epoch " | |
4961 | << map_epoch << ", but missing map. Crashing." | |
4962 | << dendl; | |
11fdf7f2 | 4963 | ceph_abort_msg("Missing map in load_pgs"); |
7c673cae FG |
4964 | } |
4965 | } | |
11fdf7f2 | 4966 | pg = _make_pg(pgosdmap, pgid); |
7c673cae | 4967 | } else { |
9f95a23c | 4968 | pg = _make_pg(get_osdmap(), pgid); |
7c673cae | 4969 | } |
11fdf7f2 | 4970 | if (!pg) { |
20effc67 | 4971 | recursive_remove_collection(cct, store.get(), pgid, *it); |
11fdf7f2 TL |
4972 | continue; |
4973 | } | |
4974 | ||
4975 | // there can be no waiters here, so we don't call _wake_pg_slot | |
7c673cae | 4976 | |
11fdf7f2 | 4977 | pg->lock(); |
7c673cae FG |
4978 | pg->ch = store->open_collection(pg->coll); |
4979 | ||
4980 | // read pg state, log | |
20effc67 | 4981 | pg->read_state(store.get()); |
7c673cae | 4982 | |
94b18763 FG |
4983 | if (pg->dne()) { |
4984 | dout(10) << "load_pgs " << *it << " deleting dne" << dendl; | |
4985 | pg->ch = nullptr; | |
94b18763 | 4986 | pg->unlock(); |
20effc67 | 4987 | recursive_remove_collection(cct, store.get(), pgid, *it); |
94b18763 FG |
4988 | continue; |
4989 | } | |
11fdf7f2 TL |
4990 | { |
4991 | uint32_t shard_index = pgid.hash_to_shard(shards.size()); | |
4992 | assert(NULL != shards[shard_index]); | |
4993 | store->set_collection_commit_queue(pg->coll, &(shards[shard_index]->context_queue)); | |
4994 | } | |
7c673cae | 4995 | |
11fdf7f2 | 4996 | dout(10) << __func__ << " loaded " << *pg << dendl; |
7c673cae | 4997 | pg->unlock(); |
7c673cae | 4998 | |
11fdf7f2 TL |
4999 | register_pg(pg); |
5000 | ++num; | |
7c673cae | 5001 | } |
11fdf7f2 | 5002 | dout(0) << __func__ << " opened " << num << " pgs" << dendl; |
7c673cae FG |
5003 | } |
5004 | ||
5005 | ||
11fdf7f2 TL |
5006 | PGRef OSD::handle_pg_create_info(const OSDMapRef& osdmap, |
5007 | const PGCreateInfo *info) | |
5008 | { | |
5009 | spg_t pgid = info->pgid; | |
7c673cae | 5010 | |
11fdf7f2 TL |
5011 | if (maybe_wait_for_max_pg(osdmap, pgid, info->by_mon)) { |
5012 | dout(10) << __func__ << " hit max pg, dropping" << dendl; | |
5013 | return nullptr; | |
5014 | } | |
3efd9988 | 5015 | |
11fdf7f2 | 5016 | OSDMapRef startmap = get_map(info->epoch); |
7c673cae | 5017 | |
11fdf7f2 TL |
5018 | if (info->by_mon) { |
5019 | int64_t pool_id = pgid.pgid.pool(); | |
5020 | const pg_pool_t *pool = osdmap->get_pg_pool(pool_id); | |
5021 | if (!pool) { | |
5022 | dout(10) << __func__ << " ignoring " << pgid << ", pool dne" << dendl; | |
5023 | return nullptr; | |
5024 | } | |
9f95a23c | 5025 | if (osdmap->require_osd_release >= ceph_release_t::nautilus && |
11fdf7f2 TL |
5026 | !pool->has_flag(pg_pool_t::FLAG_CREATING)) { |
5027 | // this ensures we do not process old creating messages after the | |
5028 | // pool's initial pgs have been created (and pg are subsequently | |
5029 | // allowed to split or merge). | |
5030 | dout(20) << __func__ << " dropping " << pgid | |
5031 | << "create, pool does not have CREATING flag set" << dendl; | |
5032 | return nullptr; | |
7c673cae FG |
5033 | } |
5034 | } | |
7c673cae | 5035 | |
11fdf7f2 TL |
5036 | int up_primary, acting_primary; |
5037 | vector<int> up, acting; | |
5038 | startmap->pg_to_up_acting_osds( | |
5039 | pgid.pgid, &up, &up_primary, &acting, &acting_primary); | |
7c673cae | 5040 | |
11fdf7f2 TL |
5041 | const pg_pool_t* pp = startmap->get_pg_pool(pgid.pool()); |
5042 | if (pp->has_flag(pg_pool_t::FLAG_EC_OVERWRITES) && | |
5043 | store->get_type() != "bluestore") { | |
5044 | clog->warn() << "pg " << pgid | |
5045 | << " is at risk of silent data corruption: " | |
5046 | << "the pool allows ec overwrites but is not stored in " | |
5047 | << "bluestore, so deep scrubbing will not detect bitrot"; | |
7c673cae | 5048 | } |
20effc67 | 5049 | PeeringCtx rctx; |
9f95a23c TL |
5050 | create_pg_collection( |
5051 | rctx.transaction, pgid, pgid.get_split_bits(pp->get_pg_num())); | |
5052 | init_pg_ondisk(rctx.transaction, pgid, pp); | |
7c673cae | 5053 | |
9f95a23c | 5054 | int role = startmap->calc_pg_role(pg_shard_t(whoami, pgid.shard), acting); |
7c673cae | 5055 | |
11fdf7f2 TL |
5056 | PGRef pg = _make_pg(startmap, pgid); |
5057 | pg->ch = store->create_new_collection(pg->coll); | |
7c673cae | 5058 | |
11fdf7f2 TL |
5059 | { |
5060 | uint32_t shard_index = pgid.hash_to_shard(shards.size()); | |
5061 | assert(NULL != shards[shard_index]); | |
5062 | store->set_collection_commit_queue(pg->coll, &(shards[shard_index]->context_queue)); | |
7c673cae | 5063 | } |
7c673cae | 5064 | |
11fdf7f2 | 5065 | pg->lock(true); |
7c673cae | 5066 | |
11fdf7f2 TL |
5067 | // we are holding the shard lock |
5068 | ceph_assert(!pg->is_deleted()); | |
5069 | ||
5070 | pg->init( | |
5071 | role, | |
5072 | up, | |
5073 | up_primary, | |
5074 | acting, | |
5075 | acting_primary, | |
5076 | info->history, | |
5077 | info->past_intervals, | |
11fdf7f2 | 5078 | rctx.transaction); |
7c673cae | 5079 | |
92f5a8d4 TL |
5080 | pg->init_collection_pool_opts(); |
5081 | ||
11fdf7f2 | 5082 | if (pg->is_primary()) { |
9f95a23c | 5083 | std::lock_guard locker{m_perf_queries_lock}; |
11fdf7f2 TL |
5084 | pg->set_dynamic_perf_stats_queries(m_perf_queries); |
5085 | } | |
7c673cae | 5086 | |
9f95a23c TL |
5087 | pg->handle_initialize(rctx); |
5088 | pg->handle_activate_map(rctx); | |
7c673cae | 5089 | |
11fdf7f2 | 5090 | dispatch_context(rctx, pg.get(), osdmap, nullptr); |
7c673cae | 5091 | |
11fdf7f2 TL |
5092 | dout(10) << __func__ << " new pg " << *pg << dendl; |
5093 | return pg; | |
7c673cae FG |
5094 | } |
5095 | ||
11fdf7f2 TL |
5096 | bool OSD::maybe_wait_for_max_pg(const OSDMapRef& osdmap, |
5097 | spg_t pgid, | |
5098 | bool is_mon_create) | |
3efd9988 FG |
5099 | { |
5100 | const auto max_pgs_per_osd = | |
11fdf7f2 TL |
5101 | (cct->_conf.get_val<uint64_t>("mon_max_pg_per_osd") * |
5102 | cct->_conf.get_val<double>("osd_max_pg_per_osd_hard_ratio")); | |
3efd9988 | 5103 | |
11fdf7f2 | 5104 | if (num_pgs < max_pgs_per_osd) { |
3efd9988 FG |
5105 | return false; |
5106 | } | |
11fdf7f2 TL |
5107 | |
5108 | std::lock_guard l(pending_creates_lock); | |
3efd9988 FG |
5109 | if (is_mon_create) { |
5110 | pending_creates_from_mon++; | |
5111 | } else { | |
9f95a23c TL |
5112 | bool is_primary = osdmap->get_pg_acting_role(pgid, whoami) == 0; |
5113 | pending_creates_from_osd.emplace(pgid, is_primary); | |
3efd9988 | 5114 | } |
1adf2230 | 5115 | dout(1) << __func__ << " withhold creation of pg " << pgid |
11fdf7f2 | 5116 | << ": " << num_pgs << " >= "<< max_pgs_per_osd << dendl; |
3efd9988 FG |
5117 | return true; |
5118 | } | |
5119 | ||
5120 | // to re-trigger a peering, we have to twiddle the pg mapping a little bit, | |
5121 | // see PG::should_restart_peering(). OSDMap::pg_to_up_acting_osds() will turn | |
5122 | // to up set if pg_temp is empty. so an empty pg_temp won't work. | |
5123 | static vector<int32_t> twiddle(const vector<int>& acting) { | |
5124 | if (acting.size() > 1) { | |
5125 | return {acting[0]}; | |
5126 | } else { | |
5127 | vector<int32_t> twiddled(acting.begin(), acting.end()); | |
5128 | twiddled.push_back(-1); | |
5129 | return twiddled; | |
5130 | } | |
5131 | } | |
5132 | ||
5133 | void OSD::resume_creating_pg() | |
5134 | { | |
5135 | bool do_sub_pg_creates = false; | |
b32b8144 | 5136 | bool have_pending_creates = false; |
3efd9988 FG |
5137 | { |
5138 | const auto max_pgs_per_osd = | |
11fdf7f2 TL |
5139 | (cct->_conf.get_val<uint64_t>("mon_max_pg_per_osd") * |
5140 | cct->_conf.get_val<double>("osd_max_pg_per_osd_hard_ratio")); | |
5141 | if (max_pgs_per_osd <= num_pgs) { | |
3efd9988 FG |
5142 | // this could happen if admin decreases this setting before a PG is removed |
5143 | return; | |
5144 | } | |
11fdf7f2 TL |
5145 | unsigned spare_pgs = max_pgs_per_osd - num_pgs; |
5146 | std::lock_guard l(pending_creates_lock); | |
3efd9988 | 5147 | if (pending_creates_from_mon > 0) { |
11fdf7f2 TL |
5148 | dout(20) << __func__ << " pending_creates_from_mon " |
5149 | << pending_creates_from_mon << dendl; | |
3efd9988 FG |
5150 | do_sub_pg_creates = true; |
5151 | if (pending_creates_from_mon >= spare_pgs) { | |
5152 | spare_pgs = pending_creates_from_mon = 0; | |
5153 | } else { | |
5154 | spare_pgs -= pending_creates_from_mon; | |
5155 | pending_creates_from_mon = 0; | |
5156 | } | |
5157 | } | |
5158 | auto pg = pending_creates_from_osd.cbegin(); | |
5159 | while (spare_pgs > 0 && pg != pending_creates_from_osd.cend()) { | |
94b18763 | 5160 | dout(20) << __func__ << " pg " << pg->first << dendl; |
3efd9988 | 5161 | vector<int> acting; |
9f95a23c TL |
5162 | get_osdmap()->pg_to_up_acting_osds(pg->first.pgid, nullptr, nullptr, &acting, nullptr); |
5163 | service.queue_want_pg_temp(pg->first.pgid, twiddle(acting), true); | |
3efd9988 | 5164 | pg = pending_creates_from_osd.erase(pg); |
94b18763 | 5165 | do_sub_pg_creates = true; |
3efd9988 FG |
5166 | spare_pgs--; |
5167 | } | |
b32b8144 FG |
5168 | have_pending_creates = (pending_creates_from_mon > 0 || |
5169 | !pending_creates_from_osd.empty()); | |
3efd9988 | 5170 | } |
b32b8144 FG |
5171 | |
5172 | bool do_renew_subs = false; | |
3efd9988 FG |
5173 | if (do_sub_pg_creates) { |
5174 | if (monc->sub_want("osd_pg_creates", last_pg_create_epoch, 0)) { | |
5175 | dout(4) << __func__ << ": resolicit pg creates from mon since " | |
5176 | << last_pg_create_epoch << dendl; | |
b32b8144 | 5177 | do_renew_subs = true; |
3efd9988 FG |
5178 | } |
5179 | } | |
9f95a23c | 5180 | version_t start = get_osdmap_epoch() + 1; |
b32b8144 FG |
5181 | if (have_pending_creates) { |
5182 | // don't miss any new osdmap deleting PGs | |
5183 | if (monc->sub_want("osdmap", start, 0)) { | |
5184 | dout(4) << __func__ << ": resolicit osdmap from mon since " | |
5185 | << start << dendl; | |
5186 | do_renew_subs = true; | |
5187 | } | |
94b18763 | 5188 | } else if (do_sub_pg_creates) { |
b32b8144 FG |
5189 | // no need to subscribe the osdmap continuously anymore |
5190 | // once the pgtemp and/or mon_subscribe(pg_creates) is sent | |
5191 | if (monc->sub_want_increment("osdmap", start, CEPH_SUBSCRIBE_ONETIME)) { | |
11fdf7f2 | 5192 | dout(4) << __func__ << ": re-subscribe osdmap(onetime) since " |
b32b8144 FG |
5193 | << start << dendl; |
5194 | do_renew_subs = true; | |
5195 | } | |
5196 | } | |
5197 | ||
5198 | if (do_renew_subs) { | |
5199 | monc->renew_subs(); | |
5200 | } | |
5201 | ||
94b18763 | 5202 | service.send_pg_temp(); |
3efd9988 | 5203 | } |
7c673cae FG |
5204 | |
5205 | void OSD::build_initial_pg_history( | |
5206 | spg_t pgid, | |
5207 | epoch_t created, | |
5208 | utime_t created_stamp, | |
5209 | pg_history_t *h, | |
5210 | PastIntervals *pi) | |
5211 | { | |
5212 | dout(10) << __func__ << " " << pgid << " created " << created << dendl; | |
9f95a23c | 5213 | *h = pg_history_t(created, created_stamp); |
7c673cae FG |
5214 | |
5215 | OSDMapRef lastmap = service.get_map(created); | |
5216 | int up_primary, acting_primary; | |
5217 | vector<int> up, acting; | |
5218 | lastmap->pg_to_up_acting_osds( | |
5219 | pgid.pgid, &up, &up_primary, &acting, &acting_primary); | |
5220 | ||
5221 | ostringstream debug; | |
9f95a23c | 5222 | for (epoch_t e = created + 1; e <= get_osdmap_epoch(); ++e) { |
7c673cae FG |
5223 | OSDMapRef osdmap = service.get_map(e); |
5224 | int new_up_primary, new_acting_primary; | |
5225 | vector<int> new_up, new_acting; | |
5226 | osdmap->pg_to_up_acting_osds( | |
5227 | pgid.pgid, &new_up, &new_up_primary, &new_acting, &new_acting_primary); | |
5228 | ||
5229 | // this is a bit imprecise, but sufficient? | |
5230 | struct min_size_predicate_t : public IsPGRecoverablePredicate { | |
5231 | const pg_pool_t *pi; | |
5232 | bool operator()(const set<pg_shard_t> &have) const { | |
5233 | return have.size() >= pi->min_size; | |
5234 | } | |
11fdf7f2 | 5235 | explicit min_size_predicate_t(const pg_pool_t *i) : pi(i) {} |
7c673cae FG |
5236 | } min_size_predicate(osdmap->get_pg_pool(pgid.pgid.pool())); |
5237 | ||
5238 | bool new_interval = PastIntervals::check_new_interval( | |
5239 | acting_primary, | |
5240 | new_acting_primary, | |
5241 | acting, new_acting, | |
5242 | up_primary, | |
5243 | new_up_primary, | |
5244 | up, new_up, | |
5245 | h->same_interval_since, | |
5246 | h->last_epoch_clean, | |
9f95a23c TL |
5247 | osdmap.get(), |
5248 | lastmap.get(), | |
7c673cae | 5249 | pgid.pgid, |
9f95a23c | 5250 | min_size_predicate, |
7c673cae FG |
5251 | pi, |
5252 | &debug); | |
5253 | if (new_interval) { | |
5254 | h->same_interval_since = e; | |
181888fb FG |
5255 | if (up != new_up) { |
5256 | h->same_up_since = e; | |
5257 | } | |
5258 | if (acting_primary != new_acting_primary) { | |
5259 | h->same_primary_since = e; | |
5260 | } | |
5261 | if (pgid.pgid.is_split(lastmap->get_pg_num(pgid.pgid.pool()), | |
5262 | osdmap->get_pg_num(pgid.pgid.pool()), | |
5263 | nullptr)) { | |
5264 | h->last_epoch_split = e; | |
5265 | } | |
5266 | up = new_up; | |
5267 | acting = new_acting; | |
5268 | up_primary = new_up_primary; | |
5269 | acting_primary = new_acting_primary; | |
c07f9fc5 | 5270 | } |
7c673cae FG |
5271 | lastmap = osdmap; |
5272 | } | |
5273 | dout(20) << __func__ << " " << debug.str() << dendl; | |
5274 | dout(10) << __func__ << " " << *h << " " << *pi | |
5275 | << " [" << (pi->empty() ? pair<epoch_t,epoch_t>(0,0) : | |
5276 | pi->get_bounds()) << ")" | |
5277 | << dendl; | |
5278 | } | |
5279 | ||
7c673cae FG |
5280 | void OSD::_add_heartbeat_peer(int p) |
5281 | { | |
5282 | if (p == whoami) | |
5283 | return; | |
5284 | HeartbeatInfo *hi; | |
5285 | ||
5286 | map<int,HeartbeatInfo>::iterator i = heartbeat_peers.find(p); | |
5287 | if (i == heartbeat_peers.end()) { | |
9f95a23c | 5288 | pair<ConnectionRef,ConnectionRef> cons = service.get_con_osd_hb(p, get_osdmap_epoch()); |
7c673cae FG |
5289 | if (!cons.first) |
5290 | return; | |
9f95a23c TL |
5291 | assert(cons.second); |
5292 | ||
7c673cae FG |
5293 | hi = &heartbeat_peers[p]; |
5294 | hi->peer = p; | |
9f95a23c TL |
5295 | |
5296 | auto stamps = service.get_hb_stamps(p); | |
5297 | ||
5298 | auto sb = ceph::make_ref<Session>(cct, cons.first.get()); | |
5299 | sb->peer = p; | |
5300 | sb->stamps = stamps; | |
eafe8130 | 5301 | hi->hb_interval_start = ceph_clock_now(); |
7c673cae | 5302 | hi->con_back = cons.first.get(); |
9f95a23c TL |
5303 | hi->con_back->set_priv(sb); |
5304 | ||
5305 | auto sf = ceph::make_ref<Session>(cct, cons.second.get()); | |
5306 | sf->peer = p; | |
5307 | sf->stamps = stamps; | |
5308 | hi->con_front = cons.second.get(); | |
5309 | hi->con_front->set_priv(sf); | |
5310 | ||
5311 | dout(10) << "_add_heartbeat_peer: new peer osd." << p | |
5312 | << " " << hi->con_back->get_peer_addr() | |
5313 | << " " << hi->con_front->get_peer_addr() | |
5314 | << dendl; | |
7c673cae FG |
5315 | } else { |
5316 | hi = &i->second; | |
5317 | } | |
9f95a23c | 5318 | hi->epoch = get_osdmap_epoch(); |
7c673cae FG |
5319 | } |
5320 | ||
5321 | void OSD::_remove_heartbeat_peer(int n) | |
5322 | { | |
5323 | map<int,HeartbeatInfo>::iterator q = heartbeat_peers.find(n); | |
11fdf7f2 | 5324 | ceph_assert(q != heartbeat_peers.end()); |
7c673cae FG |
5325 | dout(20) << " removing heartbeat peer osd." << n |
5326 | << " " << q->second.con_back->get_peer_addr() | |
5327 | << " " << (q->second.con_front ? q->second.con_front->get_peer_addr() : entity_addr_t()) | |
5328 | << dendl; | |
9f95a23c | 5329 | q->second.clear_mark_down(); |
7c673cae FG |
5330 | heartbeat_peers.erase(q); |
5331 | } | |
5332 | ||
5333 | void OSD::need_heartbeat_peer_update() | |
5334 | { | |
5335 | if (is_stopping()) | |
5336 | return; | |
5337 | dout(20) << "need_heartbeat_peer_update" << dendl; | |
5338 | heartbeat_set_peers_need_update(); | |
5339 | } | |
5340 | ||
5341 | void OSD::maybe_update_heartbeat_peers() | |
5342 | { | |
9f95a23c | 5343 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae | 5344 | |
11fdf7f2 | 5345 | if (is_waiting_for_healthy() || is_active()) { |
7c673cae FG |
5346 | utime_t now = ceph_clock_now(); |
5347 | if (last_heartbeat_resample == utime_t()) { | |
5348 | last_heartbeat_resample = now; | |
5349 | heartbeat_set_peers_need_update(); | |
5350 | } else if (!heartbeat_peers_need_update()) { | |
5351 | utime_t dur = now - last_heartbeat_resample; | |
5352 | if (dur > cct->_conf->osd_heartbeat_grace) { | |
5353 | dout(10) << "maybe_update_heartbeat_peers forcing update after " << dur << " seconds" << dendl; | |
5354 | heartbeat_set_peers_need_update(); | |
5355 | last_heartbeat_resample = now; | |
494da23a TL |
5356 | // automatically clean up any stale heartbeat peers |
5357 | // if we are unhealthy, then clean all | |
5358 | reset_heartbeat_peers(is_waiting_for_healthy()); | |
7c673cae FG |
5359 | } |
5360 | } | |
5361 | } | |
5362 | ||
5363 | if (!heartbeat_peers_need_update()) | |
5364 | return; | |
5365 | heartbeat_clear_peers_need_update(); | |
5366 | ||
11fdf7f2 | 5367 | std::lock_guard l(heartbeat_lock); |
7c673cae FG |
5368 | |
5369 | dout(10) << "maybe_update_heartbeat_peers updating" << dendl; | |
5370 | ||
5371 | ||
5372 | // build heartbeat from set | |
5373 | if (is_active()) { | |
11fdf7f2 TL |
5374 | vector<PGRef> pgs; |
5375 | _get_pgs(&pgs); | |
5376 | for (auto& pg : pgs) { | |
5377 | pg->with_heartbeat_peers([&](int peer) { | |
9f95a23c | 5378 | if (get_osdmap()->is_up(peer)) { |
11fdf7f2 TL |
5379 | _add_heartbeat_peer(peer); |
5380 | } | |
5381 | }); | |
7c673cae FG |
5382 | } |
5383 | } | |
5384 | ||
5385 | // include next and previous up osds to ensure we have a fully-connected set | |
5386 | set<int> want, extras; | |
9f95a23c | 5387 | const int next = get_osdmap()->get_next_up_osd_after(whoami); |
7c673cae FG |
5388 | if (next >= 0) |
5389 | want.insert(next); | |
9f95a23c | 5390 | int prev = get_osdmap()->get_previous_up_osd_before(whoami); |
7c673cae FG |
5391 | if (prev >= 0 && prev != next) |
5392 | want.insert(prev); | |
5393 | ||
11fdf7f2 TL |
5394 | // make sure we have at least **min_down** osds coming from different |
5395 | // subtree level (e.g., hosts) for fast failure detection. | |
5396 | auto min_down = cct->_conf.get_val<uint64_t>("mon_osd_min_down_reporters"); | |
5397 | auto subtree = cct->_conf.get_val<string>("mon_osd_reporter_subtree_level"); | |
9f95a23c TL |
5398 | auto limit = std::max(min_down, (uint64_t)cct->_conf->osd_heartbeat_min_peers); |
5399 | get_osdmap()->get_random_up_osds_by_subtree( | |
5400 | whoami, subtree, limit, want, &want); | |
11fdf7f2 | 5401 | |
7c673cae FG |
5402 | for (set<int>::iterator p = want.begin(); p != want.end(); ++p) { |
5403 | dout(10) << " adding neighbor peer osd." << *p << dendl; | |
5404 | extras.insert(*p); | |
5405 | _add_heartbeat_peer(*p); | |
5406 | } | |
5407 | ||
5408 | // remove down peers; enumerate extras | |
5409 | map<int,HeartbeatInfo>::iterator p = heartbeat_peers.begin(); | |
5410 | while (p != heartbeat_peers.end()) { | |
9f95a23c | 5411 | if (!get_osdmap()->is_up(p->first)) { |
7c673cae FG |
5412 | int o = p->first; |
5413 | ++p; | |
5414 | _remove_heartbeat_peer(o); | |
5415 | continue; | |
5416 | } | |
9f95a23c | 5417 | if (p->second.epoch < get_osdmap_epoch()) { |
7c673cae FG |
5418 | extras.insert(p->first); |
5419 | } | |
5420 | ++p; | |
5421 | } | |
5422 | ||
5423 | // too few? | |
11fdf7f2 | 5424 | for (int n = next; n >= 0; ) { |
7c673cae FG |
5425 | if ((int)heartbeat_peers.size() >= cct->_conf->osd_heartbeat_min_peers) |
5426 | break; | |
5427 | if (!extras.count(n) && !want.count(n) && n != whoami) { | |
5428 | dout(10) << " adding random peer osd." << n << dendl; | |
5429 | extras.insert(n); | |
5430 | _add_heartbeat_peer(n); | |
5431 | } | |
9f95a23c | 5432 | n = get_osdmap()->get_next_up_osd_after(n); |
11fdf7f2 | 5433 | if (n == next) |
7c673cae FG |
5434 | break; // came full circle; stop |
5435 | } | |
5436 | ||
5437 | // too many? | |
5438 | for (set<int>::iterator p = extras.begin(); | |
5439 | (int)heartbeat_peers.size() > cct->_conf->osd_heartbeat_min_peers && p != extras.end(); | |
5440 | ++p) { | |
5441 | if (want.count(*p)) | |
5442 | continue; | |
5443 | _remove_heartbeat_peer(*p); | |
5444 | } | |
5445 | ||
5446 | dout(10) << "maybe_update_heartbeat_peers " << heartbeat_peers.size() << " peers, extras " << extras << dendl; | |
9f95a23c TL |
5447 | |
5448 | // clean up stale failure pending | |
5449 | for (auto it = failure_pending.begin(); it != failure_pending.end();) { | |
5450 | if (heartbeat_peers.count(it->first) == 0) { | |
5451 | send_still_alive(get_osdmap_epoch(), it->first, it->second.second); | |
5452 | failure_pending.erase(it++); | |
5453 | } else { | |
5454 | it++; | |
5455 | } | |
5456 | } | |
7c673cae FG |
5457 | } |
5458 | ||
494da23a | 5459 | void OSD::reset_heartbeat_peers(bool all) |
7c673cae | 5460 | { |
9f95a23c | 5461 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae | 5462 | dout(10) << "reset_heartbeat_peers" << dendl; |
494da23a TL |
5463 | utime_t stale = ceph_clock_now(); |
5464 | stale -= cct->_conf.get_val<int64_t>("osd_heartbeat_stale"); | |
11fdf7f2 | 5465 | std::lock_guard l(heartbeat_lock); |
494da23a | 5466 | for (auto it = heartbeat_peers.begin(); it != heartbeat_peers.end();) { |
b3b6e05e | 5467 | auto& [peer, hi] = *it; |
494da23a | 5468 | if (all || hi.is_stale(stale)) { |
9f95a23c | 5469 | hi.clear_mark_down(); |
494da23a | 5470 | // stop sending failure_report to mon too |
b3b6e05e TL |
5471 | failure_queue.erase(peer); |
5472 | failure_pending.erase(peer); | |
5473 | it = heartbeat_peers.erase(it); | |
494da23a | 5474 | } else { |
b3b6e05e | 5475 | ++it; |
7c673cae | 5476 | } |
7c673cae | 5477 | } |
7c673cae FG |
5478 | } |
5479 | ||
5480 | void OSD::handle_osd_ping(MOSDPing *m) | |
5481 | { | |
5482 | if (superblock.cluster_fsid != m->fsid) { | |
5483 | dout(20) << "handle_osd_ping from " << m->get_source_inst() | |
9f95a23c TL |
5484 | << " bad fsid " << m->fsid << " != " << superblock.cluster_fsid |
5485 | << dendl; | |
7c673cae FG |
5486 | m->put(); |
5487 | return; | |
5488 | } | |
5489 | ||
5490 | int from = m->get_source().num(); | |
5491 | ||
9f95a23c | 5492 | heartbeat_lock.lock(); |
7c673cae | 5493 | if (is_stopping()) { |
9f95a23c | 5494 | heartbeat_lock.unlock(); |
7c673cae FG |
5495 | m->put(); |
5496 | return; | |
5497 | } | |
5498 | ||
9f95a23c TL |
5499 | utime_t now = ceph_clock_now(); |
5500 | auto mnow = service.get_mnow(); | |
5501 | ConnectionRef con(m->get_connection()); | |
7c673cae | 5502 | OSDMapRef curmap = service.get_osdmap(); |
c07f9fc5 | 5503 | if (!curmap) { |
9f95a23c | 5504 | heartbeat_lock.unlock(); |
c07f9fc5 FG |
5505 | m->put(); |
5506 | return; | |
5507 | } | |
7c673cae | 5508 | |
9f95a23c TL |
5509 | auto sref = con->get_priv(); |
5510 | Session *s = static_cast<Session*>(sref.get()); | |
5511 | if (!s) { | |
5512 | heartbeat_lock.unlock(); | |
5513 | m->put(); | |
5514 | return; | |
5515 | } | |
5516 | if (!s->stamps) { | |
5517 | s->peer = from; | |
5518 | s->stamps = service.get_hb_stamps(from); | |
5519 | } | |
5520 | ||
7c673cae FG |
5521 | switch (m->op) { |
5522 | ||
5523 | case MOSDPing::PING: | |
5524 | { | |
5525 | if (cct->_conf->osd_debug_drop_ping_probability > 0) { | |
5526 | auto heartbeat_drop = debug_heartbeat_drops_remaining.find(from); | |
5527 | if (heartbeat_drop != debug_heartbeat_drops_remaining.end()) { | |
5528 | if (heartbeat_drop->second == 0) { | |
5529 | debug_heartbeat_drops_remaining.erase(heartbeat_drop); | |
5530 | } else { | |
5531 | --heartbeat_drop->second; | |
5532 | dout(5) << "Dropping heartbeat from " << from | |
5533 | << ", " << heartbeat_drop->second | |
5534 | << " remaining to drop" << dendl; | |
5535 | break; | |
5536 | } | |
5537 | } else if (cct->_conf->osd_debug_drop_ping_probability > | |
5538 | ((((double)(rand()%100))/100.0))) { | |
5539 | heartbeat_drop = | |
5540 | debug_heartbeat_drops_remaining.insert(std::make_pair(from, | |
5541 | cct->_conf->osd_debug_drop_ping_duration)).first; | |
5542 | dout(5) << "Dropping heartbeat from " << from | |
5543 | << ", " << heartbeat_drop->second | |
5544 | << " remaining to drop" << dendl; | |
5545 | break; | |
5546 | } | |
5547 | } | |
5548 | ||
9f95a23c TL |
5549 | ceph::signedspan sender_delta_ub{}; |
5550 | s->stamps->got_ping( | |
5551 | m->up_from, | |
5552 | mnow, | |
5553 | m->mono_send_stamp, | |
5554 | m->delta_ub, | |
5555 | &sender_delta_ub); | |
5556 | dout(20) << __func__ << " new stamps " << *s->stamps << dendl; | |
5557 | ||
7c673cae | 5558 | if (!cct->get_heartbeat_map()->is_healthy()) { |
9f95a23c TL |
5559 | dout(10) << "internal heartbeat not healthy, dropping ping request" |
5560 | << dendl; | |
7c673cae FG |
5561 | break; |
5562 | } | |
5563 | ||
5564 | Message *r = new MOSDPing(monc->get_fsid(), | |
5565 | curmap->get_epoch(), | |
9f95a23c TL |
5566 | MOSDPing::PING_REPLY, |
5567 | m->ping_stamp, | |
5568 | m->mono_ping_stamp, | |
5569 | mnow, | |
5570 | service.get_up_epoch(), | |
5571 | cct->_conf->osd_heartbeat_min_size, | |
5572 | sender_delta_ub); | |
5573 | con->send_message(r); | |
7c673cae FG |
5574 | |
5575 | if (curmap->is_up(from)) { | |
7c673cae | 5576 | if (is_active()) { |
9f95a23c TL |
5577 | ConnectionRef cluster_con = service.get_con_osd_cluster( |
5578 | from, curmap->get_epoch()); | |
5579 | if (cluster_con) { | |
5580 | service.maybe_share_map(cluster_con.get(), curmap, m->map_epoch); | |
7c673cae FG |
5581 | } |
5582 | } | |
5583 | } else if (!curmap->exists(from) || | |
5584 | curmap->get_down_at(from) > m->map_epoch) { | |
5585 | // tell them they have died | |
5586 | Message *r = new MOSDPing(monc->get_fsid(), | |
5587 | curmap->get_epoch(), | |
5588 | MOSDPing::YOU_DIED, | |
9f95a23c TL |
5589 | m->ping_stamp, |
5590 | m->mono_ping_stamp, | |
5591 | mnow, | |
5592 | service.get_up_epoch(), | |
31f18b77 | 5593 | cct->_conf->osd_heartbeat_min_size); |
9f95a23c | 5594 | con->send_message(r); |
7c673cae FG |
5595 | } |
5596 | } | |
5597 | break; | |
5598 | ||
5599 | case MOSDPing::PING_REPLY: | |
5600 | { | |
5601 | map<int,HeartbeatInfo>::iterator i = heartbeat_peers.find(from); | |
5602 | if (i != heartbeat_peers.end()) { | |
9f95a23c | 5603 | auto acked = i->second.ping_history.find(m->ping_stamp); |
11fdf7f2 | 5604 | if (acked != i->second.ping_history.end()) { |
11fdf7f2 | 5605 | int &unacknowledged = acked->second.second; |
9f95a23c | 5606 | if (con == i->second.con_back) { |
11fdf7f2 TL |
5607 | dout(25) << "handle_osd_ping got reply from osd." << from |
5608 | << " first_tx " << i->second.first_tx | |
5609 | << " last_tx " << i->second.last_tx | |
9f95a23c TL |
5610 | << " last_rx_back " << i->second.last_rx_back |
5611 | << " -> " << now | |
11fdf7f2 TL |
5612 | << " last_rx_front " << i->second.last_rx_front |
5613 | << dendl; | |
5614 | i->second.last_rx_back = now; | |
5615 | ceph_assert(unacknowledged > 0); | |
5616 | --unacknowledged; | |
5617 | // if there is no front con, set both stamps. | |
5618 | if (i->second.con_front == NULL) { | |
5619 | i->second.last_rx_front = now; | |
5620 | ceph_assert(unacknowledged > 0); | |
5621 | --unacknowledged; | |
5622 | } | |
9f95a23c | 5623 | } else if (con == i->second.con_front) { |
11fdf7f2 TL |
5624 | dout(25) << "handle_osd_ping got reply from osd." << from |
5625 | << " first_tx " << i->second.first_tx | |
5626 | << " last_tx " << i->second.last_tx | |
5627 | << " last_rx_back " << i->second.last_rx_back | |
9f95a23c TL |
5628 | << " last_rx_front " << i->second.last_rx_front |
5629 | << " -> " << now | |
11fdf7f2 TL |
5630 | << dendl; |
5631 | i->second.last_rx_front = now; | |
5632 | ceph_assert(unacknowledged > 0); | |
5633 | --unacknowledged; | |
5634 | } | |
7c673cae | 5635 | |
11fdf7f2 TL |
5636 | if (unacknowledged == 0) { |
5637 | // succeeded in getting all replies | |
5638 | dout(25) << "handle_osd_ping got all replies from osd." << from | |
9f95a23c | 5639 | << " , erase pending ping(sent at " << m->ping_stamp << ")" |
11fdf7f2 TL |
5640 | << " and older pending ping(s)" |
5641 | << dendl; | |
eafe8130 TL |
5642 | |
5643 | #define ROUND_S_TO_USEC(sec) (uint32_t)((sec) * 1000 * 1000 + 0.5) | |
5644 | ++i->second.hb_average_count; | |
9f95a23c | 5645 | uint32_t back_pingtime = ROUND_S_TO_USEC(i->second.last_rx_back - m->ping_stamp); |
eafe8130 TL |
5646 | i->second.hb_total_back += back_pingtime; |
5647 | if (back_pingtime < i->second.hb_min_back) | |
5648 | i->second.hb_min_back = back_pingtime; | |
5649 | if (back_pingtime > i->second.hb_max_back) | |
5650 | i->second.hb_max_back = back_pingtime; | |
9f95a23c | 5651 | uint32_t front_pingtime = ROUND_S_TO_USEC(i->second.last_rx_front - m->ping_stamp); |
eafe8130 TL |
5652 | i->second.hb_total_front += front_pingtime; |
5653 | if (front_pingtime < i->second.hb_min_front) | |
5654 | i->second.hb_min_front = front_pingtime; | |
5655 | if (front_pingtime > i->second.hb_max_front) | |
5656 | i->second.hb_max_front = front_pingtime; | |
5657 | ||
5658 | ceph_assert(i->second.hb_interval_start != utime_t()); | |
5659 | if (i->second.hb_interval_start == utime_t()) | |
5660 | i->second.hb_interval_start = now; | |
5661 | int64_t hb_avg_time_period = 60; | |
5662 | if (cct->_conf.get_val<int64_t>("debug_heartbeat_testing_span")) { | |
5663 | hb_avg_time_period = cct->_conf.get_val<int64_t>("debug_heartbeat_testing_span"); | |
5664 | } | |
5665 | if (now - i->second.hb_interval_start >= utime_t(hb_avg_time_period, 0)) { | |
5666 | uint32_t back_avg = i->second.hb_total_back / i->second.hb_average_count; | |
5667 | uint32_t back_min = i->second.hb_min_back; | |
5668 | uint32_t back_max = i->second.hb_max_back; | |
5669 | uint32_t front_avg = i->second.hb_total_front / i->second.hb_average_count; | |
5670 | uint32_t front_min = i->second.hb_min_front; | |
5671 | uint32_t front_max = i->second.hb_max_front; | |
5672 | ||
5673 | // Reset for new interval | |
5674 | i->second.hb_average_count = 0; | |
5675 | i->second.hb_interval_start = now; | |
5676 | i->second.hb_total_back = i->second.hb_max_back = 0; | |
5677 | i->second.hb_min_back = UINT_MAX; | |
5678 | i->second.hb_total_front = i->second.hb_max_front = 0; | |
5679 | i->second.hb_min_front = UINT_MAX; | |
5680 | ||
5681 | // Record per osd interace ping times | |
5682 | // Based on osd_heartbeat_interval ignoring that it is randomly short than this interval | |
5683 | if (i->second.hb_back_pingtime.size() == 0) { | |
5684 | ceph_assert(i->second.hb_front_pingtime.size() == 0); | |
5685 | for (unsigned k = 0 ; k < hb_vector_size; ++k) { | |
5686 | i->second.hb_back_pingtime.push_back(back_avg); | |
5687 | i->second.hb_back_min.push_back(back_min); | |
5688 | i->second.hb_back_max.push_back(back_max); | |
5689 | i->second.hb_front_pingtime.push_back(front_avg); | |
5690 | i->second.hb_front_min.push_back(front_min); | |
5691 | i->second.hb_front_max.push_back(front_max); | |
5692 | ++i->second.hb_index; | |
5693 | } | |
5694 | } else { | |
5695 | int index = i->second.hb_index & (hb_vector_size - 1); | |
5696 | i->second.hb_back_pingtime[index] = back_avg; | |
5697 | i->second.hb_back_min[index] = back_min; | |
5698 | i->second.hb_back_max[index] = back_max; | |
5699 | i->second.hb_front_pingtime[index] = front_avg; | |
5700 | i->second.hb_front_min[index] = front_min; | |
5701 | i->second.hb_front_max[index] = front_max; | |
5702 | ++i->second.hb_index; | |
5703 | } | |
5704 | ||
5705 | { | |
5706 | std::lock_guard l(service.stat_lock); | |
5707 | service.osd_stat.hb_pingtime[from].last_update = now.sec(); | |
5708 | service.osd_stat.hb_pingtime[from].back_last = back_pingtime; | |
5709 | ||
5710 | uint32_t total = 0; | |
5711 | uint32_t min = UINT_MAX; | |
5712 | uint32_t max = 0; | |
5713 | uint32_t count = 0; | |
5714 | uint32_t which = 0; | |
5715 | uint32_t size = (uint32_t)i->second.hb_back_pingtime.size(); | |
5716 | for (int32_t k = size - 1 ; k >= 0; --k) { | |
5717 | ++count; | |
5718 | int index = (i->second.hb_index + k) % size; | |
5719 | total += i->second.hb_back_pingtime[index]; | |
5720 | if (i->second.hb_back_min[index] < min) | |
5721 | min = i->second.hb_back_min[index]; | |
5722 | if (i->second.hb_back_max[index] > max) | |
5723 | max = i->second.hb_back_max[index]; | |
5724 | if (count == 1 || count == 5 || count == 15) { | |
5725 | service.osd_stat.hb_pingtime[from].back_pingtime[which] = total / count; | |
5726 | service.osd_stat.hb_pingtime[from].back_min[which] = min; | |
5727 | service.osd_stat.hb_pingtime[from].back_max[which] = max; | |
5728 | which++; | |
5729 | if (count == 15) | |
5730 | break; | |
