]>
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> |
7c673cae | 27 | #include <boost/scoped_ptr.hpp> |
eafe8130 | 28 | #include <boost/range/adaptor/reversed.hpp> |
7c673cae FG |
29 | |
30 | #ifdef HAVE_SYS_PARAM_H | |
31 | #include <sys/param.h> | |
32 | #endif | |
33 | ||
34 | #ifdef HAVE_SYS_MOUNT_H | |
35 | #include <sys/mount.h> | |
36 | #endif | |
37 | ||
38 | #include "osd/PG.h" | |
39 | ||
40 | #include "include/types.h" | |
41 | #include "include/compat.h" | |
11fdf7f2 | 42 | #include "include/random.h" |
7c673cae FG |
43 | |
44 | #include "OSD.h" | |
45 | #include "OSDMap.h" | |
46 | #include "Watch.h" | |
47 | #include "osdc/Objecter.h" | |
48 | ||
49 | #include "common/errno.h" | |
50 | #include "common/ceph_argparse.h" | |
9f95a23c | 51 | #include "common/ceph_releases.h" |
224ce89b | 52 | #include "common/ceph_time.h" |
7c673cae | 53 | #include "common/version.h" |
b5b8bbf5 | 54 | #include "common/pick_address.h" |
11fdf7f2 TL |
55 | #include "common/blkdev.h" |
56 | #include "common/numa.h" | |
7c673cae FG |
57 | |
58 | #include "os/ObjectStore.h" | |
59 | #ifdef HAVE_LIBFUSE | |
60 | #include "os/FuseStore.h" | |
61 | #endif | |
62 | ||
63 | #include "PrimaryLogPG.h" | |
64 | ||
7c673cae FG |
65 | #include "msg/Messenger.h" |
66 | #include "msg/Message.h" | |
67 | ||
68 | #include "mon/MonClient.h" | |
69 | ||
70 | #include "messages/MLog.h" | |
71 | ||
72 | #include "messages/MGenericMessage.h" | |
7c673cae FG |
73 | #include "messages/MOSDPing.h" |
74 | #include "messages/MOSDFailure.h" | |
75 | #include "messages/MOSDMarkMeDown.h" | |
9f95a23c | 76 | #include "messages/MOSDMarkMeDead.h" |
7c673cae FG |
77 | #include "messages/MOSDFull.h" |
78 | #include "messages/MOSDOp.h" | |
79 | #include "messages/MOSDOpReply.h" | |
80 | #include "messages/MOSDBackoff.h" | |
81 | #include "messages/MOSDBeacon.h" | |
82 | #include "messages/MOSDRepOp.h" | |
83 | #include "messages/MOSDRepOpReply.h" | |
84 | #include "messages/MOSDBoot.h" | |
85 | #include "messages/MOSDPGTemp.h" | |
11fdf7f2 | 86 | #include "messages/MOSDPGReadyToMerge.h" |
7c673cae FG |
87 | |
88 | #include "messages/MOSDMap.h" | |
89 | #include "messages/MMonGetOSDMap.h" | |
90 | #include "messages/MOSDPGNotify.h" | |
9f95a23c | 91 | #include "messages/MOSDPGNotify2.h" |
7c673cae | 92 | #include "messages/MOSDPGQuery.h" |
9f95a23c | 93 | #include "messages/MOSDPGQuery2.h" |
7c673cae FG |
94 | #include "messages/MOSDPGLog.h" |
95 | #include "messages/MOSDPGRemove.h" | |
96 | #include "messages/MOSDPGInfo.h" | |
9f95a23c | 97 | #include "messages/MOSDPGInfo2.h" |
7c673cae | 98 | #include "messages/MOSDPGCreate.h" |
11fdf7f2 | 99 | #include "messages/MOSDPGCreate2.h" |
7c673cae | 100 | #include "messages/MOSDPGScan.h" |
7c673cae FG |
101 | #include "messages/MBackfillReserve.h" |
102 | #include "messages/MRecoveryReserve.h" | |
c07f9fc5 | 103 | #include "messages/MOSDForceRecovery.h" |
7c673cae FG |
104 | #include "messages/MOSDECSubOpWrite.h" |
105 | #include "messages/MOSDECSubOpWriteReply.h" | |
106 | #include "messages/MOSDECSubOpRead.h" | |
107 | #include "messages/MOSDECSubOpReadReply.h" | |
108 | #include "messages/MOSDPGCreated.h" | |
109 | #include "messages/MOSDPGUpdateLogMissing.h" | |
110 | #include "messages/MOSDPGUpdateLogMissingReply.h" | |
111 | ||
11fdf7f2 TL |
112 | #include "messages/MOSDPeeringOp.h" |
113 | ||
7c673cae FG |
114 | #include "messages/MOSDAlive.h" |
115 | ||
116 | #include "messages/MOSDScrub.h" | |
11fdf7f2 | 117 | #include "messages/MOSDScrub2.h" |
7c673cae FG |
118 | #include "messages/MOSDRepScrub.h" |
119 | ||
7c673cae FG |
120 | #include "messages/MCommand.h" |
121 | #include "messages/MCommandReply.h" | |
122 | ||
123 | #include "messages/MPGStats.h" | |
124 | #include "messages/MPGStatsAck.h" | |
125 | ||
126 | #include "messages/MWatchNotify.h" | |
127 | #include "messages/MOSDPGPush.h" | |
128 | #include "messages/MOSDPGPushReply.h" | |
129 | #include "messages/MOSDPGPull.h" | |
130 | ||
9f95a23c TL |
131 | #include "messages/MMonGetPurgedSnaps.h" |
132 | #include "messages/MMonGetPurgedSnapsReply.h" | |
133 | ||
7c673cae FG |
134 | #include "common/perf_counters.h" |
135 | #include "common/Timer.h" | |
136 | #include "common/LogClient.h" | |
137 | #include "common/AsyncReserver.h" | |
138 | #include "common/HeartbeatMap.h" | |
139 | #include "common/admin_socket.h" | |
140 | #include "common/ceph_context.h" | |
141 | ||
142 | #include "global/signal_handler.h" | |
143 | #include "global/pidfile.h" | |
144 | ||
145 | #include "include/color.h" | |
146 | #include "perfglue/cpu_profiler.h" | |
147 | #include "perfglue/heap_profiler.h" | |
148 | ||
149 | #include "osd/OpRequest.h" | |
150 | ||
151 | #include "auth/AuthAuthorizeHandler.h" | |
152 | #include "auth/RotatingKeyRing.h" | |
7c673cae FG |
153 | |
154 | #include "objclass/objclass.h" | |
155 | ||
156 | #include "common/cmdparse.h" | |
157 | #include "include/str_list.h" | |
158 | #include "include/util.h" | |
159 | ||
11fdf7f2 | 160 | #include "include/ceph_assert.h" |
7c673cae FG |
161 | #include "common/config.h" |
162 | #include "common/EventTrace.h" | |
163 | ||
11fdf7f2 TL |
164 | #include "json_spirit/json_spirit_reader.h" |
165 | #include "json_spirit/json_spirit_writer.h" | |
166 | ||
7c673cae FG |
167 | #ifdef WITH_LTTNG |
168 | #define TRACEPOINT_DEFINE | |
169 | #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
170 | #include "tracing/osd.h" | |
171 | #undef TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
172 | #undef TRACEPOINT_DEFINE | |
173 | #else | |
174 | #define tracepoint(...) | |
175 | #endif | |
176 | ||
177 | #define dout_context cct | |
178 | #define dout_subsys ceph_subsys_osd | |
179 | #undef dout_prefix | |
180 | #define dout_prefix _prefix(_dout, whoami, get_osdmap_epoch()) | |
181 | ||
9f95a23c TL |
182 | using namespace ceph::osd::scheduler; |
183 | using TOPNSPC::common::cmd_getval; | |
224ce89b | 184 | |
7c673cae FG |
185 | static ostream& _prefix(std::ostream* _dout, int whoami, epoch_t epoch) { |
186 | return *_dout << "osd." << whoami << " " << epoch << " "; | |
187 | } | |
188 | ||
7c673cae FG |
189 | //Initial features in new superblock. |
190 | //Features here are also automatically upgraded | |
191 | CompatSet OSD::get_osd_initial_compat_set() { | |
192 | CompatSet::FeatureSet ceph_osd_feature_compat; | |
193 | CompatSet::FeatureSet ceph_osd_feature_ro_compat; | |
194 | CompatSet::FeatureSet ceph_osd_feature_incompat; | |
195 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_BASE); | |
196 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_PGINFO); | |
197 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_OLOC); | |
198 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_LEC); | |
199 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_CATEGORIES); | |
200 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_HOBJECTPOOL); | |
201 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_BIGINFO); | |
202 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_LEVELDBINFO); | |
203 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_LEVELDBLOG); | |
204 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_SNAPMAPPER); | |
205 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_HINTS); | |
206 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_PGMETA); | |
207 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_MISSING); | |
208 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_FASTINFO); | |
c07f9fc5 | 209 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_RECOVERY_DELETES); |
9f95a23c | 210 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_SNAPMAPPER2); |
7c673cae FG |
211 | return CompatSet(ceph_osd_feature_compat, ceph_osd_feature_ro_compat, |
212 | ceph_osd_feature_incompat); | |
213 | } | |
214 | ||
215 | //Features are added here that this OSD supports. | |
216 | CompatSet OSD::get_osd_compat_set() { | |
217 | CompatSet compat = get_osd_initial_compat_set(); | |
218 | //Any features here can be set in code, but not in initial superblock | |
219 | compat.incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_SHARDS); | |
220 | return compat; | |
221 | } | |
222 | ||
223 | OSDService::OSDService(OSD *osd) : | |
224 | osd(osd), | |
225 | cct(osd->cct), | |
7c673cae FG |
226 | whoami(osd->whoami), store(osd->store), |
227 | log_client(osd->log_client), clog(osd->clog), | |
228 | pg_recovery_stats(osd->pg_recovery_stats), | |
229 | cluster_messenger(osd->cluster_messenger), | |
230 | client_messenger(osd->client_messenger), | |
231 | logger(osd->logger), | |
232 | recoverystate_perf(osd->recoverystate_perf), | |
233 | monc(osd->monc), | |
11fdf7f2 TL |
234 | osd_max_object_size(cct->_conf, "osd_max_object_size"), |
235 | osd_skip_data_digest(cct->_conf, "osd_skip_data_digest"), | |
236 | publish_lock{ceph::make_mutex("OSDService::publish_lock")}, | |
237 | pre_publish_lock{ceph::make_mutex("OSDService::pre_publish_lock")}, | |
7c673cae | 238 | max_oldest_map(0), |
eafe8130 TL |
239 | scrubs_local(0), |
240 | scrubs_remote(0), | |
7c673cae FG |
241 | agent_valid_iterator(false), |
242 | agent_ops(0), | |
243 | flush_mode_high_count(0), | |
244 | agent_active(true), | |
245 | agent_thread(this), | |
246 | agent_stop_flag(false), | |
7c673cae FG |
247 | agent_timer(osd->client_messenger->cct, agent_timer_lock), |
248 | last_recalibrate(ceph_clock_now()), | |
249 | promote_max_objects(0), | |
250 | promote_max_bytes(0), | |
9f95a23c TL |
251 | objecter(make_unique<Objecter>(osd->client_messenger->cct, |
252 | osd->objecter_messenger, | |
253 | osd->monc, nullptr, 0, 0)), | |
11fdf7f2 | 254 | m_objecter_finishers(cct->_conf->osd_objecter_finishers), |
7c673cae FG |
255 | watch_timer(osd->client_messenger->cct, watch_lock), |
256 | next_notif_id(0), | |
7c673cae | 257 | recovery_request_timer(cct, recovery_request_lock, false), |
11fdf7f2 | 258 | sleep_timer(cct, sleep_lock, false), |
7c673cae | 259 | reserver_finisher(cct), |
3efd9988 | 260 | local_reserver(cct, &reserver_finisher, cct->_conf->osd_max_backfills, |
7c673cae | 261 | cct->_conf->osd_min_recovery_priority), |
3efd9988 | 262 | remote_reserver(cct, &reserver_finisher, cct->_conf->osd_max_backfills, |
7c673cae | 263 | cct->_conf->osd_min_recovery_priority), |
3efd9988 | 264 | snap_reserver(cct, &reserver_finisher, |
7c673cae | 265 | cct->_conf->osd_max_trimming_pgs), |
7c673cae FG |
266 | recovery_ops_active(0), |
267 | recovery_ops_reserved(0), | |
268 | recovery_paused(false), | |
7c673cae FG |
269 | map_cache(cct, cct->_conf->osd_map_cache_size), |
270 | map_bl_cache(cct->_conf->osd_map_cache_size), | |
271 | map_bl_inc_cache(cct->_conf->osd_map_cache_size), | |
7c673cae | 272 | cur_state(NONE), |
11fdf7f2 | 273 | cur_ratio(0), physical_ratio(0), |
9f95a23c | 274 | boot_epoch(0), up_epoch(0), bind_epoch(0) |
7c673cae FG |
275 | { |
276 | objecter->init(); | |
11fdf7f2 TL |
277 | |
278 | for (int i = 0; i < m_objecter_finishers; i++) { | |
279 | ostringstream str; | |
280 | str << "objecter-finisher-" << i; | |
9f95a23c TL |
281 | auto fin = make_unique<Finisher>(osd->client_messenger->cct, str.str(), "finisher"); |
282 | objecter_finishers.push_back(std::move(fin)); | |
11fdf7f2 | 283 | } |
7c673cae FG |
284 | } |
285 | ||
31f18b77 FG |
286 | #ifdef PG_DEBUG_REFS |
287 | void OSDService::add_pgid(spg_t pgid, PG *pg){ | |
11fdf7f2 | 288 | std::lock_guard l(pgid_lock); |
31f18b77 FG |
289 | if (!pgid_tracker.count(pgid)) { |
290 | live_pgs[pgid] = pg; | |
291 | } | |
292 | pgid_tracker[pgid]++; | |
293 | } | |
294 | void OSDService::remove_pgid(spg_t pgid, PG *pg) | |
295 | { | |
11fdf7f2 TL |
296 | std::lock_guard l(pgid_lock); |
297 | ceph_assert(pgid_tracker.count(pgid)); | |
298 | ceph_assert(pgid_tracker[pgid] > 0); | |
31f18b77 FG |
299 | pgid_tracker[pgid]--; |
300 | if (pgid_tracker[pgid] == 0) { | |
301 | pgid_tracker.erase(pgid); | |
302 | live_pgs.erase(pgid); | |
303 | } | |
304 | } | |
305 | void OSDService::dump_live_pgids() | |
306 | { | |
11fdf7f2 | 307 | std::lock_guard l(pgid_lock); |
31f18b77 FG |
308 | derr << "live pgids:" << dendl; |
309 | for (map<spg_t, int>::const_iterator i = pgid_tracker.cbegin(); | |
310 | i != pgid_tracker.cend(); | |
311 | ++i) { | |
312 | derr << "\t" << *i << dendl; | |