5731 | } | |
5732 | } | |
5733 | ||
5734 | if (i->second.con_front != NULL) { | |
5735 | service.osd_stat.hb_pingtime[from].front_last = front_pingtime; | |
5736 | ||
5737 | total = 0; | |
5738 | min = UINT_MAX; | |
5739 | max = 0; | |
5740 | count = 0; | |
5741 | which = 0; | |
5742 | for (int32_t k = size - 1 ; k >= 0; --k) { | |
5743 | ++count; | |
5744 | int index = (i->second.hb_index + k) % size; | |
5745 | total += i->second.hb_front_pingtime[index]; | |
5746 | if (i->second.hb_front_min[index] < min) | |
5747 | min = i->second.hb_front_min[index]; | |
5748 | if (i->second.hb_front_max[index] > max) | |
5749 | max = i->second.hb_front_max[index]; | |
5750 | if (count == 1 || count == 5 || count == 15) { | |
5751 | service.osd_stat.hb_pingtime[from].front_pingtime[which] = total / count; | |
5752 | service.osd_stat.hb_pingtime[from].front_min[which] = min; | |
5753 | service.osd_stat.hb_pingtime[from].front_max[which] = max; | |
5754 | which++; | |
5755 | if (count == 15) | |
5756 | break; | |
5757 | } | |
5758 | } | |
5759 | } | |
5760 | } | |
5761 | } else { | |
5762 | std::lock_guard l(service.stat_lock); | |
5763 | service.osd_stat.hb_pingtime[from].back_last = back_pingtime; | |
5764 | if (i->second.con_front != NULL) | |
5765 | service.osd_stat.hb_pingtime[from].front_last = front_pingtime; | |
5766 | } | |
11fdf7f2 | 5767 | i->second.ping_history.erase(i->second.ping_history.begin(), ++acked); |
7c673cae FG |
5768 | } |
5769 | ||
11fdf7f2 TL |
5770 | if (i->second.is_healthy(now)) { |
5771 | // Cancel false reports | |
5772 | auto failure_queue_entry = failure_queue.find(from); | |
5773 | if (failure_queue_entry != failure_queue.end()) { | |
5774 | dout(10) << "handle_osd_ping canceling queued " | |
5775 | << "failure report for osd." << from << dendl; | |
5776 | failure_queue.erase(failure_queue_entry); | |
5777 | } | |
5778 | ||
5779 | auto failure_pending_entry = failure_pending.find(from); | |
5780 | if (failure_pending_entry != failure_pending.end()) { | |
5781 | dout(10) << "handle_osd_ping canceling in-flight " | |
5782 | << "failure report for osd." << from << dendl; | |
5783 | send_still_alive(curmap->get_epoch(), | |
5784 | from, | |
5785 | failure_pending_entry->second.second); | |
5786 | failure_pending.erase(failure_pending_entry); | |
5787 | } | |
7c673cae | 5788 | } |
11fdf7f2 TL |
5789 | } else { |
5790 | // old replies, deprecated by newly sent pings. | |
9f95a23c | 5791 | dout(10) << "handle_osd_ping no pending ping(sent at " << m->ping_stamp |
11fdf7f2 TL |
5792 | << ") is found, treat as covered by newly sent pings " |
5793 | << "and ignore" | |
5794 | << dendl; | |
7c673cae FG |
5795 | } |
5796 | } | |
5797 | ||
5798 | if (m->map_epoch && | |
5799 | curmap->is_up(from)) { | |
7c673cae | 5800 | if (is_active()) { |
9f95a23c TL |
5801 | ConnectionRef cluster_con = service.get_con_osd_cluster( |
5802 | from, curmap->get_epoch()); | |
5803 | if (cluster_con) { | |
5804 | service.maybe_share_map(cluster_con.get(), curmap, m->map_epoch); | |
7c673cae FG |
5805 | } |
5806 | } | |
5807 | } | |
9f95a23c TL |
5808 | |
5809 | s->stamps->got_ping_reply( | |
5810 | mnow, | |
5811 | m->mono_send_stamp, | |
5812 | m->delta_ub); | |
5813 | dout(20) << __func__ << " new stamps " << *s->stamps << dendl; | |
7c673cae FG |
5814 | } |
5815 | break; | |
5816 | ||
5817 | case MOSDPing::YOU_DIED: | |
5818 | dout(10) << "handle_osd_ping " << m->get_source_inst() | |
5819 | << " says i am down in " << m->map_epoch << dendl; | |
5820 | osdmap_subscribe(curmap->get_epoch()+1, false); | |
5821 | break; | |
5822 | } | |
5823 | ||
9f95a23c | 5824 | heartbeat_lock.unlock(); |
7c673cae FG |
5825 | m->put(); |
5826 | } | |
5827 | ||
5828 | void OSD::heartbeat_entry() | |
5829 | { | |
9f95a23c | 5830 | std::unique_lock l(heartbeat_lock); |
7c673cae FG |
5831 | if (is_stopping()) |
5832 | return; | |
5833 | while (!heartbeat_stop) { | |
5834 | heartbeat(); | |
5835 | ||
eafe8130 TL |
5836 | double wait; |
5837 | if (cct->_conf.get_val<bool>("debug_disable_randomized_ping")) { | |
5838 | wait = (float)cct->_conf->osd_heartbeat_interval; | |
5839 | } else { | |
5840 | wait = .5 + ((float)(rand() % 10)/10.0) * (float)cct->_conf->osd_heartbeat_interval; | |
5841 | } | |
9f95a23c | 5842 | auto w = ceph::make_timespan(wait); |
7c673cae | 5843 | dout(30) << "heartbeat_entry sleeping for " << wait << dendl; |
9f95a23c | 5844 | heartbeat_cond.wait_for(l, w); |
7c673cae FG |
5845 | if (is_stopping()) |
5846 | return; | |
5847 | dout(30) << "heartbeat_entry woke up" << dendl; | |
5848 | } | |
5849 | } | |
5850 | ||
5851 | void OSD::heartbeat_check() | |
5852 | { | |
9f95a23c | 5853 | ceph_assert(ceph_mutex_is_locked(heartbeat_lock)); |
7c673cae FG |
5854 | utime_t now = ceph_clock_now(); |
5855 | ||
11fdf7f2 | 5856 | // check for incoming heartbeats (move me elsewhere?) |
7c673cae FG |
5857 | for (map<int,HeartbeatInfo>::iterator p = heartbeat_peers.begin(); |
5858 | p != heartbeat_peers.end(); | |
5859 | ++p) { | |
5860 | ||
5861 | if (p->second.first_tx == utime_t()) { | |
5862 | dout(25) << "heartbeat_check we haven't sent ping to osd." << p->first | |
11fdf7f2 | 5863 | << " yet, skipping" << dendl; |
7c673cae FG |
5864 | continue; |
5865 | } | |
5866 | ||
5867 | dout(25) << "heartbeat_check osd." << p->first | |
5868 | << " first_tx " << p->second.first_tx | |
5869 | << " last_tx " << p->second.last_tx | |
5870 | << " last_rx_back " << p->second.last_rx_back | |
5871 | << " last_rx_front " << p->second.last_rx_front | |
5872 | << dendl; | |
11fdf7f2 TL |
5873 | if (p->second.is_unhealthy(now)) { |
5874 | utime_t oldest_deadline = p->second.ping_history.begin()->second.first; | |
7c673cae FG |
5875 | if (p->second.last_rx_back == utime_t() || |
5876 | p->second.last_rx_front == utime_t()) { | |
11fdf7f2 TL |
5877 | derr << "heartbeat_check: no reply from " |
5878 | << p->second.con_front->get_peer_addr().get_sockaddr() | |
5879 | << " osd." << p->first | |
5880 | << " ever on either front or back, first ping sent " | |
5881 | << p->second.first_tx | |
5882 | << " (oldest deadline " << oldest_deadline << ")" | |
5883 | << dendl; | |
7c673cae | 5884 | // fail |
11fdf7f2 | 5885 | failure_queue[p->first] = p->second.first_tx; |
7c673cae | 5886 | } else { |
11fdf7f2 TL |
5887 | derr << "heartbeat_check: no reply from " |
5888 | << p->second.con_front->get_peer_addr().get_sockaddr() | |
7c673cae FG |
5889 | << " osd." << p->first << " since back " << p->second.last_rx_back |
5890 | << " front " << p->second.last_rx_front | |
11fdf7f2 TL |
5891 | << " (oldest deadline " << oldest_deadline << ")" |
5892 | << dendl; | |
7c673cae | 5893 | // fail |
11fdf7f2 | 5894 | failure_queue[p->first] = std::min(p->second.last_rx_back, p->second.last_rx_front); |
7c673cae FG |
5895 | } |
5896 | } | |
5897 | } | |
5898 | } | |
5899 | ||
5900 | void OSD::heartbeat() | |
5901 | { | |
9f95a23c | 5902 | ceph_assert(ceph_mutex_is_locked_by_me(heartbeat_lock)); |
7c673cae FG |
5903 | dout(30) << "heartbeat" << dendl; |
5904 | ||
20effc67 TL |
5905 | auto load_for_logger = service.get_scrub_services().update_load_average(); |
5906 | if (load_for_logger) { | |
5907 | logger->set(l_osd_loadavg, load_for_logger.value()); | |
7c673cae | 5908 | } |
7c673cae FG |
5909 | dout(30) << "heartbeat checking stats" << dendl; |
5910 | ||
11fdf7f2 | 5911 | // refresh peer list and osd stats |
7c673cae FG |
5912 | vector<int> hb_peers; |
5913 | for (map<int,HeartbeatInfo>::iterator p = heartbeat_peers.begin(); | |
5914 | p != heartbeat_peers.end(); | |
5915 | ++p) | |
5916 | hb_peers.push_back(p->first); | |
7c673cae | 5917 | |
11fdf7f2 TL |
5918 | auto new_stat = service.set_osd_stat(hb_peers, get_num_pgs()); |
5919 | dout(5) << __func__ << " " << new_stat << dendl; | |
5920 | ceph_assert(new_stat.statfs.total); | |
5921 | ||
5922 | float pratio; | |
5923 | float ratio = service.compute_adjusted_ratio(new_stat, &pratio); | |
5924 | ||
5925 | service.check_full_status(ratio, pratio); | |
7c673cae FG |
5926 | |
5927 | utime_t now = ceph_clock_now(); | |
9f95a23c | 5928 | auto mnow = service.get_mnow(); |
11fdf7f2 TL |
5929 | utime_t deadline = now; |
5930 | deadline += cct->_conf->osd_heartbeat_grace; | |
7c673cae FG |
5931 | |
5932 | // send heartbeats | |
5933 | for (map<int,HeartbeatInfo>::iterator i = heartbeat_peers.begin(); | |
5934 | i != heartbeat_peers.end(); | |
5935 | ++i) { | |
5936 | int peer = i->first; | |
f67539c2 TL |
5937 | Session *s = static_cast<Session*>(i->second.con_back->get_priv().get()); |
5938 | if (!s) { | |
5939 | dout(30) << "heartbeat osd." << peer << " has no open con" << dendl; | |
5940 | continue; | |
5941 | } | |
9f95a23c TL |
5942 | dout(30) << "heartbeat sending ping to osd." << peer << dendl; |
5943 | ||
7c673cae FG |
5944 | i->second.last_tx = now; |
5945 | if (i->second.first_tx == utime_t()) | |
5946 | i->second.first_tx = now; | |
11fdf7f2 TL |
5947 | i->second.ping_history[now] = make_pair(deadline, |
5948 | HeartbeatInfo::HEARTBEAT_MAX_CONN); | |
eafe8130 TL |
5949 | if (i->second.hb_interval_start == utime_t()) |
5950 | i->second.hb_interval_start = now; | |
9f95a23c | 5951 | |
9f95a23c TL |
5952 | std::optional<ceph::signedspan> delta_ub; |
5953 | s->stamps->sent_ping(&delta_ub); | |
5954 | ||
5955 | i->second.con_back->send_message( | |
5956 | new MOSDPing(monc->get_fsid(), | |
5957 | service.get_osdmap_epoch(), | |
5958 | MOSDPing::PING, | |
5959 | now, | |
5960 | mnow, | |
5961 | mnow, | |
5962 | service.get_up_epoch(), | |
5963 | cct->_conf->osd_heartbeat_min_size, | |
5964 | delta_ub)); | |
7c673cae FG |
5965 | |
5966 | if (i->second.con_front) | |
9f95a23c TL |
5967 | i->second.con_front->send_message( |
5968 | new MOSDPing(monc->get_fsid(), | |
5969 | service.get_osdmap_epoch(), | |
5970 | MOSDPing::PING, | |
5971 | now, | |
5972 | mnow, | |
5973 | mnow, | |
5974 | service.get_up_epoch(), | |
5975 | cct->_conf->osd_heartbeat_min_size, | |
5976 | delta_ub)); | |
7c673cae FG |
5977 | } |
5978 | ||
5979 | logger->set(l_osd_hb_to, heartbeat_peers.size()); | |
5980 | ||
5981 | // hmm.. am i all alone? | |
5982 | dout(30) << "heartbeat lonely?" << dendl; | |
5983 | if (heartbeat_peers.empty()) { | |
5984 | if (now - last_mon_heartbeat > cct->_conf->osd_mon_heartbeat_interval && is_active()) { | |
5985 | last_mon_heartbeat = now; | |
5986 | dout(10) << "i have no heartbeat peers; checking mon for new map" << dendl; | |
9f95a23c | 5987 | osdmap_subscribe(get_osdmap_epoch() + 1, false); |
7c673cae FG |
5988 | } |
5989 | } | |
5990 | ||
5991 | dout(30) << "heartbeat done" << dendl; | |
5992 | } | |
5993 | ||
5994 | bool OSD::heartbeat_reset(Connection *con) | |
5995 | { | |
11fdf7f2 TL |
5996 | std::lock_guard l(heartbeat_lock); |
5997 | auto s = con->get_priv(); | |
9f95a23c | 5998 | dout(20) << __func__ << " con " << con << " s " << s.get() << dendl; |
11fdf7f2 | 5999 | con->set_priv(nullptr); |
7c673cae | 6000 | if (s) { |
7c673cae | 6001 | if (is_stopping()) { |
7c673cae FG |
6002 | return true; |
6003 | } | |
9f95a23c TL |
6004 | auto session = static_cast<Session*>(s.get()); |
6005 | auto p = heartbeat_peers.find(session->peer); | |
7c673cae FG |
6006 | if (p != heartbeat_peers.end() && |
6007 | (p->second.con_back == con || | |
6008 | p->second.con_front == con)) { | |
6009 | dout(10) << "heartbeat_reset failed hb con " << con << " for osd." << p->second.peer | |
6010 | << ", reopening" << dendl; | |
9f95a23c | 6011 | p->second.clear_mark_down(con); |
7c673cae FG |
6012 | pair<ConnectionRef,ConnectionRef> newcon = service.get_con_osd_hb(p->second.peer, p->second.epoch); |
6013 | if (newcon.first) { | |
6014 | p->second.con_back = newcon.first.get(); | |
11fdf7f2 | 6015 | p->second.con_back->set_priv(s); |
7c673cae FG |
6016 | if (newcon.second) { |
6017 | p->second.con_front = newcon.second.get(); | |
11fdf7f2 | 6018 | p->second.con_front->set_priv(s); |
7c673cae | 6019 | } |
11fdf7f2 | 6020 | p->second.ping_history.clear(); |
7c673cae FG |
6021 | } else { |
6022 | dout(10) << "heartbeat_reset failed hb con " << con << " for osd." << p->second.peer | |
6023 | << ", raced with osdmap update, closing out peer" << dendl; | |
6024 | heartbeat_peers.erase(p); | |
6025 | } | |
6026 | } else { | |
6027 | dout(10) << "heartbeat_reset closing (old) failed hb con " << con << dendl; | |
6028 | } | |
7c673cae FG |
6029 | } |
6030 | return true; | |
6031 | } | |
6032 | ||
6033 | ||
6034 | ||
6035 | // ========================================= | |
6036 | ||
6037 | void OSD::tick() | |
6038 | { | |
9f95a23c | 6039 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae FG |
6040 | dout(10) << "tick" << dendl; |
6041 | ||
9f95a23c TL |
6042 | utime_t now = ceph_clock_now(); |
6043 | // throw out any obsolete markdown log | |
6044 | utime_t grace = utime_t(cct->_conf->osd_max_markdown_period, 0); | |
6045 | while (!osd_markdown_log.empty() && | |
6046 | osd_markdown_log.front() + grace < now) | |
6047 | osd_markdown_log.pop_front(); | |
6048 | ||
7c673cae FG |
6049 | if (is_active() || is_waiting_for_healthy()) { |
6050 | maybe_update_heartbeat_peers(); | |
6051 | } | |
6052 | ||
6053 | if (is_waiting_for_healthy()) { | |
6054 | start_boot(); | |
494da23a TL |
6055 | } |
6056 | ||
6057 | if (is_waiting_for_healthy() || is_booting()) { | |
6058 | std::lock_guard l(heartbeat_lock); | |
494da23a TL |
6059 | if (now - last_mon_heartbeat > cct->_conf->osd_mon_heartbeat_interval) { |
6060 | last_mon_heartbeat = now; | |
6061 | dout(1) << __func__ << " checking mon for new map" << dendl; | |
9f95a23c | 6062 | osdmap_subscribe(get_osdmap_epoch() + 1, false); |
11fdf7f2 | 6063 | } |
7c673cae FG |
6064 | } |
6065 | ||
6066 | do_waiters(); | |
6067 | ||
9f95a23c TL |
6068 | // scrub purged_snaps every deep scrub interval |
6069 | { | |
6070 | const utime_t last = superblock.last_purged_snaps_scrub; | |
6071 | utime_t next = last; | |
6072 | next += cct->_conf->osd_scrub_min_interval; | |
6073 | std::mt19937 rng; | |
6074 | // use a seed that is stable for each scrub interval, but varies | |
6075 | // by OSD to avoid any herds. | |
6076 | rng.seed(whoami + superblock.last_purged_snaps_scrub.sec()); | |
20effc67 | 6077 | double r = (rng() % 1024) / 1024.0; |
9f95a23c TL |
6078 | next += |
6079 | cct->_conf->osd_scrub_min_interval * | |
6080 | cct->_conf->osd_scrub_interval_randomize_ratio * r; | |
6081 | if (next < ceph_clock_now()) { | |
6082 | dout(20) << __func__ << " last_purged_snaps_scrub " << last | |
6083 | << " next " << next << " ... now" << dendl; | |
6084 | scrub_purged_snaps(); | |
6085 | } else { | |
6086 | dout(20) << __func__ << " last_purged_snaps_scrub " << last | |
6087 | << " next " << next << dendl; | |
6088 | } | |
6089 | } | |
6090 | ||
91327a77 | 6091 | tick_timer.add_event_after(get_tick_interval(), new C_Tick(this)); |
7c673cae FG |
6092 | } |
6093 | ||
6094 | void OSD::tick_without_osd_lock() | |
6095 | { | |
9f95a23c | 6096 | ceph_assert(ceph_mutex_is_locked(tick_timer_lock)); |
7c673cae FG |
6097 | dout(10) << "tick_without_osd_lock" << dendl; |
6098 | ||
f67539c2 TL |
6099 | logger->set(l_osd_cached_crc, ceph::buffer::get_cached_crc()); |
6100 | logger->set(l_osd_cached_crc_adjusted, ceph::buffer::get_cached_crc_adjusted()); | |
6101 | logger->set(l_osd_missed_crc, ceph::buffer::get_missed_crc()); | |
11fdf7f2 TL |
6102 | |
6103 | // refresh osd stats | |
6104 | struct store_statfs_t stbuf; | |
6105 | osd_alert_list_t alerts; | |
6106 | int r = store->statfs(&stbuf, &alerts); | |
6107 | ceph_assert(r == 0); | |
6108 | service.set_statfs(stbuf, alerts); | |
7c673cae FG |
6109 | |
6110 | // osd_lock is not being held, which means the OSD state | |
6111 | // might change when doing the monitor report | |
6112 | if (is_active() || is_waiting_for_healthy()) { | |
9f95a23c TL |
6113 | { |
6114 | std::lock_guard l{heartbeat_lock}; | |
6115 | heartbeat_check(); | |
6116 | } | |
6117 | map_lock.lock_shared(); | |
11fdf7f2 | 6118 | std::lock_guard l(mon_report_lock); |
7c673cae FG |
6119 | |
6120 | // mon report? | |
7c673cae | 6121 | utime_t now = ceph_clock_now(); |
11fdf7f2 TL |
6122 | if (service.need_fullness_update() || |
6123 | now - last_mon_report > cct->_conf->osd_mon_report_interval) { | |
7c673cae | 6124 | last_mon_report = now; |
7c673cae FG |
6125 | send_full_update(); |
6126 | send_failures(); | |
7c673cae | 6127 | } |
9f95a23c | 6128 | map_lock.unlock_shared(); |
11fdf7f2 TL |
6129 | |
6130 | epoch_t max_waiting_epoch = 0; | |
6131 | for (auto s : shards) { | |
6132 | max_waiting_epoch = std::max(max_waiting_epoch, | |
6133 | s->get_max_waiting_epoch()); | |
6134 | } | |
6135 | if (max_waiting_epoch > get_osdmap()->get_epoch()) { | |
6136 | dout(20) << __func__ << " max_waiting_epoch " << max_waiting_epoch | |
6137 | << ", requesting new map" << dendl; | |
6138 | osdmap_subscribe(superblock.newest_map + 1, false); | |
6139 | } | |
7c673cae FG |
6140 | } |
6141 | ||
6142 | if (is_active()) { | |
6143 | if (!scrub_random_backoff()) { | |
6144 | sched_scrub(); | |
6145 | } | |
6146 | service.promote_throttle_recalibrate(); | |
3efd9988 | 6147 | resume_creating_pg(); |
224ce89b WB |
6148 | bool need_send_beacon = false; |
6149 | const auto now = ceph::coarse_mono_clock::now(); | |
6150 | { | |
6151 | // borrow lec lock to pretect last_sent_beacon from changing | |
11fdf7f2 | 6152 | std::lock_guard l{min_last_epoch_clean_lock}; |
224ce89b | 6153 | const auto elapsed = now - last_sent_beacon; |
f67539c2 | 6154 | if (std::chrono::duration_cast<std::chrono::seconds>(elapsed).count() > |
224ce89b WB |
6155 | cct->_conf->osd_beacon_report_interval) { |
6156 | need_send_beacon = true; | |
6157 | } | |
6158 | } | |
6159 | if (need_send_beacon) { | |
6160 | send_beacon(now); | |
6161 | } | |
7c673cae FG |
6162 | } |
6163 | ||
11fdf7f2 | 6164 | mgrc.update_daemon_health(get_health_metrics()); |
7c673cae | 6165 | service.kick_recovery_queue(); |
91327a77 AA |
6166 | tick_timer_without_osd_lock.add_event_after(get_tick_interval(), |
6167 | new C_Tick_WithoutOSDLock(this)); | |
7c673cae FG |
6168 | } |
6169 | ||
7c673cae FG |
6170 | // Usage: |
6171 | // setomapval <pool-id> [namespace/]<obj-name> <key> <val> | |
6172 | // rmomapkey <pool-id> [namespace/]<obj-name> <key> | |
6173 | // setomapheader <pool-id> [namespace/]<obj-name> <header> | |
6174 | // getomap <pool> [namespace/]<obj-name> | |
6175 | // truncobj <pool-id> [namespace/]<obj-name> <newlen> | |
6176 | // injectmdataerr [namespace/]<obj-name> [shardid] | |
6177 | // injectdataerr [namespace/]<obj-name> [shardid] | |
6178 | // | |
6179 | // set_recovery_delay [utime] | |
6180 | void TestOpsSocketHook::test_ops(OSDService *service, ObjectStore *store, | |
11fdf7f2 TL |
6181 | std::string_view command, |
6182 | const cmdmap_t& cmdmap, ostream &ss) | |
7c673cae FG |
6183 | { |
6184 | //Test support | |
6185 | //Support changing the omap on a single osd by using the Admin Socket to | |
6186 | //directly request the osd make a change. | |
6187 | if (command == "setomapval" || command == "rmomapkey" || | |
6188 | command == "setomapheader" || command == "getomap" || | |
6189 | command == "truncobj" || command == "injectmdataerr" || | |
6190 | command == "injectdataerr" | |
6191 | ) { | |
6192 | pg_t rawpg; | |
6193 | int64_t pool; | |
6194 | OSDMapRef curmap = service->get_osdmap(); | |
6195 | int r = -1; | |
6196 | ||
6197 | string poolstr; | |
6198 | ||
9f95a23c | 6199 | cmd_getval(cmdmap, "pool", poolstr); |
7c673cae FG |
6200 | pool = curmap->lookup_pg_pool_name(poolstr); |
6201 | //If we can't find it by name then maybe id specified | |
6202 | if (pool < 0 && isdigit(poolstr[0])) | |
6203 | pool = atoll(poolstr.c_str()); | |
6204 | if (pool < 0) { | |
b5b8bbf5 | 6205 | ss << "Invalid pool '" << poolstr << "''"; |
7c673cae FG |
6206 | return; |
6207 | } | |
6208 | ||
6209 | string objname, nspace; | |
9f95a23c | 6210 | cmd_getval(cmdmap, "objname", objname); |
7c673cae FG |
6211 | std::size_t found = objname.find_first_of('/'); |
6212 | if (found != string::npos) { | |
6213 | nspace = objname.substr(0, found); | |
6214 | objname = objname.substr(found+1); | |
6215 | } | |
6216 | object_locator_t oloc(pool, nspace); | |
6217 | r = curmap->object_locator_to_pg(object_t(objname), oloc, rawpg); | |
6218 | ||
6219 | if (r < 0) { | |
6220 | ss << "Invalid namespace/objname"; | |
6221 | return; | |
6222 | } | |
6223 | ||
20effc67 | 6224 | int64_t shardid = cmd_getval_or<int64_t>(cmdmap, "shardid", shard_id_t::NO_SHARD); |
7c673cae FG |
6225 | hobject_t obj(object_t(objname), string(""), CEPH_NOSNAP, rawpg.ps(), pool, nspace); |
6226 | ghobject_t gobj(obj, ghobject_t::NO_GEN, shard_id_t(uint8_t(shardid))); | |
6227 | spg_t pgid(curmap->raw_pg_to_pg(rawpg), shard_id_t(shardid)); | |
6228 | if (curmap->pg_is_ec(rawpg)) { | |
6229 | if ((command != "injectdataerr") && (command != "injectmdataerr")) { | |
6230 | ss << "Must not call on ec pool, except injectdataerr or injectmdataerr"; | |
6231 | return; | |
6232 | } | |
6233 | } | |
6234 | ||
6235 | ObjectStore::Transaction t; | |
6236 | ||
6237 | if (command == "setomapval") { | |
6238 | map<string, bufferlist> newattrs; | |
6239 | bufferlist val; | |
6240 | string key, valstr; | |
9f95a23c TL |
6241 | cmd_getval(cmdmap, "key", key); |
6242 | cmd_getval(cmdmap, "val", valstr); | |
7c673cae FG |
6243 | |
6244 | val.append(valstr); | |
6245 | newattrs[key] = val; | |
6246 | t.omap_setkeys(coll_t(pgid), ghobject_t(obj), newattrs); | |
11fdf7f2 | 6247 | r = store->queue_transaction(service->meta_ch, std::move(t)); |
7c673cae FG |
6248 | if (r < 0) |
6249 | ss << "error=" << r; | |
6250 | else | |
6251 | ss << "ok"; | |
6252 | } else if (command == "rmomapkey") { | |
6253 | string key; | |
9f95a23c | 6254 | cmd_getval(cmdmap, "key", key); |
7c673cae | 6255 | |
9f95a23c | 6256 | t.omap_rmkey(coll_t(pgid), ghobject_t(obj), key); |
11fdf7f2 | 6257 | r = store->queue_transaction(service->meta_ch, std::move(t)); |
7c673cae FG |
6258 | if (r < 0) |
6259 | ss << "error=" << r; | |
6260 | else | |
6261 | ss << "ok"; | |
6262 | } else if (command == "setomapheader") { | |
6263 | bufferlist newheader; | |
6264 | string headerstr; | |
6265 | ||
9f95a23c | 6266 | cmd_getval(cmdmap, "header", headerstr); |
7c673cae FG |
6267 | newheader.append(headerstr); |
6268 | t.omap_setheader(coll_t(pgid), ghobject_t(obj), newheader); | |
11fdf7f2 | 6269 | r = store->queue_transaction(service->meta_ch, std::move(t)); |
7c673cae FG |
6270 | if (r < 0) |
6271 | ss << "error=" << r; | |
6272 | else | |
6273 | ss << "ok"; | |
6274 | } else if (command == "getomap") { | |
6275 | //Debug: Output entire omap | |
6276 | bufferlist hdrbl; | |
6277 | map<string, bufferlist> keyvals; | |
11fdf7f2 TL |
6278 | auto ch = store->open_collection(coll_t(pgid)); |
6279 | if (!ch) { | |
6280 | ss << "unable to open collection for " << pgid; | |
6281 | r = -ENOENT; | |
6282 | } else { | |
6283 | r = store->omap_get(ch, ghobject_t(obj), &hdrbl, &keyvals); | |
6284 | if (r >= 0) { | |
7c673cae FG |
6285 | ss << "header=" << string(hdrbl.c_str(), hdrbl.length()); |
6286 | for (map<string, bufferlist>::iterator it = keyvals.begin(); | |
11fdf7f2 | 6287 | it != keyvals.end(); ++it) |
7c673cae FG |
6288 | ss << " key=" << (*it).first << " val=" |
6289 | << string((*it).second.c_str(), (*it).second.length()); | |
11fdf7f2 | 6290 | } else { |
7c673cae | 6291 | ss << "error=" << r; |
11fdf7f2 | 6292 | } |
7c673cae FG |
6293 | } |
6294 | } else if (command == "truncobj") { | |
6295 | int64_t trunclen; | |
9f95a23c | 6296 | cmd_getval(cmdmap, "len", trunclen); |
7c673cae | 6297 | t.truncate(coll_t(pgid), ghobject_t(obj), trunclen); |
11fdf7f2 | 6298 | r = store->queue_transaction(service->meta_ch, std::move(t)); |
7c673cae FG |
6299 | if (r < 0) |
6300 | ss << "error=" << r; | |
6301 | else | |
6302 | ss << "ok"; | |
6303 | } else if (command == "injectdataerr") { | |
6304 | store->inject_data_error(gobj); | |
6305 | ss << "ok"; | |
6306 | } else if (command == "injectmdataerr") { | |
6307 | store->inject_mdata_error(gobj); | |
6308 | ss << "ok"; | |
6309 | } | |
6310 | return; | |
6311 | } | |
6312 | if (command == "set_recovery_delay") { | |
20effc67 | 6313 | int64_t delay = cmd_getval_or<int64_t>(cmdmap, "utime", 0); |
7c673cae FG |
6314 | ostringstream oss; |
6315 | oss << delay; | |
11fdf7f2 | 6316 | int r = service->cct->_conf.set_val("osd_recovery_delay_start", |
7c673cae FG |
6317 | oss.str().c_str()); |
6318 | if (r != 0) { | |
6319 | ss << "set_recovery_delay: error setting " | |
6320 | << "osd_recovery_delay_start to '" << delay << "': error " | |
6321 | << r; | |
6322 | return; | |
6323 | } | |
11fdf7f2 | 6324 | service->cct->_conf.apply_changes(nullptr); |
7c673cae FG |
6325 | ss << "set_recovery_delay: set osd_recovery_delay_start " |
6326 | << "to " << service->cct->_conf->osd_recovery_delay_start; | |
6327 | return; | |
6328 | } | |
7c673cae | 6329 | if (command == "injectfull") { |
20effc67 TL |
6330 | int64_t count = cmd_getval_or<int64_t>(cmdmap, "count", -1); |
6331 | string type = cmd_getval_or<string>(cmdmap, "type", "full"); | |
7c673cae | 6332 | OSDService::s_names state; |
20effc67 | 6333 | |
7c673cae FG |
6334 | if (type == "none" || count == 0) { |
6335 | type = "none"; | |
6336 | count = 0; | |
6337 | } | |
6338 | state = service->get_full_state(type); | |
6339 | if (state == OSDService::s_names::INVALID) { | |
6340 | ss << "Invalid type use (none, nearfull, backfillfull, full, failsafe)"; | |
6341 | return; | |
6342 | } | |
6343 | service->set_injectfull(state, count); | |
6344 | return; | |
6345 | } | |
6346 | ss << "Internal error - command=" << command; | |
6347 | } | |
6348 | ||
7c673cae FG |
6349 | // ========================================= |
6350 | ||
6351 | void OSD::ms_handle_connect(Connection *con) | |
6352 | { | |
6353 | dout(10) << __func__ << " con " << con << dendl; | |
6354 | if (con->get_peer_type() == CEPH_ENTITY_TYPE_MON) { | |
11fdf7f2 | 6355 | std::lock_guard l(osd_lock); |
7c673cae FG |
6356 | if (is_stopping()) |
6357 | return; | |
6358 | dout(10) << __func__ << " on mon" << dendl; | |
6359 | ||
6360 | if (is_preboot()) { | |
6361 | start_boot(); | |
6362 | } else if (is_booting()) { | |
6363 | _send_boot(); // resend boot message | |
6364 | } else { | |
9f95a23c | 6365 | map_lock.lock_shared(); |
11fdf7f2 | 6366 | std::lock_guard l2(mon_report_lock); |
7c673cae FG |
6367 | |
6368 | utime_t now = ceph_clock_now(); | |
6369 | last_mon_report = now; | |
6370 | ||
6371 | // resend everything, it's a new session | |
6372 | send_full_update(); | |
6373 | send_alive(); | |
6374 | service.requeue_pg_temp(); | |
11fdf7f2 | 6375 | service.clear_sent_ready_to_merge(); |
7c673cae | 6376 | service.send_pg_temp(); |
11fdf7f2 TL |
6377 | service.send_ready_to_merge(); |
6378 | service.send_pg_created(); | |
7c673cae FG |
6379 | requeue_failures(); |
6380 | send_failures(); | |
7c673cae | 6381 | |
9f95a23c | 6382 | map_lock.unlock_shared(); |
7c673cae FG |
6383 | if (is_active()) { |
6384 | send_beacon(ceph::coarse_mono_clock::now()); | |
6385 | } | |
6386 | } | |
6387 | ||
6388 | // full map requests may happen while active or pre-boot | |
6389 | if (requested_full_first) { | |
6390 | rerequest_full_maps(); | |
6391 | } | |
6392 | } | |
6393 | } | |
6394 | ||
6395 | void OSD::ms_handle_fast_connect(Connection *con) | |
6396 | { | |
6397 | if (con->get_peer_type() != CEPH_ENTITY_TYPE_MON && | |
6398 | con->get_peer_type() != CEPH_ENTITY_TYPE_MGR) { | |
9f95a23c TL |
6399 | if (auto s = ceph::ref_cast<Session>(con->get_priv()); !s) { |
6400 | s = ceph::make_ref<Session>(cct, con); | |
6401 | con->set_priv(s); | |
7c673cae FG |
6402 | dout(10) << " new session (outgoing) " << s << " con=" << s->con |
6403 | << " addr=" << s->con->get_peer_addr() << dendl; | |
6404 | // we don't connect to clients | |
11fdf7f2 | 6405 | ceph_assert(con->get_peer_type() == CEPH_ENTITY_TYPE_OSD); |
7c673cae FG |
6406 | s->entity_name.set_type(CEPH_ENTITY_TYPE_OSD); |
6407 | } | |
7c673cae FG |
6408 | } |
6409 | } | |
6410 | ||
6411 | void OSD::ms_handle_fast_accept(Connection *con) | |
6412 | { | |
6413 | if (con->get_peer_type() != CEPH_ENTITY_TYPE_MON && | |
6414 | con->get_peer_type() != CEPH_ENTITY_TYPE_MGR) { | |
9f95a23c TL |
6415 | if (auto s = ceph::ref_cast<Session>(con->get_priv()); !s) { |
6416 | s = ceph::make_ref<Session>(cct, con); | |
6417 | con->set_priv(s); | |
7c673cae FG |
6418 | dout(10) << "new session (incoming)" << s << " con=" << con |
6419 | << " addr=" << con->get_peer_addr() | |
6420 | << " must have raced with connect" << dendl; | |
11fdf7f2 | 6421 | ceph_assert(con->get_peer_type() == CEPH_ENTITY_TYPE_OSD); |
7c673cae FG |
6422 | s->entity_name.set_type(CEPH_ENTITY_TYPE_OSD); |
6423 | } | |
7c673cae FG |
6424 | } |
6425 | } | |
6426 | ||
6427 | bool OSD::ms_handle_reset(Connection *con) | |
6428 | { | |
9f95a23c TL |
6429 | auto session = ceph::ref_cast<Session>(con->get_priv()); |
6430 | dout(2) << "ms_handle_reset con " << con << " session " << session.get() << dendl; | |
7c673cae FG |
6431 | if (!session) |
6432 | return false; | |
6433 | session->wstate.reset(con); | |
11fdf7f2 TL |
6434 | session->con->set_priv(nullptr); |
6435 | session->con.reset(); // break con <-> session ref cycle | |
7c673cae FG |
6436 | // note that we break session->con *before* the session_handle_reset |
6437 | // cleanup below. this avoids a race between us and | |
6438 | // PG::add_backoff, Session::check_backoff, etc. | |
9f95a23c | 6439 | session_handle_reset(session); |
7c673cae FG |
6440 | return true; |
6441 | } | |
6442 | ||
6443 | bool OSD::ms_handle_refused(Connection *con) | |
6444 | { | |
6445 | if (!cct->_conf->osd_fast_fail_on_connection_refused) | |
6446 | return false; | |
6447 | ||
9f95a23c TL |
6448 | auto session = ceph::ref_cast<Session>(con->get_priv()); |
6449 | dout(2) << "ms_handle_refused con " << con << " session " << session.get() << dendl; | |
7c673cae FG |
6450 | if (!session) |
6451 | return false; | |
6452 | int type = con->get_peer_type(); | |
6453 | // handle only OSD failures here | |
6454 | if (monc && (type == CEPH_ENTITY_TYPE_OSD)) { | |
6455 | OSDMapRef osdmap = get_osdmap(); | |
6456 | if (osdmap) { | |
6457 | int id = osdmap->identify_osd_on_all_channels(con->get_peer_addr()); | |
6458 | if (id >= 0 && osdmap->is_up(id)) { | |
6459 | // I'm cheating mon heartbeat grace logic, because we know it's not going | |
6460 | // to respawn alone. +1 so we won't hit any boundary case. | |
11fdf7f2 TL |
6461 | monc->send_mon_message( |
6462 | new MOSDFailure( | |
6463 | monc->get_fsid(), | |
6464 | id, | |
6465 | osdmap->get_addrs(id), | |
6466 | cct->_conf->osd_heartbeat_grace + 1, | |
6467 | osdmap->get_epoch(), | |
6468 | MOSDFailure::FLAG_IMMEDIATE | MOSDFailure::FLAG_FAILED | |
6469 | )); | |
7c673cae FG |
6470 | } |
6471 | } | |
6472 | } | |
7c673cae FG |
6473 | return true; |
6474 | } | |
6475 | ||
f67539c2 | 6476 | struct CB_OSD_GetVersion { |
7c673cae | 6477 | OSD *osd; |
f67539c2 TL |
6478 | explicit CB_OSD_GetVersion(OSD *o) : osd(o) {} |
6479 | void operator ()(boost::system::error_code ec, version_t newest, | |
6480 | version_t oldest) { | |
6481 | if (!ec) | |
7c673cae FG |
6482 | osd->_got_mon_epochs(oldest, newest); |
6483 | } | |
6484 | }; | |
6485 | ||
6486 | void OSD::start_boot() | |
6487 | { | |
6488 | if (!_is_healthy()) { | |
6489 | // if we are not healthy, do not mark ourselves up (yet) | |
6490 | dout(1) << "not healthy; waiting to boot" << dendl; | |
6491 | if (!is_waiting_for_healthy()) | |
6492 | start_waiting_for_healthy(); | |
6493 | // send pings sooner rather than later | |
6494 | heartbeat_kick(); | |
6495 | return; | |
6496 | } | |
6497 | dout(1) << __func__ << dendl; | |
6498 | set_state(STATE_PREBOOT); | |
6499 | dout(10) << "start_boot - have maps " << superblock.oldest_map | |
6500 | << ".." << superblock.newest_map << dendl; | |
f67539c2 | 6501 | monc->get_version("osdmap", CB_OSD_GetVersion(this)); |
7c673cae FG |
6502 | } |
6503 | ||
6504 | void OSD::_got_mon_epochs(epoch_t oldest, epoch_t newest) | |
6505 | { | |
11fdf7f2 | 6506 | std::lock_guard l(osd_lock); |
7c673cae FG |
6507 | if (is_preboot()) { |
6508 | _preboot(oldest, newest); | |
6509 | } | |
6510 | } | |
6511 | ||
6512 | void OSD::_preboot(epoch_t oldest, epoch_t newest) | |
6513 | { | |
11fdf7f2 | 6514 | ceph_assert(is_preboot()); |
7c673cae FG |
6515 | dout(10) << __func__ << " _preboot mon has osdmaps " |
6516 | << oldest << ".." << newest << dendl; | |
6517 | ||
6518 | // ensure our local fullness awareness is accurate | |
81eedcae TL |
6519 | { |
6520 | std::lock_guard l(heartbeat_lock); | |
6521 | heartbeat(); | |
6522 | } | |
7c673cae | 6523 | |
9f95a23c TL |
6524 | const auto& monmap = monc->monmap; |
6525 | const auto osdmap = get_osdmap(); | |
7c673cae | 6526 | // if our map within recent history, try to add ourselves to the osdmap. |
31f18b77 FG |
6527 | if (osdmap->get_epoch() == 0) { |
6528 | derr << "waiting for initial osdmap" << dendl; | |
c07f9fc5 | 6529 | } else if (osdmap->is_destroyed(whoami)) { |
b32b8144 FG |
6530 | derr << "osdmap says I am destroyed" << dendl; |
6531 | // provide a small margin so we don't livelock seeing if we | |
6532 | // un-destroyed ourselves. | |
6533 | if (osdmap->get_epoch() > newest - 1) { | |
6534 | exit(0); | |
6535 | } | |
81eedcae | 6536 | } else if (osdmap->is_noup(whoami)) { |
7c673cae FG |
6537 | derr << "osdmap NOUP flag is set, waiting for it to clear" << dendl; |
6538 | } else if (!osdmap->test_flag(CEPH_OSDMAP_SORTBITWISE)) { | |
6539 | derr << "osdmap SORTBITWISE OSDMap flag is NOT set; please set it" | |
6540 | << dendl; | |
7c673cae FG |
6541 | } else if (service.need_fullness_update()) { |
6542 | derr << "osdmap fullness state needs update" << dendl; | |
6543 | send_full_update(); | |
9f95a23c TL |
6544 | } else if (monmap.min_mon_release >= ceph_release_t::octopus && |
6545 | superblock.purged_snaps_last < superblock.current_epoch) { | |
6546 | dout(10) << __func__ << " purged_snaps_last " << superblock.purged_snaps_last | |
6547 | << " < newest_map " << superblock.current_epoch << dendl; | |
6548 | _get_purged_snaps(); | |
7c673cae FG |
6549 | } else if (osdmap->get_epoch() >= oldest - 1 && |
6550 | osdmap->get_epoch() + cct->_conf->osd_map_message_max > newest) { | |
11fdf7f2 TL |
6551 | |
6552 | // wait for pgs to fully catch up in a different thread, since | |
6553 | // this thread might be required for splitting and merging PGs to | |
6554 | // make progress. | |
6555 | boot_finisher.queue( | |
9f95a23c | 6556 | new LambdaContext( |
11fdf7f2 | 6557 | [this](int r) { |
9f95a23c | 6558 | std::unique_lock l(osd_lock); |
11fdf7f2 TL |
6559 | if (is_preboot()) { |
6560 | dout(10) << __func__ << " waiting for peering work to drain" | |
6561 | << dendl; | |
9f95a23c | 6562 | l.unlock(); |
11fdf7f2 | 6563 | for (auto shard : shards) { |
9f95a23c | 6564 | shard->wait_min_pg_epoch(get_osdmap_epoch()); |
11fdf7f2 | 6565 | } |
9f95a23c | 6566 | l.lock(); |
11fdf7f2 TL |
6567 | } |
6568 | if (is_preboot()) { | |
6569 | _send_boot(); | |
6570 | } | |
6571 | })); | |
6572 | return; | |
7c673cae FG |
6573 | } |
6574 | ||
6575 | // get all the latest maps | |
6576 | if (osdmap->get_epoch() + 1 >= oldest) | |
6577 | osdmap_subscribe(osdmap->get_epoch() + 1, false); | |
6578 | else | |
6579 | osdmap_subscribe(oldest - 1, true); | |
6580 | } | |
6581 | ||
9f95a23c TL |
6582 | void OSD::_get_purged_snaps() |
6583 | { | |
6584 | // NOTE: this is a naive, stateless implementaiton. it may send multiple | |
6585 | // overlapping requests to the mon, which will be somewhat inefficient, but | |
6586 | // it should be reliable. | |
6587 | dout(10) << __func__ << " purged_snaps_last " << superblock.purged_snaps_last | |
6588 | << ", newest_map " << superblock.current_epoch << dendl; | |
6589 | MMonGetPurgedSnaps *m = new MMonGetPurgedSnaps( | |
6590 | superblock.purged_snaps_last + 1, | |
6591 | superblock.current_epoch + 1); | |
6592 | monc->send_mon_message(m); | |
6593 | } | |
6594 | ||
6595 | void OSD::handle_get_purged_snaps_reply(MMonGetPurgedSnapsReply *m) | |
6596 | { | |
6597 | dout(10) << __func__ << " " << *m << dendl; | |
6598 | ObjectStore::Transaction t; | |
6599 | if (!is_preboot() || | |
6600 | m->last < superblock.purged_snaps_last) { | |
6601 | goto out; | |
6602 | } | |
20effc67 | 6603 | SnapMapper::record_purged_snaps(cct, store.get(), service.meta_ch, |
9f95a23c TL |
6604 | make_purged_snaps_oid(), &t, |
6605 | m->purged_snaps); | |
6606 | superblock.purged_snaps_last = m->last; | |
6607 | write_superblock(t); | |
6608 | store->queue_transaction( | |
6609 | service.meta_ch, | |
6610 | std::move(t)); | |
6611 | service.publish_superblock(superblock); | |
6612 | if (m->last < superblock.current_epoch) { | |
6613 | _get_purged_snaps(); | |
6614 | } else { | |
6615 | start_boot(); | |
6616 | } | |
6617 | out: | |
6618 | m->put(); | |
6619 | } | |
6620 | ||
7c673cae FG |
6621 | void OSD::send_full_update() |
6622 | { | |
6623 | if (!service.need_fullness_update()) | |
6624 | return; | |
6625 | unsigned state = 0; | |
6626 | if (service.is_full()) { | |
6627 | state = CEPH_OSD_FULL; | |
6628 | } else if (service.is_backfillfull()) { | |
6629 | state = CEPH_OSD_BACKFILLFULL; | |
6630 | } else if (service.is_nearfull()) { | |
6631 | state = CEPH_OSD_NEARFULL; | |
6632 | } | |
6633 | set<string> s; | |
6634 | OSDMap::calc_state_set(state, s); | |
6635 | dout(10) << __func__ << " want state " << s << dendl; | |
9f95a23c | 6636 | monc->send_mon_message(new MOSDFull(get_osdmap_epoch(), state)); |
7c673cae FG |
6637 | } |
6638 | ||
6639 | void OSD::start_waiting_for_healthy() | |
6640 | { | |
6641 | dout(1) << "start_waiting_for_healthy" << dendl; | |
6642 | set_state(STATE_WAITING_FOR_HEALTHY); | |
6643 | last_heartbeat_resample = utime_t(); | |
181888fb FG |
6644 | |
6645 | // subscribe to osdmap updates, in case our peers really are known to be dead | |
9f95a23c | 6646 | osdmap_subscribe(get_osdmap_epoch() + 1, false); |
7c673cae FG |
6647 | } |
6648 | ||
6649 | bool OSD::_is_healthy() | |
6650 | { | |
6651 | if (!cct->get_heartbeat_map()->is_healthy()) { | |
6652 | dout(1) << "is_healthy false -- internal heartbeat failed" << dendl; | |
6653 | return false; | |
6654 | } | |
6655 | ||
6656 | if (is_waiting_for_healthy()) { | |
11fdf7f2 | 6657 | utime_t now = ceph_clock_now(); |
9f95a23c TL |
6658 | if (osd_markdown_log.empty()) { |
6659 | dout(5) << __func__ << " force returning true since last markdown" | |
6660 | << " was " << cct->_conf->osd_max_markdown_period | |
6661 | << "s ago" << dendl; | |
11fdf7f2 TL |
6662 | return true; |
6663 | } | |
6664 | std::lock_guard l(heartbeat_lock); | |
7c673cae FG |
6665 | int num = 0, up = 0; |
6666 | for (map<int,HeartbeatInfo>::iterator p = heartbeat_peers.begin(); | |
6667 | p != heartbeat_peers.end(); | |
6668 | ++p) { | |
11fdf7f2 | 6669 | if (p->second.is_healthy(now)) |
7c673cae FG |
6670 | ++up; |
6671 | ++num; | |
6672 | } | |
6673 | if ((float)up < (float)num * cct->_conf->osd_heartbeat_min_healthy_ratio) { | |
6674 | dout(1) << "is_healthy false -- only " << up << "/" << num << " up peers (less than " | |
6675 | << int(cct->_conf->osd_heartbeat_min_healthy_ratio * 100.0) << "%)" << dendl; | |
6676 | return false; | |
6677 | } | |
6678 | } | |
6679 | ||
6680 | return true; | |
6681 | } | |
6682 | ||
6683 | void OSD::_send_boot() | |
6684 | { | |
6685 | dout(10) << "_send_boot" << dendl; | |
11fdf7f2 TL |
6686 | Connection *local_connection = |
6687 | cluster_messenger->get_loopback_connection().get(); | |
6688 | entity_addrvec_t client_addrs = client_messenger->get_myaddrs(); | |
6689 | entity_addrvec_t cluster_addrs = cluster_messenger->get_myaddrs(); | |
6690 | entity_addrvec_t hb_back_addrs = hb_back_server_messenger->get_myaddrs(); | |
6691 | entity_addrvec_t hb_front_addrs = hb_front_server_messenger->get_myaddrs(); | |
6692 | ||
6693 | dout(20) << " initial client_addrs " << client_addrs | |
6694 | << ", cluster_addrs " << cluster_addrs | |
6695 | << ", hb_back_addrs " << hb_back_addrs | |
6696 | << ", hb_front_addrs " << hb_front_addrs | |
6697 | << dendl; | |
6698 | if (cluster_messenger->set_addr_unknowns(client_addrs)) { | |
6699 | dout(10) << " assuming cluster_addrs match client_addrs " | |
6700 | << client_addrs << dendl; | |
6701 | cluster_addrs = cluster_messenger->get_myaddrs(); | |
6702 | } | |
6703 | if (auto session = local_connection->get_priv(); !session) { | |
6704 | cluster_messenger->ms_deliver_handle_fast_connect(local_connection); | |
7c673cae FG |
6705 | } |
6706 | ||
7c673cae | 6707 | local_connection = hb_back_server_messenger->get_loopback_connection().get(); |
11fdf7f2 TL |
6708 | if (hb_back_server_messenger->set_addr_unknowns(cluster_addrs)) { |
6709 | dout(10) << " assuming hb_back_addrs match cluster_addrs " | |
6710 | << cluster_addrs << dendl; | |
6711 | hb_back_addrs = hb_back_server_messenger->get_myaddrs(); | |
7c673cae | 6712 | } |
11fdf7f2 TL |
6713 | if (auto session = local_connection->get_priv(); !session) { |
6714 | hb_back_server_messenger->ms_deliver_handle_fast_connect(local_connection); | |
7c673cae FG |
6715 | } |
6716 | ||
11fdf7f2 TL |
6717 | local_connection = hb_front_server_messenger->get_loopback_connection().get(); |
6718 | if (hb_front_server_messenger->set_addr_unknowns(client_addrs)) { | |
6719 | dout(10) << " assuming hb_front_addrs match client_addrs " | |
6720 | << client_addrs << dendl; | |
6721 | hb_front_addrs = hb_front_server_messenger->get_myaddrs(); | |
6722 | } | |
6723 | if (auto session = local_connection->get_priv(); !session) { | |
6724 | hb_front_server_messenger->ms_deliver_handle_fast_connect(local_connection); | |
6725 | } | |
6726 | ||
6727 | // we now know what our front and back addrs will be, and we are | |
6728 | // about to tell the mon what our metadata (including numa bindings) | |
6729 | // are, so now is a good time! | |
6730 | set_numa_affinity(); | |
6731 | ||
6732 | MOSDBoot *mboot = new MOSDBoot( | |
6733 | superblock, get_osdmap_epoch(), service.get_boot_epoch(), | |
6734 | hb_back_addrs, hb_front_addrs, cluster_addrs, | |
6735 | CEPH_FEATURES_ALL); | |
6736 | dout(10) << " final client_addrs " << client_addrs | |
6737 | << ", cluster_addrs " << cluster_addrs | |
6738 | << ", hb_back_addrs " << hb_back_addrs | |
6739 | << ", hb_front_addrs " << hb_front_addrs | |
7c673cae FG |
6740 | << dendl; |
6741 | _collect_metadata(&mboot->metadata); | |
6742 | monc->send_mon_message(mboot); | |
6743 | set_state(STATE_BOOTING); | |
6744 | } | |
6745 | ||
6746 | void OSD::_collect_metadata(map<string,string> *pm) | |
6747 | { | |
6748 | // config info | |
6749 | (*pm)["osd_data"] = dev_path; | |
c07f9fc5 FG |
6750 | if (store->get_type() == "filestore") { |
6751 | // not applicable for bluestore | |
6752 | (*pm)["osd_journal"] = journal_path; | |
6753 | } | |
11fdf7f2 TL |
6754 | (*pm)["front_addr"] = stringify(client_messenger->get_myaddrs()); |
6755 | (*pm)["back_addr"] = stringify(cluster_messenger->get_myaddrs()); | |
6756 | (*pm)["hb_front_addr"] = stringify(hb_front_server_messenger->get_myaddrs()); | |
6757 | (*pm)["hb_back_addr"] = stringify(hb_back_server_messenger->get_myaddrs()); | |
7c673cae FG |
6758 | |
6759 | // backend | |
6760 | (*pm)["osd_objectstore"] = store->get_type(); | |
31f18b77 | 6761 | (*pm)["rotational"] = store_is_rotational ? "1" : "0"; |
d2e6a577 | 6762 | (*pm)["journal_rotational"] = journal_is_rotational ? "1" : "0"; |
c07f9fc5 | 6763 | (*pm)["default_device_class"] = store->get_default_device_class(); |
f6b5b4d7 TL |
6764 | string osdspec_affinity; |
6765 | int r = store->read_meta("osdspec_affinity", &osdspec_affinity); | |
6766 | if (r < 0 || osdspec_affinity.empty()) { | |
6767 | osdspec_affinity = ""; | |
6768 | } | |
6769 | (*pm)["osdspec_affinity"] = osdspec_affinity; | |
7c673cae FG |
6770 | store->collect_metadata(pm); |
6771 | ||
6772 | collect_sys_info(pm, cct); | |
6773 | ||
11fdf7f2 TL |
6774 | (*pm)["front_iface"] = pick_iface( |
6775 | cct, | |
6776 | client_messenger->get_myaddrs().front().get_sockaddr_storage()); | |
6777 | (*pm)["back_iface"] = pick_iface( | |
6778 | cct, | |
6779 | cluster_messenger->get_myaddrs().front().get_sockaddr_storage()); | |
6780 | ||
6781 | // network numa | |
6782 | { | |
6783 | int node = -1; | |
6784 | set<int> nodes; | |
6785 | set<string> unknown; | |
6786 | for (auto nm : { "front_iface", "back_iface" }) { | |
6787 | if (!(*pm)[nm].size()) { | |
6788 | unknown.insert(nm); | |
6789 | continue; | |
6790 | } | |
6791 | int n = -1; | |
6792 | int r = get_iface_numa_node((*pm)[nm], &n); | |
6793 | if (r < 0) { | |
6794 | unknown.insert((*pm)[nm]); | |
6795 | continue; | |
6796 | } | |
6797 | nodes.insert(n); | |
6798 | if (node < 0) { | |
6799 | node = n; | |
6800 | } | |
6801 | } | |
6802 | if (unknown.size()) { | |
6803 | (*pm)["network_numa_unknown_ifaces"] = stringify(unknown); | |
6804 | } | |
6805 | if (!nodes.empty()) { | |
6806 | (*pm)["network_numa_nodes"] = stringify(nodes); | |
6807 | } | |
6808 | if (node >= 0 && nodes.size() == 1 && unknown.empty()) { | |
6809 | (*pm)["network_numa_node"] = stringify(node); | |
6810 | } | |
6811 | } | |
6812 | ||
6813 | if (numa_node >= 0) { | |
6814 | (*pm)["numa_node"] = stringify(numa_node); | |
6815 | (*pm)["numa_node_cpus"] = cpu_set_to_str_list(numa_cpu_set_size, | |
6816 | &numa_cpu_set); | |
6817 | } | |
6818 | ||
6819 | set<string> devnames; | |
6820 | store->get_devices(&devnames); | |
9f95a23c TL |
6821 | map<string,string> errs; |
6822 | get_device_metadata(devnames, pm, &errs); | |
6823 | for (auto& i : errs) { | |
6824 | dout(1) << __func__ << " " << i.first << ": " << i.second << dendl; | |
11fdf7f2 | 6825 | } |
7c673cae FG |
6826 | dout(10) << __func__ << " " << *pm << dendl; |
6827 | } | |
6828 | ||
6829 | void OSD::queue_want_up_thru(epoch_t want) | |
6830 | { | |
9f95a23c TL |
6831 | std::shared_lock map_locker{map_lock}; |
6832 | epoch_t cur = get_osdmap()->get_up_thru(whoami); | |
6833 | std::lock_guard report_locker(mon_report_lock); | |
7c673cae FG |
6834 | if (want > up_thru_wanted) { |
6835 | dout(10) << "queue_want_up_thru now " << want << " (was " << up_thru_wanted << ")" | |
6836 | << ", currently " << cur | |
6837 | << dendl; | |
6838 | up_thru_wanted = want; | |
6839 | send_alive(); | |
6840 | } else { | |
6841 | dout(10) << "queue_want_up_thru want " << want << " <= queued " << up_thru_wanted | |
6842 | << ", currently " << cur | |
6843 | << dendl; | |
6844 | } | |
7c673cae FG |
6845 | } |
6846 | ||
6847 | void OSD::send_alive() | |
6848 | { | |
9f95a23c TL |
6849 | ceph_assert(ceph_mutex_is_locked(mon_report_lock)); |
6850 | const auto osdmap = get_osdmap(); | |
7c673cae FG |
6851 | if (!osdmap->exists(whoami)) |
6852 | return; | |
6853 | epoch_t up_thru = osdmap->get_up_thru(whoami); | |
6854 | dout(10) << "send_alive up_thru currently " << up_thru << " want " << up_thru_wanted << dendl; | |
6855 | if (up_thru_wanted > up_thru) { | |
6856 | dout(10) << "send_alive want " << up_thru_wanted << dendl; | |
6857 | monc->send_mon_message(new MOSDAlive(osdmap->get_epoch(), up_thru_wanted)); | |
6858 | } | |
6859 | } | |
6860 | ||
6861 | void OSD::request_full_map(epoch_t first, epoch_t last) | |
6862 | { | |
6863 | dout(10) << __func__ << " " << first << ".." << last | |
6864 | << ", previously requested " | |
6865 | << requested_full_first << ".." << requested_full_last << dendl; | |
9f95a23c | 6866 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
11fdf7f2 TL |
6867 | ceph_assert(first > 0 && last > 0); |
6868 | ceph_assert(first <= last); | |
6869 | ceph_assert(first >= requested_full_first); // we shouldn't ever ask for older maps | |
7c673cae FG |
6870 | if (requested_full_first == 0) { |
6871 | // first request | |
6872 | requested_full_first = first; | |
6873 | requested_full_last = last; | |
6874 | } else if (last <= requested_full_last) { | |
6875 | // dup | |
6876 | return; | |
6877 | } else { | |
6878 | // additional request | |
6879 | first = requested_full_last + 1; | |
6880 | requested_full_last = last; | |
6881 | } | |
6882 | MMonGetOSDMap *req = new MMonGetOSDMap; | |
6883 | req->request_full(first, last); | |
6884 | monc->send_mon_message(req); | |
6885 | } | |
6886 | ||
6887 | void OSD::got_full_map(epoch_t e) | |
6888 | { | |
11fdf7f2 | 6889 | ceph_assert(requested_full_first <= requested_full_last); |
9f95a23c | 6890 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae FG |
6891 | if (requested_full_first == 0) { |
6892 | dout(20) << __func__ << " " << e << ", nothing requested" << dendl; | |
6893 | return; | |
6894 | } | |
6895 | if (e < requested_full_first) { | |
6896 | dout(10) << __func__ << " " << e << ", requested " << requested_full_first | |
6897 | << ".." << requested_full_last | |
6898 | << ", ignoring" << dendl; | |
6899 | return; | |
6900 | } | |
6901 | if (e >= requested_full_last) { | |
6902 | dout(10) << __func__ << " " << e << ", requested " << requested_full_first | |
6903 | << ".." << requested_full_last << ", resetting" << dendl; | |
6904 | requested_full_first = requested_full_last = 0; | |
6905 | return; | |
6906 | } | |
f67539c2 | 6907 | |
7c673cae FG |
6908 | requested_full_first = e + 1; |
6909 | ||
6910 | dout(10) << __func__ << " " << e << ", requested " << requested_full_first | |
6911 | << ".." << requested_full_last | |
6912 | << ", still need more" << dendl; | |
6913 | } | |
6914 | ||
6915 | void OSD::requeue_failures() | |
6916 | { | |
11fdf7f2 | 6917 | std::lock_guard l(heartbeat_lock); |
7c673cae FG |
6918 | unsigned old_queue = failure_queue.size(); |
6919 | unsigned old_pending = failure_pending.size(); | |
11fdf7f2 | 6920 | for (auto p = failure_pending.begin(); p != failure_pending.end(); ) { |
7c673cae FG |
6921 | failure_queue[p->first] = p->second.first; |
6922 | failure_pending.erase(p++); | |
6923 | } | |
6924 | dout(10) << __func__ << " " << old_queue << " + " << old_pending << " -> " | |
6925 | << failure_queue.size() << dendl; | |
6926 | } | |
6927 | ||
6928 | void OSD::send_failures() | |
6929 | { | |
9f95a23c TL |
6930 | ceph_assert(ceph_mutex_is_locked(map_lock)); |
6931 | ceph_assert(ceph_mutex_is_locked(mon_report_lock)); | |
11fdf7f2 | 6932 | std::lock_guard l(heartbeat_lock); |
7c673cae | 6933 | utime_t now = ceph_clock_now(); |
9f95a23c | 6934 | const auto osdmap = get_osdmap(); |
7c673cae FG |
6935 | while (!failure_queue.empty()) { |
6936 | int osd = failure_queue.begin()->first; | |
7c673cae FG |
6937 | if (!failure_pending.count(osd)) { |
6938 | int failed_for = (int)(double)(now - failure_queue.begin()->second); | |
11fdf7f2 TL |
6939 | monc->send_mon_message( |
6940 | new MOSDFailure( | |
6941 | monc->get_fsid(), | |
6942 | osd, | |
6943 | osdmap->get_addrs(osd), | |
6944 | failed_for, | |
6945 | osdmap->get_epoch())); | |
6946 | failure_pending[osd] = make_pair(failure_queue.begin()->second, | |
6947 | osdmap->get_addrs(osd)); | |
7c673cae FG |
6948 | } |
6949 | failure_queue.erase(osd); | |
6950 | } | |
6951 | } | |
6952 | ||
11fdf7f2 | 6953 | void OSD::send_still_alive(epoch_t epoch, int osd, const entity_addrvec_t &addrs) |
7c673cae | 6954 | { |
11fdf7f2 TL |
6955 | MOSDFailure *m = new MOSDFailure(monc->get_fsid(), osd, addrs, 0, epoch, |
6956 | MOSDFailure::FLAG_ALIVE); | |
7c673cae FG |
6957 | monc->send_mon_message(m); |
6958 | } | |
6959 | ||
11fdf7f2 | 6960 | void OSD::cancel_pending_failures() |
7c673cae | 6961 | { |
11fdf7f2 TL |
6962 | std::lock_guard l(heartbeat_lock); |
6963 | auto it = failure_pending.begin(); | |
6964 | while (it != failure_pending.end()) { | |
6965 | dout(10) << __func__ << " canceling in-flight failure report for osd." | |
6966 | << it->first << dendl; | |
9f95a23c | 6967 | send_still_alive(get_osdmap_epoch(), it->first, it->second.second); |
11fdf7f2 | 6968 | failure_pending.erase(it++); |
7c673cae | 6969 | } |
7c673cae FG |
6970 | } |
6971 | ||
6972 | void OSD::send_beacon(const ceph::coarse_mono_clock::time_point& now) | |
6973 | { | |
6974 | const auto& monmap = monc->monmap; | |
6975 | // send beacon to mon even if we are just connected, and the monmap is not | |
6976 | // initialized yet by then. | |
6977 | if (monmap.epoch > 0 && | |
6978 | monmap.get_required_features().contains_all( | |
6979 | ceph::features::mon::FEATURE_LUMINOUS)) { | |
6980 | dout(20) << __func__ << " sending" << dendl; | |
7c673cae FG |
6981 | MOSDBeacon* beacon = nullptr; |
6982 | { | |
11fdf7f2 | 6983 | std::lock_guard l{min_last_epoch_clean_lock}; |
9f95a23c TL |
6984 | beacon = new MOSDBeacon(get_osdmap_epoch(), |
6985 | min_last_epoch_clean, | |
f67539c2 TL |
6986 | superblock.last_purged_snaps_scrub, |
6987 | cct->_conf->osd_beacon_report_interval); | |
494da23a | 6988 | beacon->pgs = min_last_epoch_clean_pgs; |
224ce89b | 6989 | last_sent_beacon = now; |
7c673cae FG |
6990 | } |
6991 | monc->send_mon_message(beacon); | |
6992 | } else { | |
6993 | dout(20) << __func__ << " not sending" << dendl; | |
6994 | } | |
6995 | } | |
6996 | ||
7c673cae FG |
6997 | void OSD::handle_command(MCommand *m) |
6998 | { | |
6999 | ConnectionRef con = m->get_connection(); | |
9f95a23c | 7000 | auto session = ceph::ref_cast<Session>(con->get_priv()); |
7c673cae | 7001 | if (!session) { |
9f95a23c | 7002 | con->send_message(new MCommandReply(m, -EACCES)); |
7c673cae FG |
7003 | m->put(); |
7004 | return; | |
7005 | } | |
9f95a23c TL |
7006 | if (!session->caps.allow_all()) { |
7007 | con->send_message(new MCommandReply(m, -EACCES)); | |
7c673cae FG |
7008 | m->put(); |
7009 | return; | |
7010 | } | |
9f95a23c | 7011 | cct->get_admin_socket()->queue_tell_command(m); |
7c673cae FG |
7012 | m->put(); |
7013 | } | |
7014 | ||
f64942e4 AA |
7015 | namespace { |
7016 | class unlock_guard { | |
9f95a23c | 7017 | ceph::mutex& m; |
f64942e4 | 7018 | public: |
9f95a23c | 7019 | explicit unlock_guard(ceph::mutex& mutex) |
f64942e4 AA |
7020 | : m(mutex) |
7021 | { | |
11fdf7f2 | 7022 | m.unlock(); |
f64942e4 AA |
7023 | } |
7024 | unlock_guard(unlock_guard&) = delete; | |
7025 | ~unlock_guard() { | |
11fdf7f2 | 7026 | m.lock(); |
f64942e4 AA |
7027 | } |
7028 | }; | |
7029 | } | |
7030 | ||
9f95a23c | 7031 | void OSD::scrub_purged_snaps() |
7c673cae | 7032 | { |
9f95a23c TL |
7033 | dout(10) << __func__ << dendl; |
7034 | ceph_assert(ceph_mutex_is_locked(osd_lock)); | |
20effc67 | 7035 | SnapMapper::Scrubber s(cct, store.get(), service.meta_ch, |
9f95a23c TL |
7036 | make_snapmapper_oid(), |
7037 | make_purged_snaps_oid()); | |
7038 | clog->debug() << "purged_snaps scrub starts"; | |
7039 | osd_lock.unlock(); | |
7040 | s.run(); | |
7041 | if (s.stray.size()) { | |
7042 | clog->debug() << "purged_snaps scrub found " << s.stray.size() << " strays"; | |
7043 | } else { | |
7044 | clog->debug() << "purged_snaps scrub ok"; | |
224ce89b | 7045 | } |
9f95a23c TL |
7046 | set<pair<spg_t,snapid_t>> queued; |
7047 | for (auto& [pool, snap, hash, shard] : s.stray) { | |
7048 | const pg_pool_t *pi = get_osdmap()->get_pg_pool(pool); | |
7049 | if (!pi) { | |
7050 | dout(20) << __func__ << " pool " << pool << " dne" << dendl; | |
7051 | continue; | |
11fdf7f2 | 7052 | } |
9f95a23c TL |
7053 | pg_t pgid(pi->raw_hash_to_pg(hash), pool); |
7054 | spg_t spgid(pgid, shard); | |
7055 | pair<spg_t,snapid_t> p(spgid, snap); | |
7056 | if (queued.count(p)) { | |
7057 | dout(20) << __func__ << " pg " << spgid << " snap " << snap | |
7058 | << " already queued" << dendl; | |
7059 | continue; | |
11fdf7f2 | 7060 | } |
9f95a23c TL |
7061 | PGRef pg = lookup_lock_pg(spgid); |
7062 | if (!pg) { | |
7063 | dout(20) << __func__ << " pg " << spgid << " not found" << dendl; | |
7064 | continue; | |
11fdf7f2 | 7065 | } |
9f95a23c TL |
7066 | queued.insert(p); |
7067 | dout(10) << __func__ << " requeue pg " << spgid << " " << pg << " snap " | |
7068 | << snap << dendl; | |
7069 | pg->queue_snap_retrim(snap); | |
7070 | pg->unlock(); | |
7c673cae | 7071 | } |
9f95a23c TL |
7072 | osd_lock.lock(); |
7073 | if (is_stopping()) { | |
7074 | return; | |
7075 | } | |
7076 | dout(10) << __func__ << " done queueing pgs, updating superblock" << dendl; | |
7077 | ObjectStore::Transaction t; | |
7078 | superblock.last_purged_snaps_scrub = ceph_clock_now(); | |
7079 | write_superblock(t); | |
7080 | int tr = store->queue_transaction(service.meta_ch, std::move(t), nullptr); | |
7081 | ceph_assert(tr == 0); | |
7082 | if (is_active()) { | |
7083 | send_beacon(ceph::coarse_mono_clock::now()); | |
7084 | } | |
7085 | dout(10) << __func__ << " done" << dendl; | |
11fdf7f2 TL |
7086 | } |
7087 | ||
7088 | void OSD::probe_smart(const string& only_devid, ostream& ss) | |
7089 | { | |
7090 | set<string> devnames; | |
7091 | store->get_devices(&devnames); | |
7092 | uint64_t smart_timeout = cct->_conf.get_val<uint64_t>( | |
7093 | "osd_smart_report_timeout"); | |
7094 | ||
7095 | // == typedef std::map<std::string, mValue> mObject; | |
7096 | json_spirit::mObject json_map; | |
7097 | ||
7098 | for (auto dev : devnames) { | |
7099 | // smartctl works only on physical devices; filter out any logical device | |
7100 | if (dev.find("dm-") == 0) { | |
7101 | continue; | |
7102 | } | |
7103 | ||
7104 | string err; | |
7105 | string devid = get_device_id(dev, &err); | |
7106 | if (devid.size() == 0) { | |
7107 | dout(10) << __func__ << " no unique id for dev " << dev << " (" | |
7108 | << err << "), skipping" << dendl; | |
7109 | continue; | |
7110 | } | |
7111 | if (only_devid.size() && devid != only_devid) { | |
7112 | continue; | |
7113 | } | |
7114 | ||
7115 | json_spirit::mValue smart_json; | |
7116 | if (block_device_get_metrics(dev, smart_timeout, | |
7117 | &smart_json)) { | |
7118 | dout(10) << "block_device_get_metrics failed for /dev/" << dev << dendl; | |
7119 | continue; | |
7120 | } | |
7121 | json_map[devid] = smart_json; | |
7c673cae | 7122 | } |
11fdf7f2 | 7123 | json_spirit::write(json_map, ss, json_spirit::pretty_print); |
7c673cae FG |
7124 | } |
7125 | ||
7126 | bool OSD::heartbeat_dispatch(Message *m) | |
7127 | { | |
7128 | dout(30) << "heartbeat_dispatch " << m << dendl; | |
7129 | switch (m->get_type()) { | |
7130 | ||
7131 | case CEPH_MSG_PING: | |
7132 | dout(10) << "ping from " << m->get_source_inst() << dendl; | |
7133 | m->put(); | |
7134 | break; | |
7135 | ||
7136 | case MSG_OSD_PING: | |
7137 | handle_osd_ping(static_cast<MOSDPing*>(m)); | |
7138 | break; | |
7139 | ||
7140 | default: | |
7141 | dout(0) << "dropping unexpected message " << *m << " from " << m->get_source_inst() << dendl; | |
7142 | m->put(); | |
7143 | } | |
7144 | ||
7145 | return true; | |
7146 | } | |
7147 | ||
7148 | bool OSD::ms_dispatch(Message *m) | |
7149 | { | |
7150 | dout(20) << "OSD::ms_dispatch: " << *m << dendl; | |
7151 | if (m->get_type() == MSG_OSD_MARK_ME_DOWN) { | |
7152 | service.got_stop_ack(); | |
7153 | m->put(); | |
7154 | return true; | |
7155 | } | |
7156 | ||
7157 | // lock! | |
7158 | ||
9f95a23c | 7159 | osd_lock.lock(); |
7c673cae | 7160 | if (is_stopping()) { |
9f95a23c | 7161 | osd_lock.unlock(); |
7c673cae FG |
7162 | m->put(); |
7163 | return true; | |
7164 | } | |
7165 | ||
7166 | do_waiters(); | |
7167 | _dispatch(m); | |
7168 | ||
9f95a23c | 7169 | osd_lock.unlock(); |
7c673cae FG |
7170 | |
7171 | return true; | |
7172 | } | |
7173 | ||
9f95a23c TL |
7174 | void OSDService::maybe_share_map( |
7175 | Connection *con, | |
7176 | const OSDMapRef& osdmap, | |
7177 | epoch_t peer_epoch_lb) | |
7c673cae | 7178 | { |
9f95a23c TL |
7179 | // NOTE: we assume caller hold something that keeps the Connection itself |
7180 | // pinned (e.g., an OpRequest's MessageRef). | |
7181 | auto session = ceph::ref_cast<Session>(con->get_priv()); | |
7182 | if (!session) { | |
7c673cae FG |
7183 | return; |
7184 | } | |
7c673cae | 7185 | |
9f95a23c TL |
7186 | // assume the peer has the newer of the op's sent_epoch and what |
7187 | // we think we sent them. | |
7c673cae | 7188 | session->sent_epoch_lock.lock(); |
9f95a23c TL |
7189 | if (peer_epoch_lb > session->last_sent_epoch) { |
7190 | dout(10) << __func__ << " con " << con | |
7191 | << " " << con->get_peer_addr() | |
7192 | << " map epoch " << session->last_sent_epoch | |
7193 | << " -> " << peer_epoch_lb << " (as per caller)" << dendl; | |
7194 | session->last_sent_epoch = peer_epoch_lb; | |
7195 | } | |
7196 | epoch_t last_sent_epoch = session->last_sent_epoch; | |
7c673cae FG |
7197 | session->sent_epoch_lock.unlock(); |
7198 | ||
9f95a23c TL |
7199 | if (osdmap->get_epoch() <= last_sent_epoch) { |
7200 | return; | |
7201 | } | |
11fdf7f2 | 7202 | |
9f95a23c TL |
7203 | send_incremental_map(last_sent_epoch, con, osdmap); |
7204 | last_sent_epoch = osdmap->get_epoch(); | |
7c673cae FG |
7205 | |
7206 | session->sent_epoch_lock.lock(); | |
7207 | if (session->last_sent_epoch < last_sent_epoch) { | |
9f95a23c TL |
7208 | dout(10) << __func__ << " con " << con |
7209 | << " " << con->get_peer_addr() | |
7210 | << " map epoch " << session->last_sent_epoch | |
7211 | << " -> " << last_sent_epoch << " (shared)" << dendl; | |
7c673cae FG |
7212 | session->last_sent_epoch = last_sent_epoch; |
7213 | } | |
7214 | session->sent_epoch_lock.unlock(); | |
7c673cae FG |
7215 | } |
7216 | ||
9f95a23c | 7217 | void OSD::dispatch_session_waiting(const ceph::ref_t<Session>& session, OSDMapRef osdmap) |
7c673cae | 7218 | { |
9f95a23c | 7219 | ceph_assert(ceph_mutex_is_locked(session->session_dispatch_lock)); |
7c673cae FG |
7220 | |
7221 | auto i = session->waiting_on_map.begin(); | |
7222 | while (i != session->waiting_on_map.end()) { | |
7223 | OpRequestRef op = &(*i); | |
11fdf7f2 | 7224 | ceph_assert(ms_can_fast_dispatch(op->get_req())); |
9f95a23c | 7225 | auto m = op->get_req<MOSDFastDispatchOp>(); |
7c673cae FG |
7226 | if (m->get_min_epoch() > osdmap->get_epoch()) { |
7227 | break; | |
7228 | } | |
7229 | session->waiting_on_map.erase(i++); | |
7230 | op->put(); | |
7231 | ||
7232 | spg_t pgid; | |
7233 | if (m->get_type() == CEPH_MSG_OSD_OP) { | |
7234 | pg_t actual_pgid = osdmap->raw_pg_to_pg( | |
7235 | static_cast<const MOSDOp*>(m)->get_pg()); | |
7236 | if (!osdmap->get_primary_shard(actual_pgid, &pgid)) { | |
7237 | continue; | |
7238 | } | |
7239 | } else { | |
7240 | pgid = m->get_spg(); | |
7241 | } | |
11fdf7f2 | 7242 | enqueue_op(pgid, std::move(op), m->get_map_epoch()); |
7c673cae FG |
7243 | } |
7244 | ||
7245 | if (session->waiting_on_map.empty()) { | |
7246 | clear_session_waiting_on_map(session); | |
7247 | } else { | |
7248 | register_session_waiting_on_map(session); | |
7249 | } | |
7250 | } | |
7251 | ||
7252 | void OSD::ms_fast_dispatch(Message *m) | |
7253 | { | |
20effc67 | 7254 | auto dispatch_span = tracing::osd::tracer.start_trace(__func__); |
11fdf7f2 | 7255 | FUNCTRACE(cct); |
7c673cae FG |
7256 | if (service.is_stopping()) { |
7257 | m->put(); | |
7258 | return; | |
7259 | } | |
11fdf7f2 TL |
7260 | // peering event? |
7261 | switch (m->get_type()) { | |
7262 | case CEPH_MSG_PING: | |
7263 | dout(10) << "ping from " << m->get_source() << dendl; | |
7264 | m->put(); | |
7265 | return; | |
11fdf7f2 TL |
7266 | case MSG_OSD_FORCE_RECOVERY: |
7267 | handle_fast_force_recovery(static_cast<MOSDForceRecovery*>(m)); | |
7268 | return; | |
7269 | case MSG_OSD_SCRUB2: | |
7270 | handle_fast_scrub(static_cast<MOSDScrub2*>(m)); | |
7271 | return; | |
11fdf7f2 TL |
7272 | case MSG_OSD_PG_CREATE2: |
7273 | return handle_fast_pg_create(static_cast<MOSDPGCreate2*>(m)); | |
11fdf7f2 TL |
7274 | case MSG_OSD_PG_NOTIFY: |
7275 | return handle_fast_pg_notify(static_cast<MOSDPGNotify*>(m)); | |
7276 | case MSG_OSD_PG_INFO: | |
7277 | return handle_fast_pg_info(static_cast<MOSDPGInfo*>(m)); | |
7278 | case MSG_OSD_PG_REMOVE: | |
7279 | return handle_fast_pg_remove(static_cast<MOSDPGRemove*>(m)); | |
11fdf7f2 TL |
7280 | // these are single-pg messages that handle themselves |
7281 | case MSG_OSD_PG_LOG: | |
7282 | case MSG_OSD_PG_TRIM: | |
9f95a23c TL |
7283 | case MSG_OSD_PG_NOTIFY2: |
7284 | case MSG_OSD_PG_QUERY2: | |
7285 | case MSG_OSD_PG_INFO2: | |
11fdf7f2 TL |
7286 | case MSG_OSD_BACKFILL_RESERVE: |
7287 | case MSG_OSD_RECOVERY_RESERVE: | |
9f95a23c TL |
7288 | case MSG_OSD_PG_LEASE: |
7289 | case MSG_OSD_PG_LEASE_ACK: | |
11fdf7f2 TL |
7290 | { |
7291 | MOSDPeeringOp *pm = static_cast<MOSDPeeringOp*>(m); | |
7292 | if (require_osd_peer(pm)) { | |
7293 | enqueue_peering_evt( | |
7294 | pm->get_spg(), | |
7295 | PGPeeringEventRef(pm->get_event())); | |
7296 | } | |
7297 | pm->put(); | |
7298 | return; | |
7299 | } | |
7300 | } | |
7301 | ||
7c673cae FG |
7302 | OpRequestRef op = op_tracker.create_request<OpRequest, Message*>(m); |
7303 | { | |
7304 | #ifdef WITH_LTTNG | |
7305 | osd_reqid_t reqid = op->get_reqid(); | |
7306 | #endif | |
7307 | tracepoint(osd, ms_fast_dispatch, reqid.name._type, | |
7308 | reqid.name._num, reqid.tid, reqid.inc); | |