313 | live_pgs[i->first]->dump_live_ids(); | |
314 | } | |
315 | } | |
316 | #endif | |
317 | ||
318 | ||
9f95a23c TL |
319 | ceph::signedspan OSDService::get_mnow() |
320 | { | |
321 | return ceph::mono_clock::now() - osd->startup_time; | |
322 | } | |
7c673cae | 323 | |
11fdf7f2 TL |
324 | void OSDService::identify_splits_and_merges( |
325 | OSDMapRef old_map, | |
326 | OSDMapRef new_map, | |
327 | spg_t pgid, | |
328 | set<pair<spg_t,epoch_t>> *split_children, | |
329 | set<pair<spg_t,epoch_t>> *merge_pgs) | |
7c673cae | 330 | { |
11fdf7f2 | 331 | if (!old_map->have_pg_pool(pgid.pool())) { |
7c673cae | 332 | return; |
7c673cae | 333 | } |
7c673cae | 334 | int old_pgnum = old_map->get_pg_num(pgid.pool()); |
11fdf7f2 TL |
335 | auto p = osd->pg_num_history.pg_nums.find(pgid.pool()); |
336 | if (p == osd->pg_num_history.pg_nums.end()) { | |
337 | return; | |
338 | } | |
339 | dout(20) << __func__ << " " << pgid << " e" << old_map->get_epoch() | |
340 | << " to e" << new_map->get_epoch() | |
341 | << " pg_nums " << p->second << dendl; | |
342 | deque<spg_t> queue; | |
343 | queue.push_back(pgid); | |
eafe8130 | 344 | set<spg_t> did; |
11fdf7f2 TL |
345 | while (!queue.empty()) { |
346 | auto cur = queue.front(); | |
347 | queue.pop_front(); | |
eafe8130 | 348 | did.insert(cur); |
11fdf7f2 TL |
349 | unsigned pgnum = old_pgnum; |
350 | for (auto q = p->second.lower_bound(old_map->get_epoch()); | |
351 | q != p->second.end() && | |
352 | q->first <= new_map->get_epoch(); | |
353 | ++q) { | |
354 | if (pgnum < q->second) { | |
355 | // split? | |
356 | if (cur.ps() < pgnum) { | |
357 | set<spg_t> children; | |
358 | if (cur.is_split(pgnum, q->second, &children)) { | |
359 | dout(20) << __func__ << " " << cur << " e" << q->first | |
360 | << " pg_num " << pgnum << " -> " << q->second | |
361 | << " children " << children << dendl; | |
362 | for (auto i : children) { | |
363 | split_children->insert(make_pair(i, q->first)); | |
eafe8130 TL |
364 | if (!did.count(i)) |
365 | queue.push_back(i); | |
11fdf7f2 TL |
366 | } |
367 | } | |
368 | } else if (cur.ps() < q->second) { | |
369 | dout(20) << __func__ << " " << cur << " e" << q->first | |
370 | << " pg_num " << pgnum << " -> " << q->second | |
371 | << " is a child" << dendl; | |
372 | // normally we'd capture this from the parent, but it's | |
373 | // possible the parent doesn't exist yet (it will be | |
374 | // fabricated to allow an intervening merge). note this PG | |
375 | // as a split child here to be sure we catch it. | |
376 | split_children->insert(make_pair(cur, q->first)); | |
377 | } else { | |
378 | dout(20) << __func__ << " " << cur << " e" << q->first | |
379 | << " pg_num " << pgnum << " -> " << q->second | |
380 | << " is post-split, skipping" << dendl; | |
381 | } | |
382 | } else if (merge_pgs) { | |
383 | // merge? | |
384 | if (cur.ps() >= q->second) { | |
385 | if (cur.ps() < pgnum) { | |
386 | spg_t parent; | |
387 | if (cur.is_merge_source(pgnum, q->second, &parent)) { | |
388 | set<spg_t> children; | |
389 | parent.is_split(q->second, pgnum, &children); | |
390 | dout(20) << __func__ << " " << cur << " e" << q->first | |
391 | << " pg_num " << pgnum << " -> " << q->second | |
392 | << " is merge source, target " << parent | |
393 | << ", source(s) " << children << dendl; | |
394 | merge_pgs->insert(make_pair(parent, q->first)); | |
eafe8130 TL |
395 | if (!did.count(parent)) { |
396 | // queue (and re-scan) parent in case it might not exist yet | |
397 | // and there are some future splits pending on it | |
398 | queue.push_back(parent); | |
399 | } | |
11fdf7f2 TL |
400 | for (auto c : children) { |
401 | merge_pgs->insert(make_pair(c, q->first)); | |
eafe8130 TL |
402 | if (!did.count(c)) |
403 | queue.push_back(c); | |
11fdf7f2 TL |
404 | } |
405 | } | |
406 | } else { | |
407 | dout(20) << __func__ << " " << cur << " e" << q->first | |
408 | << " pg_num " << pgnum << " -> " << q->second | |
409 | << " is beyond old pgnum, skipping" << dendl; | |
410 | } | |
411 | } else { | |
412 | set<spg_t> children; | |
413 | if (cur.is_split(q->second, pgnum, &children)) { | |
414 | dout(20) << __func__ << " " << cur << " e" << q->first | |
415 | << " pg_num " << pgnum << " -> " << q->second | |
416 | << " is merge target, source " << children << dendl; | |
417 | for (auto c : children) { | |
418 | merge_pgs->insert(make_pair(c, q->first)); | |
eafe8130 TL |
419 | if (!did.count(c)) |
420 | queue.push_back(c); | |
11fdf7f2 TL |
421 | } |
422 | merge_pgs->insert(make_pair(cur, q->first)); | |
423 | } | |
7c673cae FG |
424 | } |
425 | } | |
11fdf7f2 | 426 | pgnum = q->second; |
7c673cae FG |
427 | } |
428 | } | |
429 | } | |
430 | ||
7c673cae FG |
431 | void OSDService::need_heartbeat_peer_update() |
432 | { | |
433 | osd->need_heartbeat_peer_update(); | |
434 | } | |
435 | ||
9f95a23c TL |
436 | HeartbeatStampsRef OSDService::get_hb_stamps(unsigned peer) |
437 | { | |
438 | std::lock_guard l(hb_stamp_lock); | |
439 | if (peer >= hb_stamps.size()) { | |
440 | hb_stamps.resize(peer + 1); | |
441 | } | |
442 | if (!hb_stamps[peer]) { | |
443 | hb_stamps[peer] = ceph::make_ref<HeartbeatStamps>(peer); | |
444 | } | |
445 | return hb_stamps[peer]; | |
446 | } | |
447 | ||
448 | void OSDService::queue_renew_lease(epoch_t epoch, spg_t spgid) | |
449 | { | |
450 | osd->enqueue_peering_evt( | |
451 | spgid, | |
452 | PGPeeringEventRef( | |
453 | std::make_shared<PGPeeringEvent>( | |
454 | epoch, epoch, | |
455 | RenewLease()))); | |
456 | } | |
457 | ||
7c673cae FG |
458 | void OSDService::start_shutdown() |
459 | { | |
460 | { | |
11fdf7f2 | 461 | std::lock_guard l(agent_timer_lock); |
7c673cae FG |
462 | agent_timer.shutdown(); |
463 | } | |
31f18b77 FG |
464 | |
465 | { | |
11fdf7f2 TL |
466 | std::lock_guard l(sleep_lock); |
467 | sleep_timer.shutdown(); | |
31f18b77 | 468 | } |
81eedcae TL |
469 | |
470 | { | |
471 | std::lock_guard l(recovery_request_lock); | |
472 | recovery_request_timer.shutdown(); | |
473 | } | |
7c673cae FG |
474 | } |
475 | ||
31f18b77 | 476 | void OSDService::shutdown_reserver() |
7c673cae FG |
477 | { |
478 | reserver_finisher.wait_for_empty(); | |
479 | reserver_finisher.stop(); | |
31f18b77 FG |
480 | } |
481 | ||
482 | void OSDService::shutdown() | |
483 | { | |
9f95a23c TL |
484 | mono_timer.suspend(); |
485 | ||
7c673cae | 486 | { |
11fdf7f2 | 487 | std::lock_guard l(watch_lock); |
7c673cae FG |
488 | watch_timer.shutdown(); |
489 | } | |
490 | ||
491 | objecter->shutdown(); | |
9f95a23c | 492 | for (auto& f : objecter_finishers) { |
11fdf7f2 TL |
493 | f->wait_for_empty(); |
494 | f->stop(); | |
7c673cae FG |
495 | } |
496 | ||
11fdf7f2 | 497 | publish_map(OSDMapRef()); |
7c673cae FG |
498 | next_osdmap = OSDMapRef(); |
499 | } | |
500 | ||
501 | void OSDService::init() | |
502 | { | |
503 | reserver_finisher.start(); | |
9f95a23c | 504 | for (auto& f : objecter_finishers) { |
11fdf7f2 TL |
505 | f->start(); |
506 | } | |
7c673cae FG |
507 | objecter->set_client_incarnation(0); |
508 | ||
509 | // deprioritize objecter in daemonperf output | |
510 | objecter->get_logger()->set_prio_adjust(-3); | |
511 | ||
512 | watch_timer.init(); | |
513 | agent_timer.init(); | |
9f95a23c | 514 | mono_timer.resume(); |
7c673cae FG |
515 | |
516 | agent_thread.create("osd_srv_agent"); | |
517 | ||
518 | if (cct->_conf->osd_recovery_delay_start) | |
519 | defer_recovery(cct->_conf->osd_recovery_delay_start); | |
520 | } | |
521 | ||
522 | void OSDService::final_init() | |
523 | { | |
524 | objecter->start(osdmap.get()); | |
525 | } | |
526 | ||
527 | void OSDService::activate_map() | |
528 | { | |
529 | // wake/unwake the tiering agent | |
9f95a23c | 530 | std::lock_guard l{agent_lock}; |
7c673cae FG |
531 | agent_active = |
532 | !osdmap->test_flag(CEPH_OSDMAP_NOTIERAGENT) && | |
533 | osd->is_active(); | |
9f95a23c | 534 | agent_cond.notify_all(); |
7c673cae FG |
535 | } |
536 | ||
181888fb FG |
537 | void OSDService::request_osdmap_update(epoch_t e) |
538 | { | |
539 | osd->osdmap_subscribe(e, false); | |
540 | } | |
541 | ||
9f95a23c | 542 | |
7c673cae FG |
543 | class AgentTimeoutCB : public Context { |
544 | PGRef pg; | |
545 | public: | |
546 | explicit AgentTimeoutCB(PGRef _pg) : pg(_pg) {} | |
547 | void finish(int) override { | |
548 | pg->agent_choose_mode_restart(); | |
549 | } | |
550 | }; | |
551 | ||
552 | void OSDService::agent_entry() | |
553 | { | |
554 | dout(10) << __func__ << " start" << dendl; | |
9f95a23c | 555 | std::unique_lock agent_locker{agent_lock}; |
7c673cae FG |
556 | |
557 | while (!agent_stop_flag) { | |
558 | if (agent_queue.empty()) { | |
559 | dout(20) << __func__ << " empty queue" << dendl; | |
9f95a23c | 560 | agent_cond.wait(agent_locker); |
7c673cae FG |
561 | continue; |
562 | } | |
563 | uint64_t level = agent_queue.rbegin()->first; | |
564 | set<PGRef>& top = agent_queue.rbegin()->second; | |
565 | dout(10) << __func__ | |
566 | << " tiers " << agent_queue.size() | |
567 | << ", top is " << level | |
568 | << " with pgs " << top.size() | |
569 | << ", ops " << agent_ops << "/" | |
570 | << cct->_conf->osd_agent_max_ops | |
571 | << (agent_active ? " active" : " NOT ACTIVE") | |
572 | << dendl; | |
573 | dout(20) << __func__ << " oids " << agent_oids << dendl; | |
574 | int max = cct->_conf->osd_agent_max_ops - agent_ops; | |
575 | int agent_flush_quota = max; | |
576 | if (!flush_mode_high_count) | |
577 | agent_flush_quota = cct->_conf->osd_agent_max_low_ops - agent_ops; | |
578 | if (agent_flush_quota <= 0 || top.empty() || !agent_active) { | |
9f95a23c | 579 | agent_cond.wait(agent_locker); |
7c673cae FG |
580 | continue; |
581 | } | |
582 | ||
583 | if (!agent_valid_iterator || agent_queue_pos == top.end()) { | |
584 | agent_queue_pos = top.begin(); | |
585 | agent_valid_iterator = true; | |
586 | } | |
587 | PGRef pg = *agent_queue_pos; | |
588 | dout(10) << "high_count " << flush_mode_high_count | |
589 | << " agent_ops " << agent_ops | |
590 | << " flush_quota " << agent_flush_quota << dendl; | |
9f95a23c | 591 | agent_locker.unlock(); |
7c673cae | 592 | if (!pg->agent_work(max, agent_flush_quota)) { |
11fdf7f2 | 593 | dout(10) << __func__ << " " << pg->pg_id |
7c673cae FG |
594 | << " no agent_work, delay for " << cct->_conf->osd_agent_delay_time |
595 | << " seconds" << dendl; | |
596 | ||
597 | osd->logger->inc(l_osd_tier_delay); | |
598 | // Queue a timer to call agent_choose_mode for this pg in 5 seconds | |
9f95a23c | 599 | std::lock_guard timer_locker{agent_timer_lock}; |
7c673cae FG |
600 | Context *cb = new AgentTimeoutCB(pg); |
601 | agent_timer.add_event_after(cct->_conf->osd_agent_delay_time, cb); | |
7c673cae | 602 | } |
9f95a23c | 603 | agent_locker.lock(); |
7c673cae | 604 | } |
7c673cae FG |
605 | dout(10) << __func__ << " finish" << dendl; |
606 | } | |
607 | ||
608 | void OSDService::agent_stop() | |
609 | { | |
610 | { | |
11fdf7f2 | 611 | std::lock_guard l(agent_lock); |
7c673cae FG |
612 | |
613 | // By this time all ops should be cancelled | |
11fdf7f2 | 614 | ceph_assert(agent_ops == 0); |
7c673cae FG |
615 | // By this time all PGs are shutdown and dequeued |
616 | if (!agent_queue.empty()) { | |
617 | set<PGRef>& top = agent_queue.rbegin()->second; | |
11fdf7f2 TL |
618 | derr << "agent queue not empty, for example " << (*top.begin())->get_pgid() << dendl; |
619 | ceph_abort_msg("agent queue not empty"); | |
7c673cae FG |
620 | } |
621 | ||
622 | agent_stop_flag = true; | |
9f95a23c | 623 | agent_cond.notify_all(); |
7c673cae FG |
624 | } |
625 | agent_thread.join(); | |
626 | } | |
627 | ||
628 | // ------------------------------------- | |
629 | ||
630 | void OSDService::promote_throttle_recalibrate() | |
631 | { | |
632 | utime_t now = ceph_clock_now(); | |
633 | double dur = now - last_recalibrate; | |
634 | last_recalibrate = now; | |
635 | unsigned prob = promote_probability_millis; | |
636 | ||
637 | uint64_t target_obj_sec = cct->_conf->osd_tier_promote_max_objects_sec; | |
638 | uint64_t target_bytes_sec = cct->_conf->osd_tier_promote_max_bytes_sec; | |
639 | ||
640 | unsigned min_prob = 1; | |
641 | ||
642 | uint64_t attempts, obj, bytes; | |
643 | promote_counter.sample_and_attenuate(&attempts, &obj, &bytes); | |