7309 | } | |
20effc67 TL |
7310 | op->osd_parent_span = tracing::osd::tracer.add_span("op-request-created", dispatch_span); |
7311 | ||
7c673cae FG |
7312 | if (m->trace) |
7313 | op->osd_trace.init("osd op", &trace_endpoint, &m->trace); | |
7314 | ||
11fdf7f2 | 7315 | // note sender epoch, min req's epoch |
7c673cae FG |
7316 | op->sent_epoch = static_cast<MOSDFastDispatchOp*>(m)->get_map_epoch(); |
7317 | op->min_epoch = static_cast<MOSDFastDispatchOp*>(m)->get_min_epoch(); | |
11fdf7f2 | 7318 | ceph_assert(op->min_epoch <= op->sent_epoch); // sanity check! |
7c673cae FG |
7319 | |
7320 | service.maybe_inject_dispatch_delay(); | |
7321 | ||
7322 | if (m->get_connection()->has_features(CEPH_FEATUREMASK_RESEND_ON_SPLIT) || | |
7323 | m->get_type() != CEPH_MSG_OSD_OP) { | |
7324 | // queue it directly | |
7325 | enqueue_op( | |
7326 | static_cast<MOSDFastDispatchOp*>(m)->get_spg(), | |
11fdf7f2 | 7327 | std::move(op), |
7c673cae FG |
7328 | static_cast<MOSDFastDispatchOp*>(m)->get_map_epoch()); |
7329 | } else { | |
7330 | // legacy client, and this is an MOSDOp (the *only* fast dispatch | |
7331 | // message that didn't have an explicit spg_t); we need to map | |
7332 | // them to an spg_t while preserving delivery order. | |
11fdf7f2 TL |
7333 | auto priv = m->get_connection()->get_priv(); |
7334 | if (auto session = static_cast<Session*>(priv.get()); session) { | |
7335 | std::lock_guard l{session->session_dispatch_lock}; | |
7336 | op->get(); | |
7337 | session->waiting_on_map.push_back(*op); | |
7338 | OSDMapRef nextmap = service.get_nextmap_reserved(); | |
7339 | dispatch_session_waiting(session, nextmap); | |
7340 | service.release_map(nextmap); | |
7c673cae FG |
7341 | } |
7342 | } | |
f67539c2 | 7343 | OID_EVENT_TRACE_WITH_MSG(m, "MS_FAST_DISPATCH_END", false); |
7c673cae FG |
7344 | } |
7345 | ||
11fdf7f2 | 7346 | int OSD::ms_handle_authentication(Connection *con) |
7c673cae | 7347 | { |
11fdf7f2 | 7348 | int ret = 0; |
9f95a23c | 7349 | auto s = ceph::ref_cast<Session>(con->get_priv()); |
11fdf7f2 | 7350 | if (!s) { |
9f95a23c TL |
7351 | s = ceph::make_ref<Session>(cct, con); |
7352 | con->set_priv(s); | |
11fdf7f2 TL |
7353 | s->entity_name = con->get_peer_entity_name(); |
7354 | dout(10) << __func__ << " new session " << s << " con " << s->con | |
7355 | << " entity " << s->entity_name | |
7356 | << " addr " << con->get_peer_addrs() << dendl; | |
7357 | } else { | |
7358 | dout(10) << __func__ << " existing session " << s << " con " << s->con | |
7359 | << " entity " << s->entity_name | |
7360 | << " addr " << con->get_peer_addrs() << dendl; | |
7c673cae FG |
7361 | } |
7362 | ||
11fdf7f2 | 7363 | AuthCapsInfo &caps_info = con->get_peer_caps_info(); |
9f95a23c | 7364 | if (caps_info.allow_all) { |
11fdf7f2 | 7365 | s->caps.set_allow_all(); |
9f95a23c | 7366 | } else if (caps_info.caps.length() > 0) { |
11fdf7f2 TL |
7367 | bufferlist::const_iterator p = caps_info.caps.cbegin(); |
7368 | string str; | |
7369 | try { | |
7370 | decode(str, p); | |
7371 | } | |
f67539c2 | 7372 | catch (ceph::buffer::error& e) { |
11fdf7f2 TL |
7373 | dout(10) << __func__ << " session " << s << " " << s->entity_name |
7374 | << " failed to decode caps string" << dendl; | |
9f95a23c | 7375 | ret = -EACCES; |
11fdf7f2 TL |
7376 | } |
7377 | if (!ret) { | |
7c673cae | 7378 | bool success = s->caps.parse(str); |
11fdf7f2 TL |
7379 | if (success) { |
7380 | dout(10) << __func__ << " session " << s | |
7381 | << " " << s->entity_name | |
7382 | << " has caps " << s->caps << " '" << str << "'" << dendl; | |
7383 | ret = 1; | |
7384 | } else { | |
7385 | dout(10) << __func__ << " session " << s << " " << s->entity_name | |
7386 | << " failed to parse caps '" << str << "'" << dendl; | |
9f95a23c | 7387 | ret = -EACCES; |
11fdf7f2 | 7388 | } |
7c673cae | 7389 | } |
7c673cae | 7390 | } |
11fdf7f2 | 7391 | return ret; |
7c673cae FG |
7392 | } |
7393 | ||
7394 | void OSD::do_waiters() | |
7395 | { | |
9f95a23c | 7396 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae FG |
7397 | |
7398 | dout(10) << "do_waiters -- start" << dendl; | |
7399 | while (!finished.empty()) { | |
7400 | OpRequestRef next = finished.front(); | |
7401 | finished.pop_front(); | |
7402 | dispatch_op(next); | |
7403 | } | |
7404 | dout(10) << "do_waiters -- finish" << dendl; | |
7405 | } | |
7406 | ||
7407 | void OSD::dispatch_op(OpRequestRef op) | |
7408 | { | |
7409 | switch (op->get_req()->get_type()) { | |
7410 | ||
7411 | case MSG_OSD_PG_CREATE: | |
7412 | handle_pg_create(op); | |
7413 | break; | |
7c673cae FG |
7414 | } |
7415 | } | |
7416 | ||
7417 | void OSD::_dispatch(Message *m) | |
7418 | { | |
9f95a23c | 7419 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae FG |
7420 | dout(20) << "_dispatch " << m << " " << *m << dendl; |
7421 | ||
7422 | switch (m->get_type()) { | |
7c673cae FG |
7423 | // -- don't need OSDMap -- |
7424 | ||
7425 | // map and replication | |
7426 | case CEPH_MSG_OSD_MAP: | |
7427 | handle_osd_map(static_cast<MOSDMap*>(m)); | |
7428 | break; | |
9f95a23c TL |
7429 | case MSG_MON_GET_PURGED_SNAPS_REPLY: |
7430 | handle_get_purged_snaps_reply(static_cast<MMonGetPurgedSnapsReply*>(m)); | |
7431 | break; | |
7c673cae FG |
7432 | |
7433 | // osd | |
7c673cae FG |
7434 | case MSG_OSD_SCRUB: |
7435 | handle_scrub(static_cast<MOSDScrub*>(m)); | |
7436 | break; | |
7437 | ||
11fdf7f2 TL |
7438 | case MSG_COMMAND: |
7439 | handle_command(static_cast<MCommand*>(m)); | |
7440 | return; | |
c07f9fc5 | 7441 | |
7c673cae FG |
7442 | // -- need OSDMap -- |
7443 | ||
7444 | case MSG_OSD_PG_CREATE: | |
7c673cae FG |
7445 | { |
7446 | OpRequestRef op = op_tracker.create_request<OpRequest, Message*>(m); | |
7447 | if (m->trace) | |
7448 | op->osd_trace.init("osd op", &trace_endpoint, &m->trace); | |
7449 | // no map? starting up? | |
9f95a23c | 7450 | if (!get_osdmap()) { |
7c673cae FG |
7451 | dout(7) << "no OSDMap, not booted" << dendl; |
7452 | logger->inc(l_osd_waiting_for_map); | |
7453 | waiting_for_osdmap.push_back(op); | |
7454 | op->mark_delayed("no osdmap"); | |
7455 | break; | |
7456 | } | |
7457 | ||
7458 | // need OSDMap | |
7459 | dispatch_op(op); | |
7460 | } | |
7461 | } | |
7462 | } | |
7463 | ||
11fdf7f2 | 7464 | // remove me post-nautilus |
7c673cae FG |
7465 | void OSD::handle_scrub(MOSDScrub *m) |
7466 | { | |
7467 | dout(10) << "handle_scrub " << *m << dendl; | |
7468 | if (!require_mon_or_mgr_peer(m)) { | |
7469 | m->put(); | |
7470 | return; | |
7471 | } | |
7472 | if (m->fsid != monc->get_fsid()) { | |
11fdf7f2 TL |
7473 | dout(0) << "handle_scrub fsid " << m->fsid << " != " << monc->get_fsid() |
7474 | << dendl; | |
7c673cae FG |
7475 | m->put(); |
7476 | return; | |
7477 | } | |
7478 | ||
11fdf7f2 TL |
7479 | vector<spg_t> spgs; |
7480 | _get_pgids(&spgs); | |
7481 | ||
7482 | if (!m->scrub_pgs.empty()) { | |
7483 | vector<spg_t> v; | |
7484 | for (auto pgid : m->scrub_pgs) { | |
7c673cae | 7485 | spg_t pcand; |
9f95a23c | 7486 | if (get_osdmap()->get_primary_shard(pgid, &pcand) && |
11fdf7f2 TL |
7487 | std::find(spgs.begin(), spgs.end(), pcand) != spgs.end()) { |
7488 | v.push_back(pcand); | |
7c673cae FG |
7489 | } |
7490 | } | |
11fdf7f2 TL |
7491 | spgs.swap(v); |
7492 | } | |
7493 | ||
7494 | for (auto pgid : spgs) { | |
7495 | enqueue_peering_evt( | |
7496 | pgid, | |
7497 | PGPeeringEventRef( | |
7498 | std::make_shared<PGPeeringEvent>( | |
7499 | get_osdmap_epoch(), | |
7500 | get_osdmap_epoch(), | |
9f95a23c | 7501 | PeeringState::RequestScrub(m->deep, m->repair)))); |
7c673cae FG |
7502 | } |
7503 | ||
7504 | m->put(); | |
7505 | } | |
7506 | ||
11fdf7f2 TL |
7507 | void OSD::handle_fast_scrub(MOSDScrub2 *m) |
7508 | { | |
7509 | dout(10) << __func__ << " " << *m << dendl; | |
7510 | if (!require_mon_or_mgr_peer(m)) { | |
7511 | m->put(); | |
7512 | return; | |
7513 | } | |
7514 | if (m->fsid != monc->get_fsid()) { | |
7515 | dout(0) << __func__ << " fsid " << m->fsid << " != " << monc->get_fsid() | |
7516 | << dendl; | |
7517 | m->put(); | |
7518 | return; | |
7519 | } | |
7520 | for (auto pgid : m->scrub_pgs) { | |
7521 | enqueue_peering_evt( | |
7522 | pgid, | |
7523 | PGPeeringEventRef( | |
7524 | std::make_shared<PGPeeringEvent>( | |
7525 | m->epoch, | |
7526 | m->epoch, | |
9f95a23c | 7527 | PeeringState::RequestScrub(m->deep, m->repair)))); |
11fdf7f2 TL |
7528 | } |
7529 | m->put(); | |
7530 | } | |
7531 | ||
7c673cae FG |
7532 | bool OSD::scrub_random_backoff() |
7533 | { | |
7534 | bool coin_flip = (rand() / (double)RAND_MAX >= | |
7535 | cct->_conf->osd_scrub_backoff_ratio); | |
7536 | if (!coin_flip) { | |
7537 | dout(20) << "scrub_random_backoff lost coin flip, randomly backing off" << dendl; | |
7538 | return true; | |
7539 | } | |
7540 | return false; | |
7541 | } | |
7542 | ||
7c673cae | 7543 | |
20effc67 | 7544 | void OSD::sched_scrub() |
f67539c2 | 7545 | { |
20effc67 | 7546 | auto& scrub_scheduler = service.get_scrub_services(); |
f67539c2 | 7547 | |
20effc67 TL |
7548 | // fail fast if no resources are available |
7549 | if (!scrub_scheduler.can_inc_scrubs()) { | |
7550 | dout(20) << __func__ << ": OSD cannot inc scrubs" << dendl; | |
7551 | return; | |
f67539c2 | 7552 | } |
f67539c2 | 7553 | |
20effc67 TL |
7554 | // if there is a PG that is just now trying to reserve scrub replica resources - |
7555 | // we should wait and not initiate a new scrub | |
7556 | if (scrub_scheduler.is_reserving_now()) { | |
7557 | dout(20) << __func__ << ": scrub resources reservation in progress" << dendl; | |
7558 | return; | |
9f95a23c | 7559 | } |
9f95a23c | 7560 | |
20effc67 | 7561 | Scrub::ScrubPreconds env_conditions; |
28e407b8 | 7562 | |
20effc67 TL |
7563 | if (service.is_recovery_active() && !cct->_conf->osd_scrub_during_recovery) { |
7564 | if (!cct->_conf->osd_repair_during_recovery) { | |
7565 | dout(15) << __func__ << ": not scheduling scrubs due to active recovery" | |
7566 | << dendl; | |
7567 | return; | |
28e407b8 | 7568 | } |
20effc67 TL |
7569 | dout(10) << __func__ |
7570 | << " will only schedule explicitly requested repair due to active recovery" | |
7571 | << dendl; | |
7572 | env_conditions.allow_requested_repair_only = true; | |
28e407b8 AA |
7573 | } |
7574 | ||
20effc67 TL |
7575 | if (g_conf()->subsys.should_gather<ceph_subsys_osd, 20>()) { |
7576 | dout(20) << __func__ << " sched_scrub starts" << dendl; | |
7577 | auto all_jobs = scrub_scheduler.list_registered_jobs(); | |
7578 | for (const auto& sj : all_jobs) { | |
7579 | dout(20) << "sched_scrub scrub-queue jobs: " << *sj << dendl; | |
7c673cae FG |
7580 | } |
7581 | } | |
20effc67 TL |
7582 | |
7583 | auto was_started = scrub_scheduler.select_pg_and_scrub(env_conditions); | |
7584 | dout(20) << "sched_scrub done (" << ScrubQueue::attempt_res_text(was_started) | |
7585 | << ")" << dendl; | |
7c673cae FG |
7586 | } |
7587 | ||
20effc67 TL |
7588 | Scrub::schedule_result_t OSDService::initiate_a_scrub(spg_t pgid, |
7589 | bool allow_requested_repair_only) | |
7c673cae | 7590 | { |
20effc67 | 7591 | dout(20) << __func__ << " trying " << pgid << dendl; |
7c673cae | 7592 | |
20effc67 TL |
7593 | // we have a candidate to scrub. We need some PG information to know if scrubbing is |
7594 | // allowed | |
7c673cae | 7595 | |
20effc67 TL |
7596 | PGRef pg = osd->lookup_lock_pg(pgid); |
7597 | if (!pg) { | |
7598 | // the PG was dequeued in the short timespan between creating the candidates list | |
7599 | // (collect_ripe_jobs()) and here | |
7600 | dout(5) << __func__ << " pg " << pgid << " not found" << dendl; | |
7601 | return Scrub::schedule_result_t::no_such_pg; | |
7c673cae FG |
7602 | } |
7603 | ||
20effc67 TL |
7604 | // This has already started, so go on to the next scrub job |
7605 | if (pg->is_scrub_queued_or_active()) { | |
7606 | pg->unlock(); | |
7607 | dout(20) << __func__ << ": already in progress pgid " << pgid << dendl; | |
7608 | return Scrub::schedule_result_t::already_started; | |
7c673cae | 7609 | } |
20effc67 TL |
7610 | // Skip other kinds of scrubbing if only explicitly requested repairing is allowed |
7611 | if (allow_requested_repair_only && !pg->m_planned_scrub.must_repair) { | |
7612 | pg->unlock(); | |
7613 | dout(10) << __func__ << " skip " << pgid | |
7614 | << " because repairing is not explicitly requested on it" << dendl; | |
7615 | return Scrub::schedule_result_t::preconditions; | |
b5b8bbf5 FG |
7616 | } |
7617 | ||
20effc67 TL |
7618 | auto scrub_attempt = pg->sched_scrub(); |
7619 | pg->unlock(); | |
7620 | return scrub_attempt; | |
7c673cae FG |
7621 | } |
7622 | ||
494da23a TL |
7623 | void OSD::resched_all_scrubs() |
7624 | { | |
7625 | dout(10) << __func__ << ": start" << dendl; | |
20effc67 TL |
7626 | auto all_jobs = service.get_scrub_services().list_registered_jobs(); |
7627 | for (auto& e : all_jobs) { | |
7628 | ||
7629 | auto& job = *e; | |
7630 | dout(20) << __func__ << ": examine " << job.pgid << dendl; | |
7631 | ||
7632 | PGRef pg = _lookup_lock_pg(job.pgid); | |
7633 | if (!pg) | |
7634 | continue; | |
7635 | ||
7636 | if (!pg->m_planned_scrub.must_scrub && !pg->m_planned_scrub.need_auto) { | |
7637 | dout(15) << __func__ << ": reschedule " << job.pgid << dendl; | |
7638 | pg->reschedule_scrub(); | |
7639 | } | |
7640 | pg->unlock(); | |
494da23a TL |
7641 | } |
7642 | dout(10) << __func__ << ": done" << dendl; | |
7643 | } | |
7644 | ||
11fdf7f2 TL |
7645 | MPGStats* OSD::collect_pg_stats() |
7646 | { | |
20effc67 | 7647 | dout(15) << __func__ << dendl; |
11fdf7f2 TL |
7648 | // This implementation unconditionally sends every is_primary PG's |
7649 | // stats every time we're called. This has equivalent cost to the | |
7650 | // previous implementation's worst case where all PGs are busy and | |
7651 | // their stats are always enqueued for sending. | |
9f95a23c | 7652 | std::shared_lock l{map_lock}; |
11fdf7f2 | 7653 | |
11fdf7f2 TL |
7654 | osd_stat_t cur_stat = service.get_osd_stat(); |
7655 | cur_stat.os_perf_stat = store->get_cur_stats(); | |
7656 | ||
9f95a23c | 7657 | auto m = new MPGStats(monc->get_fsid(), get_osdmap_epoch()); |
11fdf7f2 TL |
7658 | m->osd_stat = cur_stat; |
7659 | ||
7660 | std::lock_guard lec{min_last_epoch_clean_lock}; | |
9f95a23c | 7661 | min_last_epoch_clean = get_osdmap_epoch(); |
11fdf7f2 TL |
7662 | min_last_epoch_clean_pgs.clear(); |
7663 | ||
7664 | std::set<int64_t> pool_set; | |
7665 | vector<PGRef> pgs; | |
7666 | _get_pgs(&pgs); | |
7667 | for (auto& pg : pgs) { | |
7668 | auto pool = pg->pg_id.pgid.pool(); | |
7669 | pool_set.emplace((int64_t)pool); | |
7670 | if (!pg->is_primary()) { | |
7671 | continue; | |
7672 | } | |
20effc67 | 7673 | pg->with_pg_stats([&](const pg_stat_t& s, epoch_t lec) { |
11fdf7f2 | 7674 | m->pg_stat[pg->pg_id.pgid] = s; |
f67539c2 | 7675 | min_last_epoch_clean = std::min(min_last_epoch_clean, lec); |
11fdf7f2 TL |
7676 | min_last_epoch_clean_pgs.push_back(pg->pg_id.pgid); |
7677 | }); | |
7678 | } | |
7679 | store_statfs_t st; | |
81eedcae | 7680 | bool per_pool_stats = false; |
9f95a23c | 7681 | bool per_pool_omap_stats = false; |
11fdf7f2 | 7682 | for (auto p : pool_set) { |
9f95a23c | 7683 | int r = store->pool_statfs(p, &st, &per_pool_omap_stats); |
11fdf7f2 TL |
7684 | if (r == -ENOTSUP) { |
7685 | break; | |
7686 | } else { | |
7687 | assert(r >= 0); | |
7688 | m->pool_stat[p] = st; | |
81eedcae | 7689 | per_pool_stats = true; |
11fdf7f2 TL |
7690 | } |
7691 | } | |
7c673cae | 7692 | |
81eedcae TL |
7693 | // indicate whether we are reporting per-pool stats |
7694 | m->osd_stat.num_osds = 1; | |
7695 | m->osd_stat.num_per_pool_osds = per_pool_stats ? 1 : 0; | |
9f95a23c | 7696 | m->osd_stat.num_per_pool_omap_osds = per_pool_omap_stats ? 1 : 0; |
81eedcae | 7697 | |
11fdf7f2 TL |
7698 | return m; |
7699 | } | |
7c673cae | 7700 | |
11fdf7f2 | 7701 | vector<DaemonHealthMetric> OSD::get_health_metrics() |
b32b8144 | 7702 | { |
11fdf7f2 TL |
7703 | vector<DaemonHealthMetric> metrics; |
7704 | { | |
7705 | utime_t oldest_secs; | |
7706 | const utime_t now = ceph_clock_now(); | |
7707 | auto too_old = now; | |
7708 | too_old -= cct->_conf.get_val<double>("osd_op_complaint_time"); | |
7709 | int slow = 0; | |
7710 | TrackedOpRef oldest_op; | |
20effc67 TL |
7711 | OSDMapRef osdmap = get_osdmap(); |
7712 | // map of slow op counts by slow op event type for an aggregated logging to | |
7713 | // the cluster log. | |
7714 | map<uint8_t, int> slow_op_types; | |
7715 | // map of slow op counts by pool for reporting a pool name with highest | |
7716 | // slow ops. | |
7717 | map<uint64_t, int> slow_op_pools; | |
7718 | bool log_aggregated_slow_op = | |
7719 | cct->_conf.get_val<bool>("osd_aggregated_slow_ops_logging"); | |
11fdf7f2 TL |
7720 | auto count_slow_ops = [&](TrackedOp& op) { |
7721 | if (op.get_initiated() < too_old) { | |
9f95a23c TL |
7722 | stringstream ss; |
7723 | ss << "slow request " << op.get_desc() | |
7724 | << " initiated " | |
7725 | << op.get_initiated() | |
7726 | << " currently " | |
7727 | << op.state_string(); | |
7728 | lgeneric_subdout(cct,osd,20) << ss.str() << dendl; | |
20effc67 TL |
7729 | if (log_aggregated_slow_op) { |
7730 | if (const OpRequest *req = dynamic_cast<const OpRequest *>(&op)) { | |
7731 | uint8_t op_type = req->state_flag(); | |
7732 | auto m = req->get_req<MOSDFastDispatchOp>(); | |
7733 | uint64_t poolid = m->get_spg().pgid.m_pool; | |
7734 | slow_op_types[op_type]++; | |
7735 | if (poolid > 0 && poolid <= (uint64_t) osdmap->get_pool_max()) { | |
7736 | slow_op_pools[poolid]++; | |
7737 | } | |
7738 | } | |
7739 | } else { | |
7740 | clog->warn() << ss.str(); | |
7741 | } | |
11fdf7f2 TL |
7742 | slow++; |
7743 | if (!oldest_op || op.get_initiated() < oldest_op->get_initiated()) { | |
7744 | oldest_op = &op; | |
7745 | } | |
7746 | return true; | |
7747 | } else { | |
7748 | return false; | |
7749 | } | |
7750 | }; | |
7751 | if (op_tracker.visit_ops_in_flight(&oldest_secs, count_slow_ops)) { | |
7752 | if (slow) { | |
7753 | derr << __func__ << " reporting " << slow << " slow ops, oldest is " | |
7754 | << oldest_op->get_desc() << dendl; | |
20effc67 TL |
7755 | if (log_aggregated_slow_op && |
7756 | slow_op_types.size() > 0) { | |
7757 | stringstream ss; | |
7758 | ss << slow << " slow requests (by type [ "; | |
7759 | for (const auto& [op_type, count] : slow_op_types) { | |
7760 | ss << "'" << OpRequest::get_state_string(op_type) | |
7761 | << "' : " << count | |
7762 | << " "; | |
7763 | } | |
7764 | auto slow_pool_it = std::max_element(slow_op_pools.begin(), slow_op_pools.end(), | |
7765 | [](std::pair<uint64_t, int> p1, std::pair<uint64_t, int> p2) { | |
7766 | return p1.second < p2.second; | |
7767 | }); | |
7768 | if (osdmap->get_pools().find(slow_pool_it->first) != osdmap->get_pools().end()) { | |
7769 | string pool_name = osdmap->get_pool_name(slow_pool_it->first); | |
7770 | ss << "] most affected pool [ '" | |
7771 | << pool_name | |
7772 | << "' : " | |
7773 | << slow_pool_it->second | |
7774 | << " ])"; | |
7775 | } else { | |
7776 | ss << "])"; | |
7777 | } | |
7778 | lgeneric_subdout(cct,osd,20) << ss.str() << dendl; | |
7779 | clog->warn() << ss.str(); | |
7780 | } | |
11fdf7f2 TL |
7781 | } |
7782 | metrics.emplace_back(daemon_metric::SLOW_OPS, slow, oldest_secs); | |
7783 | } else { | |
7784 | // no news is not good news. | |
7785 | metrics.emplace_back(daemon_metric::SLOW_OPS, 0, 0); | |
7786 | } | |
7787 | } | |
7788 | { | |
7789 | std::lock_guard l(pending_creates_lock); | |
7790 | auto n_primaries = pending_creates_from_mon; | |
7791 | for (const auto& create : pending_creates_from_osd) { | |
7792 | if (create.second) { | |
7793 | n_primaries++; | |
7794 | } | |
b32b8144 | 7795 | } |
11fdf7f2 | 7796 | metrics.emplace_back(daemon_metric::PENDING_CREATING_PGS, n_primaries); |
b32b8144 | 7797 | } |
b32b8144 FG |
7798 | return metrics; |
7799 | } | |
7800 | ||
7c673cae FG |
7801 | // ===================================================== |
7802 | // MAP | |
7803 | ||
7804 | void OSD::wait_for_new_map(OpRequestRef op) | |
7805 | { | |
7806 | // ask? | |
7807 | if (waiting_for_osdmap.empty()) { | |
9f95a23c | 7808 | osdmap_subscribe(get_osdmap_epoch() + 1, false); |
7c673cae FG |
7809 | } |
7810 | ||
7811 | logger->inc(l_osd_waiting_for_map); | |
7812 | waiting_for_osdmap.push_back(op); | |
7813 | op->mark_delayed("wait for new map"); | |
7814 | } | |
7815 | ||
7816 | ||
7817 | /** update_map | |
7818 | * assimilate new OSDMap(s). scan pgs, etc. | |
7819 | */ | |
7820 | ||
7821 | void OSD::note_down_osd(int peer) | |
7822 | { | |
9f95a23c TL |
7823 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7824 | cluster_messenger->mark_down_addrs(get_osdmap()->get_cluster_addrs(peer)); | |
7c673cae | 7825 | |
9f95a23c | 7826 | std::lock_guard l{heartbeat_lock}; |
7c673cae FG |
7827 | failure_queue.erase(peer); |
7828 | failure_pending.erase(peer); | |
7829 | map<int,HeartbeatInfo>::iterator p = heartbeat_peers.find(peer); | |
7830 | if (p != heartbeat_peers.end()) { | |
9f95a23c | 7831 | p->second.clear_mark_down(); |
7c673cae FG |
7832 | heartbeat_peers.erase(p); |
7833 | } | |
7c673cae FG |
7834 | } |
7835 | ||
7836 | void OSD::note_up_osd(int peer) | |
7837 | { | |
7c673cae FG |
7838 | heartbeat_set_peers_need_update(); |
7839 | } | |
7840 | ||
7841 | struct C_OnMapCommit : public Context { | |
7842 | OSD *osd; | |
7843 | epoch_t first, last; | |
7844 | MOSDMap *msg; | |
7845 | C_OnMapCommit(OSD *o, epoch_t f, epoch_t l, MOSDMap *m) | |
7846 | : osd(o), first(f), last(l), msg(m) {} | |
7847 | void finish(int r) override { | |
7848 | osd->_committed_osd_maps(first, last, msg); | |
7849 | msg->put(); | |
7850 | } | |
7851 | }; | |
7852 | ||
7c673cae FG |
7853 | void OSD::osdmap_subscribe(version_t epoch, bool force_request) |
7854 | { | |
11fdf7f2 | 7855 | std::lock_guard l(osdmap_subscribe_lock); |
181888fb | 7856 | if (latest_subscribed_epoch >= epoch && !force_request) |
7c673cae FG |
7857 | return; |
7858 | ||
11fdf7f2 | 7859 | latest_subscribed_epoch = std::max<uint64_t>(epoch, latest_subscribed_epoch); |
181888fb | 7860 | |
7c673cae FG |
7861 | if (monc->sub_want_increment("osdmap", epoch, CEPH_SUBSCRIBE_ONETIME) || |
7862 | force_request) { | |
7863 | monc->renew_subs(); | |
7864 | } | |
7865 | } | |
7866 | ||
7867 | void OSD::trim_maps(epoch_t oldest, int nreceived, bool skip_maps) | |
7868 | { | |
7869 | epoch_t min = std::min(oldest, service.map_cache.cached_key_lower_bound()); | |
7870 | if (min <= superblock.oldest_map) | |
7871 | return; | |
7872 | ||
7873 | int num = 0; | |
7874 | ObjectStore::Transaction t; | |
7875 | for (epoch_t e = superblock.oldest_map; e < min; ++e) { | |
7876 | dout(20) << " removing old osdmap epoch " << e << dendl; | |
7877 | t.remove(coll_t::meta(), get_osdmap_pobject_name(e)); | |
7878 | t.remove(coll_t::meta(), get_inc_osdmap_pobject_name(e)); | |
7879 | superblock.oldest_map = e + 1; | |
7880 | num++; | |
7881 | if (num >= cct->_conf->osd_target_transaction_size && num >= nreceived) { | |
7882 | service.publish_superblock(superblock); | |
7883 | write_superblock(t); | |
11fdf7f2 TL |
7884 | int tr = store->queue_transaction(service.meta_ch, std::move(t), nullptr); |
7885 | ceph_assert(tr == 0); | |
7c673cae FG |
7886 | num = 0; |
7887 | if (!skip_maps) { | |
7888 | // skip_maps leaves us with a range of old maps if we fail to remove all | |
7889 | // of them before moving superblock.oldest_map forward to the first map | |
7890 | // in the incoming MOSDMap msg. so we should continue removing them in | |
7891 | // this case, even we could do huge series of delete transactions all at | |
7892 | // once. | |
7893 | break; | |
7894 | } | |
7895 | } | |
7896 | } | |
7897 | if (num > 0) { | |
7898 | service.publish_superblock(superblock); | |
7899 | write_superblock(t); | |
11fdf7f2 TL |
7900 | int tr = store->queue_transaction(service.meta_ch, std::move(t), nullptr); |
7901 | ceph_assert(tr == 0); | |
7c673cae FG |
7902 | } |
7903 | // we should not remove the cached maps | |
11fdf7f2 | 7904 | ceph_assert(min <= service.map_cache.cached_key_lower_bound()); |
7c673cae FG |
7905 | } |
7906 | ||
7907 | void OSD::handle_osd_map(MOSDMap *m) | |
7908 | { | |
11fdf7f2 TL |
7909 | // wait for pgs to catch up |
7910 | { | |
7911 | // we extend the map cache pins to accomodate pgs slow to consume maps | |
7912 | // for some period, until we hit the max_lag_factor bound, at which point | |
7913 | // we block here to stop injesting more maps than they are able to keep | |
7914 | // up with. | |
7915 | epoch_t max_lag = cct->_conf->osd_map_cache_size * | |
7916 | m_osd_pg_epoch_max_lag_factor; | |
7917 | ceph_assert(max_lag > 0); | |
7918 | epoch_t osd_min = 0; | |
7919 | for (auto shard : shards) { | |
7920 | epoch_t min = shard->get_min_pg_epoch(); | |
7921 | if (osd_min == 0 || min < osd_min) { | |
7922 | osd_min = min; | |
7923 | } | |
7924 | } | |
9f95a23c | 7925 | epoch_t osdmap_epoch = get_osdmap_epoch(); |
11fdf7f2 | 7926 | if (osd_min > 0 && |
9f95a23c TL |
7927 | osdmap_epoch > max_lag && |
7928 | osdmap_epoch - max_lag > osd_min) { | |
7929 | epoch_t need = osdmap_epoch - max_lag; | |
11fdf7f2 TL |
7930 | dout(10) << __func__ << " waiting for pgs to catch up (need " << need |
7931 | << " max_lag " << max_lag << ")" << dendl; | |
7932 | for (auto shard : shards) { | |
7933 | epoch_t min = shard->get_min_pg_epoch(); | |
7934 | if (need > min) { | |
7935 | dout(10) << __func__ << " waiting for pgs to consume " << need | |
7936 | << " (shard " << shard->shard_id << " min " << min | |
7937 | << ", map cache is " << cct->_conf->osd_map_cache_size | |
7938 | << ", max_lag_factor " << m_osd_pg_epoch_max_lag_factor | |
7939 | << ")" << dendl; | |
7940 | unlock_guard unlock{osd_lock}; | |
7941 | shard->wait_min_pg_epoch(need); | |
7942 | } | |
7943 | } | |
7944 | } | |
7945 | } | |
7946 | ||
9f95a23c | 7947 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
11fdf7f2 TL |
7948 | map<epoch_t,OSDMapRef> added_maps; |
7949 | map<epoch_t,bufferlist> added_maps_bl; | |
7c673cae FG |
7950 | if (m->fsid != monc->get_fsid()) { |
7951 | dout(0) << "handle_osd_map fsid " << m->fsid << " != " | |
7952 | << monc->get_fsid() << dendl; | |
7953 | m->put(); | |
7954 | return; | |
7955 | } | |
7956 | if (is_initializing()) { | |
7957 | dout(0) << "ignoring osdmap until we have initialized" << dendl; | |
7958 | m->put(); | |
7959 | return; | |
7960 | } | |
7961 | ||
9f95a23c TL |
7962 | auto session = ceph::ref_cast<Session>(m->get_connection()->get_priv()); |
7963 | if (session && !(session->entity_name.is_mon() || | |
7c673cae FG |
7964 | session->entity_name.is_osd())) { |
7965 | //not enough perms! | |
7966 | dout(10) << "got osd map from Session " << session | |
7967 | << " which we can't take maps from (not a mon or osd)" << dendl; | |
7968 | m->put(); | |
7c673cae FG |
7969 | return; |
7970 | } | |
7c673cae FG |
7971 | |
7972 | // share with the objecter | |
7973 | if (!is_preboot()) | |
7974 | service.objecter->handle_osd_map(m); | |
7975 | ||
7976 | epoch_t first = m->get_first(); | |
7977 | epoch_t last = m->get_last(); | |
7978 | dout(3) << "handle_osd_map epochs [" << first << "," << last << "], i have " | |
7979 | << superblock.newest_map | |
7980 | << ", src has [" << m->oldest_map << "," << m->newest_map << "]" | |
7981 | << dendl; | |
7982 | ||
7983 | logger->inc(l_osd_map); | |
7984 | logger->inc(l_osd_mape, last - first + 1); | |
7985 | if (first <= superblock.newest_map) | |
7986 | logger->inc(l_osd_mape_dup, superblock.newest_map - first + 1); | |
7987 | if (service.max_oldest_map < m->oldest_map) { | |
7988 | service.max_oldest_map = m->oldest_map; | |
11fdf7f2 | 7989 | ceph_assert(service.max_oldest_map >= superblock.oldest_map); |
7c673cae FG |
7990 | } |
7991 | ||
7992 | // make sure there is something new, here, before we bother flushing | |
7993 | // the queues and such | |
7994 | if (last <= superblock.newest_map) { | |
7995 | dout(10) << " no new maps here, dropping" << dendl; | |
7996 | m->put(); | |
7997 | return; | |
7998 | } | |
7999 | ||
8000 | // missing some? | |
8001 | bool skip_maps = false; | |
8002 | if (first > superblock.newest_map + 1) { | |
8003 | dout(10) << "handle_osd_map message skips epochs " | |
8004 | << superblock.newest_map + 1 << ".." << (first-1) << dendl; | |
8005 | if (m->oldest_map <= superblock.newest_map + 1) { | |
8006 | osdmap_subscribe(superblock.newest_map + 1, false); | |
8007 | m->put(); | |
8008 | return; | |
8009 | } | |
8010 | // always try to get the full range of maps--as many as we can. this | |
8011 | // 1- is good to have | |
8012 | // 2- is at present the only way to ensure that we get a *full* map as | |
8013 | // the first map! | |
8014 | if (m->oldest_map < first) { | |
8015 | osdmap_subscribe(m->oldest_map - 1, true); | |
8016 | m->put(); | |
8017 | return; | |
8018 | } | |
8019 | skip_maps = true; | |
8020 | } | |
8021 | ||
8022 | ObjectStore::Transaction t; | |
8023 | uint64_t txn_size = 0; | |
8024 | ||
9f95a23c TL |
8025 | map<epoch_t,mempool::osdmap::map<int64_t,snap_interval_set_t>> purged_snaps; |
8026 | ||
7c673cae | 8027 | // store new maps: queue for disk and put in the osdmap cache |
11fdf7f2 | 8028 | epoch_t start = std::max(superblock.newest_map + 1, first); |
7c673cae FG |
8029 | for (epoch_t e = start; e <= last; e++) { |
8030 | if (txn_size >= t.get_num_bytes()) { | |
8031 | derr << __func__ << " transaction size overflowed" << dendl; | |
11fdf7f2 | 8032 | ceph_assert(txn_size < t.get_num_bytes()); |
7c673cae FG |
8033 | } |
8034 | txn_size = t.get_num_bytes(); | |
8035 | map<epoch_t,bufferlist>::iterator p; | |
8036 | p = m->maps.find(e); | |
8037 | if (p != m->maps.end()) { | |
8038 | dout(10) << "handle_osd_map got full map for epoch " << e << dendl; | |
8039 | OSDMap *o = new OSDMap; | |
8040 | bufferlist& bl = p->second; | |
8041 | ||
8042 | o->decode(bl); | |
8043 | ||
9f95a23c TL |
8044 | purged_snaps[e] = o->get_new_purged_snaps(); |
8045 | ||
7c673cae FG |
8046 | ghobject_t fulloid = get_osdmap_pobject_name(e); |
8047 | t.write(coll_t::meta(), fulloid, 0, bl.length(), bl); | |
11fdf7f2 TL |
8048 | added_maps[e] = add_map(o); |
8049 | added_maps_bl[e] = bl; | |
7c673cae FG |
8050 | got_full_map(e); |
8051 | continue; | |
8052 | } | |
8053 | ||
8054 | p = m->incremental_maps.find(e); | |
8055 | if (p != m->incremental_maps.end()) { | |
8056 | dout(10) << "handle_osd_map got inc map for epoch " << e << dendl; | |
8057 | bufferlist& bl = p->second; | |
8058 | ghobject_t oid = get_inc_osdmap_pobject_name(e); | |
8059 | t.write(coll_t::meta(), oid, 0, bl.length(), bl); | |
7c673cae FG |
8060 | |
8061 | OSDMap *o = new OSDMap; | |
8062 | if (e > 1) { | |
8063 | bufferlist obl; | |
8064 | bool got = get_map_bl(e - 1, obl); | |
11fdf7f2 TL |
8065 | if (!got) { |
8066 | auto p = added_maps_bl.find(e - 1); | |
8067 | ceph_assert(p != added_maps_bl.end()); | |
8068 | obl = p->second; | |
8069 | } | |
7c673cae FG |
8070 | o->decode(obl); |
8071 | } | |
8072 | ||
8073 | OSDMap::Incremental inc; | |
11fdf7f2 | 8074 | auto p = bl.cbegin(); |
7c673cae | 8075 | inc.decode(p); |
494da23a | 8076 | |
7c673cae | 8077 | if (o->apply_incremental(inc) < 0) { |
9f95a23c | 8078 | derr << "ERROR: bad fsid? i have " << get_osdmap()->get_fsid() << " and inc has " << inc.fsid << dendl; |
11fdf7f2 | 8079 | ceph_abort_msg("bad fsid"); |
7c673cae FG |
8080 | } |
8081 | ||
8082 | bufferlist fbl; | |
8083 | o->encode(fbl, inc.encode_features | CEPH_FEATURE_RESERVED); | |
8084 | ||
8085 | bool injected_failure = false; | |
8086 | if (cct->_conf->osd_inject_bad_map_crc_probability > 0 && | |
8087 | (rand() % 10000) < cct->_conf->osd_inject_bad_map_crc_probability*10000.0) { | |
8088 | derr << __func__ << " injecting map crc failure" << dendl; | |
8089 | injected_failure = true; | |
8090 | } | |
8091 | ||
8092 | if ((inc.have_crc && o->get_crc() != inc.full_crc) || injected_failure) { | |
8093 | dout(2) << "got incremental " << e | |
8094 | << " but failed to encode full with correct crc; requesting" | |
8095 | << dendl; | |
8096 | clog->warn() << "failed to encode map e" << e << " with expected crc"; | |
8097 | dout(20) << "my encoded map was:\n"; | |
8098 | fbl.hexdump(*_dout); | |
8099 | *_dout << dendl; | |
8100 | delete o; | |
8101 | request_full_map(e, last); | |
8102 | last = e - 1; | |
f6b5b4d7 TL |
8103 | |
8104 | // don't continue committing if we failed to enc the first inc map | |
8105 | if (last < start) { | |
8106 | dout(10) << __func__ << " bailing because last < start (" << last << "<" << start << ")" << dendl; | |
8107 | m->put(); | |
8108 | return; | |
8109 | } | |
7c673cae FG |
8110 | break; |
8111 | } | |
8112 | got_full_map(e); | |
9f95a23c | 8113 | purged_snaps[e] = o->get_new_purged_snaps(); |
7c673cae FG |
8114 | |
8115 | ghobject_t fulloid = get_osdmap_pobject_name(e); | |
8116 | t.write(coll_t::meta(), fulloid, 0, fbl.length(), fbl); | |
11fdf7f2 TL |
8117 | added_maps[e] = add_map(o); |
8118 | added_maps_bl[e] = fbl; | |
7c673cae FG |
8119 | continue; |
8120 | } | |
8121 | ||
11fdf7f2 | 8122 | ceph_abort_msg("MOSDMap lied about what maps it had?"); |
7c673cae FG |
8123 | } |
8124 | ||
8125 | // even if this map isn't from a mon, we may have satisfied our subscription | |
8126 | monc->sub_got("osdmap", last); | |
8127 | ||
8128 | if (!m->maps.empty() && requested_full_first) { | |
8129 | dout(10) << __func__ << " still missing full maps " << requested_full_first | |
8130 | << ".." << requested_full_last << dendl; | |
8131 | rerequest_full_maps(); | |
8132 | } | |
8133 | ||
7c673cae FG |
8134 | if (superblock.oldest_map) { |
8135 | // make sure we at least keep pace with incoming maps | |
8136 | trim_maps(m->oldest_map, last - first + 1, skip_maps); | |
11fdf7f2 | 8137 | pg_num_history.prune(superblock.oldest_map); |
7c673cae FG |
8138 | } |
8139 | ||
8140 | if (!superblock.oldest_map || skip_maps) | |
8141 | superblock.oldest_map = first; | |
8142 | superblock.newest_map = last; | |
8143 | superblock.current_epoch = last; | |
8144 | ||
8145 | // note in the superblock that we were clean thru the prior epoch | |
8146 | epoch_t boot_epoch = service.get_boot_epoch(); | |
8147 | if (boot_epoch && boot_epoch >= superblock.mounted) { | |
8148 | superblock.mounted = boot_epoch; | |
8149 | superblock.clean_thru = last; | |
8150 | } | |
8151 | ||
11fdf7f2 TL |
8152 | // check for pg_num changes and deleted pools |
8153 | OSDMapRef lastmap; | |
8154 | for (auto& i : added_maps) { | |
8155 | if (!lastmap) { | |
8156 | if (!(lastmap = service.try_get_map(i.first - 1))) { | |
8157 | dout(10) << __func__ << " can't get previous map " << i.first - 1 | |
8158 | << " probably first start of this osd" << dendl; | |
8159 | continue; | |
8160 | } | |
8161 | } | |
8162 | ceph_assert(lastmap->get_epoch() + 1 == i.second->get_epoch()); | |
8163 | for (auto& j : lastmap->get_pools()) { | |
8164 | if (!i.second->have_pg_pool(j.first)) { | |
8165 | pg_num_history.log_pool_delete(i.first, j.first); | |
8166 | dout(10) << __func__ << " recording final pg_pool_t for pool " | |
8167 | << j.first << dendl; | |
8168 | // this information is needed by _make_pg() if have to restart before | |
8169 | // the pool is deleted and need to instantiate a new (zombie) PG[Pool]. | |
8170 | ghobject_t obj = make_final_pool_info_oid(j.first); | |
8171 | bufferlist bl; | |
8172 | encode(j.second, bl, CEPH_FEATURES_ALL); | |
8173 | string name = lastmap->get_pool_name(j.first); | |
8174 | encode(name, bl); | |
8175 | map<string,string> profile; | |
8176 | if (lastmap->get_pg_pool(j.first)->is_erasure()) { | |
8177 | profile = lastmap->get_erasure_code_profile( | |
8178 | lastmap->get_pg_pool(j.first)->erasure_code_profile); | |
8179 | } | |
8180 | encode(profile, bl); | |
8181 | t.write(coll_t::meta(), obj, 0, bl.length(), bl); | |
11fdf7f2 TL |
8182 | } else if (unsigned new_pg_num = i.second->get_pg_num(j.first); |
8183 | new_pg_num != j.second.get_pg_num()) { | |
8184 | dout(10) << __func__ << " recording pool " << j.first << " pg_num " | |
8185 | << j.second.get_pg_num() << " -> " << new_pg_num << dendl; | |
8186 | pg_num_history.log_pg_num_change(i.first, j.first, new_pg_num); | |
8187 | } | |
8188 | } | |
8189 | for (auto& j : i.second->get_pools()) { | |
8190 | if (!lastmap->have_pg_pool(j.first)) { | |
8191 | dout(10) << __func__ << " recording new pool " << j.first << " pg_num " | |
8192 | << j.second.get_pg_num() << dendl; | |
8193 | pg_num_history.log_pg_num_change(i.first, j.first, | |
8194 | j.second.get_pg_num()); | |
8195 | } | |
8196 | } | |
8197 | lastmap = i.second; | |
8198 | } | |
8199 | pg_num_history.epoch = last; | |
8200 | { | |
8201 | bufferlist bl; | |
8202 | ::encode(pg_num_history, bl); | |
8203 | t.write(coll_t::meta(), make_pg_num_history_oid(), 0, bl.length(), bl); | |
8204 | dout(20) << __func__ << " pg_num_history " << pg_num_history << dendl; | |
8205 | } | |
8206 | ||
9f95a23c TL |
8207 | // record new purged_snaps |
8208 | if (superblock.purged_snaps_last == start - 1) { | |
20effc67 | 8209 | SnapMapper::record_purged_snaps(cct, store.get(), service.meta_ch, |
9f95a23c TL |
8210 | make_purged_snaps_oid(), &t, |
8211 | purged_snaps); | |
8212 | superblock.purged_snaps_last = last; | |
8213 | } else { | |
8214 | dout(10) << __func__ << " superblock purged_snaps_last is " | |
8215 | << superblock.purged_snaps_last | |
8216 | << ", not recording new purged_snaps" << dendl; | |
8217 | } | |
8218 | ||
7c673cae FG |
8219 | // superblock and commit |
8220 | write_superblock(t); | |
11fdf7f2 | 8221 | t.register_on_commit(new C_OnMapCommit(this, start, last, m)); |
7c673cae | 8222 | store->queue_transaction( |
11fdf7f2 TL |
8223 | service.meta_ch, |
8224 | std::move(t)); | |
7c673cae FG |
8225 | service.publish_superblock(superblock); |
8226 | } | |
8227 | ||
8228 | void OSD::_committed_osd_maps(epoch_t first, epoch_t last, MOSDMap *m) | |
8229 | { | |
8230 | dout(10) << __func__ << " " << first << ".." << last << dendl; | |
8231 | if (is_stopping()) { | |
8232 | dout(10) << __func__ << " bailing, we are shutting down" << dendl; | |
8233 | return; | |
8234 | } | |
11fdf7f2 | 8235 | std::lock_guard l(osd_lock); |
31f18b77 FG |
8236 | if (is_stopping()) { |
8237 | dout(10) << __func__ << " bailing, we are shutting down" << dendl; | |
8238 | return; | |
8239 | } | |
9f95a23c | 8240 | map_lock.lock(); |
7c673cae | 8241 | |
f6b5b4d7 TL |
8242 | ceph_assert(first <= last); |
8243 | ||
7c673cae FG |
8244 | bool do_shutdown = false; |
8245 | bool do_restart = false; | |
8246 | bool network_error = false; | |
f6b5b4d7 | 8247 | OSDMapRef osdmap = get_osdmap(); |
7c673cae FG |
8248 | |
8249 | // advance through the new maps | |
8250 | for (epoch_t cur = first; cur <= last; cur++) { | |
8251 | dout(10) << " advance to epoch " << cur | |
8252 | << " (<= last " << last | |
8253 | << " <= newest_map " << superblock.newest_map | |
8254 | << ")" << dendl; | |
8255 | ||
8256 | OSDMapRef newmap = get_map(cur); | |
11fdf7f2 | 8257 | ceph_assert(newmap); // we just cached it above! |
7c673cae | 8258 | |
f67539c2 | 8259 | // start blocklisting messages sent to peers that go down. |
7c673cae FG |
8260 | service.pre_publish_map(newmap); |
8261 | ||
8262 | // kill connections to newly down osds | |
8263 | bool waited_for_reservations = false; | |
8264 | set<int> old; | |
9f95a23c | 8265 | osdmap = get_osdmap(); |
7c673cae FG |
8266 | osdmap->get_all_osds(old); |
8267 | for (set<int>::iterator p = old.begin(); p != old.end(); ++p) { | |
8268 | if (*p != whoami && | |
8269 | osdmap->is_up(*p) && // in old map | |
8270 | newmap->is_down(*p)) { // but not the new one | |
8271 | if (!waited_for_reservations) { | |
8272 | service.await_reserved_maps(); | |
8273 | waited_for_reservations = true; | |
8274 | } | |
8275 | note_down_osd(*p); | |
8276 | } else if (*p != whoami && | |
8277 | osdmap->is_down(*p) && | |
8278 | newmap->is_up(*p)) { | |
8279 | note_up_osd(*p); | |
8280 | } | |
8281 | } | |
8282 | ||
81eedcae | 8283 | if (osdmap->is_noup(whoami) != newmap->is_noup(whoami)) { |
7c673cae FG |
8284 | dout(10) << __func__ << " NOUP flag changed in " << newmap->get_epoch() |
8285 | << dendl; | |
8286 | if (is_booting()) { | |
8287 | // this captures the case where we sent the boot message while | |
8288 | // NOUP was being set on the mon and our boot request was | |
8289 | // dropped, and then later it is cleared. it imperfectly | |
8290 | // handles the case where our original boot message was not | |
8291 | // dropped and we restart even though we might have booted, but | |
8292 | // that is harmless (boot will just take slightly longer). | |
8293 | do_restart = true; | |
8294 | } | |
8295 | } | |
8296 | ||
9f95a23c TL |
8297 | osdmap = std::move(newmap); |
8298 | set_osdmap(osdmap); | |
7c673cae FG |
8299 | epoch_t up_epoch; |
8300 | epoch_t boot_epoch; | |
8301 | service.retrieve_epochs(&boot_epoch, &up_epoch, NULL); | |
8302 | if (!up_epoch && | |
8303 | osdmap->is_up(whoami) && | |
11fdf7f2 | 8304 | osdmap->get_addrs(whoami) == client_messenger->get_myaddrs()) { |
7c673cae FG |
8305 | up_epoch = osdmap->get_epoch(); |
8306 | dout(10) << "up_epoch is " << up_epoch << dendl; | |
8307 | if (!boot_epoch) { | |
8308 | boot_epoch = osdmap->get_epoch(); | |
8309 | dout(10) << "boot_epoch is " << boot_epoch << dendl; | |
8310 | } | |
8311 | service.set_epochs(&boot_epoch, &up_epoch, NULL); | |
8312 | } | |
8313 | } | |
8314 | ||
7c673cae FG |
8315 | epoch_t _bind_epoch = service.get_bind_epoch(); |
8316 | if (osdmap->is_up(whoami) && | |
11fdf7f2 TL |
8317 | osdmap->get_addrs(whoami).legacy_equals( |
8318 | client_messenger->get_myaddrs()) && | |
7c673cae FG |
8319 | _bind_epoch < osdmap->get_up_from(whoami)) { |
8320 | ||
8321 | if (is_booting()) { | |
8322 | dout(1) << "state: booting -> active" << dendl; | |
8323 | set_state(STATE_ACTIVE); | |
11fdf7f2 | 8324 | do_restart = false; |
7c673cae FG |
8325 | |
8326 | // set incarnation so that osd_reqid_t's we generate for our | |
8327 | // objecter requests are unique across restarts. | |
8328 | service.objecter->set_client_incarnation(osdmap->get_epoch()); | |
11fdf7f2 | 8329 | cancel_pending_failures(); |
7c673cae FG |
8330 | } |
8331 | } | |
8332 | ||
8333 | if (osdmap->get_epoch() > 0 && | |
8334 | is_active()) { | |
8335 | if (!osdmap->exists(whoami)) { | |
9f95a23c | 8336 | derr << "map says i do not exist. shutting down." << dendl; |
7c673cae FG |
8337 | do_shutdown = true; // don't call shutdown() while we have |
8338 | // everything paused | |
9f95a23c TL |
8339 | } else if (osdmap->is_stop(whoami)) { |
8340 | derr << "map says i am stopped by admin. shutting down." << dendl; | |
8341 | do_shutdown = true; | |
7c673cae | 8342 | } else if (!osdmap->is_up(whoami) || |
11fdf7f2 TL |
8343 | !osdmap->get_addrs(whoami).legacy_equals( |
8344 | client_messenger->get_myaddrs()) || | |
8345 | !osdmap->get_cluster_addrs(whoami).legacy_equals( | |
8346 | cluster_messenger->get_myaddrs()) || | |
8347 | !osdmap->get_hb_back_addrs(whoami).legacy_equals( | |
8348 | hb_back_server_messenger->get_myaddrs()) || | |
8349 | !osdmap->get_hb_front_addrs(whoami).legacy_equals( | |
8350 | hb_front_server_messenger->get_myaddrs())) { | |
7c673cae FG |
8351 | if (!osdmap->is_up(whoami)) { |
8352 | if (service.is_preparing_to_stop() || service.is_stopping()) { | |
8353 | service.got_stop_ack(); | |
8354 | } else { | |
c07f9fc5 FG |
8355 | clog->warn() << "Monitor daemon marked osd." << whoami << " down, " |
8356 | "but it is still running"; | |
8357 | clog->debug() << "map e" << osdmap->get_epoch() | |
8358 | << " wrongly marked me down at e" | |
8359 | << osdmap->get_down_at(whoami); | |
7c673cae | 8360 | } |
9f95a23c TL |
8361 | if (monc->monmap.min_mon_release >= ceph_release_t::octopus) { |
8362 | // note that this is best-effort... | |
8363 | monc->send_mon_message( | |
8364 | new MOSDMarkMeDead( | |
8365 | monc->get_fsid(), | |
8366 | whoami, | |
8367 | osdmap->get_epoch())); | |
8368 | } | |
11fdf7f2 TL |
8369 | } else if (!osdmap->get_addrs(whoami).legacy_equals( |
8370 | client_messenger->get_myaddrs())) { | |
7c673cae | 8371 | clog->error() << "map e" << osdmap->get_epoch() |
11fdf7f2 TL |
8372 | << " had wrong client addr (" << osdmap->get_addrs(whoami) |
8373 | << " != my " << client_messenger->get_myaddrs() << ")"; | |
8374 | } else if (!osdmap->get_cluster_addrs(whoami).legacy_equals( | |
8375 | cluster_messenger->get_myaddrs())) { | |
7c673cae FG |
8376 | clog->error() << "map e" << osdmap->get_epoch() |
8377 | << " had wrong cluster addr (" | |
11fdf7f2 TL |
8378 | << osdmap->get_cluster_addrs(whoami) |
8379 | << " != my " << cluster_messenger->get_myaddrs() << ")"; | |
8380 | } else if (!osdmap->get_hb_back_addrs(whoami).legacy_equals( | |
8381 | hb_back_server_messenger->get_myaddrs())) { | |
7c673cae | 8382 | clog->error() << "map e" << osdmap->get_epoch() |
c07f9fc5 | 8383 | << " had wrong heartbeat back addr (" |
11fdf7f2 TL |
8384 | << osdmap->get_hb_back_addrs(whoami) |
8385 | << " != my " << hb_back_server_messenger->get_myaddrs() | |
7c673cae | 8386 | << ")"; |
11fdf7f2 TL |
8387 | } else if (!osdmap->get_hb_front_addrs(whoami).legacy_equals( |
8388 | hb_front_server_messenger->get_myaddrs())) { | |
7c673cae | 8389 | clog->error() << "map e" << osdmap->get_epoch() |
c07f9fc5 | 8390 | << " had wrong heartbeat front addr (" |
11fdf7f2 TL |
8391 | << osdmap->get_hb_front_addrs(whoami) |
8392 | << " != my " << hb_front_server_messenger->get_myaddrs() | |
7c673cae FG |
8393 | << ")"; |
8394 | } | |
8395 | ||
8396 | if (!service.is_stopping()) { | |
8397 | epoch_t up_epoch = 0; | |
8398 | epoch_t bind_epoch = osdmap->get_epoch(); | |
8399 | service.set_epochs(NULL,&up_epoch, &bind_epoch); | |
8400 | do_restart = true; | |
8401 | ||
8402 | //add markdown log | |
8403 | utime_t now = ceph_clock_now(); | |
8404 | utime_t grace = utime_t(cct->_conf->osd_max_markdown_period, 0); | |
8405 | osd_markdown_log.push_back(now); | |
7c673cae | 8406 | if ((int)osd_markdown_log.size() > cct->_conf->osd_max_markdown_count) { |
9f95a23c TL |
8407 | derr << __func__ << " marked down " |
8408 | << osd_markdown_log.size() | |
8409 | << " > osd_max_markdown_count " | |
8410 | << cct->_conf->osd_max_markdown_count | |
8411 | << " in last " << grace << " seconds, shutting down" | |
8412 | << dendl; | |
7c673cae FG |
8413 | do_restart = false; |
8414 | do_shutdown = true; | |
8415 | } | |
8416 | ||
8417 | start_waiting_for_healthy(); | |
8418 | ||
8419 | set<int> avoid_ports; | |
8420 | #if defined(__FreeBSD__) | |
8421 | // prevent FreeBSD from grabbing the client_messenger port during | |
f67539c2 | 8422 | // rebinding. In which case a cluster_meesneger will connect also |
7c673cae | 8423 | // to the same port |
11fdf7f2 | 8424 | client_messenger->get_myaddrs().get_ports(&avoid_ports); |
7c673cae | 8425 | #endif |
11fdf7f2 | 8426 | cluster_messenger->get_myaddrs().get_ports(&avoid_ports); |
7c673cae FG |
8427 | |
8428 | int r = cluster_messenger->rebind(avoid_ports); | |
8429 | if (r != 0) { | |
8430 | do_shutdown = true; // FIXME: do_restart? | |
8431 | network_error = true; | |
9f95a23c TL |
8432 | derr << __func__ << " marked down:" |
8433 | << " rebind cluster_messenger failed" << dendl; | |
7c673cae FG |
8434 | } |
8435 | ||
9f95a23c TL |
8436 | hb_back_server_messenger->mark_down_all(); |
8437 | hb_front_server_messenger->mark_down_all(); | |
7c673cae FG |
8438 | hb_front_client_messenger->mark_down_all(); |
8439 | hb_back_client_messenger->mark_down_all(); | |
8440 | ||
494da23a | 8441 | reset_heartbeat_peers(true); |
7c673cae FG |
8442 | } |
8443 | } | |
20effc67 TL |
8444 | } else if (osdmap->get_epoch() > 0 && osdmap->is_stop(whoami)) { |
8445 | derr << "map says i am stopped by admin. shutting down." << dendl; | |
8446 | do_shutdown = true; | |
7c673cae FG |
8447 | } |
8448 | ||
9f95a23c | 8449 | map_lock.unlock(); |
7c673cae | 8450 | |
11fdf7f2 | 8451 | check_osdmap_features(); |
7c673cae FG |
8452 | |
8453 | // yay! | |
8454 | consume_map(); | |
8455 | ||
8456 | if (is_active() || is_waiting_for_healthy()) | |
8457 | maybe_update_heartbeat_peers(); | |
8458 | ||
11fdf7f2 | 8459 | if (is_active()) { |
7c673cae FG |
8460 | activate_map(); |
8461 | } | |
8462 | ||
31f18b77 | 8463 | if (do_shutdown) { |
7c673cae | 8464 | if (network_error) { |
11fdf7f2 | 8465 | cancel_pending_failures(); |
7c673cae FG |
8466 | } |
8467 | // trigger shutdown in a different thread | |
8468 | dout(0) << __func__ << " shutdown OSD via async signal" << dendl; | |
8469 | queue_async_signal(SIGINT); | |
8470 | } | |
31f18b77 FG |
8471 | else if (m->newest_map && m->newest_map > last) { |
8472 | dout(10) << " msg say newest map is " << m->newest_map | |
8473 | << ", requesting more" << dendl; | |
8474 | osdmap_subscribe(osdmap->get_epoch()+1, false); | |
8475 | } | |
7c673cae FG |
8476 | else if (is_preboot()) { |
8477 | if (m->get_source().is_mon()) | |
8478 | _preboot(m->oldest_map, m->newest_map); | |
8479 | else | |
8480 | start_boot(); | |
8481 | } | |
8482 | else if (do_restart) | |
8483 | start_boot(); | |
8484 | ||
8485 | } | |
8486 | ||
11fdf7f2 | 8487 | void OSD::check_osdmap_features() |
7c673cae FG |
8488 | { |
8489 | // adjust required feature bits? | |
8490 | ||
8491 | // we have to be a bit careful here, because we are accessing the | |
8492 | // Policy structures without taking any lock. in particular, only | |
8493 | // modify integer values that can safely be read by a racing CPU. | |
8494 | // since we are only accessing existing Policy structures a their | |
8495 | // current memory location, and setting or clearing bits in integer | |
8496 | // fields, and we are the only writer, this is not a problem. | |
8497 | ||
9f95a23c | 8498 | const auto osdmap = get_osdmap(); |
7c673cae FG |
8499 | { |
8500 | Messenger::Policy p = client_messenger->get_default_policy(); | |
8501 | uint64_t mask; | |
8502 | uint64_t features = osdmap->get_features(entity_name_t::TYPE_CLIENT, &mask); | |
8503 | if ((p.features_required & mask) != features) { | |
8504 | dout(0) << "crush map has features " << features | |
8505 | << ", adjusting msgr requires for clients" << dendl; | |
8506 | p.features_required = (p.features_required & ~mask) | features; | |
8507 | client_messenger->set_default_policy(p); | |
8508 | } | |
8509 | } | |
8510 | { | |
8511 | Messenger::Policy p = client_messenger->get_policy(entity_name_t::TYPE_MON); | |
8512 | uint64_t mask; | |
8513 | uint64_t features = osdmap->get_features(entity_name_t::TYPE_MON, &mask); | |
8514 | if ((p.features_required & mask) != features) { | |
8515 | dout(0) << "crush map has features " << features | |
8516 | << " was " << p.features_required | |
8517 | << ", adjusting msgr requires for mons" << dendl; | |
8518 | p.features_required = (p.features_required & ~mask) | features; | |
8519 | client_messenger->set_policy(entity_name_t::TYPE_MON, p); | |
8520 | } | |
8521 | } | |
8522 | { | |
8523 | Messenger::Policy p = cluster_messenger->get_policy(entity_name_t::TYPE_OSD); | |
8524 | uint64_t mask; | |
8525 | uint64_t features = osdmap->get_features(entity_name_t::TYPE_OSD, &mask); | |
8526 | ||
8527 | if ((p.features_required & mask) != features) { | |
8528 | dout(0) << "crush map has features " << features | |
8529 | << ", adjusting msgr requires for osds" << dendl; | |
8530 | p.features_required = (p.features_required & ~mask) | features; | |
8531 | cluster_messenger->set_policy(entity_name_t::TYPE_OSD, p); | |
8532 | } | |
8533 | ||
11fdf7f2 | 8534 | if (!superblock.compat_features.incompat.contains(CEPH_OSD_FEATURE_INCOMPAT_SHARDS)) { |
7c673cae FG |
8535 | dout(0) << __func__ << " enabling on-disk ERASURE CODES compat feature" << dendl; |
8536 | superblock.compat_features.incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_SHARDS); | |
8537 | ObjectStore::Transaction t; | |
8538 | write_superblock(t); | |
11fdf7f2 TL |
8539 | int err = store->queue_transaction(service.meta_ch, std::move(t), NULL); |
8540 | ceph_assert(err == 0); | |
7c673cae FG |
8541 | } |
8542 | } | |
11fdf7f2 | 8543 | |
9f95a23c TL |
8544 | if (osdmap->require_osd_release < ceph_release_t::nautilus) { |
8545 | hb_front_server_messenger->set_require_authorizer(false); | |
8546 | hb_back_server_messenger->set_require_authorizer(false); | |
8547 | } else { | |
8548 | hb_front_server_messenger->set_require_authorizer(true); | |
8549 | hb_back_server_messenger->set_require_authorizer(true); | |
11fdf7f2 TL |
8550 | } |
8551 | ||
8552 | if (osdmap->require_osd_release != last_require_osd_release) { | |
8553 | dout(1) << __func__ << " require_osd_release " << last_require_osd_release | |
8554 | << " -> " << to_string(osdmap->require_osd_release) << dendl; | |
8555 | store->write_meta("require_osd_release", | |
8556 | stringify((int)osdmap->require_osd_release)); | |
8557 | last_require_osd_release = osdmap->require_osd_release; | |
8558 | } | |
7c673cae FG |
8559 | } |
8560 | ||
11fdf7f2 TL |
8561 | struct C_FinishSplits : public Context { |
8562 | OSD *osd; | |
8563 | set<PGRef> pgs; | |
8564 | C_FinishSplits(OSD *osd, const set<PGRef> &in) | |
8565 | : osd(osd), pgs(in) {} | |
8566 | void finish(int r) override { | |
8567 | osd->_finish_splits(pgs); | |
8568 | } | |
8569 | }; | |
8570 | ||
8571 | void OSD::_finish_splits(set<PGRef>& pgs) | |
7c673cae | 8572 | { |
11fdf7f2 TL |
8573 | dout(10) << __func__ << " " << pgs << dendl; |
8574 | if (is_stopping()) | |
8575 | return; | |
11fdf7f2 TL |
8576 | for (set<PGRef>::iterator i = pgs.begin(); |
8577 | i != pgs.end(); | |
8578 | ++i) { | |
8579 | PG *pg = i->get(); | |
7c673cae | 8580 | |
20effc67 | 8581 | PeeringCtx rctx; |
11fdf7f2 TL |
8582 | pg->lock(); |
8583 | dout(10) << __func__ << " " << *pg << dendl; | |
8584 | epoch_t e = pg->get_osdmap_epoch(); | |
9f95a23c | 8585 | pg->handle_initialize(rctx); |
11fdf7f2 | 8586 | pg->queue_null(e, e); |
9f95a23c | 8587 | dispatch_context(rctx, pg, service.get_osdmap()); |
11fdf7f2 | 8588 | pg->unlock(); |
7c673cae | 8589 | |
11fdf7f2 TL |
8590 | unsigned shard_index = pg->pg_id.hash_to_shard(num_shards); |
8591 | shards[shard_index]->register_and_wake_split_child(pg); | |
7c673cae | 8592 | } |
11fdf7f2 TL |
8593 | }; |
8594 | ||
8595 | bool OSD::add_merge_waiter(OSDMapRef nextmap, spg_t target, PGRef src, | |
8596 | unsigned need) | |
8597 | { | |
8598 | std::lock_guard l(merge_lock); | |
8599 | auto& p = merge_waiters[nextmap->get_epoch()][target]; | |
8600 | p[src->pg_id] = src; | |
8601 | dout(10) << __func__ << " added merge_waiter " << src->pg_id | |
8602 | << " for " << target << ", have " << p.size() << "/" << need | |
8603 | << dendl; | |
8604 | return p.size() == need; | |
8605 | } | |
8606 | ||
8607 | bool OSD::advance_pg( | |
8608 | epoch_t osd_epoch, | |
8609 | PG *pg, | |
8610 | ThreadPool::TPHandle &handle, | |
9f95a23c | 8611 | PeeringCtx &rctx) |
11fdf7f2 TL |
8612 | { |
8613 | if (osd_epoch <= pg->get_osdmap_epoch()) { | |
8614 | return true; | |
8615 | } | |
8616 | ceph_assert(pg->is_locked()); | |
8617 | OSDMapRef lastmap = pg->get_osdmap(); | |
11fdf7f2 TL |
8618 | set<PGRef> new_pgs; // any split children |
8619 | bool ret = true; | |
8620 | ||
8621 | unsigned old_pg_num = lastmap->have_pg_pool(pg->pg_id.pool()) ? | |
8622 | lastmap->get_pg_num(pg->pg_id.pool()) : 0; | |
8623 | for (epoch_t next_epoch = pg->get_osdmap_epoch() + 1; | |
8624 | next_epoch <= osd_epoch; | |
7c673cae FG |
8625 | ++next_epoch) { |
8626 | OSDMapRef nextmap = service.try_get_map(next_epoch); | |
8627 | if (!nextmap) { | |
8628 | dout(20) << __func__ << " missing map " << next_epoch << dendl; | |
7c673cae FG |
8629 | continue; |
8630 | } | |
8631 | ||
11fdf7f2 TL |
8632 | unsigned new_pg_num = |
8633 | (old_pg_num && nextmap->have_pg_pool(pg->pg_id.pool())) ? | |
8634 | nextmap->get_pg_num(pg->pg_id.pool()) : 0; | |
8635 | if (old_pg_num && new_pg_num && old_pg_num != new_pg_num) { | |
8636 | // check for merge | |
8637 | if (nextmap->have_pg_pool(pg->pg_id.pool())) { | |
8638 | spg_t parent; | |
8639 | if (pg->pg_id.is_merge_source( | |
8640 | old_pg_num, | |
8641 | new_pg_num, | |
8642 | &parent)) { | |
8643 | // we are merge source | |
8644 | PGRef spg = pg; // carry a ref | |
8645 | dout(1) << __func__ << " " << pg->pg_id | |
8646 | << " is merge source, target is " << parent | |
8647 | << dendl; | |
8648 | pg->write_if_dirty(rctx); | |
9f95a23c TL |
8649 | if (!new_pgs.empty()) { |
8650 | rctx.transaction.register_on_applied(new C_FinishSplits(this, | |
8651 | new_pgs)); | |
8652 | new_pgs.clear(); | |
8653 | } | |
8654 | dispatch_context(rctx, pg, pg->get_osdmap(), &handle); | |
11fdf7f2 | 8655 | pg->ch->flush(); |
eafe8130 TL |
8656 | // release backoffs explicitly, since the on_shutdown path |
8657 | // aggressively tears down backoff state. | |
8658 | if (pg->is_primary()) { | |
8659 | pg->release_pg_backoffs(); | |
8660 | } | |
11fdf7f2 TL |
8661 | pg->on_shutdown(); |
8662 | OSDShard *sdata = pg->osd_shard; | |
8663 | { | |
8664 | std::lock_guard l(sdata->shard_lock); | |
8665 | if (pg->pg_slot) { | |
8666 | sdata->_detach_pg(pg->pg_slot); | |
8667 | // update pg count now since we might not get an osdmap | |
8668 | // any time soon. | |
8669 | if (pg->is_primary()) | |
8670 | logger->dec(l_osd_pg_primary); | |
9f95a23c TL |
8671 | else if (pg->is_nonprimary()) |
8672 | logger->dec(l_osd_pg_replica); // misnomer | |
11fdf7f2 TL |
8673 | else |
8674 | logger->dec(l_osd_pg_stray); | |
8675 | } | |
8676 | } | |
8677 | pg->unlock(); | |
8678 | ||
8679 | set<spg_t> children; | |
8680 | parent.is_split(new_pg_num, old_pg_num, &children); | |
8681 | if (add_merge_waiter(nextmap, parent, pg, children.size())) { | |
8682 | enqueue_peering_evt( | |
8683 | parent, | |
8684 | PGPeeringEventRef( | |
8685 | std::make_shared<PGPeeringEvent>( | |
8686 | nextmap->get_epoch(), | |
8687 | nextmap->get_epoch(), | |
8688 | NullEvt()))); | |
8689 | } | |
8690 | ret = false; | |
8691 | goto out; | |
8692 | } else if (pg->pg_id.is_merge_target(old_pg_num, new_pg_num)) { | |
8693 | // we are merge target | |
8694 | set<spg_t> children; | |
8695 | pg->pg_id.is_split(new_pg_num, old_pg_num, &children); | |
8696 | dout(20) << __func__ << " " << pg->pg_id | |
8697 | << " is merge target, sources are " << children | |
8698 | << dendl; | |
8699 | map<spg_t,PGRef> sources; | |
8700 | { | |
8701 | std::lock_guard l(merge_lock); | |
8702 | auto& s = merge_waiters[nextmap->get_epoch()][pg->pg_id]; | |
8703 | unsigned need = children.size(); | |
8704 | dout(20) << __func__ << " have " << s.size() << "/" | |
8705 | << need << dendl; | |
8706 | if (s.size() == need) { | |
8707 | sources.swap(s); | |
8708 | merge_waiters[nextmap->get_epoch()].erase(pg->pg_id); | |
8709 | if (merge_waiters[nextmap->get_epoch()].empty()) { | |
8710 | merge_waiters.erase(nextmap->get_epoch()); | |
8711 | } | |
8712 | } | |
8713 | } | |
8714 | if (!sources.empty()) { | |
8715 | unsigned new_pg_num = nextmap->get_pg_num(pg->pg_id.pool()); | |
8716 | unsigned split_bits = pg->pg_id.get_split_bits(new_pg_num); | |
8717 | dout(1) << __func__ << " merging " << pg->pg_id << dendl; | |
8718 | pg->merge_from( | |
8719 | sources, rctx, split_bits, | |
8720 | nextmap->get_pg_pool( | |
8721 | pg->pg_id.pool())->last_pg_merge_meta); | |
8722 | pg->pg_slot->waiting_for_merge_epoch = 0; | |
8723 | } else { | |
8724 | dout(20) << __func__ << " not ready to merge yet" << dendl; | |
8725 | pg->write_if_dirty(rctx); | |
9f95a23c TL |
8726 | if (!new_pgs.empty()) { |
8727 | rctx.transaction.register_on_applied(new C_FinishSplits(this, | |
8728 | new_pgs)); | |
8729 | new_pgs.clear(); | |
8730 | } | |
8731 | dispatch_context(rctx, pg, pg->get_osdmap(), &handle); | |
11fdf7f2 TL |
8732 | pg->unlock(); |
8733 | // kick source(s) to get them ready | |
8734 | for (auto& i : children) { | |
8735 | dout(20) << __func__ << " kicking source " << i << dendl; | |
8736 | enqueue_peering_evt( | |
8737 | i, | |
8738 | PGPeeringEventRef( | |
8739 | std::make_shared<PGPeeringEvent>( | |
8740 | nextmap->get_epoch(), | |
8741 | nextmap->get_epoch(), | |
8742 | NullEvt()))); | |
8743 | } | |
8744 | ret = false; | |
8745 | goto out; | |
8746 | } | |
8747 | } | |
8748 | } | |
8749 | } | |
8750 | ||
7c673cae FG |
8751 | vector<int> newup, newacting; |
8752 | int up_primary, acting_primary; | |
8753 | nextmap->pg_to_up_acting_osds( | |
11fdf7f2 | 8754 | pg->pg_id.pgid, |
7c673cae FG |
8755 | &newup, &up_primary, |
8756 | &newacting, &acting_primary); | |
8757 | pg->handle_advance_map( | |
8758 | nextmap, lastmap, newup, up_primary, | |
8759 | newacting, acting_primary, rctx); | |
8760 | ||
494da23a TL |
8761 | auto oldpool = lastmap->get_pools().find(pg->pg_id.pool()); |
8762 | auto newpool = nextmap->get_pools().find(pg->pg_id.pool()); | |
8763 | if (oldpool != lastmap->get_pools().end() | |
8764 | && newpool != nextmap->get_pools().end()) { | |
8765 | dout(20) << __func__ | |
8766 | << " new pool opts " << newpool->second.opts | |
8767 | << " old pool opts " << oldpool->second.opts | |
8768 | << dendl; | |
8769 | ||
8770 | double old_min_interval = 0, new_min_interval = 0; | |
8771 | oldpool->second.opts.get(pool_opts_t::SCRUB_MIN_INTERVAL, &old_min_interval); | |
8772 | newpool->second.opts.get(pool_opts_t::SCRUB_MIN_INTERVAL, &new_min_interval); | |
8773 | ||
8774 | double old_max_interval = 0, new_max_interval = 0; | |
8775 | oldpool->second.opts.get(pool_opts_t::SCRUB_MAX_INTERVAL, &old_max_interval); | |
8776 | newpool->second.opts.get(pool_opts_t::SCRUB_MAX_INTERVAL, &new_max_interval); | |
8777 | ||
8778 | // Assume if an interval is change from set to unset or vice versa the actual config | |
8779 | // is different. Keep it simple even if it is possible to call resched_all_scrub() | |
8780 | // unnecessarily. | |
8781 | if (old_min_interval != new_min_interval || old_max_interval != new_max_interval) { | |
8782 | pg->on_info_history_change(); | |
8783 | } | |
8784 | } | |
8785 | ||
11fdf7f2 TL |
8786 | if (new_pg_num && old_pg_num != new_pg_num) { |
8787 | // check for split | |
8788 | set<spg_t> children; | |
8789 | if (pg->pg_id.is_split( | |
8790 | old_pg_num, | |
8791 | new_pg_num, | |
8792 | &children)) { | |
8793 | split_pgs( | |
8794 | pg, children, &new_pgs, lastmap, nextmap, | |
8795 | rctx); | |
8796 | } | |
7c673cae FG |
8797 | } |
8798 | ||
8799 | lastmap = nextmap; | |
11fdf7f2 | 8800 | old_pg_num = new_pg_num; |
7c673cae FG |
8801 | handle.reset_tp_timeout(); |
8802 | } | |
7c673cae | 8803 | pg->handle_activate_map(rctx); |
11fdf7f2 TL |
8804 | |
8805 | ret = true; | |
8806 | out: | |
8807 | if (!new_pgs.empty()) { | |
9f95a23c | 8808 | rctx.transaction.register_on_applied(new C_FinishSplits(this, new_pgs)); |
7c673cae | 8809 | } |
11fdf7f2 | 8810 | return ret; |
7c673cae FG |
8811 | } |
8812 | ||
8813 | void OSD::consume_map() | |
8814 | { | |
9f95a23c TL |
8815 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
8816 | auto osdmap = get_osdmap(); | |
7c673cae FG |
8817 | dout(7) << "consume_map version " << osdmap->get_epoch() << dendl; |
8818 | ||
3efd9988 FG |
8819 | /** make sure the cluster is speaking in SORTBITWISE, because we don't |
8820 | * speak the older sorting version any more. Be careful not to force | |
8821 | * a shutdown if we are merely processing old maps, though. | |
8822 | */ | |
8823 | if (!osdmap->test_flag(CEPH_OSDMAP_SORTBITWISE) && is_active()) { | |
8824 | derr << __func__ << " SORTBITWISE flag is not set" << dendl; | |
8825 | ceph_abort(); | |
8826 | } | |
8827 | ||
11fdf7f2 TL |
8828 | service.pre_publish_map(osdmap); |
8829 | service.await_reserved_maps(); | |
8830 | service.publish_map(osdmap); | |
7c673cae | 8831 | |
11fdf7f2 TL |
8832 | // prime splits and merges |
8833 | set<pair<spg_t,epoch_t>> newly_split; // splits, and when | |
8834 | set<pair<spg_t,epoch_t>> merge_pgs; // merge participants, and when | |
8835 | for (auto& shard : shards) { | |
8836 | shard->identify_splits_and_merges(osdmap, &newly_split, &merge_pgs); | |
8837 | } | |
8838 | if (!newly_split.empty()) { | |
8839 | for (auto& shard : shards) { | |
8840 | shard->prime_splits(osdmap, &newly_split); | |
8841 | } | |
8842 | ceph_assert(newly_split.empty()); | |
8843 | } | |
7c673cae | 8844 | |
11fdf7f2 TL |
8845 | // prune sent_ready_to_merge |
8846 | service.prune_sent_ready_to_merge(osdmap); | |
7c673cae | 8847 | |
11fdf7f2 TL |
8848 | // FIXME, maybe: We could race against an incoming peering message |
8849 | // that instantiates a merge PG after identify_merges() below and | |
8850 | // never set up its peer to complete the merge. An OSD restart | |
8851 | // would clear it up. This is a hard race to resolve, | |
8852 | // extraordinarily rare (we only merge PGs that are stable and | |
8853 | // clean, so it'd have to be an imported PG to an OSD with a | |
8854 | // slightly stale OSDMap...), so I'm ignoring it for now. We plan to | |
8855 | // replace all of this with a seastar-based code soon anyway. | |