644 | dout(10) << __func__ << " " << attempts << " attempts, promoted " | |
1adf2230 | 645 | << obj << " objects and " << byte_u_t(bytes) << "; target " |
7c673cae | 646 | << target_obj_sec << " obj/sec or " |
1adf2230 | 647 | << byte_u_t(target_bytes_sec) << "/sec" |
7c673cae FG |
648 | << dendl; |
649 | ||
650 | // calculate what the probability *should* be, given the targets | |
651 | unsigned new_prob; | |
652 | if (attempts && dur > 0) { | |
653 | uint64_t avg_size = 1; | |
654 | if (obj) | |
11fdf7f2 | 655 | avg_size = std::max<uint64_t>(bytes / obj, 1); |
7c673cae FG |
656 | unsigned po = (double)target_obj_sec * dur * 1000.0 / (double)attempts; |
657 | unsigned pb = (double)target_bytes_sec / (double)avg_size * dur * 1000.0 | |
658 | / (double)attempts; | |
659 | dout(20) << __func__ << " po " << po << " pb " << pb << " avg_size " | |
660 | << avg_size << dendl; | |
661 | if (target_obj_sec && target_bytes_sec) | |
11fdf7f2 | 662 | new_prob = std::min(po, pb); |
7c673cae FG |
663 | else if (target_obj_sec) |
664 | new_prob = po; | |
665 | else if (target_bytes_sec) | |
666 | new_prob = pb; | |
667 | else | |
668 | new_prob = 1000; | |
669 | } else { | |
670 | new_prob = 1000; | |
671 | } | |
672 | dout(20) << __func__ << " new_prob " << new_prob << dendl; | |
673 | ||
674 | // correct for persistent skew between target rate and actual rate, adjust | |
675 | double ratio = 1.0; | |
676 | unsigned actual = 0; | |
677 | if (attempts && obj) { | |
678 | actual = obj * 1000 / attempts; | |
679 | ratio = (double)actual / (double)prob; | |
680 | new_prob = (double)new_prob / ratio; | |
681 | } | |
11fdf7f2 TL |
682 | new_prob = std::max(new_prob, min_prob); |
683 | new_prob = std::min(new_prob, 1000u); | |
7c673cae FG |
684 | |
685 | // adjust | |
686 | prob = (prob + new_prob) / 2; | |
11fdf7f2 TL |
687 | prob = std::max(prob, min_prob); |
688 | prob = std::min(prob, 1000u); | |
7c673cae FG |
689 | dout(10) << __func__ << " actual " << actual |
690 | << ", actual/prob ratio " << ratio | |
691 | << ", adjusted new_prob " << new_prob | |
692 | << ", prob " << promote_probability_millis << " -> " << prob | |
693 | << dendl; | |
694 | promote_probability_millis = prob; | |
695 | ||
696 | // set hard limits for this interval to mitigate stampedes | |
91327a77 AA |
697 | promote_max_objects = target_obj_sec * osd->OSD_TICK_INTERVAL * 2; |
698 | promote_max_bytes = target_bytes_sec * osd->OSD_TICK_INTERVAL * 2; | |
7c673cae FG |
699 | } |
700 | ||
701 | // ------------------------------------- | |
702 | ||
703 | float OSDService::get_failsafe_full_ratio() | |
704 | { | |
705 | float full_ratio = cct->_conf->osd_failsafe_full_ratio; | |
706 | if (full_ratio > 1.0) full_ratio /= 100.0; | |
707 | return full_ratio; | |
708 | } | |
709 | ||
11fdf7f2 | 710 | OSDService::s_names OSDService::recalc_full_state(float ratio, float pratio, string &inject) |
7c673cae | 711 | { |
7c673cae FG |
712 | // The OSDMap ratios take precendence. So if the failsafe is .95 and |
713 | // the admin sets the cluster full to .96, the failsafe moves up to .96 | |
714 | // too. (Not that having failsafe == full is ideal, but it's better than | |
715 | // dropping writes before the clusters appears full.) | |
716 | OSDMapRef osdmap = get_osdmap(); | |
717 | if (!osdmap || osdmap->get_epoch() == 0) { | |
11fdf7f2 | 718 | return NONE; |
7c673cae FG |
719 | } |
720 | float nearfull_ratio = osdmap->get_nearfull_ratio(); | |
721 | float backfillfull_ratio = std::max(osdmap->get_backfillfull_ratio(), nearfull_ratio); | |
722 | float full_ratio = std::max(osdmap->get_full_ratio(), backfillfull_ratio); | |
723 | float failsafe_ratio = std::max(get_failsafe_full_ratio(), full_ratio); | |
724 | ||
9f95a23c | 725 | if (osdmap->require_osd_release < ceph_release_t::luminous) { |
7c673cae FG |
726 | // use the failsafe for nearfull and full; the mon isn't using the |
727 | // flags anyway because we're mid-upgrade. | |
728 | full_ratio = failsafe_ratio; | |
729 | backfillfull_ratio = failsafe_ratio; | |
730 | nearfull_ratio = failsafe_ratio; | |
731 | } else if (full_ratio <= 0 || | |
732 | backfillfull_ratio <= 0 || | |
733 | nearfull_ratio <= 0) { | |
734 | derr << __func__ << " full_ratio, backfillfull_ratio or nearfull_ratio is <= 0" << dendl; | |
735 | // use failsafe flag. ick. the monitor did something wrong or the user | |
736 | // did something stupid. | |
737 | full_ratio = failsafe_ratio; | |
738 | backfillfull_ratio = failsafe_ratio; | |
739 | nearfull_ratio = failsafe_ratio; | |
740 | } | |
741 | ||
7c673cae | 742 | if (injectfull_state > NONE && injectfull) { |
7c673cae | 743 | inject = "(Injected)"; |
11fdf7f2 TL |
744 | return injectfull_state; |
745 | } else if (pratio > failsafe_ratio) { | |
746 | return FAILSAFE; | |
7c673cae | 747 | } else if (ratio > full_ratio) { |
11fdf7f2 | 748 | return FULL; |
7c673cae | 749 | } else if (ratio > backfillfull_ratio) { |
11fdf7f2 | 750 | return BACKFILLFULL; |
92f5a8d4 | 751 | } else if (pratio > nearfull_ratio) { |
11fdf7f2 | 752 | return NEARFULL; |
7c673cae | 753 | } |
11fdf7f2 TL |
754 | return NONE; |
755 | } | |
756 | ||
757 | void OSDService::check_full_status(float ratio, float pratio) | |
758 | { | |
759 | std::lock_guard l(full_status_lock); | |
760 | ||
761 | cur_ratio = ratio; | |
762 | physical_ratio = pratio; | |
763 | ||
764 | string inject; | |
765 | s_names new_state; | |
766 | new_state = recalc_full_state(ratio, pratio, inject); | |
767 | ||
7c673cae | 768 | dout(20) << __func__ << " cur ratio " << ratio |
11fdf7f2 | 769 | << ", physical ratio " << pratio |
7c673cae FG |
770 | << ", new state " << get_full_state_name(new_state) |
771 | << " " << inject | |
772 | << dendl; | |
773 | ||
774 | // warn | |
775 | if (cur_state != new_state) { | |
776 | dout(10) << __func__ << " " << get_full_state_name(cur_state) | |
777 | << " -> " << get_full_state_name(new_state) << dendl; | |
778 | if (new_state == FAILSAFE) { | |
c07f9fc5 | 779 | clog->error() << "full status failsafe engaged, dropping updates, now " |
7c673cae FG |
780 | << (int)roundf(ratio * 100) << "% full"; |
781 | } else if (cur_state == FAILSAFE) { | |
c07f9fc5 FG |
782 | clog->error() << "full status failsafe disengaged, no longer dropping " |
783 | << "updates, now " << (int)roundf(ratio * 100) << "% full"; | |
7c673cae FG |
784 | } |
785 | cur_state = new_state; | |
786 | } | |
787 | } | |
788 | ||
789 | bool OSDService::need_fullness_update() | |
790 | { | |
791 | OSDMapRef osdmap = get_osdmap(); | |
792 | s_names cur = NONE; | |
793 | if (osdmap->exists(whoami)) { | |
794 | if (osdmap->get_state(whoami) & CEPH_OSD_FULL) { | |
795 | cur = FULL; | |
796 | } else if (osdmap->get_state(whoami) & CEPH_OSD_BACKFILLFULL) { | |
797 | cur = BACKFILLFULL; | |
798 | } else if (osdmap->get_state(whoami) & CEPH_OSD_NEARFULL) { | |
799 | cur = NEARFULL; | |
800 | } | |
801 | } | |
802 | s_names want = NONE; | |
803 | if (is_full()) | |
804 | want = FULL; | |
805 | else if (is_backfillfull()) | |
806 | want = BACKFILLFULL; | |
807 | else if (is_nearfull()) | |
808 | want = NEARFULL; | |
809 | return want != cur; | |
810 | } | |
811 | ||
11fdf7f2 | 812 | bool OSDService::_check_inject_full(DoutPrefixProvider *dpp, s_names type) const |
7c673cae | 813 | { |
7c673cae FG |
814 | if (injectfull && injectfull_state >= type) { |
815 | // injectfull is either a count of the number of times to return failsafe full | |
816 | // or if -1 then always return full | |
817 | if (injectfull > 0) | |
818 | --injectfull; | |
11fdf7f2 TL |
819 | ldpp_dout(dpp, 10) << __func__ << " Injected " << get_full_state_name(type) << " OSD (" |
820 | << (injectfull < 0 ? "set" : std::to_string(injectfull)) << ")" | |
821 | << dendl; | |
7c673cae FG |
822 | return true; |
823 | } | |
11fdf7f2 TL |
824 | return false; |
825 | } | |
826 | ||
827 | bool OSDService::_check_full(DoutPrefixProvider *dpp, s_names type) const | |
828 | { | |
829 | std::lock_guard l(full_status_lock); | |
830 | ||
831 | if (_check_inject_full(dpp, type)) | |
832 | return true; | |
833 | ||
834 | if (cur_state >= type) | |
835 | ldpp_dout(dpp, 10) << __func__ << " current usage is " << cur_ratio | |
836 | << " physical " << physical_ratio << dendl; | |
7c673cae | 837 | |
7c673cae FG |
838 | return cur_state >= type; |
839 | } | |
840 | ||
11fdf7f2 TL |
841 | bool OSDService::_tentative_full(DoutPrefixProvider *dpp, s_names type, uint64_t adjust_used, osd_stat_t adjusted_stat) |
842 | { | |
843 | ldpp_dout(dpp, 20) << __func__ << " type " << get_full_state_name(type) << " adjust_used " << (adjust_used >> 10) << "KiB" << dendl; | |
844 | { | |
845 | std::lock_guard l(full_status_lock); | |
846 | if (_check_inject_full(dpp, type)) { | |
847 | return true; | |
848 | } | |
849 | } | |
850 | ||
851 | float pratio; | |
852 | float ratio = compute_adjusted_ratio(adjusted_stat, &pratio, adjust_used); | |
853 | ||
854 | string notused; | |
855 | s_names tentative_state = recalc_full_state(ratio, pratio, notused); | |
856 | ||
857 | if (tentative_state >= type) | |
858 | ldpp_dout(dpp, 10) << __func__ << " tentative usage is " << ratio << dendl; | |
859 | ||
860 | return tentative_state >= type; | |
861 | } | |
862 | ||
863 | bool OSDService::check_failsafe_full(DoutPrefixProvider *dpp) const | |
864 | { | |
865 | return _check_full(dpp, FAILSAFE); | |
866 | } | |
867 | ||
868 | bool OSDService::check_full(DoutPrefixProvider *dpp) const | |
7c673cae | 869 | { |
11fdf7f2 | 870 | return _check_full(dpp, FULL); |
7c673cae FG |
871 | } |
872 | ||
11fdf7f2 | 873 | bool OSDService::tentative_backfill_full(DoutPrefixProvider *dpp, uint64_t adjust_used, osd_stat_t stats) |
7c673cae | 874 | { |
11fdf7f2 | 875 | return _tentative_full(dpp, BACKFILLFULL, adjust_used, stats); |
7c673cae FG |
876 | } |
877 | ||
11fdf7f2 | 878 | bool OSDService::check_backfill_full(DoutPrefixProvider *dpp) const |
7c673cae | 879 | { |
11fdf7f2 | 880 | return _check_full(dpp, BACKFILLFULL); |
7c673cae FG |
881 | } |
882 | ||
11fdf7f2 | 883 | bool OSDService::check_nearfull(DoutPrefixProvider *dpp) const |
7c673cae | 884 | { |
11fdf7f2 | 885 | return _check_full(dpp, NEARFULL); |
7c673cae FG |
886 | } |
887 | ||
888 | bool OSDService::is_failsafe_full() const | |
889 | { | |
11fdf7f2 | 890 | std::lock_guard l(full_status_lock); |
7c673cae FG |
891 | return cur_state == FAILSAFE; |
892 | } | |
893 | ||
894 | bool OSDService::is_full() const | |
895 | { | |
11fdf7f2 | 896 | std::lock_guard l(full_status_lock); |
7c673cae FG |
897 | return cur_state >= FULL; |
898 | } | |
899 | ||
900 | bool OSDService::is_backfillfull() const | |
901 | { | |
11fdf7f2 | 902 | std::lock_guard l(full_status_lock); |
7c673cae FG |
903 | return cur_state >= BACKFILLFULL; |
904 | } | |
905 | ||
906 | bool OSDService::is_nearfull() const | |
907 | { | |
11fdf7f2 | 908 | std::lock_guard l(full_status_lock); |
7c673cae FG |
909 | return cur_state >= NEARFULL; |
910 | } | |
911 | ||
912 | void OSDService::set_injectfull(s_names type, int64_t count) | |
913 | { | |
11fdf7f2 | 914 | std::lock_guard l(full_status_lock); |
7c673cae FG |
915 | injectfull_state = type; |
916 | injectfull = count; | |
917 | } | |
918 | ||
11fdf7f2 TL |
919 | void OSDService::set_statfs(const struct store_statfs_t &stbuf, |
920 | osd_alert_list_t& alerts) | |
7c673cae | 921 | { |
224ce89b | 922 | uint64_t bytes = stbuf.total; |
224ce89b | 923 | uint64_t avail = stbuf.available; |
11fdf7f2 TL |
924 | uint64_t used = stbuf.get_used_raw(); |
925 | ||
926 | // For testing fake statfs values so it doesn't matter if all | |
927 | // OSDs are using the same partition. | |
928 | if (cct->_conf->fake_statfs_for_testing) { | |
929 | uint64_t total_num_bytes = 0; | |
930 | vector<PGRef> pgs; | |
931 | osd->_get_pgs(&pgs); | |
932 | for (auto p : pgs) { | |
933 | total_num_bytes += p->get_stats_num_bytes(); | |
934 | } | |
935 | bytes = cct->_conf->fake_statfs_for_testing; | |
936 | if (total_num_bytes < bytes) | |
937 | avail = bytes - total_num_bytes; | |
938 | else | |
939 | avail = 0; | |
940 | dout(0) << __func__ << " fake total " << cct->_conf->fake_statfs_for_testing | |
941 | << " adjust available " << avail | |
942 | << dendl; | |
943 | used = bytes - avail; | |
944 | } | |
7c673cae | 945 | |
224ce89b WB |
946 | osd->logger->set(l_osd_stat_bytes, bytes); |