8856 | if (!merge_pgs.empty()) { | |
8857 | // mark the pgs we already have, or create new and empty merge | |
8858 | // participants for those we are missing. do this all under the | |
8859 | // shard lock so we don't have to worry about racing pg creates | |
8860 | // via _process. | |
8861 | for (auto& shard : shards) { | |
8862 | shard->prime_merges(osdmap, &merge_pgs); | |
7c673cae | 8863 | } |
11fdf7f2 TL |
8864 | ceph_assert(merge_pgs.empty()); |
8865 | } | |
8866 | ||
8867 | service.prune_pg_created(); | |
8868 | ||
8869 | unsigned pushes_to_free = 0; | |
8870 | for (auto& shard : shards) { | |
8871 | shard->consume_map(osdmap, &pushes_to_free); | |
8872 | } | |
8873 | ||
8874 | vector<spg_t> pgids; | |
8875 | _get_pgids(&pgids); | |
8876 | ||
8877 | // count (FIXME, probably during seastar rewrite) | |
8878 | int num_pg_primary = 0, num_pg_replica = 0, num_pg_stray = 0; | |
8879 | vector<PGRef> pgs; | |
8880 | _get_pgs(&pgs); | |
8881 | for (auto& pg : pgs) { | |
8882 | // FIXME (probably during seastar rewrite): this is lockless and | |
8883 | // racy, but we don't want to take pg lock here. | |
8884 | if (pg->is_primary()) | |
8885 | num_pg_primary++; | |
9f95a23c TL |
8886 | else if (pg->is_nonprimary()) |
8887 | num_pg_replica++; // misnomer | |
11fdf7f2 TL |
8888 | else |
8889 | num_pg_stray++; | |
8890 | } | |
3efd9988 | 8891 | |
11fdf7f2 TL |
8892 | { |
8893 | // FIXME (as part of seastar rewrite): move to OSDShard | |
8894 | std::lock_guard l(pending_creates_lock); | |
8895 | for (auto pg = pending_creates_from_osd.begin(); | |
8896 | pg != pending_creates_from_osd.end();) { | |
9f95a23c | 8897 | if (osdmap->get_pg_acting_role(pg->first, whoami) < 0) { |
11fdf7f2 TL |
8898 | dout(10) << __func__ << " pg " << pg->first << " doesn't map here, " |
8899 | << "discarding pending_create_from_osd" << dendl; | |
3efd9988 FG |
8900 | pg = pending_creates_from_osd.erase(pg); |
8901 | } else { | |
8902 | ++pg; | |
8903 | } | |
8904 | } | |
7c673cae FG |
8905 | } |
8906 | ||
7c673cae FG |
8907 | service.maybe_inject_dispatch_delay(); |
8908 | ||
8909 | dispatch_sessions_waiting_on_map(); | |
8910 | ||
8911 | service.maybe_inject_dispatch_delay(); | |
8912 | ||
11fdf7f2 | 8913 | service.release_reserved_pushes(pushes_to_free); |
7c673cae | 8914 | |
11fdf7f2 TL |
8915 | // queue null events to push maps down to individual PGs |
8916 | for (auto pgid : pgids) { | |
8917 | enqueue_peering_evt( | |
8918 | pgid, | |
8919 | PGPeeringEventRef( | |
8920 | std::make_shared<PGPeeringEvent>( | |
8921 | osdmap->get_epoch(), | |
8922 | osdmap->get_epoch(), | |
8923 | NullEvt()))); | |
7c673cae | 8924 | } |
11fdf7f2 | 8925 | logger->set(l_osd_pg, pgids.size()); |
7c673cae FG |
8926 | logger->set(l_osd_pg_primary, num_pg_primary); |
8927 | logger->set(l_osd_pg_replica, num_pg_replica); | |
8928 | logger->set(l_osd_pg_stray, num_pg_stray); | |
8929 | } | |
8930 | ||
8931 | void OSD::activate_map() | |
8932 | { | |
9f95a23c TL |
8933 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
8934 | auto osdmap = get_osdmap(); | |
7c673cae FG |
8935 | |
8936 | dout(7) << "activate_map version " << osdmap->get_epoch() << dendl; | |
8937 | ||
7c673cae FG |
8938 | // norecover? |
8939 | if (osdmap->test_flag(CEPH_OSDMAP_NORECOVER)) { | |
8940 | if (!service.recovery_is_paused()) { | |
8941 | dout(1) << "pausing recovery (NORECOVER flag set)" << dendl; | |
8942 | service.pause_recovery(); | |
8943 | } | |
8944 | } else { | |
8945 | if (service.recovery_is_paused()) { | |
8946 | dout(1) << "unpausing recovery (NORECOVER flag unset)" << dendl; | |
8947 | service.unpause_recovery(); | |
8948 | } | |
8949 | } | |
8950 | ||
8951 | service.activate_map(); | |
8952 | ||
8953 | // process waiters | |
8954 | take_waiters(waiting_for_osdmap); | |
8955 | } | |
8956 | ||
8957 | bool OSD::require_mon_peer(const Message *m) | |
8958 | { | |
8959 | if (!m->get_connection()->peer_is_mon()) { | |
8960 | dout(0) << "require_mon_peer received from non-mon " | |
8961 | << m->get_connection()->get_peer_addr() | |
8962 | << " " << *m << dendl; | |
8963 | return false; | |
8964 | } | |
8965 | return true; | |
8966 | } | |
8967 | ||
8968 | bool OSD::require_mon_or_mgr_peer(const Message *m) | |
8969 | { | |
8970 | if (!m->get_connection()->peer_is_mon() && | |
8971 | !m->get_connection()->peer_is_mgr()) { | |
8972 | dout(0) << "require_mon_or_mgr_peer received from non-mon, non-mgr " | |
8973 | << m->get_connection()->get_peer_addr() | |
8974 | << " " << *m << dendl; | |
8975 | return false; | |
8976 | } | |
8977 | return true; | |
8978 | } | |
8979 | ||
8980 | bool OSD::require_osd_peer(const Message *m) | |
8981 | { | |
8982 | if (!m->get_connection()->peer_is_osd()) { | |
8983 | dout(0) << "require_osd_peer received from non-osd " | |
8984 | << m->get_connection()->get_peer_addr() | |
8985 | << " " << *m << dendl; | |
8986 | return false; | |
8987 | } | |
8988 | return true; | |
8989 | } | |
8990 | ||
8991 | bool OSD::require_self_aliveness(const Message *m, epoch_t epoch) | |
8992 | { | |
8993 | epoch_t up_epoch = service.get_up_epoch(); | |
8994 | if (epoch < up_epoch) { | |
8995 | dout(7) << "from pre-up epoch " << epoch << " < " << up_epoch << dendl; | |
8996 | return false; | |
8997 | } | |
8998 | ||
8999 | if (!is_active()) { | |
9000 | dout(7) << "still in boot state, dropping message " << *m << dendl; | |
9001 | return false; | |
9002 | } | |
9003 | ||
9004 | return true; | |
9005 | } | |
9006 | ||
9f95a23c | 9007 | bool OSD::require_same_peer_instance(const Message *m, const OSDMapRef& map, |
7c673cae FG |
9008 | bool is_fast_dispatch) |
9009 | { | |
9010 | int from = m->get_source().num(); | |
9011 | ||
9012 | if (map->is_down(from) || | |
11fdf7f2 | 9013 | (map->get_cluster_addrs(from) != m->get_source_addrs())) { |
7c673cae FG |
9014 | dout(5) << "from dead osd." << from << ", marking down, " |
9015 | << " msg was " << m->get_source_inst().addr | |
11fdf7f2 TL |
9016 | << " expected " |
9017 | << (map->is_up(from) ? | |
9018 | map->get_cluster_addrs(from) : entity_addrvec_t()) | |
7c673cae FG |
9019 | << dendl; |
9020 | ConnectionRef con = m->get_connection(); | |
9021 | con->mark_down(); | |
9f95a23c | 9022 | if (auto s = ceph::ref_cast<Session>(con->get_priv()); s) { |
7c673cae | 9023 | if (!is_fast_dispatch) |
9f95a23c | 9024 | s->session_dispatch_lock.lock(); |
7c673cae | 9025 | clear_session_waiting_on_map(s); |
11fdf7f2 TL |
9026 | con->set_priv(nullptr); // break ref <-> session cycle, if any |
9027 | s->con.reset(); | |
7c673cae | 9028 | if (!is_fast_dispatch) |
9f95a23c | 9029 | s->session_dispatch_lock.unlock(); |
7c673cae FG |
9030 | } |
9031 | return false; | |
9032 | } | |
9033 | return true; | |
9034 | } | |
9035 | ||
9036 | ||
9037 | /* | |
9038 | * require that we have same (or newer) map, and that | |
9039 | * the source is the pg primary. | |
9040 | */ | |
9041 | bool OSD::require_same_or_newer_map(OpRequestRef& op, epoch_t epoch, | |
9042 | bool is_fast_dispatch) | |
9043 | { | |
9044 | const Message *m = op->get_req(); | |
9f95a23c | 9045 | const auto osdmap = get_osdmap(); |
7c673cae FG |
9046 | dout(15) << "require_same_or_newer_map " << epoch |
9047 | << " (i am " << osdmap->get_epoch() << ") " << m << dendl; | |
9048 | ||
9f95a23c | 9049 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae FG |
9050 | |
9051 | // do they have a newer map? | |
9052 | if (epoch > osdmap->get_epoch()) { | |
9053 | dout(7) << "waiting for newer map epoch " << epoch | |
9054 | << " > my " << osdmap->get_epoch() << " with " << m << dendl; | |
9055 | wait_for_new_map(op); | |
9056 | return false; | |
9057 | } | |
9058 | ||
9059 | if (!require_self_aliveness(op->get_req(), epoch)) { | |
9060 | return false; | |
9061 | } | |
9062 | ||
9063 | // ok, our map is same or newer.. do they still exist? | |
9064 | if (m->get_connection()->get_messenger() == cluster_messenger && | |
9065 | !require_same_peer_instance(op->get_req(), osdmap, is_fast_dispatch)) { | |
9066 | return false; | |
9067 | } | |
9068 | ||
9069 | return true; | |
9070 | } | |
9071 | ||
9072 | ||
9073 | ||
9074 | ||
9075 | ||
9076 | // ---------------------------------------- | |
9077 | // pg creation | |
9078 | ||
9079 | void OSD::split_pgs( | |
9080 | PG *parent, | |
31f18b77 | 9081 | const set<spg_t> &childpgids, set<PGRef> *out_pgs, |
7c673cae FG |
9082 | OSDMapRef curmap, |
9083 | OSDMapRef nextmap, | |
9f95a23c | 9084 | PeeringCtx &rctx) |
7c673cae | 9085 | { |
11fdf7f2 TL |
9086 | unsigned pg_num = nextmap->get_pg_num(parent->pg_id.pool()); |
9087 | parent->update_snap_mapper_bits(parent->get_pgid().get_split_bits(pg_num)); | |
7c673cae | 9088 | |
11fdf7f2 TL |
9089 | vector<object_stat_sum_t> updated_stats; |
9090 | parent->start_split_stats(childpgids, &updated_stats); | |
7c673cae FG |
9091 | |
9092 | vector<object_stat_sum_t>::iterator stat_iter = updated_stats.begin(); | |
9093 | for (set<spg_t>::const_iterator i = childpgids.begin(); | |
9094 | i != childpgids.end(); | |
9095 | ++i, ++stat_iter) { | |
11fdf7f2 TL |
9096 | ceph_assert(stat_iter != updated_stats.end()); |
9097 | dout(10) << __func__ << " splitting " << *parent << " into " << *i << dendl; | |
7c673cae FG |
9098 | PG* child = _make_pg(nextmap, *i); |
9099 | child->lock(true); | |
9100 | out_pgs->insert(child); | |
11fdf7f2 | 9101 | child->ch = store->create_new_collection(child->coll); |
7c673cae | 9102 | |
11fdf7f2 TL |
9103 | { |
9104 | uint32_t shard_index = i->hash_to_shard(shards.size()); | |
9105 | assert(NULL != shards[shard_index]); | |
9106 | store->set_collection_commit_queue(child->coll, &(shards[shard_index]->context_queue)); | |
9107 | } | |
7c673cae | 9108 | |
11fdf7f2 TL |
9109 | unsigned split_bits = i->get_split_bits(pg_num); |
9110 | dout(10) << " pg_num is " << pg_num | |
9111 | << ", m_seed " << i->ps() | |
9112 | << ", split_bits is " << split_bits << dendl; | |
7c673cae FG |
9113 | parent->split_colls( |
9114 | *i, | |
9115 | split_bits, | |
9116 | i->ps(), | |
11fdf7f2 | 9117 | &child->get_pool().info, |
9f95a23c | 9118 | rctx.transaction); |
7c673cae FG |
9119 | parent->split_into( |
9120 | i->pgid, | |
9121 | child, | |
9122 | split_bits); | |
7c673cae | 9123 | |
92f5a8d4 TL |
9124 | child->init_collection_pool_opts(); |
9125 | ||
9f95a23c | 9126 | child->finish_split_stats(*stat_iter, rctx.transaction); |
7c673cae FG |
9127 | child->unlock(); |
9128 | } | |
11fdf7f2 | 9129 | ceph_assert(stat_iter != updated_stats.end()); |
9f95a23c | 9130 | parent->finish_split_stats(*stat_iter, rctx.transaction); |
7c673cae FG |
9131 | } |
9132 | ||
9133 | /* | |
9134 | * holding osd_lock | |
9135 | */ | |
9136 | void OSD::handle_pg_create(OpRequestRef op) | |
9137 | { | |
9f95a23c TL |
9138 | // NOTE: this can be removed in P release (mimic is the last version to |
9139 | // send MOSDPGCreate messages). | |
9140 | ||
9141 | auto m = op->get_req<MOSDPGCreate>(); | |
11fdf7f2 | 9142 | ceph_assert(m->get_type() == MSG_OSD_PG_CREATE); |
7c673cae FG |
9143 | |
9144 | dout(10) << "handle_pg_create " << *m << dendl; | |
9145 | ||
9146 | if (!require_mon_peer(op->get_req())) { | |
9147 | return; | |
9148 | } | |
9149 | ||
9150 | if (!require_same_or_newer_map(op, m->epoch, false)) | |
9151 | return; | |
9152 | ||
9153 | op->mark_started(); | |
9154 | ||
9f95a23c | 9155 | const auto osdmap = get_osdmap(); |
7c673cae FG |
9156 | map<pg_t,utime_t>::const_iterator ci = m->ctimes.begin(); |
9157 | for (map<pg_t,pg_create_t>::const_iterator p = m->mkpg.begin(); | |
9158 | p != m->mkpg.end(); | |
9159 | ++p, ++ci) { | |
11fdf7f2 | 9160 | ceph_assert(ci != m->ctimes.end() && ci->first == p->first); |
7c673cae FG |
9161 | epoch_t created = p->second.created; |
9162 | if (p->second.split_bits) // Skip split pgs | |
9163 | continue; | |
9164 | pg_t on = p->first; | |
9165 | ||
7c673cae FG |
9166 | if (!osdmap->have_pg_pool(on.pool())) { |
9167 | dout(20) << "ignoring pg on deleted pool " << on << dendl; | |
9168 | continue; | |
9169 | } | |
9170 | ||
9171 | dout(20) << "mkpg " << on << " e" << created << "@" << ci->second << dendl; | |
9172 | ||
9f95a23c TL |
9173 | spg_t pgid; |
9174 | bool mapped = osdmap->get_primary_shard(on, &pgid); | |
9175 | ceph_assert(mapped); | |
9176 | ||
7c673cae FG |
9177 | // is it still ours? |
9178 | vector<int> up, acting; | |
9179 | int up_primary = -1; | |
9180 | int acting_primary = -1; | |
9181 | osdmap->pg_to_up_acting_osds(on, &up, &up_primary, &acting, &acting_primary); | |
9f95a23c | 9182 | int role = osdmap->calc_pg_role(pg_shard_t(whoami, pgid.shard), acting); |
7c673cae FG |
9183 | |
9184 | if (acting_primary != whoami) { | |
9185 | dout(10) << "mkpg " << on << " not acting_primary (" << acting_primary | |
9186 | << "), my role=" << role << ", skipping" << dendl; | |
9187 | continue; | |
9188 | } | |
9189 | ||
7c673cae | 9190 | |
11fdf7f2 | 9191 | PastIntervals pi; |
7c673cae FG |
9192 | pg_history_t history; |
9193 | build_initial_pg_history(pgid, created, ci->second, &history, &pi); | |
9194 | ||
11fdf7f2 TL |
9195 | // The mon won't resend unless the primary changed, so we ignore |
9196 | // same_interval_since. We'll pass this history with the current | |
9197 | // epoch as the event. | |
7c673cae FG |
9198 | if (history.same_primary_since > m->epoch) { |
9199 | dout(10) << __func__ << ": got obsolete pg create on pgid " | |
9200 | << pgid << " from epoch " << m->epoch | |
9201 | << ", primary changed in " << history.same_primary_since | |
9202 | << dendl; | |
9203 | continue; | |
9204 | } | |
11fdf7f2 TL |
9205 | enqueue_peering_evt( |
9206 | pgid, | |
9207 | PGPeeringEventRef( | |
9208 | std::make_shared<PGPeeringEvent>( | |
9209 | osdmap->get_epoch(), | |
9210 | osdmap->get_epoch(), | |
9211 | NullEvt(), | |
9212 | true, | |
9213 | new PGCreateInfo( | |
9214 | pgid, | |
9215 | osdmap->get_epoch(), | |
9216 | history, | |
9217 | pi, | |
9218 | true) | |
9219 | ))); | |
7c673cae | 9220 | } |
7c673cae | 9221 | |
3efd9988 | 9222 | { |
11fdf7f2 | 9223 | std::lock_guard l(pending_creates_lock); |
3efd9988 FG |
9224 | if (pending_creates_from_mon == 0) { |
9225 | last_pg_create_epoch = m->epoch; | |
9226 | } | |
9227 | } | |
11fdf7f2 | 9228 | |
7c673cae FG |
9229 | maybe_update_heartbeat_peers(); |
9230 | } | |
9231 | ||
9232 | ||
9233 | // ---------------------------------------- | |
9234 | // peering and recovery | |
9235 | ||
9f95a23c | 9236 | void OSD::dispatch_context(PeeringCtx &ctx, PG *pg, OSDMapRef curmap, |
7c673cae FG |
9237 | ThreadPool::TPHandle *handle) |
9238 | { | |
11fdf7f2 TL |
9239 | if (!service.get_osdmap()->is_up(whoami)) { |
9240 | dout(20) << __func__ << " not up in osdmap" << dendl; | |
9241 | } else if (!is_active()) { | |
9242 | dout(20) << __func__ << " not active" << dendl; | |
9243 | } else { | |
9f95a23c TL |
9244 | for (auto& [osd, ls] : ctx.message_map) { |
9245 | if (!curmap->is_up(osd)) { | |
9246 | dout(20) << __func__ << " skipping down osd." << osd << dendl; | |
9247 | continue; | |
9248 | } | |
9249 | ConnectionRef con = service.get_con_osd_cluster( | |
9250 | osd, curmap->get_epoch()); | |
9251 | if (!con) { | |
9252 | dout(20) << __func__ << " skipping osd." << osd << " (NULL con)" | |
9253 | << dendl; | |
9254 | continue; | |
9255 | } | |
9256 | service.maybe_share_map(con.get(), curmap); | |
9257 | for (auto m : ls) { | |
9258 | con->send_message2(m); | |
9259 | } | |
9260 | ls.clear(); | |
9261 | } | |
7c673cae | 9262 | } |
9f95a23c | 9263 | if ((!ctx.transaction.empty() || ctx.transaction.has_contexts()) && pg) { |
7c673cae | 9264 | int tr = store->queue_transaction( |
11fdf7f2 | 9265 | pg->ch, |
9f95a23c | 9266 | std::move(ctx.transaction), TrackedOpRef(), |
7c673cae | 9267 | handle); |
11fdf7f2 | 9268 | ceph_assert(tr == 0); |
7c673cae | 9269 | } |
7c673cae FG |
9270 | } |
9271 | ||
11fdf7f2 | 9272 | void OSD::handle_fast_pg_create(MOSDPGCreate2 *m) |
7c673cae | 9273 | { |
11fdf7f2 TL |
9274 | dout(7) << __func__ << " " << *m << " from " << m->get_source() << dendl; |
9275 | if (!require_mon_peer(m)) { | |
9276 | m->put(); | |
7c673cae | 9277 | return; |
7c673cae | 9278 | } |
11fdf7f2 TL |
9279 | for (auto& p : m->pgs) { |
9280 | spg_t pgid = p.first; | |
9281 | epoch_t created = p.second.first; | |
9282 | utime_t created_stamp = p.second.second; | |
9f95a23c TL |
9283 | auto q = m->pg_extra.find(pgid); |
9284 | if (q == m->pg_extra.end()) { | |
9285 | dout(20) << __func__ << " " << pgid << " e" << created | |
9286 | << "@" << created_stamp | |
9287 | << " (no history or past_intervals)" << dendl; | |
9288 | // pre-octopus ... no pg history. this can be removed in Q release. | |
9289 | enqueue_peering_evt( | |
9290 | pgid, | |
9291 | PGPeeringEventRef( | |
9292 | std::make_shared<PGPeeringEvent>( | |
9293 | m->epoch, | |
9294 | m->epoch, | |
9295 | NullEvt(), | |
9296 | true, | |
9297 | new PGCreateInfo( | |
9298 | pgid, | |
9299 | created, | |
9300 | pg_history_t(created, created_stamp), | |
9301 | PastIntervals(), | |
9302 | true) | |
9303 | ))); | |
9304 | } else { | |
9305 | dout(20) << __func__ << " " << pgid << " e" << created | |
9306 | << "@" << created_stamp | |
9307 | << " history " << q->second.first | |
9308 | << " pi " << q->second.second << dendl; | |
9309 | if (!q->second.second.empty() && | |
9310 | m->epoch < q->second.second.get_bounds().second) { | |
9311 | clog->error() << "got pg_create on " << pgid << " epoch " << m->epoch | |
9312 | << " and unmatched past_intervals " << q->second.second | |
9313 | << " (history " << q->second.first << ")"; | |
9314 | } else { | |
9315 | enqueue_peering_evt( | |
9316 | pgid, | |
9317 | PGPeeringEventRef( | |
9318 | std::make_shared<PGPeeringEvent>( | |
9319 | m->epoch, | |
9320 | m->epoch, | |
9321 | NullEvt(), | |
9322 | true, | |
9323 | new PGCreateInfo( | |
9324 | pgid, | |
9325 | m->epoch, | |
9326 | q->second.first, | |
9327 | q->second.second, | |
9328 | true) | |
9329 | ))); | |
9330 | } | |
9331 | } | |
11fdf7f2 | 9332 | } |
7c673cae | 9333 | |
11fdf7f2 TL |
9334 | { |
9335 | std::lock_guard l(pending_creates_lock); | |
9336 | if (pending_creates_from_mon == 0) { | |
9337 | last_pg_create_epoch = m->epoch; | |
9338 | } | |
7c673cae FG |
9339 | } |
9340 | ||
11fdf7f2 | 9341 | m->put(); |
7c673cae FG |
9342 | } |
9343 | ||
11fdf7f2 | 9344 | void OSD::handle_fast_pg_notify(MOSDPGNotify* m) |
7c673cae | 9345 | { |
11fdf7f2 TL |
9346 | dout(7) << __func__ << " " << *m << " from " << m->get_source() << dendl; |
9347 | if (!require_osd_peer(m)) { | |
9348 | m->put(); | |
7c673cae FG |
9349 | return; |
9350 | } | |
11fdf7f2 TL |
9351 | int from = m->get_source().num(); |
9352 | for (auto& p : m->get_pg_list()) { | |
9f95a23c | 9353 | spg_t pgid(p.info.pgid.pgid, p.to); |
11fdf7f2 TL |
9354 | enqueue_peering_evt( |
9355 | pgid, | |
9356 | PGPeeringEventRef( | |
9357 | std::make_shared<PGPeeringEvent>( | |
9f95a23c TL |
9358 | p.epoch_sent, |
9359 | p.query_epoch, | |
11fdf7f2 | 9360 | MNotifyRec( |
9f95a23c TL |
9361 | pgid, pg_shard_t(from, p.from), |
9362 | p, | |
9363 | m->get_connection()->get_features()), | |
11fdf7f2 TL |
9364 | true, |
9365 | new PGCreateInfo( | |
9366 | pgid, | |
9f95a23c TL |
9367 | p.query_epoch, |
9368 | p.info.history, | |
9369 | p.past_intervals, | |
11fdf7f2 TL |
9370 | false) |
9371 | ))); | |
7c673cae | 9372 | } |
11fdf7f2 | 9373 | m->put(); |
7c673cae FG |
9374 | } |
9375 | ||
11fdf7f2 | 9376 | void OSD::handle_fast_pg_info(MOSDPGInfo* m) |
7c673cae | 9377 | { |
11fdf7f2 TL |
9378 | dout(7) << __func__ << " " << *m << " from " << m->get_source() << dendl; |
9379 | if (!require_osd_peer(m)) { | |
9380 | m->put(); | |
7c673cae FG |
9381 | return; |
9382 | } | |
11fdf7f2 TL |
9383 | int from = m->get_source().num(); |
9384 | for (auto& p : m->pg_list) { | |
9385 | enqueue_peering_evt( | |
9f95a23c | 9386 | spg_t(p.info.pgid.pgid, p.to), |
11fdf7f2 | 9387 | PGPeeringEventRef( |
20effc67 TL |
9388 | std::make_shared<PGPeeringEvent>( |
9389 | p.epoch_sent, p.query_epoch, | |
9390 | MInfoRec( | |
9391 | pg_shard_t(from, p.from), | |
9392 | p.info, | |
9393 | p.epoch_sent))) | |
11fdf7f2 | 9394 | ); |
7c673cae | 9395 | } |
11fdf7f2 | 9396 | m->put(); |
7c673cae FG |
9397 | } |
9398 | ||
11fdf7f2 | 9399 | void OSD::handle_fast_pg_remove(MOSDPGRemove *m) |
7c673cae | 9400 | { |
11fdf7f2 TL |
9401 | dout(7) << __func__ << " " << *m << " from " << m->get_source() << dendl; |
9402 | if (!require_osd_peer(m)) { | |
9403 | m->put(); | |
7c673cae FG |
9404 | return; |
9405 | } | |
11fdf7f2 TL |
9406 | for (auto& pgid : m->pg_list) { |
9407 | enqueue_peering_evt( | |
9408 | pgid, | |
9409 | PGPeeringEventRef( | |
9410 | std::make_shared<PGPeeringEvent>( | |
9411 | m->get_epoch(), m->get_epoch(), | |
9f95a23c | 9412 | PeeringState::DeleteStart()))); |
7c673cae | 9413 | } |
11fdf7f2 | 9414 | m->put(); |
7c673cae FG |
9415 | } |
9416 | ||
11fdf7f2 | 9417 | void OSD::handle_fast_force_recovery(MOSDForceRecovery *m) |
c07f9fc5 | 9418 | { |
11fdf7f2 TL |
9419 | dout(10) << __func__ << " " << *m << dendl; |
9420 | if (!require_mon_or_mgr_peer(m)) { | |
9421 | m->put(); | |
9422 | return; | |
9423 | } | |
9424 | epoch_t epoch = get_osdmap_epoch(); | |
9425 | for (auto pgid : m->forced_pgs) { | |
9426 | if (m->options & OFR_BACKFILL) { | |
9427 | if (m->options & OFR_CANCEL) { | |
9428 | enqueue_peering_evt( | |
9429 | pgid, | |
9430 | PGPeeringEventRef( | |
9431 | std::make_shared<PGPeeringEvent>( | |
9432 | epoch, epoch, | |
9f95a23c | 9433 | PeeringState::UnsetForceBackfill()))); |
11fdf7f2 TL |
9434 | } else { |
9435 | enqueue_peering_evt( | |
9436 | pgid, | |
9437 | PGPeeringEventRef( | |
9438 | std::make_shared<PGPeeringEvent>( | |
9439 | epoch, epoch, | |
9f95a23c | 9440 | PeeringState::SetForceBackfill()))); |
11fdf7f2 TL |
9441 | } |
9442 | } else if (m->options & OFR_RECOVERY) { | |
9443 | if (m->options & OFR_CANCEL) { | |
9444 | enqueue_peering_evt( | |
9445 | pgid, | |
9446 | PGPeeringEventRef( | |
9447 | std::make_shared<PGPeeringEvent>( | |
9448 | epoch, epoch, | |
9f95a23c | 9449 | PeeringState::UnsetForceRecovery()))); |
11fdf7f2 TL |
9450 | } else { |
9451 | enqueue_peering_evt( | |
9452 | pgid, | |
9453 | PGPeeringEventRef( | |
9454 | std::make_shared<PGPeeringEvent>( | |
9455 | epoch, epoch, | |
9f95a23c | 9456 | PeeringState::SetForceRecovery()))); |
c07f9fc5 FG |
9457 | } |
9458 | } | |
9459 | } | |
11fdf7f2 | 9460 | m->put(); |
c07f9fc5 | 9461 | } |
7c673cae | 9462 | |
11fdf7f2 | 9463 | void OSD::handle_pg_query_nopg(const MQuery& q) |
7c673cae | 9464 | { |
11fdf7f2 TL |
9465 | spg_t pgid = q.pgid; |
9466 | dout(10) << __func__ << " " << pgid << dendl; | |
7c673cae | 9467 | |
11fdf7f2 TL |
9468 | OSDMapRef osdmap = get_osdmap(); |
9469 | if (!osdmap->have_pg_pool(pgid.pool())) | |
7c673cae FG |
9470 | return; |
9471 | ||
11fdf7f2 TL |
9472 | dout(10) << " pg " << pgid << " dne" << dendl; |
9473 | pg_info_t empty(spg_t(pgid.pgid, q.query.to)); | |
9474 | ConnectionRef con = service.get_con_osd_cluster(q.from.osd, osdmap->get_epoch()); | |
9475 | if (con) { | |
9476 | Message *m; | |
9477 | if (q.query.type == pg_query_t::LOG || | |
9478 | q.query.type == pg_query_t::FULLLOG) { | |
9479 | m = new MOSDPGLog( | |
9480 | q.query.from, q.query.to, | |
9481 | osdmap->get_epoch(), empty, | |
9482 | q.query.epoch_sent); | |
7c673cae | 9483 | } else { |
20effc67 TL |
9484 | pg_notify_t notify{q.query.from, q.query.to, |
9485 | q.query.epoch_sent, | |
9486 | osdmap->get_epoch(), | |
9487 | empty, | |
9488 | PastIntervals()}; | |
9489 | m = new MOSDPGNotify2(spg_t{pgid.pgid, q.query.from}, | |
9490 | std::move(notify)); | |
7c673cae | 9491 | } |
9f95a23c | 9492 | service.maybe_share_map(con.get(), osdmap); |
11fdf7f2 | 9493 | con->send_message(m); |
7c673cae FG |
9494 | } |
9495 | } | |
9496 | ||
9f95a23c TL |
9497 | void OSDService::queue_check_readable(spg_t spgid, |
9498 | epoch_t lpr, | |
9499 | ceph::signedspan delay) | |
9500 | { | |
9501 | if (delay == ceph::signedspan::zero()) { | |
9502 | osd->enqueue_peering_evt( | |
9503 | spgid, | |
9504 | PGPeeringEventRef( | |
9505 | std::make_shared<PGPeeringEvent>( | |
9506 | lpr, lpr, | |
9507 | PeeringState::CheckReadable()))); | |
9508 | } else { | |
9509 | mono_timer.add_event( | |
9510 | delay, | |
9511 | [this, spgid, lpr]() { | |
9512 | queue_check_readable(spgid, lpr); | |
9513 | }); | |
9514 | } | |
9515 | } | |
9516 | ||
7c673cae | 9517 | |
7c673cae FG |
9518 | // ========================================================= |
9519 | // RECOVERY | |
9520 | ||
9521 | void OSDService::_maybe_queue_recovery() { | |
9f95a23c | 9522 | ceph_assert(ceph_mutex_is_locked_by_me(recovery_lock)); |
7c673cae FG |
9523 | uint64_t available_pushes; |
9524 | while (!awaiting_throttle.empty() && | |
9525 | _recover_now(&available_pushes)) { | |
11fdf7f2 | 9526 | uint64_t to_start = std::min( |
7c673cae FG |
9527 | available_pushes, |
9528 | cct->_conf->osd_recovery_max_single_start); | |
9529 | _queue_for_recovery(awaiting_throttle.front(), to_start); | |
9530 | awaiting_throttle.pop_front(); | |
11fdf7f2 TL |
9531 | dout(10) << __func__ << " starting " << to_start |
9532 | << ", recovery_ops_reserved " << recovery_ops_reserved | |
9533 | << " -> " << (recovery_ops_reserved + to_start) << dendl; | |
7c673cae FG |
9534 | recovery_ops_reserved += to_start; |
9535 | } | |
9536 | } | |
9537 | ||
9538 | bool OSDService::_recover_now(uint64_t *available_pushes) | |
9539 | { | |
9540 | if (available_pushes) | |
9541 | *available_pushes = 0; | |
9542 | ||
9543 | if (ceph_clock_now() < defer_recovery_until) { | |
9544 | dout(15) << __func__ << " defer until " << defer_recovery_until << dendl; | |
9545 | return false; | |
9546 | } | |
9547 | ||
9548 | if (recovery_paused) { | |
9549 | dout(15) << __func__ << " paused" << dendl; | |
9550 | return false; | |
9551 | } | |
9552 | ||
9f95a23c | 9553 | uint64_t max = osd->get_recovery_max_active(); |
7c673cae FG |
9554 | if (max <= recovery_ops_active + recovery_ops_reserved) { |
9555 | dout(15) << __func__ << " active " << recovery_ops_active | |
9556 | << " + reserved " << recovery_ops_reserved | |
9557 | << " >= max " << max << dendl; | |
9558 | return false; | |
9559 | } | |
9560 | ||
9561 | if (available_pushes) | |
9562 | *available_pushes = max - recovery_ops_active - recovery_ops_reserved; | |
9563 | ||
9564 | return true; | |
9565 | } | |
9566 | ||
9f95a23c TL |
9567 | unsigned OSDService::get_target_pg_log_entries() const |
9568 | { | |
9569 | auto num_pgs = osd->get_num_pgs(); | |
9570 | auto target = cct->_conf->osd_target_pg_log_entries_per_osd; | |
9571 | if (num_pgs > 0 && target > 0) { | |
9572 | // target an even spread of our budgeted log entries across all | |
9573 | // PGs. note that while we only get to control the entry count | |
9574 | // for primary PGs, we'll normally be responsible for a mix of | |
9575 | // primary and replica PGs (for the same pool(s) even), so this | |
9576 | // will work out. | |
9577 | return std::max<unsigned>( | |
9578 | std::min<unsigned>(target / num_pgs, | |
9579 | cct->_conf->osd_max_pg_log_entries), | |
9580 | cct->_conf->osd_min_pg_log_entries); | |
9581 | } else { | |
9582 | // fall back to a per-pg value. | |
9583 | return cct->_conf->osd_min_pg_log_entries; | |
9584 | } | |
9585 | } | |
9586 | ||
7c673cae FG |
9587 | void OSD::do_recovery( |
9588 | PG *pg, epoch_t queued, uint64_t reserved_pushes, | |
9589 | ThreadPool::TPHandle &handle) | |
9590 | { | |
9591 | uint64_t started = 0; | |
31f18b77 FG |
9592 | |
9593 | /* | |
9594 | * When the value of osd_recovery_sleep is set greater than zero, recovery | |
9595 | * ops are scheduled after osd_recovery_sleep amount of time from the previous | |
9596 | * recovery event's schedule time. This is done by adding a | |
9597 | * recovery_requeue_callback event, which re-queues the recovery op using | |
9598 | * queue_recovery_after_sleep. | |
9599 | */ | |
c07f9fc5 | 9600 | float recovery_sleep = get_osd_recovery_sleep(); |
b32b8144 | 9601 | { |
11fdf7f2 | 9602 | std::lock_guard l(service.sleep_lock); |
b32b8144 FG |
9603 | if (recovery_sleep > 0 && service.recovery_needs_sleep) { |
9604 | PGRef pgref(pg); | |
9f95a23c | 9605 | auto recovery_requeue_callback = new LambdaContext([this, pgref, queued, reserved_pushes](int r) { |
b32b8144 FG |
9606 | dout(20) << "do_recovery wake up at " |
9607 | << ceph_clock_now() | |
9608 | << ", re-queuing recovery" << dendl; | |
11fdf7f2 | 9609 | std::lock_guard l(service.sleep_lock); |
b32b8144 FG |
9610 | service.recovery_needs_sleep = false; |
9611 | service.queue_recovery_after_sleep(pgref.get(), queued, reserved_pushes); | |
9612 | }); | |
9613 | ||
9614 | // This is true for the first recovery op and when the previous recovery op | |
9615 | // has been scheduled in the past. The next recovery op is scheduled after | |
9616 | // completing the sleep from now. | |
f67539c2 | 9617 | |
9f95a23c TL |
9618 | if (auto now = ceph::real_clock::now(); |
9619 | service.recovery_schedule_time < now) { | |
9620 | service.recovery_schedule_time = now; | |
b32b8144 | 9621 | } |
9f95a23c | 9622 | service.recovery_schedule_time += ceph::make_timespan(recovery_sleep); |
11fdf7f2 | 9623 | service.sleep_timer.add_event_at(service.recovery_schedule_time, |
9f95a23c | 9624 | recovery_requeue_callback); |
b32b8144 FG |
9625 | dout(20) << "Recovery event scheduled at " |
9626 | << service.recovery_schedule_time << dendl; | |
9627 | return; | |
9628 | } | |
7c673cae FG |
9629 | } |
9630 | ||
9631 | { | |
b32b8144 | 9632 | { |
11fdf7f2 | 9633 | std::lock_guard l(service.sleep_lock); |
b32b8144 FG |
9634 | service.recovery_needs_sleep = true; |
9635 | } | |
9636 | ||
7c673cae FG |
9637 | if (pg->pg_has_reset_since(queued)) { |
9638 | goto out; | |
9639 | } | |
9640 | ||
7c673cae FG |
9641 | dout(10) << "do_recovery starting " << reserved_pushes << " " << *pg << dendl; |
9642 | #ifdef DEBUG_RECOVERY_OIDS | |
11fdf7f2 | 9643 | dout(20) << " active was " << service.recovery_oids[pg->pg_id] << dendl; |
7c673cae FG |
9644 | #endif |
9645 | ||
11fdf7f2 | 9646 | bool do_unfound = pg->start_recovery_ops(reserved_pushes, handle, &started); |
f67539c2 | 9647 | dout(10) << "do_recovery started " << started << "/" << reserved_pushes |
7c673cae FG |
9648 | << " on " << *pg << dendl; |
9649 | ||
11fdf7f2 | 9650 | if (do_unfound) { |
20effc67 | 9651 | PeeringCtx rctx; |
11fdf7f2 | 9652 | rctx.handle = &handle; |
9f95a23c | 9653 | pg->find_unfound(queued, rctx); |
11fdf7f2 | 9654 | dispatch_context(rctx, pg, pg->get_osdmap()); |
7c673cae | 9655 | } |
7c673cae FG |
9656 | } |
9657 | ||
9658 | out: | |
11fdf7f2 | 9659 | ceph_assert(started <= reserved_pushes); |
7c673cae FG |
9660 | service.release_reserved_pushes(reserved_pushes); |
9661 | } | |
9662 | ||
9663 | void OSDService::start_recovery_op(PG *pg, const hobject_t& soid) | |
9664 | { | |
11fdf7f2 | 9665 | std::lock_guard l(recovery_lock); |
7c673cae FG |
9666 | dout(10) << "start_recovery_op " << *pg << " " << soid |
9667 | << " (" << recovery_ops_active << "/" | |
9f95a23c | 9668 | << osd->get_recovery_max_active() << " rops)" |
7c673cae FG |
9669 | << dendl; |
9670 | recovery_ops_active++; | |
9671 | ||
9672 | #ifdef DEBUG_RECOVERY_OIDS | |
11fdf7f2 TL |
9673 | dout(20) << " active was " << recovery_oids[pg->pg_id] << dendl; |
9674 | ceph_assert(recovery_oids[pg->pg_id].count(soid) == 0); | |
9675 | recovery_oids[pg->pg_id].insert(soid); | |
7c673cae FG |
9676 | #endif |
9677 | } | |
9678 | ||
9679 | void OSDService::finish_recovery_op(PG *pg, const hobject_t& soid, bool dequeue) | |
9680 | { | |
11fdf7f2 | 9681 | std::lock_guard l(recovery_lock); |
7c673cae FG |
9682 | dout(10) << "finish_recovery_op " << *pg << " " << soid |
9683 | << " dequeue=" << dequeue | |
9f95a23c TL |
9684 | << " (" << recovery_ops_active << "/" |
9685 | << osd->get_recovery_max_active() << " rops)" | |
7c673cae FG |
9686 | << dendl; |
9687 | ||
9688 | // adjust count | |
11fdf7f2 | 9689 | ceph_assert(recovery_ops_active > 0); |
7c673cae FG |
9690 | recovery_ops_active--; |
9691 | ||
9692 | #ifdef DEBUG_RECOVERY_OIDS | |
11fdf7f2 TL |
9693 | dout(20) << " active oids was " << recovery_oids[pg->pg_id] << dendl; |
9694 | ceph_assert(recovery_oids[pg->pg_id].count(soid)); | |
9695 | recovery_oids[pg->pg_id].erase(soid); | |
7c673cae FG |
9696 | #endif |
9697 | ||
9698 | _maybe_queue_recovery(); | |
9699 | } | |
9700 | ||
9701 | bool OSDService::is_recovery_active() | |
9702 | { | |
eafe8130 TL |
9703 | if (cct->_conf->osd_debug_pretend_recovery_active) { |
9704 | return true; | |
9705 | } | |
b5b8bbf5 | 9706 | return local_reserver.has_reservation() || remote_reserver.has_reservation(); |
7c673cae FG |
9707 | } |
9708 | ||
11fdf7f2 TL |
9709 | void OSDService::release_reserved_pushes(uint64_t pushes) |
9710 | { | |
9711 | std::lock_guard l(recovery_lock); | |
9712 | dout(10) << __func__ << "(" << pushes << "), recovery_ops_reserved " | |
9713 | << recovery_ops_reserved << " -> " << (recovery_ops_reserved-pushes) | |
9714 | << dendl; | |
9715 | ceph_assert(recovery_ops_reserved >= pushes); | |
9716 | recovery_ops_reserved -= pushes; | |
9717 | _maybe_queue_recovery(); | |
9718 | } | |
9719 | ||
7c673cae FG |
9720 | // ========================================================= |
9721 | // OPS | |
9722 | ||
9723 | bool OSD::op_is_discardable(const MOSDOp *op) | |
9724 | { | |
9725 | // drop client request if they are not connected and can't get the | |
9726 | // reply anyway. | |
9727 | if (!op->get_connection()->is_connected()) { | |
9728 | return true; | |
9729 | } | |
9730 | return false; | |
9731 | } | |
9732 | ||
11fdf7f2 | 9733 | void OSD::enqueue_op(spg_t pg, OpRequestRef&& op, epoch_t epoch) |
7c673cae | 9734 | { |
11fdf7f2 TL |
9735 | const utime_t stamp = op->get_req()->get_recv_stamp(); |
9736 | const utime_t latency = ceph_clock_now() - stamp; | |
9737 | const unsigned priority = op->get_req()->get_priority(); | |
9738 | const int cost = op->get_req()->get_cost(); | |
9739 | const uint64_t owner = op->get_req()->get_source().num(); | |
f67539c2 | 9740 | const int type = op->get_req()->get_type(); |
11fdf7f2 TL |
9741 | |
9742 | dout(15) << "enqueue_op " << op << " prio " << priority | |
f67539c2 | 9743 | << " type " << type |
11fdf7f2 | 9744 | << " cost " << cost |
7c673cae FG |
9745 | << " latency " << latency |
9746 | << " epoch " << epoch | |
9747 | << " " << *(op->get_req()) << dendl; | |
9748 | op->osd_trace.event("enqueue op"); | |
11fdf7f2 TL |
9749 | op->osd_trace.keyval("priority", priority); |
9750 | op->osd_trace.keyval("cost", cost); | |
20effc67 TL |
9751 | |
9752 | auto enqueue_span = tracing::osd::tracer.add_span(__func__, op->osd_parent_span); | |
9753 | enqueue_span->AddEvent(__func__, { | |
9754 | {"priority", priority}, | |
9755 | {"cost", cost}, | |
9756 | {"epoch", epoch}, | |
9757 | {"owner", owner}, | |
9758 | {"type", type} | |
9759 | }); | |
9760 | ||
7c673cae | 9761 | op->mark_queued_for_pg(); |
224ce89b | 9762 | logger->tinc(l_osd_op_before_queue_op_lat, latency); |
f67539c2 TL |
9763 | if (type == MSG_OSD_PG_PUSH || |
9764 | type == MSG_OSD_PG_PUSH_REPLY) { | |
9765 | op_shardedwq.queue( | |
9766 | OpSchedulerItem( | |
9767 | unique_ptr<OpSchedulerItem::OpQueueable>(new PGRecoveryMsg(pg, std::move(op))), | |
9768 | cost, priority, stamp, owner, epoch)); | |
9769 | } else { | |
9770 | op_shardedwq.queue( | |
9771 | OpSchedulerItem( | |
9772 | unique_ptr<OpSchedulerItem::OpQueueable>(new PGOpItem(pg, std::move(op))), | |
9773 | cost, priority, stamp, owner, epoch)); | |
9774 | } | |
7c673cae FG |
9775 | } |
9776 | ||
11fdf7f2 TL |
9777 | void OSD::enqueue_peering_evt(spg_t pgid, PGPeeringEventRef evt) |
9778 | { | |
9779 | dout(15) << __func__ << " " << pgid << " " << evt->get_desc() << dendl; | |
9780 | op_shardedwq.queue( | |
9f95a23c TL |
9781 | OpSchedulerItem( |
9782 | unique_ptr<OpSchedulerItem::OpQueueable>(new PGPeeringItem(pgid, evt)), | |
11fdf7f2 TL |
9783 | 10, |
9784 | cct->_conf->osd_peering_op_priority, | |
9785 | utime_t(), | |
9786 | 0, | |
9787 | evt->get_epoch_sent())); | |
9788 | } | |
7c673cae FG |
9789 | |
9790 | /* | |
9791 | * NOTE: dequeue called in worker thread, with pg lock | |
9792 | */ | |
9793 | void OSD::dequeue_op( | |
9794 | PGRef pg, OpRequestRef op, | |
9795 | ThreadPool::TPHandle &handle) | |
9796 | { | |
9f95a23c TL |
9797 | const Message *m = op->get_req(); |
9798 | ||
11fdf7f2 | 9799 | FUNCTRACE(cct); |
9f95a23c | 9800 | OID_EVENT_TRACE_WITH_MSG(m, "DEQUEUE_OP_BEGIN", false); |
7c673cae FG |
9801 | |
9802 | utime_t now = ceph_clock_now(); | |
9803 | op->set_dequeued_time(now); | |
9f95a23c TL |
9804 | |
9805 | utime_t latency = now - m->get_recv_stamp(); | |
9806 | dout(10) << "dequeue_op " << op << " prio " << m->get_priority() | |
9807 | << " cost " << m->get_cost() | |
7c673cae | 9808 | << " latency " << latency |
9f95a23c | 9809 | << " " << *m |
7c673cae FG |
9810 | << " pg " << *pg << dendl; |
9811 | ||
224ce89b WB |
9812 | logger->tinc(l_osd_op_before_dequeue_op_lat, latency); |
9813 | ||
9f95a23c TL |
9814 | service.maybe_share_map(m->get_connection().get(), |
9815 | pg->get_osdmap(), | |
9816 | op->sent_epoch); | |
7c673cae | 9817 | |
11fdf7f2 | 9818 | if (pg->is_deleting()) |
7c673cae FG |
9819 | return; |
9820 | ||
9821 | op->mark_reached_pg(); | |
9822 | op->osd_trace.event("dequeue_op"); | |
9823 | ||
9824 | pg->do_request(op, handle); | |
9825 | ||
9826 | // finish | |
9827 | dout(10) << "dequeue_op " << op << " finish" << dendl; | |
9f95a23c | 9828 | OID_EVENT_TRACE_WITH_MSG(m, "DEQUEUE_OP_END", false); |
7c673cae FG |
9829 | } |
9830 | ||
9831 | ||
11fdf7f2 TL |
9832 | void OSD::dequeue_peering_evt( |
9833 | OSDShard *sdata, | |
9834 | PG *pg, | |
9835 | PGPeeringEventRef evt, | |
9836 | ThreadPool::TPHandle& handle) | |
7c673cae | 9837 | { |
11fdf7f2 | 9838 | auto curmap = sdata->get_osdmap(); |
9f95a23c TL |
9839 | bool need_up_thru = false; |
9840 | epoch_t same_interval_since = 0; | |
11fdf7f2 TL |
9841 | if (!pg) { |
9842 | if (const MQuery *q = dynamic_cast<const MQuery*>(evt->evt.get())) { | |
9843 | handle_pg_query_nopg(*q); | |
7c673cae | 9844 | } else { |
11fdf7f2 TL |
9845 | derr << __func__ << " unrecognized pg-less event " << evt->get_desc() << dendl; |
9846 | ceph_abort(); | |
9847 | } | |
20effc67 TL |
9848 | } else if (PeeringCtx rctx; |
9849 | advance_pg(curmap->get_epoch(), pg, handle, rctx)) { | |
9f95a23c | 9850 | pg->do_peering_event(evt, rctx); |
11fdf7f2 | 9851 | if (pg->is_deleted()) { |
11fdf7f2 TL |
9852 | pg->unlock(); |
9853 | return; | |
7c673cae | 9854 | } |
9f95a23c | 9855 | dispatch_context(rctx, pg, curmap, &handle); |
11fdf7f2 TL |
9856 | need_up_thru = pg->get_need_up_thru(); |
9857 | same_interval_since = pg->get_same_interval_since(); | |
7c673cae FG |
9858 | pg->unlock(); |
9859 | } | |
11fdf7f2 TL |
9860 | |
9861 | if (need_up_thru) { | |
7c673cae | 9862 | queue_want_up_thru(same_interval_since); |
11fdf7f2 | 9863 | } |
7c673cae FG |
9864 | |
9865 | service.send_pg_temp(); | |
9866 | } | |
9867 | ||
11fdf7f2 TL |
9868 | void OSD::dequeue_delete( |
9869 | OSDShard *sdata, | |
9870 | PG *pg, | |
9871 | epoch_t e, | |
9872 | ThreadPool::TPHandle& handle) | |
9873 | { | |
9874 | dequeue_peering_evt( | |
9875 | sdata, | |
9876 | pg, | |
9877 | PGPeeringEventRef( | |
9878 | std::make_shared<PGPeeringEvent>( | |
9879 | e, e, | |
9f95a23c | 9880 | PeeringState::DeleteSome())), |
11fdf7f2 TL |
9881 | handle); |
9882 | } | |
9883 | ||
9884 | ||
9885 | ||
7c673cae FG |
9886 | // -------------------------------- |
9887 | ||
9888 | const char** OSD::get_tracked_conf_keys() const | |
9889 | { | |
9890 | static const char* KEYS[] = { | |
9891 | "osd_max_backfills", | |
9892 | "osd_min_recovery_priority", | |
224ce89b WB |
9893 | "osd_max_trimming_pgs", |
9894 | "osd_op_complaint_time", | |
9895 | "osd_op_log_threshold", | |
9896 | "osd_op_history_size", | |
9897 | "osd_op_history_duration", | |
9898 | "osd_op_history_slow_op_size", | |
9899 | "osd_op_history_slow_op_threshold", | |
7c673cae FG |
9900 | "osd_enable_op_tracker", |
9901 | "osd_map_cache_size", | |
11fdf7f2 | 9902 | "osd_pg_epoch_max_lag_factor", |
7c673cae | 9903 | "osd_pg_epoch_persisted_max_stale", |
f67539c2 TL |
9904 | "osd_recovery_sleep", |
9905 | "osd_recovery_sleep_hdd", | |
9906 | "osd_recovery_sleep_ssd", | |
9907 | "osd_recovery_sleep_hybrid", | |
b3b6e05e TL |
9908 | "osd_delete_sleep", |
9909 | "osd_delete_sleep_hdd", | |
9910 | "osd_delete_sleep_ssd", | |
9911 | "osd_delete_sleep_hybrid", | |
9912 | "osd_snap_trim_sleep", | |
9913 | "osd_snap_trim_sleep_hdd", | |
9914 | "osd_snap_trim_sleep_ssd", | |
20effc67 | 9915 | "osd_snap_trim_sleep_hybrid", |
b3b6e05e | 9916 | "osd_scrub_sleep", |
f67539c2 TL |
9917 | "osd_recovery_max_active", |
9918 | "osd_recovery_max_active_hdd", | |
9919 | "osd_recovery_max_active_ssd", | |
7c673cae FG |
9920 | // clog & admin clog |
9921 | "clog_to_monitors", | |
9922 | "clog_to_syslog", | |
9923 | "clog_to_syslog_facility", | |
9924 | "clog_to_syslog_level", | |
9925 | "osd_objectstore_fuse", | |
9926 | "clog_to_graylog", | |
9927 | "clog_to_graylog_host", | |
9928 | "clog_to_graylog_port", | |
9929 | "host", | |
9930 | "fsid", | |
9931 | "osd_recovery_delay_start", | |
9932 | "osd_client_message_size_cap", | |
9933 | "osd_client_message_cap", | |
31f18b77 FG |
9934 | "osd_heartbeat_min_size", |
9935 | "osd_heartbeat_interval", | |
9f95a23c | 9936 | "osd_object_clean_region_max_num_intervals", |
494da23a TL |
9937 | "osd_scrub_min_interval", |
9938 | "osd_scrub_max_interval", | |
7c673cae FG |
9939 | NULL |
9940 | }; | |
9941 | return KEYS; | |
9942 | } | |
9943 | ||
11fdf7f2 | 9944 | void OSD::handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
9945 | const std::set <std::string> &changed) |
9946 | { | |
9f95a23c | 9947 | std::lock_guard l{osd_lock}; |
f67539c2 TL |
9948 | |
9949 | if (changed.count("osd_max_backfills") || | |
9950 | changed.count("osd_delete_sleep") || | |
9951 | changed.count("osd_delete_sleep_hdd") || | |
9952 | changed.count("osd_delete_sleep_ssd") || | |
9953 | changed.count("osd_delete_sleep_hybrid") || | |
9954 | changed.count("osd_snap_trim_sleep") || | |
9955 | changed.count("osd_snap_trim_sleep_hdd") || | |
9956 | changed.count("osd_snap_trim_sleep_ssd") || | |
9957 | changed.count("osd_snap_trim_sleep_hybrid") || | |
9958 | changed.count("osd_scrub_sleep") || | |
9959 | changed.count("osd_recovery_sleep") || | |
9960 | changed.count("osd_recovery_sleep_hdd") || | |
9961 | changed.count("osd_recovery_sleep_ssd") || | |
9962 | changed.count("osd_recovery_sleep_hybrid") || | |
9963 | changed.count("osd_recovery_max_active") || | |
9964 | changed.count("osd_recovery_max_active_hdd") || | |
9965 | changed.count("osd_recovery_max_active_ssd")) { | |
b3b6e05e TL |
9966 | if (!maybe_override_options_for_qos() && |
9967 | changed.count("osd_max_backfills")) { | |
9968 | // Scheduler is not "mclock". Fallback to earlier behavior | |
f67539c2 TL |
9969 | service.local_reserver.set_max(cct->_conf->osd_max_backfills); |
9970 | service.remote_reserver.set_max(cct->_conf->osd_max_backfills); | |
9971 | } | |
7c673cae FG |
9972 | } |
9973 | if (changed.count("osd_min_recovery_priority")) { | |
9974 | service.local_reserver.set_min_priority(cct->_conf->osd_min_recovery_priority); | |
9975 | service.remote_reserver.set_min_priority(cct->_conf->osd_min_recovery_priority); | |
9976 | } | |
9977 | if (changed.count("osd_max_trimming_pgs")) { | |
9978 | service.snap_reserver.set_max(cct->_conf->osd_max_trimming_pgs); | |
9979 | } | |
9980 | if (changed.count("osd_op_complaint_time") || | |
9981 | changed.count("osd_op_log_threshold")) { | |
9982 | op_tracker.set_complaint_and_threshold(cct->_conf->osd_op_complaint_time, | |
9983 | cct->_conf->osd_op_log_threshold); | |
9984 | } | |
9985 | if (changed.count("osd_op_history_size") || | |
9986 | changed.count("osd_op_history_duration")) { | |
9987 | op_tracker.set_history_size_and_duration(cct->_conf->osd_op_history_size, | |
9988 | cct->_conf->osd_op_history_duration); | |
9989 | } | |
9990 | if (changed.count("osd_op_history_slow_op_size") || | |
9991 | changed.count("osd_op_history_slow_op_threshold")) { | |
9992 | op_tracker.set_history_slow_op_size_and_threshold(cct->_conf->osd_op_history_slow_op_size, | |
9993 | cct->_conf->osd_op_history_slow_op_threshold); | |
9994 | } | |
9995 | if (changed.count("osd_enable_op_tracker")) { | |
9996 | op_tracker.set_tracking(cct->_conf->osd_enable_op_tracker); | |
9997 | } | |
7c673cae FG |
9998 | if (changed.count("osd_map_cache_size")) { |
9999 | service.map_cache.set_size(cct->_conf->osd_map_cache_size); | |
10000 | service.map_bl_cache.set_size(cct->_conf->osd_map_cache_size); | |
10001 | service.map_bl_inc_cache.set_size(cct->_conf->osd_map_cache_size); | |
10002 | } | |
10003 | if (changed.count("clog_to_monitors") || | |
10004 | changed.count("clog_to_syslog") || | |
10005 | changed.count("clog_to_syslog_level") || | |
10006 | changed.count("clog_to_syslog_facility") || | |
10007 | changed.count("clog_to_graylog") || | |
10008 | changed.count("clog_to_graylog_host") || | |
10009 | changed.count("clog_to_graylog_port") || | |
10010 | changed.count("host") || | |
10011 | changed.count("fsid")) { | |
10012 | update_log_config(); | |
10013 | } | |
11fdf7f2 TL |
10014 | if (changed.count("osd_pg_epoch_max_lag_factor")) { |
10015 | m_osd_pg_epoch_max_lag_factor = conf.get_val<double>( | |
10016 | "osd_pg_epoch_max_lag_factor"); | |
10017 | } | |
7c673cae FG |
10018 | |
10019 | #ifdef HAVE_LIBFUSE | |
10020 | if (changed.count("osd_objectstore_fuse")) { | |
10021 | if (store) { | |
10022 | enable_disable_fuse(false); | |
10023 | } | |
10024 | } | |
10025 | #endif | |
10026 | ||
10027 | if (changed.count("osd_recovery_delay_start")) { | |
10028 | service.defer_recovery(cct->_conf->osd_recovery_delay_start); | |
10029 | service.kick_recovery_queue(); | |
10030 | } | |
10031 | ||
10032 | if (changed.count("osd_client_message_cap")) { | |
10033 | uint64_t newval = cct->_conf->osd_client_message_cap; | |
10034 | Messenger::Policy pol = client_messenger->get_policy(entity_name_t::TYPE_CLIENT); | |
20effc67 | 10035 | if (pol.throttler_messages) { |
7c673cae FG |
10036 | pol.throttler_messages->reset_max(newval); |
10037 | } | |
10038 | } | |
10039 | if (changed.count("osd_client_message_size_cap")) { | |
10040 | uint64_t newval = cct->_conf->osd_client_message_size_cap; | |
10041 | Messenger::Policy pol = client_messenger->get_policy(entity_name_t::TYPE_CLIENT); | |
20effc67 | 10042 | if (pol.throttler_bytes) { |
7c673cae FG |
10043 | pol.throttler_bytes->reset_max(newval); |
10044 | } | |
10045 | } | |
9f95a23c TL |
10046 | if (changed.count("osd_object_clean_region_max_num_intervals")) { |
10047 | ObjectCleanRegions::set_max_num_intervals(cct->_conf->osd_object_clean_region_max_num_intervals); | |
10048 | } | |
7c673cae | 10049 | |
494da23a TL |
10050 | if (changed.count("osd_scrub_min_interval") || |
10051 | changed.count("osd_scrub_max_interval")) { | |
10052 | resched_all_scrubs(); | |
10053 | dout(0) << __func__ << ": scrub interval change" << dendl; | |
10054 | } | |
7c673cae | 10055 | check_config(); |
f67539c2 TL |
10056 | if (changed.count("osd_asio_thread_count")) { |
10057 | service.poolctx.stop(); | |
10058 | service.poolctx.start(conf.get_val<std::uint64_t>("osd_asio_thread_count")); | |
10059 | } | |
7c673cae FG |
10060 | } |
10061 | ||
a4b75251 TL |
10062 | void OSD::maybe_override_max_osd_capacity_for_qos() |
10063 | { | |
10064 | // If the scheduler enabled is mclock, override the default | |
10065 | // osd capacity with the value obtained from running the | |
10066 | // osd bench test. This is later used to setup mclock. | |
10067 | if ((cct->_conf.get_val<std::string>("osd_op_queue") == "mclock_scheduler") && | |
20effc67 TL |
10068 | (cct->_conf.get_val<bool>("osd_mclock_skip_benchmark") == false) && |
10069 | (!unsupported_objstore_for_qos())) { | |
a4b75251 TL |
10070 | std::string max_capacity_iops_config; |
10071 | bool force_run_benchmark = | |
10072 | cct->_conf.get_val<bool>("osd_mclock_force_run_benchmark_on_init"); | |
10073 | ||
10074 | if (store_is_rotational) { | |
10075 | max_capacity_iops_config = "osd_mclock_max_capacity_iops_hdd"; | |
10076 | } else { | |
10077 | max_capacity_iops_config = "osd_mclock_max_capacity_iops_ssd"; | |
10078 | } | |
10079 | ||
10080 | if (!force_run_benchmark) { | |
10081 | double default_iops = 0.0; | |
10082 | ||
10083 | // Get the current osd iops capacity | |
10084 | double cur_iops = cct->_conf.get_val<double>(max_capacity_iops_config); | |
10085 | ||
10086 | // Get the default max iops capacity | |
10087 | auto val = cct->_conf.get_val_default(max_capacity_iops_config); | |
10088 | if (!val.has_value()) { | |
10089 | derr << __func__ << " Unable to determine default value of " | |
10090 | << max_capacity_iops_config << dendl; | |
10091 | // Cannot determine default iops. Force a run of the OSD benchmark. | |
10092 | force_run_benchmark = true; | |
10093 | } else { | |
10094 | // Default iops | |
10095 | default_iops = std::stod(val.value()); | |
10096 | } | |
10097 | ||
10098 | // Determine if we really need to run the osd benchmark | |
10099 | if (!force_run_benchmark && (default_iops != cur_iops)) { | |
10100 | dout(1) << __func__ << std::fixed << std::setprecision(2) | |
10101 | << " default_iops: " << default_iops | |
10102 | << " cur_iops: " << cur_iops | |
10103 | << ". Skip OSD benchmark test." << dendl; | |
10104 | return; | |
10105 | } | |
10106 | } | |
10107 | ||
10108 | // Run osd bench: write 100 4MiB objects with blocksize 4KiB | |
10109 | int64_t count = 12288000; // Count of bytes to write | |
10110 | int64_t bsize = 4096; // Block size | |
10111 | int64_t osize = 4194304; // Object size | |
10112 | int64_t onum = 100; // Count of objects to write | |
10113 | double elapsed = 0.0; // Time taken to complete the test | |
10114 | double iops = 0.0; | |
10115 | stringstream ss; | |
10116 | int ret = run_osd_bench_test(count, bsize, osize, onum, &elapsed, ss); | |
10117 | if (ret != 0) { | |
10118 | derr << __func__ | |
10119 | << " osd bench err: " << ret | |
10120 | << " osd bench errstr: " << ss.str() | |
10121 | << dendl; | |
10122 | return; | |
10123 | } | |
10124 | ||
10125 | double rate = count / elapsed; | |
10126 | iops = rate / bsize; | |
10127 | dout(1) << __func__ | |
10128 | << " osd bench result -" | |
10129 | << std::fixed << std::setprecision(3) | |
10130 | << " bandwidth (MiB/sec): " << rate / (1024 * 1024) | |
10131 | << " iops: " << iops | |
10132 | << " elapsed_sec: " << elapsed | |
10133 | << dendl; | |
10134 | ||
10135 | // Persist iops to the MON store | |
10136 | ret = mon_cmd_set_config(max_capacity_iops_config, std::to_string(iops)); | |
10137 | if (ret < 0) { | |
10138 | // Fallback to setting the config within the in-memory "values" map. | |
10139 | cct->_conf.set_val(max_capacity_iops_config, std::to_string(iops)); | |
10140 | } | |
10141 | ||
10142 | // Override the max osd capacity for all shards | |
10143 | for (auto& shard : shards) { | |
10144 | shard->update_scheduler_config(); | |
10145 | } | |
10146 | } | |
10147 | } | |
10148 | ||
b3b6e05e TL |
10149 | bool OSD::maybe_override_options_for_qos() |
10150 | { | |
10151 | // If the scheduler enabled is mclock, override the recovery, backfill | |
10152 | // and sleep options so that mclock can meet the QoS goals. | |
20effc67 TL |
10153 | if (cct->_conf.get_val<std::string>("osd_op_queue") == "mclock_scheduler" && |
10154 | !unsupported_objstore_for_qos()) { | |
b3b6e05e TL |
10155 | dout(1) << __func__ |
10156 | << ": Changing recovery/backfill/sleep settings for QoS" << dendl; | |
10157 | ||
10158 | // Set high value for recovery max active | |
10159 | uint32_t rec_max_active = 1000; | |
10160 | cct->_conf.set_val( | |
10161 | "osd_recovery_max_active", std::to_string(rec_max_active)); | |
10162 | cct->_conf.set_val( | |
10163 | "osd_recovery_max_active_hdd", std::to_string(rec_max_active)); | |
10164 | cct->_conf.set_val( | |
10165 | "osd_recovery_max_active_ssd", std::to_string(rec_max_active)); | |
10166 | ||
10167 | // Set high value for osd_max_backfill | |
10168 | uint32_t max_backfills = 1000; | |
10169 | cct->_conf.set_val("osd_max_backfills", std::to_string(max_backfills)); | |
10170 | service.local_reserver.set_max(max_backfills); | |
10171 | service.remote_reserver.set_max(max_backfills); | |
10172 | ||
10173 | // Disable recovery sleep | |
10174 | cct->_conf.set_val("osd_recovery_sleep", std::to_string(0)); | |
10175 | cct->_conf.set_val("osd_recovery_sleep_hdd", std::to_string(0)); | |
10176 | cct->_conf.set_val("osd_recovery_sleep_ssd", std::to_string(0)); | |
10177 | cct->_conf.set_val("osd_recovery_sleep_hybrid", std::to_string(0)); | |
10178 | ||
10179 | // Disable delete sleep | |
10180 | cct->_conf.set_val("osd_delete_sleep", std::to_string(0)); | |
10181 | cct->_conf.set_val("osd_delete_sleep_hdd", std::to_string(0)); | |
10182 | cct->_conf.set_val("osd_delete_sleep_ssd", std::to_string(0)); | |
10183 | cct->_conf.set_val("osd_delete_sleep_hybrid", std::to_string(0)); | |
10184 | ||
10185 | // Disable snap trim sleep | |
10186 | cct->_conf.set_val("osd_snap_trim_sleep", std::to_string(0)); | |
10187 | cct->_conf.set_val("osd_snap_trim_sleep_hdd", std::to_string(0)); | |
10188 | cct->_conf.set_val("osd_snap_trim_sleep_ssd", std::to_string(0)); | |
10189 | cct->_conf.set_val("osd_snap_trim_sleep_hybrid", std::to_string(0)); | |
10190 | ||
10191 | // Disable scrub sleep | |
10192 | cct->_conf.set_val("osd_scrub_sleep", std::to_string(0)); | |
10193 | return true; | |
10194 | } | |
10195 | return false; | |
10196 | } | |
10197 | ||
a4b75251 TL |
10198 | int OSD::mon_cmd_set_config(const std::string &key, const std::string &val) |
10199 | { | |
10200 | std::string cmd = | |
10201 | "{" | |
10202 | "\"prefix\": \"config set\", " | |
10203 | "\"who\": \"osd." + std::to_string(whoami) + "\", " | |
10204 | "\"name\": \"" + key + "\", " | |
10205 | "\"value\": \"" + val + "\"" | |
10206 | "}"; | |
10207 | ||
10208 | vector<std::string> vcmd{cmd}; | |
10209 | bufferlist inbl; | |
10210 | std::string outs; | |
10211 | C_SaferCond cond; | |
10212 | monc->start_mon_command(vcmd, inbl, nullptr, &outs, &cond); | |
10213 | int r = cond.wait(); | |
10214 | if (r < 0) { | |
10215 | derr << __func__ << " Failed to set config key " << key | |
10216 | << " err: " << cpp_strerror(r) | |
10217 | << " errstr: " << outs << dendl; | |
10218 | return r; | |
10219 | } | |
10220 | ||
10221 | return 0; | |
10222 | } | |
10223 | ||
20effc67 TL |
10224 | bool OSD::unsupported_objstore_for_qos() |
10225 | { | |
10226 | static const std::vector<std::string> unsupported_objstores = { "filestore" }; | |
10227 | return std::find(unsupported_objstores.begin(), | |
10228 | unsupported_objstores.end(), | |
10229 | store->get_type()) != unsupported_objstores.end(); | |
10230 | } | |
10231 | ||
7c673cae FG |
10232 | void OSD::update_log_config() |
10233 | { | |
20effc67 TL |
10234 | auto parsed_options = clog->parse_client_options(cct); |
10235 | derr << "log_to_monitors " << parsed_options.log_to_monitors << dendl; | |
7c673cae FG |
10236 | } |
10237 | ||
10238 | void OSD::check_config() | |
10239 | { | |
10240 | // some sanity checks | |
7c673cae FG |
10241 | if (cct->_conf->osd_map_cache_size <= (int)cct->_conf->osd_pg_epoch_persisted_max_stale + 2) { |
10242 | clog->warn() << "osd_map_cache_size (" << cct->_conf->osd_map_cache_size << ")" | |
10243 | << " is not > osd_pg_epoch_persisted_max_stale (" | |
10244 | << cct->_conf->osd_pg_epoch_persisted_max_stale << ")"; | |
10245 | } | |
9f95a23c | 10246 | if (cct->_conf->osd_object_clean_region_max_num_intervals < 0) { |
f67539c2 | 10247 | clog->warn() << "osd_object_clean_region_max_num_intervals (" |
9f95a23c TL |
10248 | << cct->_conf->osd_object_clean_region_max_num_intervals |
10249 | << ") is < 0"; | |
10250 | } | |
7c673cae FG |
10251 | } |
10252 | ||
7c673cae FG |
10253 | // -------------------------------- |
10254 | ||
10255 | void OSD::get_latest_osdmap() | |
10256 | { | |
10257 | dout(10) << __func__ << " -- start" << dendl; | |
10258 | ||
f67539c2 TL |
10259 | boost::system::error_code ec; |
10260 | service.objecter->wait_for_latest_osdmap(ceph::async::use_blocked[ec]); | |
7c673cae FG |
10261 | |
10262 | dout(10) << __func__ << " -- finish" << dendl; | |
10263 | } | |
10264 | ||
10265 | // -------------------------------- | |
10266 | ||
9f95a23c TL |
10267 | void OSD::set_perf_queries(const ConfigPayload &config_payload) { |
10268 | const OSDConfigPayload &osd_config_payload = boost::get<OSDConfigPayload>(config_payload); | |
10269 | const std::map<OSDPerfMetricQuery, OSDPerfMetricLimits> &queries = osd_config_payload.config; | |
11fdf7f2 TL |
10270 | dout(10) << "setting " << queries.size() << " queries" << dendl; |
10271 | ||
10272 | std::list<OSDPerfMetricQuery> supported_queries; | |
10273 | for (auto &it : queries) { | |
10274 | auto &query = it.first; | |
10275 | if (!query.key_descriptor.empty()) { | |
10276 | supported_queries.push_back(query); | |
10277 | } | |
10278 | } | |
10279 | if (supported_queries.size() < queries.size()) { | |
10280 | dout(1) << queries.size() - supported_queries.size() | |
10281 | << " unsupported queries" << dendl; | |
10282 | } | |
11fdf7f2 | 10283 | { |
9f95a23c | 10284 | std::lock_guard locker{m_perf_queries_lock}; |
11fdf7f2 TL |
10285 | m_perf_queries = supported_queries; |
10286 | m_perf_limits = queries; | |
10287 | } | |
11fdf7f2 TL |
10288 | std::vector<PGRef> pgs; |
10289 | _get_pgs(&pgs); | |
10290 | for (auto& pg : pgs) { | |
9f95a23c | 10291 | std::scoped_lock l{*pg}; |
eafe8130 | 10292 | pg->set_dynamic_perf_stats_queries(supported_queries); |
7c673cae | 10293 | } |
7c673cae FG |
10294 | } |
10295 | ||
9f95a23c TL |
10296 | MetricPayload OSD::get_perf_reports() { |
10297 | OSDMetricPayload payload; | |
10298 | std::map<OSDPerfMetricQuery, OSDPerfMetricReport> &reports = payload.report; | |
10299 | ||
11fdf7f2 TL |
10300 | std::vector<PGRef> pgs; |
10301 | _get_pgs(&pgs); | |
10302 | DynamicPerfStats dps; | |
10303 | for (auto& pg : pgs) { | |
eafe8130 TL |
10304 | // m_perf_queries can be modified only in set_perf_queries by mgr client |
10305 | // request, and it is protected by by mgr client's lock, which is held | |
10306 | // when set_perf_queries/get_perf_reports are called, so we may not hold | |
10307 | // m_perf_queries_lock here. | |
10308 | DynamicPerfStats pg_dps(m_perf_queries); | |
10309 | pg->lock(); | |
10310 | pg->get_dynamic_perf_stats(&pg_dps); | |
10311 | pg->unlock(); | |
10312 | dps.merge(pg_dps); | |
11fdf7f2 | 10313 | } |
9f95a23c TL |
10314 | dps.add_to_reports(m_perf_limits, &reports); |
10315 | dout(20) << "reports for " << reports.size() << " queries" << dendl; | |
10316 | ||
10317 | return payload; | |
11fdf7f2 | 10318 | } |
224ce89b | 10319 | |
7c673cae FG |
10320 | // ============================================================= |
10321 | ||
10322 | #undef dout_context | |
11fdf7f2 | 10323 | #define dout_context cct |
7c673cae | 10324 | #undef dout_prefix |
11fdf7f2 | 10325 | #define dout_prefix *_dout << "osd." << osd->get_nodeid() << ":" << shard_id << "." << __func__ << " " |
7c673cae | 10326 | |
11fdf7f2 | 10327 | void OSDShard::_attach_pg(OSDShardPGSlot *slot, PG *pg) |
7c673cae | 10328 | { |
11fdf7f2 TL |
10329 | dout(10) << pg->pg_id << " " << pg << dendl; |
10330 | slot->pg = pg; | |
10331 | pg->osd_shard = this; | |
10332 | pg->pg_slot = slot; | |
10333 | osd->inc_num_pgs(); | |
10334 | ||
10335 | slot->epoch = pg->get_osdmap_epoch(); | |
10336 | pg_slots_by_epoch.insert(*slot); | |
10337 | } | |
10338 | ||
10339 | void OSDShard::_detach_pg(OSDShardPGSlot *slot) | |
10340 | { | |
10341 | dout(10) << slot->pg->pg_id << " " << slot->pg << dendl; | |
10342 | slot->pg->osd_shard = nullptr; | |
10343 | slot->pg->pg_slot = nullptr; | |
10344 | slot->pg = nullptr; | |
10345 | osd->dec_num_pgs(); | |
10346 | ||
10347 | pg_slots_by_epoch.erase(pg_slots_by_epoch.iterator_to(*slot)); | |
10348 | slot->epoch = 0; | |
10349 | if (waiting_for_min_pg_epoch) { | |
10350 | min_pg_epoch_cond.notify_all(); | |
10351 | } | |
10352 | } | |
10353 | ||
10354 | void OSDShard::update_pg_epoch(OSDShardPGSlot *slot, epoch_t e) | |
10355 | { | |
10356 | std::lock_guard l(shard_lock); | |
10357 | dout(30) << "min was " << pg_slots_by_epoch.begin()->epoch | |
10358 | << " on " << pg_slots_by_epoch.begin()->pg->pg_id << dendl; | |
10359 | pg_slots_by_epoch.erase(pg_slots_by_epoch.iterator_to(*slot)); | |
10360 | dout(20) << slot->pg->pg_id << " " << slot->epoch << " -> " << e << dendl; | |
10361 | slot->epoch = e; | |
10362 | pg_slots_by_epoch.insert(*slot); | |
10363 | dout(30) << "min is now " << pg_slots_by_epoch.begin()->epoch | |
10364 | << " on " << pg_slots_by_epoch.begin()->pg->pg_id << dendl; | |
10365 | if (waiting_for_min_pg_epoch) { | |
10366 | min_pg_epoch_cond.notify_all(); | |
10367 | } | |
10368 | } | |
10369 | ||
10370 | epoch_t OSDShard::get_min_pg_epoch() | |
10371 | { | |
10372 | std::lock_guard l(shard_lock); | |
10373 | auto p = pg_slots_by_epoch.begin(); | |
10374 | if (p == pg_slots_by_epoch.end()) { | |
10375 | return 0; | |
10376 | } | |
10377 | return p->epoch; | |
10378 | } | |
10379 | ||
10380 | void OSDShard::wait_min_pg_epoch(epoch_t need) | |
10381 | { | |
10382 | std::unique_lock l{shard_lock}; | |
10383 | ++waiting_for_min_pg_epoch; | |
10384 | min_pg_epoch_cond.wait(l, [need, this] { | |
10385 | if (pg_slots_by_epoch.empty()) { | |
10386 | return true; | |
10387 | } else if (pg_slots_by_epoch.begin()->epoch >= need) { | |
10388 | return true; | |
10389 | } else { | |
10390 | dout(10) << need << " waiting on " | |
10391 | << pg_slots_by_epoch.begin()->epoch << dendl; | |
10392 | return false; | |
10393 | } | |
10394 | }); | |
10395 | --waiting_for_min_pg_epoch; | |
10396 | } | |
10397 | ||
10398 | epoch_t OSDShard::get_max_waiting_epoch() | |
10399 | { | |
10400 | std::lock_guard l(shard_lock); | |
10401 | epoch_t r = 0; | |
10402 | for (auto& i : pg_slots) { | |
10403 | if (!i.second->waiting_peering.empty()) { | |
10404 | r = std::max(r, i.second->waiting_peering.rbegin()->first); | |
10405 | } | |
10406 | } | |
10407 | return r; | |
10408 | } | |
10409 | ||
10410 | void OSDShard::consume_map( | |
9f95a23c | 10411 | const OSDMapRef& new_osdmap, |
11fdf7f2 TL |
10412 | unsigned *pushes_to_free) |
10413 | { | |
10414 | std::lock_guard l(shard_lock); | |
10415 | OSDMapRef old_osdmap; | |
7c673cae | 10416 | { |
11fdf7f2 TL |
10417 | std::lock_guard l(osdmap_lock); |
10418 | old_osdmap = std::move(shard_osdmap); | |
10419 | shard_osdmap = new_osdmap; | |
10420 | } | |
10421 | dout(10) << new_osdmap->get_epoch() | |
10422 | << " (was " << (old_osdmap ? old_osdmap->get_epoch() : 0) << ")" | |
10423 | << dendl; | |
20effc67 | 10424 | int queued = 0; |
11fdf7f2 TL |
10425 | |
10426 | // check slots | |
10427 | auto p = pg_slots.begin(); | |
10428 | while (p != pg_slots.end()) { | |
10429 | OSDShardPGSlot *slot = p->second.get(); | |
10430 | const spg_t& pgid = p->first; | |
10431 | dout(20) << __func__ << " " << pgid << dendl; | |
10432 | if (!slot->waiting_for_split.empty()) { | |
10433 | dout(20) << __func__ << " " << pgid | |
10434 | << " waiting for split " << slot->waiting_for_split << dendl; | |
10435 | ++p; | |
10436 | continue; | |
10437 | } | |
10438 | if (slot->waiting_for_merge_epoch > new_osdmap->get_epoch()) { | |
10439 | dout(20) << __func__ << " " << pgid | |
10440 | << " waiting for merge by epoch " << slot->waiting_for_merge_epoch | |
10441 | << dendl; | |
10442 | ++p; | |
10443 | continue; | |
10444 | } | |
10445 | if (!slot->waiting_peering.empty()) { | |
10446 | epoch_t first = slot->waiting_peering.begin()->first; | |
10447 | if (first <= new_osdmap->get_epoch()) { | |
10448 | dout(20) << __func__ << " " << pgid | |
10449 | << " pending_peering first epoch " << first | |
10450 | << " <= " << new_osdmap->get_epoch() << ", requeueing" << dendl; | |
20effc67 | 10451 | queued += _wake_pg_slot(pgid, slot); |
11fdf7f2 TL |
10452 | } |
10453 | ++p; | |
10454 | continue; | |
10455 | } | |
10456 | if (!slot->waiting.empty()) { | |
10457 | if (new_osdmap->is_up_acting_osd_shard(pgid, osd->get_nodeid())) { | |
10458 | dout(20) << __func__ << " " << pgid << " maps to us, keeping" | |
10459 | << dendl; | |
10460 | ++p; | |
10461 | continue; | |
7c673cae | 10462 | } |
11fdf7f2 TL |
10463 | while (!slot->waiting.empty() && |
10464 | slot->waiting.front().get_map_epoch() <= new_osdmap->get_epoch()) { | |
10465 | auto& qi = slot->waiting.front(); | |
10466 | dout(20) << __func__ << " " << pgid | |
10467 | << " waiting item " << qi | |
10468 | << " epoch " << qi.get_map_epoch() | |
10469 | << " <= " << new_osdmap->get_epoch() | |
10470 | << ", " | |