947 | osd->logger->set(l_osd_stat_bytes_used, used); | |
948 | osd->logger->set(l_osd_stat_bytes_avail, avail); | |
7c673cae | 949 | |
11fdf7f2 TL |
950 | std::lock_guard l(stat_lock); |
951 | osd_stat.statfs = stbuf; | |
952 | osd_stat.os_alerts.clear(); | |
953 | osd_stat.os_alerts[whoami].swap(alerts); | |
954 | if (cct->_conf->fake_statfs_for_testing) { | |
955 | osd_stat.statfs.total = bytes; | |
956 | osd_stat.statfs.available = avail; | |
957 | // For testing don't want used to go negative, so clear reserved | |
958 | osd_stat.statfs.internally_reserved = 0; | |
224ce89b WB |
959 | } |
960 | } | |
7c673cae | 961 | |
11fdf7f2 TL |
962 | osd_stat_t OSDService::set_osd_stat(vector<int>& hb_peers, |
963 | int num_pgs) | |
224ce89b | 964 | { |
eafe8130 TL |
965 | utime_t now = ceph_clock_now(); |
966 | auto stale_time = g_conf().get_val<int64_t>("osd_mon_heartbeat_stat_stale"); | |
11fdf7f2 TL |
967 | std::lock_guard l(stat_lock); |
968 | osd_stat.hb_peers.swap(hb_peers); | |
969 | osd->op_tracker.get_age_ms_histogram(&osd_stat.op_queue_age_hist); | |
970 | osd_stat.num_pgs = num_pgs; | |
eafe8130 TL |
971 | // Clean entries that aren't updated |
972 | // This is called often enough that we can just remove 1 at a time | |
973 | for (auto i: osd_stat.hb_pingtime) { | |
974 | if (i.second.last_update == 0) | |
975 | continue; | |
976 | if (stale_time && now.sec() - i.second.last_update > stale_time) { | |
977 | dout(20) << __func__ << " time out heartbeat for osd " << i.first | |
978 | << " last_update " << i.second.last_update << dendl; | |
979 | osd_stat.hb_pingtime.erase(i.first); | |
980 | break; | |
981 | } | |
982 | } | |
11fdf7f2 TL |
983 | return osd_stat; |
984 | } | |
985 | ||
986 | void OSDService::inc_osd_stat_repaired() | |
987 | { | |
988 | std::lock_guard l(stat_lock); | |
989 | osd_stat.num_shards_repaired++; | |
990 | return; | |
991 | } | |
992 | ||
993 | float OSDService::compute_adjusted_ratio(osd_stat_t new_stat, float *pratio, | |
994 | uint64_t adjust_used) | |
995 | { | |
996 | *pratio = | |
997 | ((float)new_stat.statfs.get_used()) / ((float)new_stat.statfs.total); | |
998 | ||
999 | if (adjust_used) { | |
1000 | dout(20) << __func__ << " Before kb_used() " << new_stat.statfs.kb_used() << dendl; | |
1001 | if (new_stat.statfs.available > adjust_used) | |
1002 | new_stat.statfs.available -= adjust_used; | |
1003 | else | |
1004 | new_stat.statfs.available = 0; | |
1005 | dout(20) << __func__ << " After kb_used() " << new_stat.statfs.kb_used() << dendl; | |
7c673cae FG |
1006 | } |
1007 | ||
11fdf7f2 TL |
1008 | // Check all pgs and adjust kb_used to include all pending backfill data |
1009 | int backfill_adjusted = 0; | |
1010 | vector<PGRef> pgs; | |
1011 | osd->_get_pgs(&pgs); | |
1012 | for (auto p : pgs) { | |
1013 | backfill_adjusted += p->pg_stat_adjust(&new_stat); | |
1014 | } | |
1015 | if (backfill_adjusted) { | |
1016 | dout(20) << __func__ << " backfill adjusted " << new_stat << dendl; | |
1017 | } | |
1018 | return ((float)new_stat.statfs.get_used()) / ((float)new_stat.statfs.total); | |
7c673cae FG |
1019 | } |
1020 | ||
7c673cae FG |
1021 | void OSDService::send_message_osd_cluster(int peer, Message *m, epoch_t from_epoch) |
1022 | { | |
1023 | OSDMapRef next_map = get_nextmap_reserved(); | |
1024 | // service map is always newer/newest | |
11fdf7f2 | 1025 | ceph_assert(from_epoch <= next_map->get_epoch()); |
7c673cae FG |
1026 | |
1027 | if (next_map->is_down(peer) || | |
1028 | next_map->get_info(peer).up_from > from_epoch) { | |
1029 | m->put(); | |
1030 | release_map(next_map); | |
1031 | return; | |
1032 | } | |
9f95a23c TL |
1033 | ConnectionRef peer_con; |
1034 | if (peer == whoami) { | |
1035 | peer_con = osd->cluster_messenger->get_loopback_connection(); | |
1036 | } else { | |
1037 | peer_con = osd->cluster_messenger->connect_to_osd( | |
1038 | next_map->get_cluster_addrs(peer), false, true); | |
1039 | } | |
1040 | maybe_share_map(peer_con.get(), next_map); | |
7c673cae FG |
1041 | peer_con->send_message(m); |
1042 | release_map(next_map); | |
1043 | } | |
1044 | ||
9f95a23c TL |
1045 | void OSDService::send_message_osd_cluster(std::vector<std::pair<int, Message*>>& messages, epoch_t from_epoch) |
1046 | { | |
1047 | OSDMapRef next_map = get_nextmap_reserved(); | |
1048 | // service map is always newer/newest | |
1049 | ceph_assert(from_epoch <= next_map->get_epoch()); | |
1050 | ||
1051 | for (auto& iter : messages) { | |
1052 | if (next_map->is_down(iter.first) || | |
1053 | next_map->get_info(iter.first).up_from > from_epoch) { | |
1054 | iter.second->put(); | |
1055 | continue; | |
1056 | } | |
1057 | ConnectionRef peer_con; | |
1058 | if (iter.first == whoami) { | |
1059 | peer_con = osd->cluster_messenger->get_loopback_connection(); | |
1060 | } else { | |
1061 | peer_con = osd->cluster_messenger->connect_to_osd( | |
1062 | next_map->get_cluster_addrs(iter.first), false, true); | |
1063 | } | |
1064 | maybe_share_map(peer_con.get(), next_map); | |
1065 | peer_con->send_message(iter.second); | |
1066 | } | |
1067 | release_map(next_map); | |
1068 | } | |
7c673cae FG |
1069 | ConnectionRef OSDService::get_con_osd_cluster(int peer, epoch_t from_epoch) |
1070 | { | |
1071 | OSDMapRef next_map = get_nextmap_reserved(); | |
1072 | // service map is always newer/newest | |
11fdf7f2 | 1073 | ceph_assert(from_epoch <= next_map->get_epoch()); |
7c673cae FG |
1074 | |
1075 | if (next_map->is_down(peer) || | |
1076 | next_map->get_info(peer).up_from > from_epoch) { | |
1077 | release_map(next_map); | |
1078 | return NULL; | |
1079 | } | |
9f95a23c TL |
1080 | ConnectionRef con; |
1081 | if (peer == whoami) { | |
1082 | con = osd->cluster_messenger->get_loopback_connection(); | |
1083 | } else { | |
1084 | con = osd->cluster_messenger->connect_to_osd( | |
1085 | next_map->get_cluster_addrs(peer), false, true); | |
1086 | } | |
7c673cae FG |
1087 | release_map(next_map); |
1088 | return con; | |
1089 | } | |
1090 | ||
1091 | pair<ConnectionRef,ConnectionRef> OSDService::get_con_osd_hb(int peer, epoch_t from_epoch) | |
1092 | { | |
1093 | OSDMapRef next_map = get_nextmap_reserved(); | |
1094 | // service map is always newer/newest | |
11fdf7f2 | 1095 | ceph_assert(from_epoch <= next_map->get_epoch()); |
7c673cae FG |
1096 | |
1097 | pair<ConnectionRef,ConnectionRef> ret; | |
1098 | if (next_map->is_down(peer) || | |
1099 | next_map->get_info(peer).up_from > from_epoch) { | |
1100 | release_map(next_map); | |
1101 | return ret; | |
1102 | } | |
11fdf7f2 TL |
1103 | ret.first = osd->hb_back_client_messenger->connect_to_osd( |
1104 | next_map->get_hb_back_addrs(peer)); | |
1105 | ret.second = osd->hb_front_client_messenger->connect_to_osd( | |
1106 | next_map->get_hb_front_addrs(peer)); | |
7c673cae FG |
1107 | release_map(next_map); |
1108 | return ret; | |
1109 | } | |
1110 | ||
11fdf7f2 TL |
1111 | entity_name_t OSDService::get_cluster_msgr_name() const |
1112 | { | |
1113 | return cluster_messenger->get_myname(); | |
1114 | } | |
7c673cae | 1115 | |
94b18763 FG |
1116 | void OSDService::queue_want_pg_temp(pg_t pgid, |
1117 | const vector<int>& want, | |
1118 | bool forced) | |
7c673cae | 1119 | { |
11fdf7f2 | 1120 | std::lock_guard l(pg_temp_lock); |
94b18763 | 1121 | auto p = pg_temp_pending.find(pgid); |
7c673cae | 1122 | if (p == pg_temp_pending.end() || |
94b18763 FG |
1123 | p->second.acting != want || |
1124 | forced) { | |
11fdf7f2 | 1125 | pg_temp_wanted[pgid] = {want, forced}; |
7c673cae FG |
1126 | } |
1127 | } | |
1128 | ||
1129 | void OSDService::remove_want_pg_temp(pg_t pgid) | |
1130 | { | |
11fdf7f2 | 1131 | std::lock_guard l(pg_temp_lock); |
7c673cae FG |
1132 | pg_temp_wanted.erase(pgid); |
1133 | pg_temp_pending.erase(pgid); | |
1134 | } | |
1135 | ||
1136 | void OSDService::_sent_pg_temp() | |
1137 | { | |
11fdf7f2 TL |
1138 | #ifdef HAVE_STDLIB_MAP_SPLICING |
1139 | pg_temp_pending.merge(pg_temp_wanted); | |
1140 | #else | |
94b18763 FG |
1141 | pg_temp_pending.insert(make_move_iterator(begin(pg_temp_wanted)), |
1142 | make_move_iterator(end(pg_temp_wanted))); | |
11fdf7f2 | 1143 | #endif |
7c673cae FG |
1144 | pg_temp_wanted.clear(); |
1145 | } | |
1146 | ||
1147 | void OSDService::requeue_pg_temp() | |
1148 | { | |
11fdf7f2 | 1149 | std::lock_guard l(pg_temp_lock); |
7c673cae FG |
1150 | // wanted overrides pending. note that remove_want_pg_temp |
1151 | // clears the item out of both. | |
1152 | unsigned old_wanted = pg_temp_wanted.size(); | |
1153 | unsigned old_pending = pg_temp_pending.size(); | |
1154 | _sent_pg_temp(); | |
1155 | pg_temp_wanted.swap(pg_temp_pending); | |
1156 | dout(10) << __func__ << " " << old_wanted << " + " << old_pending << " -> " | |
1157 | << pg_temp_wanted.size() << dendl; | |
1158 | } | |
1159 | ||
94b18763 FG |
1160 | std::ostream& operator<<(std::ostream& out, |
1161 | const OSDService::pg_temp_t& pg_temp) | |
1162 | { | |
1163 | out << pg_temp.acting; | |
1164 | if (pg_temp.forced) { | |
1165 | out << " (forced)"; | |
1166 | } | |
1167 | return out; | |
1168 | } | |
1169 | ||
7c673cae FG |
1170 | void OSDService::send_pg_temp() |
1171 | { | |
11fdf7f2 | 1172 | std::lock_guard l(pg_temp_lock); |
7c673cae FG |
1173 | if (pg_temp_wanted.empty()) |
1174 | return; | |
1175 | dout(10) << "send_pg_temp " << pg_temp_wanted << dendl; | |
94b18763 | 1176 | MOSDPGTemp *ms[2] = {nullptr, nullptr}; |
11fdf7f2 TL |
1177 | for (auto& [pgid, pg_temp] : pg_temp_wanted) { |
1178 | auto& m = ms[pg_temp.forced]; | |
94b18763 FG |
1179 | if (!m) { |
1180 | m = new MOSDPGTemp(osdmap->get_epoch()); | |
11fdf7f2 | 1181 | m->forced = pg_temp.forced; |
94b18763 | 1182 | } |
11fdf7f2 | 1183 | m->pg_temp.emplace(pgid, pg_temp.acting); |
94b18763 FG |
1184 | } |
1185 | for (auto m : ms) { | |
1186 | if (m) { | |
1187 | monc->send_mon_message(m); | |
1188 | } | |
1189 | } | |
7c673cae FG |
1190 | _sent_pg_temp(); |
1191 | } | |
1192 | ||
1193 | void OSDService::send_pg_created(pg_t pgid) | |
1194 | { | |
11fdf7f2 | 1195 | std::lock_guard l(pg_created_lock); |
7c673cae | 1196 | dout(20) << __func__ << dendl; |
11fdf7f2 | 1197 | auto o = get_osdmap(); |
9f95a23c | 1198 | if (o->require_osd_release >= ceph_release_t::luminous) { |
11fdf7f2 | 1199 | pg_created.insert(pgid); |
c07f9fc5 FG |
1200 | monc->send_mon_message(new MOSDPGCreated(pgid)); |
1201 | } | |
7c673cae FG |
1202 | } |
1203 | ||
11fdf7f2 TL |
1204 | void OSDService::send_pg_created() |
1205 | { | |
1206 | std::lock_guard l(pg_created_lock); | |
1207 | dout(20) << __func__ << dendl; | |
1208 | auto o = get_osdmap(); | |
9f95a23c | 1209 | if (o->require_osd_release >= ceph_release_t::luminous) { |
11fdf7f2 TL |
1210 | for (auto pgid : pg_created) { |
1211 | monc->send_mon_message(new MOSDPGCreated(pgid)); | |
1212 | } | |
1213 | } | |
1214 | } | |
1215 | ||
1216 | void OSDService::prune_pg_created() | |
1217 | { | |
1218 | std::lock_guard l(pg_created_lock); | |
1219 | dout(20) << __func__ << dendl; | |
1220 | auto o = get_osdmap(); | |
1221 | auto i = pg_created.begin(); | |
1222 | while (i != pg_created.end()) { | |
1223 | auto p = o->get_pg_pool(i->pool()); | |
1224 | if (!p || !p->has_flag(pg_pool_t::FLAG_CREATING)) { | |
1225 | dout(20) << __func__ << " pruning " << *i << dendl; | |
1226 | i = pg_created.erase(i); | |
1227 | } else { | |
1228 | dout(20) << __func__ << " keeping " << *i << dendl; | |
1229 | ++i; | |
1230 | } | |
1231 | } | |
1232 | } | |
1233 | ||
1234 | ||
7c673cae FG |
1235 | // -------------------------------------- |
1236 | // dispatch | |
1237 | ||
eafe8130 | 1238 | bool OSDService::can_inc_scrubs() |
7c673cae FG |
1239 | { |
1240 | bool can_inc = false; | |
11fdf7f2 | 1241 | std::lock_guard l(sched_scrub_lock); |
7c673cae | 1242 | |
eafe8130 TL |
1243 | if (scrubs_local + scrubs_remote < cct->_conf->osd_max_scrubs) { |
1244 | dout(20) << __func__ << " == true " << scrubs_local << " local + " << scrubs_remote | |
1245 | << " remote < max " << cct->_conf->osd_max_scrubs << dendl; | |
7c673cae FG |
1246 | can_inc = true; |
1247 | } else { | |
eafe8130 TL |
1248 | dout(20) << __func__ << " == false " << scrubs_local << " local + " << scrubs_remote |
1249 | << " remote >= max " << cct->_conf->osd_max_scrubs << dendl; | |
7c673cae FG |
1250 | } |
1251 | ||