10471 | << (qi.get_map_epoch() < new_osdmap->get_epoch() ? "stale" : | |
10472 | "misdirected") | |
10473 | << ", dropping" << dendl; | |
10474 | *pushes_to_free += qi.get_reserved_pushes(); | |
10475 | slot->waiting.pop_front(); | |
10476 | } | |
10477 | } | |
10478 | if (slot->waiting.empty() && | |
10479 | slot->num_running == 0 && | |
10480 | slot->waiting_for_split.empty() && | |
10481 | !slot->pg) { | |
10482 | dout(20) << __func__ << " " << pgid << " empty, pruning" << dendl; | |
10483 | p = pg_slots.erase(p); | |
10484 | continue; | |
7c673cae | 10485 | } |
11fdf7f2 TL |
10486 | |
10487 | ++p; | |
7c673cae | 10488 | } |
7c673cae | 10489 | if (queued) { |
11fdf7f2 | 10490 | std::lock_guard l{sdata_wait_lock}; |
20effc67 TL |
10491 | if (queued == 1) |
10492 | sdata_cond.notify_one(); | |
10493 | else | |
10494 | sdata_cond.notify_all(); | |
7c673cae FG |
10495 | } |
10496 | } | |
10497 | ||
20effc67 | 10498 | int OSDShard::_wake_pg_slot( |
11fdf7f2 TL |
10499 | spg_t pgid, |
10500 | OSDShardPGSlot *slot) | |
10501 | { | |
20effc67 | 10502 | int count = 0; |
11fdf7f2 TL |
10503 | dout(20) << __func__ << " " << pgid |
10504 | << " to_process " << slot->to_process | |
10505 | << " waiting " << slot->waiting | |
10506 | << " waiting_peering " << slot->waiting_peering << dendl; | |
10507 | for (auto i = slot->to_process.rbegin(); | |
10508 | i != slot->to_process.rend(); | |
10509 | ++i) { | |
9f95a23c | 10510 | scheduler->enqueue_front(std::move(*i)); |
20effc67 | 10511 | count++; |
11fdf7f2 TL |
10512 | } |
10513 | slot->to_process.clear(); | |
10514 | for (auto i = slot->waiting.rbegin(); | |
10515 | i != slot->waiting.rend(); | |
10516 | ++i) { | |
9f95a23c | 10517 | scheduler->enqueue_front(std::move(*i)); |
20effc67 | 10518 | count++; |
11fdf7f2 TL |
10519 | } |
10520 | slot->waiting.clear(); | |
10521 | for (auto i = slot->waiting_peering.rbegin(); | |
10522 | i != slot->waiting_peering.rend(); | |
10523 | ++i) { | |
10524 | // this is overkill; we requeue everything, even if some of these | |
10525 | // items are waiting for maps we don't have yet. FIXME, maybe, | |
10526 | // someday, if we decide this inefficiency matters | |
10527 | for (auto j = i->second.rbegin(); j != i->second.rend(); ++j) { | |
9f95a23c | 10528 | scheduler->enqueue_front(std::move(*j)); |
20effc67 | 10529 | count++; |
11fdf7f2 TL |
10530 | } |
10531 | } | |
10532 | slot->waiting_peering.clear(); | |
10533 | ++slot->requeue_seq; | |
20effc67 | 10534 | return count; |
11fdf7f2 TL |
10535 | } |
10536 | ||
10537 | void OSDShard::identify_splits_and_merges( | |
10538 | const OSDMapRef& as_of_osdmap, | |
10539 | set<pair<spg_t,epoch_t>> *split_pgs, | |
10540 | set<pair<spg_t,epoch_t>> *merge_pgs) | |
10541 | { | |
10542 | std::lock_guard l(shard_lock); | |
10543 | if (shard_osdmap) { | |
10544 | for (auto& i : pg_slots) { | |
10545 | const spg_t& pgid = i.first; | |
10546 | auto *slot = i.second.get(); | |
10547 | if (slot->pg) { | |
10548 | osd->service.identify_splits_and_merges( | |
10549 | shard_osdmap, as_of_osdmap, pgid, | |
10550 | split_pgs, merge_pgs); | |
10551 | } else if (!slot->waiting_for_split.empty()) { | |
10552 | osd->service.identify_splits_and_merges( | |
10553 | shard_osdmap, as_of_osdmap, pgid, | |
10554 | split_pgs, nullptr); | |
10555 | } else { | |
10556 | dout(20) << __func__ << " slot " << pgid | |
9f95a23c | 10557 | << " has no pg and waiting_for_split " << dendl; |
7c673cae | 10558 | } |
11fdf7f2 TL |
10559 | } |
10560 | } | |
10561 | } | |
10562 | ||
10563 | void OSDShard::prime_splits(const OSDMapRef& as_of_osdmap, | |
10564 | set<pair<spg_t,epoch_t>> *pgids) | |
10565 | { | |
10566 | std::lock_guard l(shard_lock); | |
10567 | _prime_splits(pgids); | |
10568 | if (shard_osdmap->get_epoch() > as_of_osdmap->get_epoch()) { | |
10569 | set<pair<spg_t,epoch_t>> newer_children; | |
10570 | for (auto i : *pgids) { | |
10571 | osd->service.identify_splits_and_merges( | |
10572 | as_of_osdmap, shard_osdmap, i.first, | |
10573 | &newer_children, nullptr); | |
10574 | } | |
10575 | newer_children.insert(pgids->begin(), pgids->end()); | |
10576 | dout(10) << "as_of_osdmap " << as_of_osdmap->get_epoch() << " < shard " | |
10577 | << shard_osdmap->get_epoch() << ", new children " << newer_children | |
10578 | << dendl; | |
10579 | _prime_splits(&newer_children); | |
10580 | // note: we don't care what is left over here for other shards. | |
10581 | // if this shard is ahead of us and one isn't, e.g., one thread is | |
10582 | // calling into prime_splits via _process (due to a newly created | |
10583 | // pg) and this shard has a newer map due to a racing consume_map, | |
10584 | // then any grandchildren left here will be identified (or were | |
10585 | // identified) when the slower shard's osdmap is advanced. | |
10586 | // _prime_splits() will tolerate the case where the pgid is | |
10587 | // already primed. | |
10588 | } | |
10589 | } | |
10590 | ||
10591 | void OSDShard::_prime_splits(set<pair<spg_t,epoch_t>> *pgids) | |
10592 | { | |
10593 | dout(10) << *pgids << dendl; | |
10594 | auto p = pgids->begin(); | |
10595 | while (p != pgids->end()) { | |
10596 | unsigned shard_index = p->first.hash_to_shard(osd->num_shards); | |
10597 | if (shard_index == shard_id) { | |
10598 | auto r = pg_slots.emplace(p->first, nullptr); | |
10599 | if (r.second) { | |
10600 | dout(10) << "priming slot " << p->first << " e" << p->second << dendl; | |
10601 | r.first->second = make_unique<OSDShardPGSlot>(); | |
10602 | r.first->second->waiting_for_split.insert(p->second); | |
7c673cae | 10603 | } else { |
11fdf7f2 TL |
10604 | auto q = r.first; |
10605 | ceph_assert(q != pg_slots.end()); | |
10606 | dout(10) << "priming (existing) slot " << p->first << " e" << p->second | |
10607 | << dendl; | |
10608 | q->second->waiting_for_split.insert(p->second); | |
7c673cae | 10609 | } |
11fdf7f2 TL |
10610 | p = pgids->erase(p); |
10611 | } else { | |
10612 | ++p; | |
7c673cae FG |
10613 | } |
10614 | } | |
11fdf7f2 TL |
10615 | } |
10616 | ||
10617 | void OSDShard::prime_merges(const OSDMapRef& as_of_osdmap, | |
10618 | set<pair<spg_t,epoch_t>> *merge_pgs) | |
10619 | { | |
10620 | std::lock_guard l(shard_lock); | |
10621 | dout(20) << __func__ << " checking shard " << shard_id | |
10622 | << " for remaining merge pgs " << merge_pgs << dendl; | |
10623 | auto p = merge_pgs->begin(); | |
10624 | while (p != merge_pgs->end()) { | |
10625 | spg_t pgid = p->first; | |
10626 | epoch_t epoch = p->second; | |
10627 | unsigned shard_index = pgid.hash_to_shard(osd->num_shards); | |
10628 | if (shard_index != shard_id) { | |
10629 | ++p; | |
10630 | continue; | |
10631 | } | |
10632 | OSDShardPGSlot *slot; | |
10633 | auto r = pg_slots.emplace(pgid, nullptr); | |
10634 | if (r.second) { | |
10635 | r.first->second = make_unique<OSDShardPGSlot>(); | |
10636 | } | |
10637 | slot = r.first->second.get(); | |
10638 | if (slot->pg) { | |
10639 | // already have pg | |
10640 | dout(20) << __func__ << " have merge participant pg " << pgid | |
10641 | << " " << slot->pg << dendl; | |
10642 | } else if (!slot->waiting_for_split.empty() && | |
10643 | *slot->waiting_for_split.begin() < epoch) { | |
10644 | dout(20) << __func__ << " pending split on merge participant pg " << pgid | |
10645 | << " " << slot->waiting_for_split << dendl; | |
10646 | } else { | |
10647 | dout(20) << __func__ << " creating empty merge participant " << pgid | |
10648 | << " for merge in " << epoch << dendl; | |
10649 | // leave history zeroed; PG::merge_from() will fill it in. | |
10650 | pg_history_t history; | |
10651 | PGCreateInfo cinfo(pgid, epoch - 1, | |
10652 | history, PastIntervals(), false); | |
10653 | PGRef pg = osd->handle_pg_create_info(shard_osdmap, &cinfo); | |
10654 | _attach_pg(r.first->second.get(), pg.get()); | |
10655 | _wake_pg_slot(pgid, slot); | |
10656 | pg->unlock(); | |
10657 | } | |
10658 | // mark slot for merge | |
10659 | dout(20) << __func__ << " marking merge participant " << pgid << dendl; | |
10660 | slot->waiting_for_merge_epoch = epoch; | |
10661 | p = merge_pgs->erase(p); | |
7c673cae FG |
10662 | } |
10663 | } | |
10664 | ||
11fdf7f2 | 10665 | void OSDShard::register_and_wake_split_child(PG *pg) |
7c673cae | 10666 | { |
20effc67 | 10667 | dout(15) << __func__ << ": " << pg << " #:" << pg_slots.size() << dendl; |
11fdf7f2 TL |
10668 | epoch_t epoch; |
10669 | { | |
10670 | std::lock_guard l(shard_lock); | |
20effc67 | 10671 | dout(10) << __func__ << ": " << pg->pg_id << " " << pg << dendl; |
11fdf7f2 TL |
10672 | auto p = pg_slots.find(pg->pg_id); |
10673 | ceph_assert(p != pg_slots.end()); | |
10674 | auto *slot = p->second.get(); | |
20effc67 TL |
10675 | dout(20) << __func__ << ": " << pg->pg_id << " waiting_for_split " |
10676 | << slot->waiting_for_split << dendl; | |
11fdf7f2 TL |
10677 | ceph_assert(!slot->pg); |
10678 | ceph_assert(!slot->waiting_for_split.empty()); | |
10679 | _attach_pg(slot, pg); | |
10680 | ||
10681 | epoch = pg->get_osdmap_epoch(); | |
10682 | ceph_assert(slot->waiting_for_split.count(epoch)); | |
10683 | slot->waiting_for_split.erase(epoch); | |
10684 | if (slot->waiting_for_split.empty()) { | |
10685 | _wake_pg_slot(pg->pg_id, slot); | |
10686 | } else { | |
10687 | dout(10) << __func__ << " still waiting for split on " | |
10688 | << slot->waiting_for_split << dendl; | |
10689 | } | |
7c673cae | 10690 | } |
11fdf7f2 TL |
10691 | |
10692 | // kick child to ensure it pulls up to the latest osdmap | |
10693 | osd->enqueue_peering_evt( | |
10694 | pg->pg_id, | |
10695 | PGPeeringEventRef( | |
10696 | std::make_shared<PGPeeringEvent>( | |
10697 | epoch, | |
10698 | epoch, | |
10699 | NullEvt()))); | |
10700 | ||
10701 | std::lock_guard l{sdata_wait_lock}; | |
10702 | sdata_cond.notify_one(); | |
7c673cae FG |
10703 | } |
10704 | ||
11fdf7f2 | 10705 | void OSDShard::unprime_split_children(spg_t parent, unsigned old_pg_num) |
7c673cae | 10706 | { |
11fdf7f2 TL |
10707 | std::lock_guard l(shard_lock); |
10708 | vector<spg_t> to_delete; | |
10709 | for (auto& i : pg_slots) { | |
10710 | if (i.first != parent && | |
10711 | i.first.get_ancestor(old_pg_num) == parent) { | |
10712 | dout(10) << __func__ << " parent " << parent << " clearing " << i.first | |
10713 | << dendl; | |
10714 | _wake_pg_slot(i.first, i.second.get()); | |
10715 | to_delete.push_back(i.first); | |
10716 | } | |
10717 | } | |
10718 | for (auto pgid : to_delete) { | |
10719 | pg_slots.erase(pgid); | |
10720 | } | |
10721 | } | |
10722 | ||
a4b75251 TL |
10723 | void OSDShard::update_scheduler_config() |
10724 | { | |
10725 | std::lock_guard l(shard_lock); | |
10726 | scheduler->update_configuration(); | |
10727 | } | |
10728 | ||
20effc67 TL |
10729 | std::string OSDShard::get_scheduler_type() |
10730 | { | |
10731 | std::ostringstream scheduler_type; | |
10732 | scheduler_type << *scheduler; | |
10733 | return scheduler_type.str(); | |
10734 | } | |
10735 | ||
9f95a23c TL |
10736 | OSDShard::OSDShard( |
10737 | int id, | |
10738 | CephContext *cct, | |
10739 | OSD *osd) | |
10740 | : shard_id(id), | |
10741 | cct(cct), | |
10742 | osd(osd), | |
10743 | shard_name(string("OSDShard.") + stringify(id)), | |
10744 | sdata_wait_lock_name(shard_name + "::sdata_wait_lock"), | |
10745 | sdata_wait_lock{make_mutex(sdata_wait_lock_name)}, | |
10746 | osdmap_lock{make_mutex(shard_name + "::osdmap_lock")}, | |
10747 | shard_lock_name(shard_name + "::shard_lock"), | |
10748 | shard_lock{make_mutex(shard_lock_name)}, | |
f67539c2 | 10749 | scheduler(ceph::osd::scheduler::make_scheduler( |
20effc67 TL |
10750 | cct, osd->num_shards, osd->store->is_rotational(), |
10751 | osd->store->get_type())), | |
9f95a23c TL |
10752 | context_queue(sdata_wait_lock, sdata_cond) |
10753 | { | |
10754 | dout(0) << "using op scheduler " << *scheduler << dendl; | |
10755 | } | |
10756 | ||
11fdf7f2 TL |
10757 | |
10758 | // ============================================================= | |
10759 | ||
10760 | #undef dout_context | |
10761 | #define dout_context osd->cct | |
10762 | #undef dout_prefix | |
10763 | #define dout_prefix *_dout << "osd." << osd->whoami << " op_wq " | |
10764 | ||
10765 | void OSD::ShardedOpWQ::_add_slot_waiter( | |
10766 | spg_t pgid, | |
10767 | OSDShardPGSlot *slot, | |
9f95a23c | 10768 | OpSchedulerItem&& qi) |
11fdf7f2 TL |
10769 | { |
10770 | if (qi.is_peering()) { | |
10771 | dout(20) << __func__ << " " << pgid | |
10772 | << " peering, item epoch is " | |
10773 | << qi.get_map_epoch() | |
10774 | << ", will wait on " << qi << dendl; | |
10775 | slot->waiting_peering[qi.get_map_epoch()].push_back(std::move(qi)); | |
10776 | } else { | |
10777 | dout(20) << __func__ << " " << pgid | |
10778 | << " item epoch is " | |
10779 | << qi.get_map_epoch() | |
10780 | << ", will wait on " << qi << dendl; | |
10781 | slot->waiting.push_back(std::move(qi)); | |
7c673cae FG |
10782 | } |
10783 | } | |
10784 | ||
10785 | #undef dout_prefix | |
10786 | #define dout_prefix *_dout << "osd." << osd->whoami << " op_wq(" << shard_index << ") " | |
10787 | ||
10788 | void OSD::ShardedOpWQ::_process(uint32_t thread_index, heartbeat_handle_d *hb) | |
10789 | { | |
11fdf7f2 TL |
10790 | uint32_t shard_index = thread_index % osd->num_shards; |
10791 | auto& sdata = osd->shards[shard_index]; | |
10792 | ceph_assert(sdata); | |
10793 | ||
10794 | // If all threads of shards do oncommits, there is a out-of-order | |
10795 | // problem. So we choose the thread which has the smallest | |
10796 | // thread_index(thread_index < num_shards) of shard to do oncommit | |
10797 | // callback. | |
10798 | bool is_smallest_thread_index = thread_index < osd->num_shards; | |
7c673cae FG |
10799 | |
10800 | // peek at spg_t | |
11fdf7f2 | 10801 | sdata->shard_lock.lock(); |
9f95a23c | 10802 | if (sdata->scheduler->empty() && |
11fdf7f2 TL |
10803 | (!is_smallest_thread_index || sdata->context_queue.empty())) { |
10804 | std::unique_lock wait_lock{sdata->sdata_wait_lock}; | |
10805 | if (is_smallest_thread_index && !sdata->context_queue.empty()) { | |
10806 | // we raced with a context_queue addition, don't wait | |
10807 | wait_lock.unlock(); | |
10808 | } else if (!sdata->stop_waiting) { | |
10809 | dout(20) << __func__ << " empty q, waiting" << dendl; | |
10810 | osd->cct->get_heartbeat_map()->clear_timeout(hb); | |
10811 | sdata->shard_lock.unlock(); | |
10812 | sdata->sdata_cond.wait(wait_lock); | |
10813 | wait_lock.unlock(); | |
10814 | sdata->shard_lock.lock(); | |
9f95a23c | 10815 | if (sdata->scheduler->empty() && |
11fdf7f2 TL |
10816 | !(is_smallest_thread_index && !sdata->context_queue.empty())) { |
10817 | sdata->shard_lock.unlock(); | |
10818 | return; | |
10819 | } | |
e306af50 | 10820 | // found a work item; reapply default wq timeouts |
11fdf7f2 | 10821 | osd->cct->get_heartbeat_map()->reset_timeout(hb, |
e306af50 | 10822 | timeout_interval, suicide_interval); |
11fdf7f2 TL |
10823 | } else { |
10824 | dout(20) << __func__ << " need return immediately" << dendl; | |
10825 | wait_lock.unlock(); | |
10826 | sdata->shard_lock.unlock(); | |
7c673cae FG |
10827 | return; |
10828 | } | |
10829 | } | |
11fdf7f2 TL |
10830 | |
10831 | list<Context *> oncommits; | |
9f95a23c TL |
10832 | if (is_smallest_thread_index) { |
10833 | sdata->context_queue.move_to(oncommits); | |
7c673cae | 10834 | } |
11fdf7f2 | 10835 | |
f67539c2 TL |
10836 | WorkItem work_item; |
10837 | while (!std::get_if<OpSchedulerItem>(&work_item)) { | |
10838 | if (sdata->scheduler->empty()) { | |
10839 | if (osd->is_stopping()) { | |
10840 | sdata->shard_lock.unlock(); | |
10841 | for (auto c : oncommits) { | |
10842 | dout(10) << __func__ << " discarding in-flight oncommit " << c << dendl; | |
10843 | delete c; | |
10844 | } | |
10845 | return; // OSD shutdown, discard. | |
10846 | } | |
10847 | sdata->shard_lock.unlock(); | |
10848 | handle_oncommits(oncommits); | |
10849 | return; | |
10850 | } | |
10851 | ||
10852 | work_item = sdata->scheduler->dequeue(); | |
11fdf7f2 TL |
10853 | if (osd->is_stopping()) { |
10854 | sdata->shard_lock.unlock(); | |
10855 | for (auto c : oncommits) { | |
f67539c2 TL |
10856 | dout(10) << __func__ << " discarding in-flight oncommit " << c << dendl; |
10857 | delete c; | |
11fdf7f2 TL |
10858 | } |
10859 | return; // OSD shutdown, discard. | |
7c673cae | 10860 | } |
7c673cae | 10861 | |
f67539c2 TL |
10862 | // If the work item is scheduled in the future, wait until |
10863 | // the time returned in the dequeue response before retrying. | |
10864 | if (auto when_ready = std::get_if<double>(&work_item)) { | |
10865 | if (is_smallest_thread_index) { | |
10866 | sdata->shard_lock.unlock(); | |
10867 | handle_oncommits(oncommits); | |
10868 | return; | |
10869 | } | |
10870 | std::unique_lock wait_lock{sdata->sdata_wait_lock}; | |
10871 | auto future_time = ceph::real_clock::from_double(*when_ready); | |
10872 | dout(10) << __func__ << " dequeue future request at " << future_time << dendl; | |
a4b75251 TL |
10873 | // Disable heartbeat timeout until we find a non-future work item to process. |
10874 | osd->cct->get_heartbeat_map()->clear_timeout(hb); | |
f67539c2 TL |
10875 | sdata->shard_lock.unlock(); |
10876 | ++sdata->waiting_threads; | |
10877 | sdata->sdata_cond.wait_until(wait_lock, future_time); | |
10878 | --sdata->waiting_threads; | |
10879 | wait_lock.unlock(); | |
10880 | sdata->shard_lock.lock(); | |
a4b75251 TL |
10881 | // Reapply default wq timeouts |
10882 | osd->cct->get_heartbeat_map()->reset_timeout(hb, | |
10883 | timeout_interval, suicide_interval); | |
f67539c2 TL |
10884 | } |
10885 | } // while | |
10886 | ||
10887 | // Access the stored item | |
10888 | auto item = std::move(std::get<OpSchedulerItem>(work_item)); | |
11fdf7f2 TL |
10889 | if (osd->is_stopping()) { |
10890 | sdata->shard_lock.unlock(); | |
10891 | for (auto c : oncommits) { | |
10892 | dout(10) << __func__ << " discarding in-flight oncommit " << c << dendl; | |
10893 | delete c; | |
10894 | } | |
10895 | return; // OSD shutdown, discard. | |
10896 | } | |
7c673cae | 10897 | |
11fdf7f2 TL |
10898 | const auto token = item.get_ordering_token(); |
10899 | auto r = sdata->pg_slots.emplace(token, nullptr); | |
10900 | if (r.second) { | |
10901 | r.first->second = make_unique<OSDShardPGSlot>(); | |
7c673cae | 10902 | } |
11fdf7f2 TL |
10903 | OSDShardPGSlot *slot = r.first->second.get(); |
10904 | dout(20) << __func__ << " " << token | |
10905 | << (r.second ? " (new)" : "") | |
10906 | << " to_process " << slot->to_process | |
10907 | << " waiting " << slot->waiting | |
10908 | << " waiting_peering " << slot->waiting_peering | |
10909 | << dendl; | |
10910 | slot->to_process.push_back(std::move(item)); | |
10911 | dout(20) << __func__ << " " << slot->to_process.back() | |
10912 | << " queued" << dendl; | |
7c673cae | 10913 | |
11fdf7f2 TL |
10914 | retry_pg: |
10915 | PGRef pg = slot->pg; | |
7c673cae | 10916 | |
11fdf7f2 TL |
10917 | // lock pg (if we have it) |
10918 | if (pg) { | |
10919 | // note the requeue seq now... | |
10920 | uint64_t requeue_seq = slot->requeue_seq; | |
10921 | ++slot->num_running; | |
7c673cae | 10922 | |
11fdf7f2 TL |
10923 | sdata->shard_lock.unlock(); |
10924 | osd->service.maybe_inject_dispatch_delay(); | |
10925 | pg->lock(); | |
10926 | osd->service.maybe_inject_dispatch_delay(); | |
10927 | sdata->shard_lock.lock(); | |
7c673cae | 10928 | |
11fdf7f2 TL |
10929 | auto q = sdata->pg_slots.find(token); |
10930 | if (q == sdata->pg_slots.end()) { | |
10931 | // this can happen if we race with pg removal. | |
10932 | dout(20) << __func__ << " slot " << token << " no longer there" << dendl; | |
10933 | pg->unlock(); | |
10934 | sdata->shard_lock.unlock(); | |
10935 | handle_oncommits(oncommits); | |
10936 | return; | |
10937 | } | |
10938 | slot = q->second.get(); | |
10939 | --slot->num_running; | |
7c673cae | 10940 | |
11fdf7f2 TL |
10941 | if (slot->to_process.empty()) { |
10942 | // raced with _wake_pg_slot or consume_map | |
10943 | dout(20) << __func__ << " " << token | |
10944 | << " nothing queued" << dendl; | |
7c673cae | 10945 | pg->unlock(); |
11fdf7f2 TL |
10946 | sdata->shard_lock.unlock(); |
10947 | handle_oncommits(oncommits); | |
10948 | return; | |
7c673cae | 10949 | } |
11fdf7f2 TL |
10950 | if (requeue_seq != slot->requeue_seq) { |
10951 | dout(20) << __func__ << " " << token | |
10952 | << " requeue_seq " << slot->requeue_seq << " > our " | |
10953 | << requeue_seq << ", we raced with _wake_pg_slot" | |
10954 | << dendl; | |
7c673cae | 10955 | pg->unlock(); |
11fdf7f2 TL |
10956 | sdata->shard_lock.unlock(); |
10957 | handle_oncommits(oncommits); | |
10958 | return; | |
7c673cae | 10959 | } |
11fdf7f2 TL |
10960 | if (slot->pg != pg) { |
10961 | // this can happen if we race with pg removal. | |
10962 | dout(20) << __func__ << " slot " << token << " no longer attached to " | |
10963 | << pg << dendl; | |
7c673cae | 10964 | pg->unlock(); |
11fdf7f2 | 10965 | goto retry_pg; |
7c673cae | 10966 | } |
7c673cae FG |
10967 | } |
10968 | ||
11fdf7f2 TL |
10969 | dout(20) << __func__ << " " << token |
10970 | << " to_process " << slot->to_process | |
10971 | << " waiting " << slot->waiting | |
10972 | << " waiting_peering " << slot->waiting_peering << dendl; | |
10973 | ||
10974 | ThreadPool::TPHandle tp_handle(osd->cct, hb, timeout_interval, | |
10975 | suicide_interval); | |
10976 | ||
7c673cae | 10977 | // take next item |
11fdf7f2 TL |
10978 | auto qi = std::move(slot->to_process.front()); |
10979 | slot->to_process.pop_front(); | |
10980 | dout(20) << __func__ << " " << qi << " pg " << pg << dendl; | |
10981 | set<pair<spg_t,epoch_t>> new_children; | |
10982 | OSDMapRef osdmap; | |
7c673cae | 10983 | |
11fdf7f2 | 10984 | while (!pg) { |
7c673cae | 10985 | // should this pg shard exist on this osd in this (or a later) epoch? |
11fdf7f2 TL |
10986 | osdmap = sdata->shard_osdmap; |
10987 | const PGCreateInfo *create_info = qi.creates_pg(); | |
10988 | if (!slot->waiting_for_split.empty()) { | |
10989 | dout(20) << __func__ << " " << token | |
10990 | << " splitting " << slot->waiting_for_split << dendl; | |
10991 | _add_slot_waiter(token, slot, std::move(qi)); | |
10992 | } else if (qi.get_map_epoch() > osdmap->get_epoch()) { | |
10993 | dout(20) << __func__ << " " << token | |
10994 | << " map " << qi.get_map_epoch() << " > " | |
10995 | << osdmap->get_epoch() << dendl; | |
10996 | _add_slot_waiter(token, slot, std::move(qi)); | |
10997 | } else if (qi.is_peering()) { | |
10998 | if (!qi.peering_requires_pg()) { | |
10999 | // for pg-less events, we run them under the ordering lock, since | |
11000 | // we don't have the pg lock to keep them ordered. | |
11001 | qi.run(osd, sdata, pg, tp_handle); | |
11002 | } else if (osdmap->is_up_acting_osd_shard(token, osd->whoami)) { | |
11003 | if (create_info) { | |
11004 | if (create_info->by_mon && | |
11005 | osdmap->get_pg_acting_primary(token.pgid) != osd->whoami) { | |
11006 | dout(20) << __func__ << " " << token | |
11007 | << " no pg, no longer primary, ignoring mon create on " | |
11008 | << qi << dendl; | |
11009 | } else { | |
11010 | dout(20) << __func__ << " " << token | |
11011 | << " no pg, should create on " << qi << dendl; | |
11012 | pg = osd->handle_pg_create_info(osdmap, create_info); | |
11013 | if (pg) { | |
11014 | // we created the pg! drop out and continue "normally"! | |
11015 | sdata->_attach_pg(slot, pg.get()); | |
11016 | sdata->_wake_pg_slot(token, slot); | |
11017 | ||
11018 | // identify split children between create epoch and shard epoch. | |
11019 | osd->service.identify_splits_and_merges( | |
11020 | pg->get_osdmap(), osdmap, pg->pg_id, &new_children, nullptr); | |
11021 | sdata->_prime_splits(&new_children); | |
11022 | // distribute remaining split children to other shards below! | |
11023 | break; | |
11024 | } | |
11025 | dout(20) << __func__ << " ignored create on " << qi << dendl; | |
11026 | } | |
11027 | } else { | |
11028 | dout(20) << __func__ << " " << token | |
11029 | << " no pg, peering, !create, discarding " << qi << dendl; | |
11030 | } | |
11031 | } else { | |
11032 | dout(20) << __func__ << " " << token | |
11033 | << " no pg, peering, doesn't map here e" << osdmap->get_epoch() | |
11034 | << ", discarding " << qi | |
11035 | << dendl; | |
11036 | } | |
11037 | } else if (osdmap->is_up_acting_osd_shard(token, osd->whoami)) { | |
11038 | dout(20) << __func__ << " " << token | |
11039 | << " no pg, should exist e" << osdmap->get_epoch() | |
11040 | << ", will wait on " << qi << dendl; | |
11041 | _add_slot_waiter(token, slot, std::move(qi)); | |
7c673cae | 11042 | } else { |
11fdf7f2 TL |
11043 | dout(20) << __func__ << " " << token |
11044 | << " no pg, shouldn't exist e" << osdmap->get_epoch() | |
11045 | << ", dropping " << qi << dendl; | |
7c673cae | 11046 | // share map with client? |
9f95a23c TL |
11047 | if (std::optional<OpRequestRef> _op = qi.maybe_get_op()) { |
11048 | osd->service.maybe_share_map((*_op)->get_req()->get_connection().get(), | |
11049 | sdata->shard_osdmap, | |
11050 | (*_op)->sent_epoch); | |
7c673cae | 11051 | } |
11fdf7f2 | 11052 | unsigned pushes_to_free = qi.get_reserved_pushes(); |
7c673cae | 11053 | if (pushes_to_free > 0) { |
11fdf7f2 | 11054 | sdata->shard_lock.unlock(); |
7c673cae | 11055 | osd->service.release_reserved_pushes(pushes_to_free); |
11fdf7f2 | 11056 | handle_oncommits(oncommits); |
7c673cae FG |
11057 | return; |
11058 | } | |
11059 | } | |
11fdf7f2 TL |
11060 | sdata->shard_lock.unlock(); |
11061 | handle_oncommits(oncommits); | |
7c673cae FG |
11062 | return; |
11063 | } | |
11fdf7f2 TL |
11064 | if (qi.is_peering()) { |
11065 | OSDMapRef osdmap = sdata->shard_osdmap; | |
11066 | if (qi.get_map_epoch() > osdmap->get_epoch()) { | |
11067 | _add_slot_waiter(token, slot, std::move(qi)); | |
11068 | sdata->shard_lock.unlock(); | |
11069 | pg->unlock(); | |
11070 | handle_oncommits(oncommits); | |
11071 | return; | |
11072 | } | |
11073 | } | |
11074 | sdata->shard_lock.unlock(); | |
7c673cae | 11075 | |
11fdf7f2 TL |
11076 | if (!new_children.empty()) { |
11077 | for (auto shard : osd->shards) { | |
11078 | shard->prime_splits(osdmap, &new_children); | |
11079 | } | |
11080 | ceph_assert(new_children.empty()); | |
11081 | } | |
7c673cae FG |
11082 | |
11083 | // osd_opwq_process marks the point at which an operation has been dequeued | |
11084 | // and will begin to be handled by a worker thread. | |
11085 | { | |
11086 | #ifdef WITH_LTTNG | |
11087 | osd_reqid_t reqid; | |
9f95a23c | 11088 | if (std::optional<OpRequestRef> _op = qi.maybe_get_op()) { |
7c673cae FG |
11089 | reqid = (*_op)->get_reqid(); |
11090 | } | |
11091 | #endif | |
11092 | tracepoint(osd, opwq_process_start, reqid.name._type, | |
11093 | reqid.name._num, reqid.tid, reqid.inc); | |
11094 | } | |
11095 | ||
11096 | lgeneric_subdout(osd->cct, osd, 30) << "dequeue status: "; | |
11097 | Formatter *f = Formatter::create("json"); | |
11098 | f->open_object_section("q"); | |
11099 | dump(f); | |
11100 | f->close_section(); | |
11101 | f->flush(*_dout); | |
11102 | delete f; | |
11103 | *_dout << dendl; | |
11104 | ||
11fdf7f2 | 11105 | qi.run(osd, sdata, pg, tp_handle); |
7c673cae FG |
11106 | |
11107 | { | |
11108 | #ifdef WITH_LTTNG | |
11109 | osd_reqid_t reqid; | |
9f95a23c | 11110 | if (std::optional<OpRequestRef> _op = qi.maybe_get_op()) { |
7c673cae FG |
11111 | reqid = (*_op)->get_reqid(); |
11112 | } | |
11113 | #endif | |
11114 | tracepoint(osd, opwq_process_finish, reqid.name._type, | |
11115 | reqid.name._num, reqid.tid, reqid.inc); | |
11116 | } | |
11117 | ||
11fdf7f2 | 11118 | handle_oncommits(oncommits); |
7c673cae FG |
11119 | } |
11120 | ||
9f95a23c | 11121 | void OSD::ShardedOpWQ::_enqueue(OpSchedulerItem&& item) { |
1d09f67e TL |
11122 | if (unlikely(m_fast_shutdown) ) { |
11123 | // stop enqueing when we are in the middle of a fast shutdown | |
11124 | return; | |
11125 | } | |
11126 | ||
7c673cae | 11127 | uint32_t shard_index = |
11fdf7f2 | 11128 | item.get_ordering_token().hash_to_shard(osd->shards.size()); |
7c673cae | 11129 | |
11fdf7f2 | 11130 | OSDShard* sdata = osd->shards[shard_index]; |
7c673cae | 11131 | assert (NULL != sdata); |
20effc67 TL |
11132 | if (sdata->get_scheduler_type() == "mClockScheduler") { |
11133 | item.maybe_set_is_qos_item(); | |
11134 | } | |
11135 | ||
11136 | dout(20) << __func__ << " " << item << dendl; | |
7c673cae | 11137 | |
9f95a23c TL |
11138 | bool empty = true; |
11139 | { | |
11140 | std::lock_guard l{sdata->shard_lock}; | |
11141 | empty = sdata->scheduler->empty(); | |
11142 | sdata->scheduler->enqueue(std::move(item)); | |
11143 | } | |
7c673cae | 11144 | |
f67539c2 | 11145 | { |
9f95a23c | 11146 | std::lock_guard l{sdata->sdata_wait_lock}; |
f67539c2 TL |
11147 | if (empty) { |
11148 | sdata->sdata_cond.notify_all(); | |
11149 | } else if (sdata->waiting_threads) { | |
11150 | sdata->sdata_cond.notify_one(); | |
11151 | } | |
9f95a23c | 11152 | } |
7c673cae FG |
11153 | } |
11154 | ||
9f95a23c | 11155 | void OSD::ShardedOpWQ::_enqueue_front(OpSchedulerItem&& item) |
7c673cae | 11156 | { |
1d09f67e TL |
11157 | if (unlikely(m_fast_shutdown) ) { |
11158 | // stop enqueing when we are in the middle of a fast shutdown | |
11159 | return; | |
11160 | } | |
11161 | ||
11fdf7f2 TL |
11162 | auto shard_index = item.get_ordering_token().hash_to_shard(osd->shards.size()); |
11163 | auto& sdata = osd->shards[shard_index]; | |
11164 | ceph_assert(sdata); | |
11165 | sdata->shard_lock.lock(); | |
11166 | auto p = sdata->pg_slots.find(item.get_ordering_token()); | |
11167 | if (p != sdata->pg_slots.end() && | |
11168 | !p->second->to_process.empty()) { | |
7c673cae | 11169 | // we may be racing with _process, which has dequeued a new item |
9f95a23c | 11170 | // from scheduler, put it on to_process, and is now busy taking the |
7c673cae FG |
11171 | // pg lock. ensure this old requeued item is ordered before any |
11172 | // such newer item in to_process. | |
11fdf7f2 TL |
11173 | p->second->to_process.push_front(std::move(item)); |
11174 | item = std::move(p->second->to_process.back()); | |
11175 | p->second->to_process.pop_back(); | |
11176 | dout(20) << __func__ | |
11177 | << " " << p->second->to_process.front() | |
11178 | << " shuffled w/ " << item << dendl; | |
7c673cae | 11179 | } else { |
11fdf7f2 | 11180 | dout(20) << __func__ << " " << item << dendl; |
7c673cae | 11181 | } |
9f95a23c | 11182 | sdata->scheduler->enqueue_front(std::move(item)); |
11fdf7f2 TL |
11183 | sdata->shard_lock.unlock(); |
11184 | std::lock_guard l{sdata->sdata_wait_lock}; | |
11185 | sdata->sdata_cond.notify_one(); | |
7c673cae FG |
11186 | } |
11187 | ||
1d09f67e TL |
11188 | void OSD::ShardedOpWQ::stop_for_fast_shutdown() |
11189 | { | |
11190 | uint32_t shard_index = 0; | |
11191 | m_fast_shutdown = true; | |
11192 | ||
11193 | for (; shard_index < osd->num_shards; shard_index++) { | |
11194 | auto& sdata = osd->shards[shard_index]; | |
11195 | ceph_assert(sdata); | |
11196 | sdata->shard_lock.lock(); | |
11197 | int work_count = 0; | |
11198 | while(! sdata->scheduler->empty() ) { | |
11199 | auto work_item = sdata->scheduler->dequeue(); | |
11200 | work_count++; | |
11201 | } | |
11202 | sdata->shard_lock.unlock(); | |
11203 | } | |
11204 | } | |
11205 | ||
f67539c2 | 11206 | namespace ceph::osd_cmds { |
7c673cae | 11207 | |
11fdf7f2 TL |
11208 | int heap(CephContext& cct, const cmdmap_t& cmdmap, Formatter& f, |
11209 | std::ostream& os) | |
7c673cae FG |
11210 | { |
11211 | if (!ceph_using_tcmalloc()) { | |
11212 | os << "could not issue heap profiler command -- not using tcmalloc!"; | |
11213 | return -EOPNOTSUPP; | |
11214 | } | |
f67539c2 | 11215 | |
7c673cae | 11216 | string cmd; |
9f95a23c | 11217 | if (!cmd_getval(cmdmap, "heapcmd", cmd)) { |
7c673cae FG |
11218 | os << "unable to get value for command \"" << cmd << "\""; |
11219 | return -EINVAL; | |
11fdf7f2 | 11220 | } |
f67539c2 | 11221 | |
7c673cae FG |
11222 | std::vector<std::string> cmd_vec; |
11223 | get_str_vec(cmd, cmd_vec); | |
11fdf7f2 TL |
11224 | |
11225 | string val; | |
9f95a23c | 11226 | if (cmd_getval(cmdmap, "value", val)) { |
11fdf7f2 TL |
11227 | cmd_vec.push_back(val); |
11228 | } | |
f67539c2 | 11229 | |
7c673cae | 11230 | ceph_heap_profiler_handle_command(cmd_vec, os); |
f67539c2 | 11231 | |
7c673cae FG |
11232 | return 0; |
11233 | } | |
f67539c2 TL |
11234 | |
11235 | } // namespace ceph::osd_cmds |