1252 | return can_inc; | |
1253 | } | |
1254 | ||
eafe8130 | 1255 | bool OSDService::inc_scrubs_local() |
7c673cae FG |
1256 | { |
1257 | bool result = false; | |
eafe8130 TL |
1258 | std::lock_guard l{sched_scrub_lock}; |
1259 | if (scrubs_local + scrubs_remote < cct->_conf->osd_max_scrubs) { | |
1260 | dout(20) << __func__ << " " << scrubs_local << " -> " << (scrubs_local+1) | |
1261 | << " (max " << cct->_conf->osd_max_scrubs << ", remote " << scrubs_remote << ")" << dendl; | |
7c673cae | 1262 | result = true; |
eafe8130 | 1263 | ++scrubs_local; |
7c673cae | 1264 | } else { |
eafe8130 | 1265 | dout(20) << __func__ << " " << scrubs_local << " local + " << scrubs_remote << " remote >= max " << cct->_conf->osd_max_scrubs << dendl; |
7c673cae | 1266 | } |
7c673cae FG |
1267 | return result; |
1268 | } | |
1269 | ||
eafe8130 | 1270 | void OSDService::dec_scrubs_local() |
7c673cae | 1271 | { |
eafe8130 TL |
1272 | std::lock_guard l{sched_scrub_lock}; |
1273 | dout(20) << __func__ << " " << scrubs_local << " -> " << (scrubs_local-1) | |
1274 | << " (max " << cct->_conf->osd_max_scrubs << ", remote " << scrubs_remote << ")" << dendl; | |
1275 | --scrubs_local; | |
1276 | ceph_assert(scrubs_local >= 0); | |
7c673cae FG |
1277 | } |
1278 | ||
eafe8130 | 1279 | bool OSDService::inc_scrubs_remote() |
7c673cae | 1280 | { |
eafe8130 TL |
1281 | bool result = false; |
1282 | std::lock_guard l{sched_scrub_lock}; | |
1283 | if (scrubs_local + scrubs_remote < cct->_conf->osd_max_scrubs) { | |
1284 | dout(20) << __func__ << " " << scrubs_remote << " -> " << (scrubs_remote+1) | |
1285 | << " (max " << cct->_conf->osd_max_scrubs << ", local " << scrubs_local << ")" << dendl; | |
1286 | result = true; | |
1287 | ++scrubs_remote; | |
7c673cae | 1288 | } else { |
eafe8130 | 1289 | dout(20) << __func__ << " " << scrubs_local << " local + " << scrubs_remote << " remote >= max " << cct->_conf->osd_max_scrubs << dendl; |
7c673cae | 1290 | } |
eafe8130 TL |
1291 | return result; |
1292 | } | |
1293 | ||
1294 | void OSDService::dec_scrubs_remote() | |
1295 | { | |
1296 | std::lock_guard l{sched_scrub_lock}; | |
1297 | dout(20) << __func__ << " " << scrubs_remote << " -> " << (scrubs_remote-1) | |
1298 | << " (max " << cct->_conf->osd_max_scrubs << ", local " << scrubs_local << ")" << dendl; | |
1299 | --scrubs_remote; | |
1300 | ceph_assert(scrubs_remote >= 0); | |
7c673cae FG |
1301 | } |
1302 | ||
eafe8130 | 1303 | void OSDService::dump_scrub_reservations(Formatter *f) |
7c673cae | 1304 | { |
eafe8130 TL |
1305 | std::lock_guard l{sched_scrub_lock}; |
1306 | f->dump_int("scrubs_local", scrubs_local); | |
1307 | f->dump_int("scrubs_remote", scrubs_remote); | |
1308 | f->dump_int("osd_max_scrubs", cct->_conf->osd_max_scrubs); | |
7c673cae FG |
1309 | } |
1310 | ||
1311 | void OSDService::retrieve_epochs(epoch_t *_boot_epoch, epoch_t *_up_epoch, | |
1312 | epoch_t *_bind_epoch) const | |
1313 | { | |
11fdf7f2 | 1314 | std::lock_guard l(epoch_lock); |
7c673cae FG |
1315 | if (_boot_epoch) |
1316 | *_boot_epoch = boot_epoch; | |
1317 | if (_up_epoch) | |
1318 | *_up_epoch = up_epoch; | |
1319 | if (_bind_epoch) | |
1320 | *_bind_epoch = bind_epoch; | |
1321 | } | |
1322 | ||
1323 | void OSDService::set_epochs(const epoch_t *_boot_epoch, const epoch_t *_up_epoch, | |
1324 | const epoch_t *_bind_epoch) | |
1325 | { | |
11fdf7f2 | 1326 | std::lock_guard l(epoch_lock); |
7c673cae | 1327 | if (_boot_epoch) { |
11fdf7f2 | 1328 | ceph_assert(*_boot_epoch == 0 || *_boot_epoch >= boot_epoch); |
7c673cae FG |
1329 | boot_epoch = *_boot_epoch; |
1330 | } | |
1331 | if (_up_epoch) { | |
11fdf7f2 | 1332 | ceph_assert(*_up_epoch == 0 || *_up_epoch >= up_epoch); |
7c673cae FG |
1333 | up_epoch = *_up_epoch; |
1334 | } | |
1335 | if (_bind_epoch) { | |
11fdf7f2 | 1336 | ceph_assert(*_bind_epoch == 0 || *_bind_epoch >= bind_epoch); |
7c673cae FG |
1337 | bind_epoch = *_bind_epoch; |
1338 | } | |
1339 | } | |
1340 | ||
1341 | bool OSDService::prepare_to_stop() | |
1342 | { | |
9f95a23c | 1343 | std::unique_lock l(is_stopping_lock); |
7c673cae FG |
1344 | if (get_state() != NOT_STOPPING) |
1345 | return false; | |
1346 | ||
1347 | OSDMapRef osdmap = get_osdmap(); | |
1348 | if (osdmap && osdmap->is_up(whoami)) { | |
1349 | dout(0) << __func__ << " telling mon we are shutting down" << dendl; | |
1350 | set_state(PREPARING_TO_STOP); | |
11fdf7f2 TL |
1351 | monc->send_mon_message( |
1352 | new MOSDMarkMeDown( | |
1353 | monc->get_fsid(), | |
1354 | whoami, | |
1355 | osdmap->get_addrs(whoami), | |
1356 | osdmap->get_epoch(), | |
1357 | true // request ack | |
1358 | )); | |
9f95a23c TL |
1359 | const auto timeout = ceph::make_timespan(cct->_conf->osd_mon_shutdown_timeout); |
1360 | is_stopping_cond.wait_for(l, timeout, | |
1361 | [this] { return get_state() == STOPPING; }); | |
7c673cae FG |
1362 | } |
1363 | dout(0) << __func__ << " starting shutdown" << dendl; | |
1364 | set_state(STOPPING); | |
1365 | return true; | |
1366 | } | |
1367 | ||
1368 | void OSDService::got_stop_ack() | |
1369 | { | |
9f95a23c | 1370 | std::scoped_lock l(is_stopping_lock); |
7c673cae FG |
1371 | if (get_state() == PREPARING_TO_STOP) { |
1372 | dout(0) << __func__ << " starting shutdown" << dendl; | |
1373 | set_state(STOPPING); | |
9f95a23c | 1374 | is_stopping_cond.notify_all(); |
7c673cae FG |
1375 | } else { |
1376 | dout(10) << __func__ << " ignoring msg" << dendl; | |
1377 | } | |
1378 | } | |
1379 | ||
1380 | MOSDMap *OSDService::build_incremental_map_msg(epoch_t since, epoch_t to, | |
1381 | OSDSuperblock& sblock) | |
1382 | { | |
28e407b8 AA |
1383 | MOSDMap *m = new MOSDMap(monc->get_fsid(), |
1384 | osdmap->get_encoding_features()); | |
7c673cae FG |
1385 | m->oldest_map = max_oldest_map; |
1386 | m->newest_map = sblock.newest_map; | |
1387 | ||
11fdf7f2 TL |
1388 | int max = cct->_conf->osd_map_message_max; |
1389 | ssize_t max_bytes = cct->_conf->osd_map_message_max_bytes; | |
1390 | ||
1391 | if (since < m->oldest_map) { | |
1392 | // we don't have the next map the target wants, so start with a | |
1393 | // full map. | |
1394 | bufferlist bl; | |
1395 | dout(10) << __func__ << " oldest map " << max_oldest_map << " > since " | |
1396 | << since << ", starting with full map" << dendl; | |
1397 | since = m->oldest_map; | |
1398 | if (!get_map_bl(since, bl)) { | |
1399 | derr << __func__ << " missing full map " << since << dendl; | |
1400 | goto panic; | |
1401 | } | |
1402 | max--; | |
1403 | max_bytes -= bl.length(); | |
1404 | m->maps[since].claim(bl); | |
1405 | } | |
1406 | for (epoch_t e = since + 1; e <= to; ++e) { | |
7c673cae | 1407 | bufferlist bl; |
11fdf7f2 | 1408 | if (get_inc_map_bl(e, bl)) { |
7c673cae | 1409 | m->incremental_maps[e].claim(bl); |
11fdf7f2 TL |
1410 | } else { |
1411 | derr << __func__ << " missing incremental map " << e << dendl; | |
1412 | if (!get_map_bl(e, bl)) { | |
1413 | derr << __func__ << " also missing full map " << e << dendl; | |
1414 | goto panic; | |
1415 | } | |
7c673cae | 1416 | m->maps[e].claim(bl); |
11fdf7f2 TL |
1417 | } |
1418 | max--; | |
1419 | max_bytes -= bl.length(); | |
1420 | if (max <= 0 || max_bytes <= 0) { | |
7c673cae | 1421 | break; |
11fdf7f2 TL |
1422 | } |
1423 | } | |
1424 | return m; | |
1425 | ||
1426 | panic: | |
1427 | if (!m->maps.empty() || | |
1428 | !m->incremental_maps.empty()) { | |
1429 | // send what we have so far | |
1430 | return m; | |
1431 | } | |
1432 | // send something | |
1433 | bufferlist bl; | |
1434 | if (get_inc_map_bl(m->newest_map, bl)) { | |
1435 | m->incremental_maps[m->newest_map].claim(bl); | |
1436 | } else { | |
1437 | derr << __func__ << " unable to load latest map " << m->newest_map << dendl; | |
1438 | if (!get_map_bl(m->newest_map, bl)) { | |
1439 | derr << __func__ << " unable to load latest full map " << m->newest_map | |
7c673cae | 1440 | << dendl; |
11fdf7f2 | 1441 | ceph_abort(); |
7c673cae | 1442 | } |
11fdf7f2 | 1443 | m->maps[m->newest_map].claim(bl); |
7c673cae FG |
1444 | } |
1445 | return m; | |
1446 | } | |
1447 | ||
1448 | void OSDService::send_map(MOSDMap *m, Connection *con) | |
1449 | { | |
1450 | con->send_message(m); | |
1451 | } | |
1452 | ||
1453 | void OSDService::send_incremental_map(epoch_t since, Connection *con, | |
9f95a23c | 1454 | const OSDMapRef& osdmap) |
7c673cae FG |
1455 | { |
1456 | epoch_t to = osdmap->get_epoch(); | |
1457 | dout(10) << "send_incremental_map " << since << " -> " << to | |
1458 | << " to " << con << " " << con->get_peer_addr() << dendl; | |
1459 | ||
1460 | MOSDMap *m = NULL; | |
1461 | while (!m) { | |
1462 | OSDSuperblock sblock(get_superblock()); | |
1463 | if (since < sblock.oldest_map) { | |
1464 | // just send latest full map | |
28e407b8 AA |
1465 | MOSDMap *m = new MOSDMap(monc->get_fsid(), |
1466 | osdmap->get_encoding_features()); | |
7c673cae FG |
1467 | m->oldest_map = max_oldest_map; |
1468 | m->newest_map = sblock.newest_map; | |
1469 | get_map_bl(to, m->maps[to]); | |
1470 | send_map(m, con); | |
1471 | return; | |
1472 | } | |
1473 | ||
1474 | if (to > since && (int64_t)(to - since) > cct->_conf->osd_map_share_max_epochs) { | |
1475 | dout(10) << " " << (to - since) << " > max " << cct->_conf->osd_map_share_max_epochs | |
1476 | << ", only sending most recent" << dendl; | |
1477 | since = to - cct->_conf->osd_map_share_max_epochs; | |
1478 | } | |
1479 | ||
7c673cae FG |
1480 | m = build_incremental_map_msg(since, to, sblock); |
1481 | } | |
1482 | send_map(m, con); | |
1483 | } | |
1484 | ||
1485 | bool OSDService::_get_map_bl(epoch_t e, bufferlist& bl) | |
1486 | { | |
1487 | bool found = map_bl_cache.lookup(e, &bl); | |
31f18b77 FG |
1488 | if (found) { |
1489 | if (logger) | |
1490 | logger->inc(l_osd_map_bl_cache_hit); | |
7c673cae | 1491 | return true; |
31f18b77 FG |
1492 | } |
1493 | if (logger) | |
1494 | logger->inc(l_osd_map_bl_cache_miss); | |
11fdf7f2 | 1495 | found = store->read(meta_ch, |
31f18b77 FG |
1496 | OSD::get_osdmap_pobject_name(e), 0, 0, bl, |
1497 | CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) >= 0; | |
1498 | if (found) { | |
7c673cae | 1499 | _add_map_bl(e, bl); |
31f18b77 | 1500 | } |
7c673cae FG |
1501 | return found; |
1502 | } | |
1503 | ||
1504 | bool OSDService::get_inc_map_bl(epoch_t e, bufferlist& bl) | |
1505 | { | |
11fdf7f2 | 1506 | std::lock_guard l(map_cache_lock); |
7c673cae | 1507 | bool found = map_bl_inc_cache.lookup(e, &bl); |
31f18b77 FG |
1508 | if (found) { |
1509 | if (logger) | |
1510 | logger->inc(l_osd_map_bl_cache_hit); | |
7c673cae | 1511 | return true; |
31f18b77 FG |
1512 | } |
1513 | if (logger) | |
1514 | logger->inc(l_osd_map_bl_cache_miss); | |
11fdf7f2 | 1515 | found = store->read(meta_ch, |
31f18b77 FG |
1516 | OSD::get_inc_osdmap_pobject_name(e), 0, 0, bl, |
1517 | CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) >= 0; | |
1518 | if (found) { | |
7c673cae | 1519 | _add_map_inc_bl(e, bl); |
31f18b77 | 1520 | } |
7c673cae FG |
1521 | return found; |
1522 | } | |
1523 | ||
1524 | void OSDService::_add_map_bl(epoch_t e, bufferlist& bl) | |
1525 | { | |
1526 | dout(10) << "add_map_bl " << e << " " << bl.length() << " bytes" << dendl; | |
31f18b77 FG |
1527 | // cache a contiguous buffer |
1528 | if (bl.get_num_buffers() > 1) { | |
1529 | bl.rebuild(); | |
1530 | } | |
1531 | bl.try_assign_to_mempool(mempool::mempool_osd_mapbl); | |
7c673cae FG |
1532 | map_bl_cache.add(e, bl); |
1533 | } | |
1534 | ||
1535 | void OSDService::_add_map_inc_bl(epoch_t e, bufferlist& bl) | |
1536 | { | |
1537 | dout(10) << "add_map_inc_bl " << e << " " << bl.length() << " bytes" << dendl; | |
31f18b77 FG |
1538 | // cache a contiguous buffer |
1539 | if (bl.get_num_buffers() > 1) { | |
1540 | bl.rebuild(); | |
1541 | } | |
1542 | bl.try_assign_to_mempool(mempool::mempool_osd_mapbl); | |
7c673cae FG |
1543 | map_bl_inc_cache.add(e, bl); |
1544 | } | |
1545 | ||
7c673cae FG |
1546 | OSDMapRef OSDService::_add_map(OSDMap *o) |
1547 | { | |
1548 | epoch_t e = o->get_epoch(); | |
1549 | ||
1550 | if (cct->_conf->osd_map_dedup) { | |
1551 | // Dedup against an existing map at a nearby epoch | |
1552 | OSDMapRef for_dedup = map_cache.lower_bound(e); | |
1553 | if (for_dedup) { | |
1554 | OSDMap::dedup(for_dedup.get(), o); | |
1555 | } | |
1556 | } | |
1557 | bool existed; | |
1558 | OSDMapRef l = map_cache.add(e, o, &existed); | |
1559 | if (existed) { | |
1560 | delete o; | |
1561 | } | |
1562 | return l; | |
1563 | } | |
1564 | ||
1565 | OSDMapRef OSDService::try_get_map(epoch_t epoch) | |
1566 | { | |
11fdf7f2 | 1567 | std::lock_guard l(map_cache_lock); |
7c673cae FG |
1568 | OSDMapRef retval = map_cache.lookup(epoch); |
1569 | if (retval) { | |
1570 | dout(30) << "get_map " << epoch << " -cached" << dendl; | |
1571 | if (logger) { | |
1572 | logger->inc(l_osd_map_cache_hit); | |
1573 | } | |
1574 | return retval; | |
1575 | } | |
1576 | if (logger) { | |
1577 | logger->inc(l_osd_map_cache_miss); | |
1578 | epoch_t lb = map_cache.cached_key_lower_bound(); | |
1579 | if (epoch < lb) { | |
1580 | dout(30) << "get_map " << epoch << " - miss, below lower bound" << dendl; | |
1581 | logger->inc(l_osd_map_cache_miss_low); | |
1582 | logger->inc(l_osd_map_cache_miss_low_avg, lb - epoch); | |
1583 | } | |
1584 | } | |
1585 | ||
1586 | OSDMap *map = new OSDMap; | |
1587 | if (epoch > 0) { | |
1588 | dout(20) << "get_map " << epoch << " - loading and decoding " << map << dendl; | |
1589 | bufferlist bl; | |
1590 | if (!_get_map_bl(epoch, bl) || bl.length() == 0) { | |
1591 | derr << "failed to load OSD map for epoch " << epoch << ", got " << bl.length() << " bytes" << dendl; | |
1592 | delete map; | |
1593 | return OSDMapRef(); | |
1594 | } | |
1595 | map->decode(bl); | |
1596 | } else { | |
1597 | dout(20) << "get_map " << epoch << " - return initial " << map << dendl; | |
1598 | } | |
1599 | return _add_map(map); | |
1600 | } | |
1601 | ||
1602 | // ops | |
1603 | ||
1604 | ||
1605 | void OSDService::reply_op_error(OpRequestRef op, int err) | |
1606 | { | |
9f95a23c | 1607 | reply_op_error(op, err, eversion_t(), 0, {}); |
7c673cae FG |
1608 | } |
1609 | ||
1610 | void OSDService::reply_op_error(OpRequestRef op, int err, eversion_t v, | |
9f95a23c TL |
1611 | version_t uv, |
1612 | vector<pg_log_op_return_item_t> op_returns) | |
7c673cae | 1613 | { |
9f95a23c | 1614 | auto m = op->get_req<MOSDOp>(); |
11fdf7f2 | 1615 | ceph_assert(m->get_type() == CEPH_MSG_OSD_OP); |
7c673cae FG |
1616 | int flags; |
1617 | flags = m->get_flags() & (CEPH_OSD_FLAG_ACK|CEPH_OSD_FLAG_ONDISK); | |
1618 | ||
9f95a23c TL |
1619 | MOSDOpReply *reply = new MOSDOpReply(m, err, osdmap->get_epoch(), flags, |
1620 | !m->has_flag(CEPH_OSD_FLAG_RETURNVEC)); | |
7c673cae | 1621 | reply->set_reply_versions(v, uv); |
9f95a23c | 1622 | reply->set_op_returns(op_returns); |
7c673cae FG |
1623 | m->get_connection()->send_message(reply); |
1624 | } | |
1625 | ||
1626 | void OSDService::handle_misdirected_op(PG *pg, OpRequestRef op) | |
1627 | { | |
31f18b77 FG |
1628 | if (!cct->_conf->osd_debug_misdirected_ops) { |
1629 | return; | |
1630 | } | |
1631 | ||
9f95a23c | 1632 | auto m = op->get_req<MOSDOp>(); |
11fdf7f2 | 1633 | ceph_assert(m->get_type() == CEPH_MSG_OSD_OP); |
7c673cae | 1634 | |
11fdf7f2 | 1635 | ceph_assert(m->get_map_epoch() >= pg->get_history().same_primary_since); |
7c673cae FG |
1636 | |
1637 | if (pg->is_ec_pg()) { | |
1638 | /** | |
1639 | * OSD recomputes op target based on current OSDMap. With an EC pg, we | |
1640 | * can get this result: | |
1641 | * 1) client at map 512 sends an op to osd 3, pg_t 3.9 based on mapping | |
1642 | * [CRUSH_ITEM_NONE, 2, 3]/3 | |
1643 | * 2) OSD 3 at map 513 remaps op to osd 3, spg_t 3.9s0 based on mapping | |
1644 | * [3, 2, 3]/3 | |
1645 | * 3) PG 3.9s0 dequeues the op at epoch 512 and notices that it isn't primary | |
1646 | * -- misdirected op | |
1647 | * 4) client resends and this time PG 3.9s0 having caught up to 513 gets | |
1648 | * it and fulfils it | |
1649 | * | |
1650 | * We can't compute the op target based on the sending map epoch due to | |
1651 | * splitting. The simplest thing is to detect such cases here and drop | |
1652 | * them without an error (the client will resend anyway). | |
1653 | */ | |
11fdf7f2 | 1654 | ceph_assert(m->get_map_epoch() <= superblock.newest_map); |
7c673cae FG |
1655 | OSDMapRef opmap = try_get_map(m->get_map_epoch()); |
1656 | if (!opmap) { | |
1657 | dout(7) << __func__ << ": " << *pg << " no longer have map for " | |
1658 | << m->get_map_epoch() << ", dropping" << dendl; | |
1659 | return; | |
1660 | } | |
1661 | pg_t _pgid = m->get_raw_pg(); | |
1662 | spg_t pgid; | |
1663 | if ((m->get_flags() & CEPH_OSD_FLAG_PGOP) == 0) | |
1664 | _pgid = opmap->raw_pg_to_pg(_pgid); | |
1665 | if (opmap->get_primary_shard(_pgid, &pgid) && | |
11fdf7f2 | 1666 | pgid.shard != pg->pg_id.shard) { |
7c673cae FG |
1667 | dout(7) << __func__ << ": " << *pg << " primary changed since " |
1668 | << m->get_map_epoch() << ", dropping" << dendl; | |
1669 | return; | |
1670 | } | |
1671 | } | |
1672 | ||
1673 | dout(7) << *pg << " misdirected op in " << m->get_map_epoch() << dendl; | |
1674 | clog->warn() << m->get_source_inst() << " misdirected " << m->get_reqid() | |
1675 | << " pg " << m->get_raw_pg() | |
1676 | << " to osd." << whoami | |
11fdf7f2 | 1677 | << " not " << pg->get_acting() |
7c673cae | 1678 | << " in e" << m->get_map_epoch() << "/" << osdmap->get_epoch(); |
7c673cae FG |
1679 | } |
1680 | ||
9f95a23c | 1681 | void OSDService::enqueue_back(OpSchedulerItem&& qi) |
7c673cae | 1682 | { |
11fdf7f2 | 1683 | osd->op_shardedwq.queue(std::move(qi)); |
7c673cae FG |
1684 | } |
1685 | ||
9f95a23c | 1686 | void OSDService::enqueue_front(OpSchedulerItem&& qi) |
7c673cae | 1687 | { |
11fdf7f2 | 1688 | osd->op_shardedwq.queue_front(std::move(qi)); |
7c673cae FG |
1689 | } |
1690 | ||
11fdf7f2 TL |
1691 | void OSDService::queue_recovery_context( |
1692 | PG *pg, | |
1693 | GenContext<ThreadPool::TPHandle&> *c) | |
7c673cae | 1694 | { |
11fdf7f2 TL |
1695 | epoch_t e = get_osdmap_epoch(); |
1696 | enqueue_back( | |
9f95a23c TL |
1697 | OpSchedulerItem( |
1698 | unique_ptr<OpSchedulerItem::OpQueueable>( | |
11fdf7f2 TL |
1699 | new PGRecoveryContext(pg->get_pgid(), c, e)), |
1700 | cct->_conf->osd_recovery_cost, | |
1701 | cct->_conf->osd_recovery_priority, | |
1702 | ceph_clock_now(), | |
1703 | 0, | |
1704 | e)); | |
7c673cae FG |
1705 | } |
1706 | ||
1707 | void OSDService::queue_for_snap_trim(PG *pg) | |
1708 | { | |
1709 | dout(10) << "queueing " << *pg << " for snaptrim" << dendl; | |
11fdf7f2 | 1710 | enqueue_back( |
9f95a23c TL |
1711 | OpSchedulerItem( |
1712 | unique_ptr<OpSchedulerItem::OpQueueable>( | |
11fdf7f2 TL |
1713 | new PGSnapTrim(pg->get_pgid(), pg->get_osdmap_epoch())), |
1714 | cct->_conf->osd_snap_trim_cost, | |
1715 | cct->_conf->osd_snap_trim_priority, | |
1716 | ceph_clock_now(), | |
1717 | 0, | |
1718 | pg->get_osdmap_epoch())); | |
1719 | } | |
1720 | ||
1721 | void OSDService::queue_for_scrub(PG *pg, bool with_high_priority) | |
1722 | { | |
1723 | unsigned scrub_queue_priority = pg->scrubber.priority; | |
1724 | if (with_high_priority && scrub_queue_priority < cct->_conf->osd_client_op_priority) { | |
1725 | scrub_queue_priority = cct->_conf->osd_client_op_priority; | |
1726 | } | |
1727 | const auto epoch = pg->get_osdmap_epoch(); | |
1728 | enqueue_back( | |
9f95a23c TL |
1729 | OpSchedulerItem( |
1730 | unique_ptr<OpSchedulerItem::OpQueueable>(new PGScrub(pg->get_pgid(), epoch)), | |
11fdf7f2 TL |
1731 | cct->_conf->osd_scrub_cost, |
1732 | scrub_queue_priority, | |
1733 | ceph_clock_now(), | |
1734 | 0, | |
1735 | epoch)); | |
1736 | } | |
1737 | ||
1738 | void OSDService::queue_for_pg_delete(spg_t pgid, epoch_t e) | |
1739 | { | |
1740 | dout(10) << __func__ << " on " << pgid << " e " << e << dendl; | |
1741 | enqueue_back( | |
9f95a23c TL |
1742 | OpSchedulerItem( |
1743 | unique_ptr<OpSchedulerItem::OpQueueable>( | |
11fdf7f2 TL |
1744 | new PGDelete(pgid, e)), |
1745 | cct->_conf->osd_pg_delete_cost, | |
1746 | cct->_conf->osd_pg_delete_priority, | |
1747 | ceph_clock_now(), | |
1748 | 0, | |
1749 | e)); | |
1750 | } | |
1751 | ||
1752 | bool OSDService::try_finish_pg_delete(PG *pg, unsigned old_pg_num) | |
1753 | { | |
1754 | return osd->try_finish_pg_delete(pg, old_pg_num); | |
1755 | } | |
1756 | ||
1757 | // --- | |
1758 | ||
1759 | void OSDService::set_ready_to_merge_source(PG *pg, eversion_t version) | |
1760 | { | |
1761 | std::lock_guard l(merge_lock); | |
1762 | dout(10) << __func__ << " " << pg->pg_id << dendl; | |
1763 | ready_to_merge_source[pg->pg_id.pgid] = version; | |
1764 | assert(not_ready_to_merge_source.count(pg->pg_id.pgid) == 0); | |
1765 | _send_ready_to_merge(); | |
1766 | } | |
1767 | ||
1768 | void OSDService::set_ready_to_merge_target(PG *pg, | |
1769 | eversion_t version, | |
1770 | epoch_t last_epoch_started, | |
1771 | epoch_t last_epoch_clean) | |
1772 | { | |
1773 | std::lock_guard l(merge_lock); | |
1774 | dout(10) << __func__ << " " << pg->pg_id << dendl; | |
1775 | ready_to_merge_target.insert(make_pair(pg->pg_id.pgid, | |
1776 | make_tuple(version, | |
1777 | last_epoch_started, | |
1778 | last_epoch_clean))); | |
1779 | assert(not_ready_to_merge_target.count(pg->pg_id.pgid) == 0); | |
1780 | _send_ready_to_merge(); | |
1781 | } | |
1782 | ||
1783 | void OSDService::set_not_ready_to_merge_source(pg_t source) | |
1784 | { | |
1785 | std::lock_guard l(merge_lock); | |
1786 | dout(10) << __func__ << " " << source << dendl; | |
1787 | not_ready_to_merge_source.insert(source); | |
1788 | assert(ready_to_merge_source.count(source) == 0); | |
1789 | _send_ready_to_merge(); | |
1790 | } | |
1791 | ||
1792 | void OSDService::set_not_ready_to_merge_target(pg_t target, pg_t source) | |
1793 | { | |
1794 | std::lock_guard l(merge_lock); | |
1795 | dout(10) << __func__ << " " << target << " source " << source << dendl; | |
1796 | not_ready_to_merge_target[target] = source; | |
1797 | assert(ready_to_merge_target.count(target) == 0); | |
1798 | _send_ready_to_merge(); | |
1799 | } | |
1800 | ||
1801 | void OSDService::send_ready_to_merge() | |
1802 | { | |
1803 | std::lock_guard l(merge_lock); | |
1804 | _send_ready_to_merge(); | |
1805 | } | |
1806 | ||
1807 | void OSDService::_send_ready_to_merge() | |
1808 | { | |
1809 | dout(20) << __func__ | |
1810 | << " ready_to_merge_source " << ready_to_merge_source | |
1811 | << " not_ready_to_merge_source " << not_ready_to_merge_source | |
1812 | << " ready_to_merge_target " << ready_to_merge_target | |
1813 | << " not_ready_to_merge_target " << not_ready_to_merge_target | |
1814 | << " sent_ready_to_merge_source " << sent_ready_to_merge_source | |
1815 | << dendl; | |
1816 | for (auto src : not_ready_to_merge_source) { | |
1817 | if (sent_ready_to_merge_source.count(src) == 0) { | |
1818 | monc->send_mon_message(new MOSDPGReadyToMerge( | |
1819 | src, | |
1820 | {}, {}, 0, 0, | |
1821 | false, | |
1822 | osdmap->get_epoch())); | |
1823 | sent_ready_to_merge_source.insert(src); | |
1824 | } | |
1825 | } | |
1826 | for (auto p : not_ready_to_merge_target) { | |
1827 | if (sent_ready_to_merge_source.count(p.second) == 0) { | |
1828 | monc->send_mon_message(new MOSDPGReadyToMerge( | |
1829 | p.second, | |
1830 | {}, {}, 0, 0, | |
1831 | false, | |
1832 | osdmap->get_epoch())); | |
1833 | sent_ready_to_merge_source.insert(p.second); | |
1834 | } | |
1835 | } | |
1836 | for (auto src : ready_to_merge_source) { | |
1837 | if (not_ready_to_merge_source.count(src.first) || | |
1838 | not_ready_to_merge_target.count(src.first.get_parent())) { | |
1839 | continue; | |
1840 | } | |
1841 | auto p = ready_to_merge_target.find(src.first.get_parent()); | |
1842 | if (p != ready_to_merge_target.end() && | |
1843 | sent_ready_to_merge_source.count(src.first) == 0) { | |
1844 | monc->send_mon_message(new MOSDPGReadyToMerge( | |
1845 | src.first, // source pgid | |
1846 | src.second, // src version | |
1847 | std::get<0>(p->second), // target version | |
1848 | std::get<1>(p->second), // PG's last_epoch_started | |
1849 | std::get<2>(p->second), // PG's last_epoch_clean | |
1850 | true, | |
1851 | osdmap->get_epoch())); | |
1852 | sent_ready_to_merge_source.insert(src.first); | |
1853 | } | |
1854 | } | |
1855 | } | |
1856 | ||
1857 | void OSDService::clear_ready_to_merge(PG *pg) | |
1858 | { | |
1859 | std::lock_guard l(merge_lock); | |
1860 | dout(10) << __func__ << " " << pg->pg_id << dendl; | |
1861 | ready_to_merge_source.erase(pg->pg_id.pgid); | |
1862 | ready_to_merge_target.erase(pg->pg_id.pgid); | |
1863 | not_ready_to_merge_source.erase(pg->pg_id.pgid); | |
1864 | not_ready_to_merge_target.erase(pg->pg_id.pgid); | |
1865 | sent_ready_to_merge_source.erase(pg->pg_id.pgid); | |
1866 | } | |
1867 | ||
1868 | void OSDService::clear_sent_ready_to_merge() | |
1869 | { | |
1870 | std::lock_guard l(merge_lock); | |
1871 | sent_ready_to_merge_source.clear(); | |
1872 | } | |
1873 | ||
9f95a23c | 1874 | void OSDService::prune_sent_ready_to_merge(const OSDMapRef& osdmap) |
11fdf7f2 TL |
1875 | { |
1876 | std::lock_guard l(merge_lock); | |
1877 | auto i = sent_ready_to_merge_source.begin(); | |
1878 | while (i != sent_ready_to_merge_source.end()) { | |
1879 | if (!osdmap->pg_exists(*i)) { | |
1880 | dout(10) << __func__ << " " << *i << dendl; | |
1881 | i = sent_ready_to_merge_source.erase(i); | |
1882 | } else { | |
1883 | ++i; | |
1884 | } | |
1885 | } | |
7c673cae FG |
1886 | } |
1887 | ||
11fdf7f2 TL |
1888 | // --- |
1889 | ||
1890 | void OSDService::_queue_for_recovery( | |
1891 | std::pair<epoch_t, PGRef> p, | |
1892 | uint64_t reserved_pushes) | |
1893 | { | |
9f95a23c | 1894 | ceph_assert(ceph_mutex_is_locked_by_me(recovery_lock)); |
11fdf7f2 | 1895 | enqueue_back( |
9f95a23c TL |
1896 | OpSchedulerItem( |
1897 | unique_ptr<OpSchedulerItem::OpQueueable>( | |
11fdf7f2 TL |
1898 | new PGRecovery( |
1899 | p.second->get_pgid(), p.first, reserved_pushes)), | |
1900 | cct->_conf->osd_recovery_cost, | |
1901 | cct->_conf->osd_recovery_priority, | |
1902 | ceph_clock_now(), | |
1903 | 0, | |
1904 | p.first)); | |
1905 | } | |
7c673cae FG |
1906 | |
1907 | // ==================================================================== | |
1908 | // OSD | |
1909 | ||
1910 | #undef dout_prefix | |
1911 | #define dout_prefix *_dout | |
1912 | ||
1913 | // Commands shared between OSD's console and admin console: | |
1914 | namespace ceph { | |
1915 | namespace osd_cmds { | |
1916 | ||
11fdf7f2 | 1917 | int heap(CephContext& cct, const cmdmap_t& cmdmap, Formatter& f, std::ostream& os); |
7c673cae FG |
1918 | |
1919 | }} // namespace ceph::osd_cmds | |
1920 | ||
11fdf7f2 | 1921 | int OSD::mkfs(CephContext *cct, ObjectStore *store, uuid_d fsid, int whoami) |
7c673cae FG |
1922 | { |
1923 | int ret; | |
1924 | ||
7c673cae FG |
1925 | OSDSuperblock sb; |
1926 | bufferlist sbbl; | |
11fdf7f2 | 1927 | ObjectStore::CollectionHandle ch; |
7c673cae FG |
1928 | |
1929 | // if we are fed a uuid for this osd, use it. | |
1930 | store->set_fsid(cct->_conf->osd_uuid); | |
1931 | ||
1932 | ret = store->mkfs(); | |
1933 | if (ret) { | |
224ce89b WB |
1934 | derr << "OSD::mkfs: ObjectStore::mkfs failed with error " |
1935 | << cpp_strerror(ret) << dendl; | |
7c673cae FG |
1936 | goto free_store; |
1937 | } | |
1938 | ||
31f18b77 | 1939 | store->set_cache_shards(1); // doesn't matter for mkfs! |
7c673cae FG |
1940 | |
1941 | ret = store->mount(); | |
1942 | if (ret) { | |
224ce89b WB |
1943 | derr << "OSD::mkfs: couldn't mount ObjectStore: error " |
1944 | << cpp_strerror(ret) << dendl; | |
7c673cae FG |
1945 | goto free_store; |
1946 | } | |
1947 | ||
11fdf7f2 TL |
1948 | ch = store->open_collection(coll_t::meta()); |
1949 | if (ch) { | |
1950 | ret = store->read(ch, OSD_SUPERBLOCK_GOBJECT, 0, 0, sbbl); | |
1951 | if (ret < 0) { | |
1952 | derr << "OSD::mkfs: have meta collection but no superblock" << dendl; | |
1953 | goto free_store; | |
1954 | } | |
7c673cae FG |
1955 | /* if we already have superblock, check content of superblock */ |
1956 | dout(0) << " have superblock" << dendl; | |
11fdf7f2 TL |
1957 | auto p = sbbl.cbegin(); |
1958 | decode(sb, p); | |
7c673cae FG |
1959 | if (whoami != sb.whoami) { |
1960 | derr << "provided osd id " << whoami << " != superblock's " << sb.whoami | |
1961 | << dendl; | |
1962 | ret = -EINVAL; | |
1963 | goto umount_store; | |
1964 | } | |
1965 | if (fsid != sb.cluster_fsid) { | |
1966 | derr << "provided cluster fsid " << fsid | |
1967 | << " != superblock's " << sb.cluster_fsid << dendl; | |
1968 | ret = -EINVAL; | |
1969 | goto umount_store; | |
1970 | } | |
1971 | } else { | |
1972 | // create superblock | |
1973 | sb.cluster_fsid = fsid; | |
1974 | sb.osd_fsid = store->get_fsid(); | |
1975 | sb.whoami = whoami; | |
1976 | sb.compat_features = get_osd_initial_compat_set(); | |
1977 | ||
1978 | bufferlist bl; | |
11fdf7f2 | 1979 | encode(sb, bl); |
7c673cae | 1980 | |
11fdf7f2 TL |
1981 | ObjectStore::CollectionHandle ch = store->create_new_collection( |
1982 | coll_t::meta()); | |
7c673cae FG |
1983 | ObjectStore::Transaction t; |
1984 | t.create_collection(coll_t::meta(), 0); | |
1985 | t.write(coll_t::meta(), OSD_SUPERBLOCK_GOBJECT, 0, bl.length(), bl); | |
11fdf7f2 | 1986 | ret = store->queue_transaction(ch, std::move(t)); |
7c673cae FG |
1987 | if (ret) { |
1988 | derr << "OSD::mkfs: error while writing OSD_SUPERBLOCK_GOBJECT: " | |
11fdf7f2 | 1989 | << "queue_transaction returned " << cpp_strerror(ret) << dendl; |
7c673cae FG |
1990 | goto umount_store; |
1991 | } | |
1992 | } | |
1993 | ||
3efd9988 | 1994 | ret = write_meta(cct, store, sb.cluster_fsid, sb.osd_fsid, whoami); |
7c673cae | 1995 | if (ret) { |
224ce89b WB |
1996 | derr << "OSD::mkfs: failed to write fsid file: error " |
1997 | << cpp_strerror(ret) << dendl; | |
7c673cae FG |
1998 | goto umount_store; |
1999 | } | |
2000 | ||
2001 | umount_store: | |
11fdf7f2 TL |
2002 | if (ch) { |
2003 | ch.reset(); | |
2004 | } | |
7c673cae FG |
2005 | store->umount(); |
2006 | free_store: | |
2007 | delete store; | |
2008 | return ret; | |
2009 | } | |
2010 | ||
3efd9988 | 2011 | int OSD::write_meta(CephContext *cct, ObjectStore *store, uuid_d& cluster_fsid, uuid_d& osd_fsid, int whoami) |
7c673cae FG |
2012 | { |
2013 | char val[80]; | |
2014 | int r; | |
2015 | ||
2016 | snprintf(val, sizeof(val), "%s", CEPH_OSD_ONDISK_MAGIC); | |
2017 | r = store->write_meta("magic", val); | |
2018 | if (r < 0) | |
2019 | return r; | |
2020 | ||
2021 | snprintf(val, sizeof(val), "%d", whoami); | |
2022 | r = store->write_meta("whoami", val); | |
2023 | if (r < 0) | |
2024 | return r; | |
2025 | ||
2026 | cluster_fsid.print(val); | |
2027 | r = store->write_meta("ceph_fsid", val); | |
2028 | if (r < 0) | |
2029 | return r; | |
2030 | ||
11fdf7f2 | 2031 | string key = cct->_conf.get_val<string>("key"); |
3efd9988 FG |
2032 | if (key.size()) { |
2033 | r = store->write_meta("osd_key", key); | |
2034 | if (r < 0) | |
2035 | return r; | |
b32b8144 | 2036 | } else { |
11fdf7f2 | 2037 | string keyfile = cct->_conf.get_val<string>("keyfile"); |
b32b8144 FG |
2038 | if (!keyfile.empty()) { |
2039 | bufferlist keybl; | |
2040 | string err; | |
11fdf7f2 | 2041 | r = keybl.read_file(keyfile.c_str(), &err); |
b32b8144 FG |
2042 | if (r < 0) { |
2043 | derr << __func__ << " failed to read keyfile " << keyfile << ": " | |
2044 | << err << ": " << cpp_strerror(r) << dendl; | |
2045 | return r; | |
2046 | } | |
2047 | r = store->write_meta("osd_key", keybl.to_str()); | |
2048 | if (r < 0) | |
2049 | return r; | |
2050 | } | |
3efd9988 FG |
2051 | } |
2052 | ||
7c673cae FG |
2053 | r = store->write_meta("ready", "ready"); |
2054 | if (r < 0) | |
2055 | return r; | |
2056 | ||
2057 | return 0; | |
2058 | } | |
2059 | ||
11fdf7f2 TL |
2060 | int OSD::peek_meta(ObjectStore *store, |
2061 | std::string *magic, | |
2062 | uuid_d *cluster_fsid, | |
2063 | uuid_d *osd_fsid, | |
2064 | int *whoami, | |
9f95a23c | 2065 | ceph_release_t *require_osd_release) |
7c673cae FG |
2066 | { |
2067 | string val; | |
2068 | ||
2069 | int r = store->read_meta("magic", &val); | |
2070 | if (r < 0) | |
2071 | return r; | |
11fdf7f2 | 2072 | *magic = val; |
7c673cae FG |
2073 | |
2074 | r = store->read_meta("whoami", &val); | |
2075 | if (r < 0) | |
2076 | return r; | |
11fdf7f2 | 2077 | *whoami = atoi(val.c_str()); |
7c673cae FG |
2078 | |
2079 | r = store->read_meta("ceph_fsid", &val); | |
2080 | if (r < 0) | |
2081 | return r; | |
11fdf7f2 | 2082 | r = cluster_fsid->parse(val.c_str()); |
7c673cae FG |
2083 | if (!r) |
2084 | return -EINVAL; | |
2085 | ||
2086 | r = store->read_meta("fsid", &val); | |
2087 | if (r < 0) { | |
11fdf7f2 | 2088 | *osd_fsid = uuid_d(); |
7c673cae | 2089 | } else { |
11fdf7f2 | 2090 | r = osd_fsid->parse(val.c_str()); |
7c673cae FG |
2091 | if (!r) |
2092 | return -EINVAL; | |
2093 | } | |
2094 | ||
11fdf7f2 TL |
2095 | r = store->read_meta("require_osd_release", &val); |
2096 | if (r >= 0) { | |
9f95a23c | 2097 | *require_osd_release = ceph_release_from_name(val); |
11fdf7f2 TL |
2098 | } |
2099 | ||
7c673cae FG |
2100 | return 0; |
2101 | } | |
2102 | ||
2103 | ||
2104 | #undef dout_prefix | |
2105 | #define dout_prefix _prefix(_dout, whoami, get_osdmap_epoch()) | |
2106 | ||
2107 | // cons/des | |
2108 | ||
2109 | OSD::OSD(CephContext *cct_, ObjectStore *store_, | |
2110 | int id, | |
2111 | Messenger *internal_messenger, | |
2112 | Messenger *external_messenger, | |
2113 | Messenger *hb_client_front, | |
2114 | Messenger *hb_client_back, | |
2115 | Messenger *hb_front_serverm, | |
2116 | Messenger *hb_back_serverm, | |
2117 | Messenger *osdc_messenger, | |
2118 | MonClient *mc, | |
2119 | const std::string &dev, const std::string &jdev) : | |
2120 | Dispatcher(cct_), | |
7c673cae | 2121 | tick_timer(cct, osd_lock), |
7c673cae | 2122 | tick_timer_without_osd_lock(cct, tick_timer_lock), |
11fdf7f2 | 2123 | gss_ktfile_client(cct->_conf.get_val<std::string>("gss_ktab_client_file")), |
7c673cae FG |
2124 | cluster_messenger(internal_messenger), |
2125 | client_messenger(external_messenger), | |
2126 | objecter_messenger(osdc_messenger), | |
2127 | monc(mc), | |
9f95a23c | 2128 | mgrc(cct_, client_messenger, &mc->monmap), |
7c673cae FG |
2129 | logger(NULL), |
2130 | recoverystate_perf(NULL), | |
2131 | store(store_), | |
2132 | log_client(cct, client_messenger, &mc->monmap, LogClient::NO_FLAGS), | |
2133 | clog(log_client.create_channel()), | |
2134 | whoami(id), | |
2135 | dev_path(dev), journal_path(jdev), | |
31f18b77 | 2136 | store_is_rotational(store->is_rotational()), |
7c673cae FG |
2137 | trace_endpoint("0.0.0.0", 0, "osd"), |
2138 | asok_hook(NULL), | |
11fdf7f2 TL |
2139 | m_osd_pg_epoch_max_lag_factor(cct->_conf.get_val<double>( |
2140 | "osd_pg_epoch_max_lag_factor")), | |
7c673cae | 2141 | osd_compat(get_osd_compat_set()), |
7c673cae | 2142 | osd_op_tp(cct, "OSD::osd_op_tp", "tp_osd_tp", |
31f18b77 | 2143 | get_num_op_threads()), |
7c673cae FG |
2144 | heartbeat_stop(false), |
2145 | heartbeat_need_update(true), | |
2146 | hb_front_client_messenger(hb_client_front), | |
2147 | hb_back_client_messenger(hb_client_back), | |
2148 | hb_front_server_messenger(hb_front_serverm), | |
2149 | hb_back_server_messenger(hb_back_serverm), | |
2150 | daily_loadavg(0.0), | |
2151 | heartbeat_thread(this), | |
2152 | heartbeat_dispatcher(this), | |
2153 | op_tracker(cct, cct->_conf->osd_enable_op_tracker, | |
2154 | cct->_conf->osd_num_op_tracker_shard), | |
2155 | test_ops_hook(NULL), | |
7c673cae | 2156 | op_shardedwq( |
7c673cae FG |
2157 | this, |
2158 | cct->_conf->osd_op_thread_timeout, | |
2159 | cct->_conf->osd_op_thread_suicide_timeout, | |
2160 | &osd_op_tp), | |
7c673cae | 2161 | last_pg_create_epoch(0), |
11fdf7f2 | 2162 | boot_finisher(cct), |
7c673cae FG |
2163 | up_thru_wanted(0), |
2164 | requested_full_first(0), | |
2165 | requested_full_last(0), | |
7c673cae FG |
2166 | service(this) |
2167 | { | |
11fdf7f2 TL |
2168 | |
2169 | if (!gss_ktfile_client.empty()) { | |
2170 | // Assert we can export environment variable | |
2171 | /* | |
2172 | The default client keytab is used, if it is present and readable, | |
2173 | to automatically obtain initial credentials for GSSAPI client | |
2174 | applications. The principal name of the first entry in the client | |
2175 | keytab is used by default when obtaining initial credentials. | |
2176 | 1. The KRB5_CLIENT_KTNAME environment variable. | |
2177 | 2. The default_client_keytab_name profile variable in [libdefaults]. | |
2178 | 3. The hardcoded default, DEFCKTNAME. | |
2179 | */ | |
2180 | const int32_t set_result(setenv("KRB5_CLIENT_KTNAME", | |
2181 | gss_ktfile_client.c_str(), 1)); | |
2182 | ceph_assert(set_result == 0); | |
2183 | } | |
2184 | ||
7c673cae FG |
2185 | monc->set_messenger(client_messenger); |
2186 | op_tracker.set_complaint_and_threshold(cct->_conf->osd_op_complaint_time, | |
2187 | cct->_conf->osd_op_log_threshold); | |
2188 | op_tracker.set_history_size_and_duration(cct->_conf->osd_op_history_size, | |
2189 | cct->_conf->osd_op_history_duration); | |
2190 | op_tracker.set_history_slow_op_size_and_threshold(cct->_conf->osd_op_history_slow_op_size, | |
2191 | cct->_conf->osd_op_history_slow_op_threshold); | |
9f95a23c | 2192 | ObjectCleanRegions::set_max_num_intervals(cct->_conf->osd_object_clean_region_max_num_intervals); |
7c673cae FG |
2193 | #ifdef WITH_BLKIN |
2194 | std::stringstream ss; | |
2195 | ss << "osd." << whoami; | |
2196 | trace_endpoint.copy_name(ss.str()); | |
2197 | #endif | |
11fdf7f2 TL |
2198 | |
2199 | // initialize shards | |
2200 | num_shards = get_num_op_shards(); | |
2201 | for (uint32_t i = 0; i < num_shards; i++) { | |
2202 | OSDShard *one_shard = new OSDShard( | |
2203 | i, | |
2204 | cct, | |
9f95a23c | 2205 | this); |
11fdf7f2 TL |
2206 | shards.push_back(one_shard); |
2207 | } | |
7c673cae FG |
2208 | } |
2209 | ||
2210 | OSD::~OSD() | |
2211 | { | |
11fdf7f2 TL |
2212 | while (!shards.empty()) { |
2213 | delete shards.back(); | |
2214 | shards.pop_back(); | |
2215 | } | |
7c673cae FG |
2216 | cct->get_perfcounters_collection()->remove(recoverystate_perf); |
2217 | cct->get_perfcounters_collection()->remove(logger); | |
2218 | delete recoverystate_perf; | |
2219 | delete logger; | |
2220 | delete store; | |
2221 | } | |
2222 | ||
91327a77 AA |
2223 | double OSD::get_tick_interval() const |
2224 | { | |
2225 | // vary +/- 5% to avoid scrub scheduling livelocks | |
2226 | constexpr auto delta = 0.05; | |
91327a77 | 2227 | return (OSD_TICK_INTERVAL * |
11fdf7f2 | 2228 | ceph::util::generate_random_number(1.0 - delta, 1.0 + delta)); |
91327a77 AA |
2229 | } |
2230 | ||
7c673cae FG |
2231 | void OSD::handle_signal(int signum) |
2232 | { | |
11fdf7f2 | 2233 | ceph_assert(signum == SIGINT || signum == SIGTERM); |
7c673cae FG |
2234 | derr << "*** Got signal " << sig_str(signum) << " ***" << dendl; |
2235 | shutdown(); | |
2236 | } | |
2237 | ||
2238 | int OSD::pre_init() | |
2239 | { | |
11fdf7f2 | 2240 | std::lock_guard lock(osd_lock); |
7c673cae FG |
2241 | if (is_stopping()) |
2242 | return 0; | |
2243 | ||
2244 | if (store->test_mount_in_use()) { | |
2245 | derr << "OSD::pre_init: object store '" << dev_path << "' is " | |
2246 | << "currently in use. (Is ceph-osd already running?)" << dendl; | |
2247 | return -EBUSY; | |
2248 | } | |
2249 | ||
11fdf7f2 TL |
2250 | cct->_conf.add_observer(this); |
2251 | return 0; | |
2252 | } | |
2253 | ||
2254 | int OSD::set_numa_affinity() | |
2255 | { | |
2256 | // storage numa node | |
2257 | int store_node = -1; | |
2258 | store->get_numa_node(&store_node, nullptr, nullptr); | |
2259 | if (store_node >= 0) { | |
2260 | dout(1) << __func__ << " storage numa node " << store_node << dendl; | |
2261 | } | |
2262 | ||
2263 | // check network numa node(s) | |
2264 | int front_node = -1, back_node = -1; | |
2265 | string front_iface = pick_iface( | |
2266 | cct, | |
2267 | client_messenger->get_myaddrs().front().get_sockaddr_storage()); | |
2268 | string back_iface = pick_iface( | |
2269 | cct, | |
2270 | cluster_messenger->get_myaddrs().front().get_sockaddr_storage()); | |
2271 | int r = get_iface_numa_node(front_iface, &front_node); | |
92f5a8d4 | 2272 | if (r >= 0 && front_node >= 0) { |
11fdf7f2 | 2273 | dout(1) << __func__ << " public network " << front_iface << " numa node " |
92f5a8d4 | 2274 | << front_node << dendl; |
11fdf7f2 | 2275 | r = get_iface_numa_node(back_iface, &back_node); |
92f5a8d4 | 2276 | if (r >= 0 && back_node >= 0) { |
11fdf7f2 TL |
2277 | dout(1) << __func__ << " cluster network " << back_iface << " numa node " |
2278 | << back_node << dendl; | |
2279 | if (front_node == back_node && | |
2280 | front_node == store_node) { | |
2281 | dout(1) << " objectstore and network numa nodes all match" << dendl; | |
2282 | if (g_conf().get_val<bool>("osd_numa_auto_affinity")) { | |
2283 | numa_node = front_node; | |
2284 | } | |
92f5a8d4 TL |
2285 | } else if (front_node != back_node) { |
2286 | dout(1) << __func__ << " public and cluster network numa nodes do not match" | |
2287 | << dendl; | |
11fdf7f2 TL |
2288 | } else { |
2289 | dout(1) << __func__ << " objectstore and network numa nodes do not match" | |
2290 | << dendl; | |
2291 | } | |
92f5a8d4 TL |
2292 | } else if (back_node == -2) { |
2293 | dout(1) << __func__ << " cluster network " << back_iface | |
2294 | << " ports numa nodes do not match" << dendl; | |
2295 | } else { | |
2296 | derr << __func__ << " unable to identify cluster interface '" << back_iface | |
2297 | << "' numa node: " << cpp_strerror(r) << dendl; | |
11fdf7f2 | 2298 | } |
92f5a8d4 TL |
2299 | } else if (front_node == -2) { |
2300 | dout(1) << __func__ << " public network " << front_iface | |
2301 | << " ports numa nodes do not match" << dendl; | |
11fdf7f2 TL |
2302 | } else { |
2303 | derr << __func__ << " unable to identify public interface '" << front_iface | |
2304 | << "' numa node: " << cpp_strerror(r) << dendl; | |
2305 | } | |
2306 | if (int node = g_conf().get_val<int64_t>("osd_numa_node"); node >= 0) { | |
2307 | // this takes precedence over the automagic logic above | |
2308 | numa_node = node; | |
2309 | } | |
2310 | if (numa_node >= 0) { | |
2311 | int r = get_numa_node_cpu_set(numa_node, &numa_cpu_set_size, &numa_cpu_set); | |
2312 | if (r < 0) { | |
2313 | dout(1) << __func__ << " unable to determine numa node " << numa_node | |
2314 | << " CPUs" << dendl; | |
2315 | numa_node = -1; | |
2316 | } else { | |
2317 | dout(1) << __func__ << " setting numa affinity to node " << numa_node | |
2318 | << " cpus " | |
2319 | << cpu_set_to_str_list(numa_cpu_set_size, &numa_cpu_set) | |
2320 | << dendl; | |
92f5a8d4 | 2321 | r = set_cpu_affinity_all_threads(numa_cpu_set_size, &numa_cpu_set); |
11fdf7f2 TL |
2322 | if (r < 0) { |
2323 | r = -errno; | |
2324 | derr << __func__ << " failed to set numa affinity: " << cpp_strerror(r) | |
2325 | << dendl; | |
2326 | numa_node = -1; | |
2327 | } | |
2328 | } | |
2329 | } else { | |
2330 | dout(1) << __func__ << " not setting numa affinity" << dendl; | |
2331 | } | |
7c673cae FG |
2332 | return 0; |
2333 | } | |
2334 | ||
2335 | // asok | |
2336 | ||
2337 | class OSDSocketHook : public AdminSocketHook { | |
2338 | OSD *osd; | |
2339 | public: | |
2340 | explicit OSDSocketHook(OSD *o) : osd(o) {} | |
9f95a23c TL |
2341 | int call(std::string_view prefix, const cmdmap_t& cmdmap, |
2342 | Formatter *f, | |
2343 | std::ostream& ss, | |
2344 | bufferlist& out) override { | |
2345 | ceph_abort("should use async hook"); | |
2346 | } | |
2347 | void call_async( | |
2348 | std::string_view prefix, | |
2349 | const cmdmap_t& cmdmap, | |
2350 | Formatter *f, | |
2351 | const bufferlist& inbl, | |
2352 | std::function<void(int,const std::string&,bufferlist&)> on_finish) override { | |
11fdf7f2 | 2353 | try { |
9f95a23c TL |
2354 | osd->asok_command(prefix, cmdmap, f, inbl, on_finish); |
2355 | } catch (const TOPNSPC::common::bad_cmd_get& e) { | |
2356 | bufferlist empty; | |
2357 | on_finish(-EINVAL, e.what(), empty); | |
11fdf7f2 | 2358 | } |
7c673cae FG |
2359 | } |
2360 | }; | |
2361 | ||
11fdf7f2 TL |
2362 | std::set<int64_t> OSD::get_mapped_pools() |
2363 | { | |
2364 | std::set<int64_t> pools; | |
2365 | std::vector<spg_t> pgids; | |
2366 | _get_pgids(&pgids); | |
2367 | for (const auto &pgid : pgids) { | |
2368 | pools.insert(pgid.pool()); | |
2369 | } | |
2370 | return pools; | |
2371 | } | |
2372 | ||
9f95a23c TL |
2373 | void OSD::asok_command( |
2374 | std::string_view prefix, const cmdmap_t& cmdmap, | |
2375 | Formatter *f, | |
2376 | const bufferlist& inbl, | |
2377 | std::function<void(int,const std::string&,bufferlist&)> on_finish) | |
7c673cae | 2378 | { |
9f95a23c TL |
2379 | int ret = 0; |
2380 | stringstream ss; // stderr error message stream | |
2381 | bufferlist outbl; // if empty at end, we'll dump formatter as output | |
2382 | ||
2383 | // --- PG commands are routed here to PG::do_command --- | |
2384 | if (prefix == "pg" || | |
2385 | prefix == "query" || | |
2386 | prefix == "mark_unfound_lost" || | |
2387 | prefix == "list_unfound" || | |
2388 | prefix == "scrub" || | |
2389 | prefix == "deep_scrub" | |
2390 | ) { | |
2391 | string pgidstr; | |
2392 | pg_t pgid; | |
2393 | if (!cmd_getval(cmdmap, "pgid", pgidstr)) { | |
2394 | ss << "no pgid specified"; | |
2395 | ret = -EINVAL; | |
2396 | goto out; | |
2397 | } | |
2398 | if (!pgid.parse(pgidstr.c_str())) { | |
2399 | ss << "couldn't parse pgid '" << pgidstr << "'"; | |
2400 | ret = -EINVAL; | |
2401 | goto out; | |
2402 | } | |
2403 | spg_t pcand; | |
2404 | PGRef pg; | |
2405 | if (get_osdmap()->get_primary_shard(pgid, &pcand) && | |
2406 | (pg = _lookup_lock_pg(pcand))) { | |
2407 | if (pg->is_primary()) { | |
2408 | cmdmap_t new_cmdmap = cmdmap; | |
2409 | try { | |
2410 | pg->do_command(prefix, new_cmdmap, inbl, on_finish); | |
2411 | pg->unlock(); | |
2412 | return; // the pg handler calls on_finish directly | |
2413 | } catch (const TOPNSPC::common::bad_cmd_get& e) { | |
2414 | pg->unlock(); | |
2415 | ss << e.what(); | |
2416 | ret = -EINVAL; | |
2417 | goto out; | |
2418 | } | |
2419 | } else { | |
2420 | ss << "not primary for pgid " << pgid; | |
2421 | // do not reply; they will get newer maps and realize they | |
2422 | // need to resend. | |
2423 | pg->unlock(); | |
2424 | ret = -EAGAIN; | |
2425 | goto out; | |
2426 | } | |
2427 | } else { | |
2428 | ss << "i don't have pgid " << pgid; | |
2429 | ret = -ENOENT; | |
2430 | } | |
2431 | } | |
2432 | ||
2433 | // --- OSD commands follow --- | |
2434 | ||
2435 | else if (prefix == "status") { | |
2436 | lock_guard l(osd_lock); | |
7c673cae FG |
2437 | f->open_object_section("status"); |
2438 | f->dump_stream("cluster_fsid") << superblock.cluster_fsid; | |
2439 | f->dump_stream("osd_fsid") << superblock.osd_fsid; | |
2440 | f->dump_unsigned("whoami", superblock.whoami); | |
2441 | f->dump_string("state", get_state_name(get_state())); | |
2442 | f->dump_unsigned("oldest_map", superblock.oldest_map); | |
2443 | f->dump_unsigned("newest_map", superblock.newest_map); | |
11fdf7f2 | 2444 | f->dump_unsigned("num_pgs", num_pgs); |
7c673cae | 2445 | f->close_section(); |
9f95a23c | 2446 | } else if (prefix == "flush_journal") { |
7c673cae | 2447 | store->flush_journal(); |
9f95a23c TL |
2448 | } else if (prefix == "dump_ops_in_flight" || |
2449 | prefix == "ops" || | |
2450 | prefix == "dump_blocked_ops" || | |
2451 | prefix == "dump_historic_ops" || | |
2452 | prefix == "dump_historic_ops_by_duration" || | |
2453 | prefix == "dump_historic_slow_ops") { | |
c07f9fc5 FG |
2454 | |
2455 | const string error_str = "op_tracker tracking is not enabled now, so no ops are tracked currently, \ | |
2456 | even those get stuck. Please enable \"osd_enable_op_tracker\", and the tracker \ | |
2457 | will start to track new ops received afterwards."; | |
2458 | ||
2459 | set<string> filters; | |
2460 | vector<string> filter_str; | |
9f95a23c | 2461 | if (cmd_getval(cmdmap, "filterstr", filter_str)) { |
c07f9fc5 FG |
2462 | copy(filter_str.begin(), filter_str.end(), |
2463 | inserter(filters, filters.end())); | |
2464 | } | |
2465 | ||
9f95a23c TL |
2466 | if (prefix == "dump_ops_in_flight" || |
2467 | prefix == "ops") { | |
c07f9fc5 FG |
2468 | if (!op_tracker.dump_ops_in_flight(f, false, filters)) { |
2469 | ss << error_str; | |
9f95a23c TL |
2470 | ret = -EINVAL; |
2471 | goto out; | |
c07f9fc5 FG |
2472 | } |
2473 | } | |
9f95a23c | 2474 | if (prefix == "dump_blocked_ops") { |
c07f9fc5 FG |
2475 | if (!op_tracker.dump_ops_in_flight(f, true, filters)) { |
2476 | ss << error_str; | |
9f95a23c TL |
2477 | ret = -EINVAL; |
2478 | goto out; | |
c07f9fc5 FG |
2479 | } |
2480 | } | |
9f95a23c | 2481 | if (prefix == "dump_historic_ops") { |
c07f9fc5 FG |
2482 | if (!op_tracker.dump_historic_ops(f, false, filters)) { |
2483 | ss << error_str; | |
9f95a23c TL |
2484 | ret = -EINVAL; |
2485 | goto out; | |
c07f9fc5 FG |
2486 | } |
2487 | } | |
9f95a23c | 2488 | if (prefix == "dump_historic_ops_by_duration") { |
c07f9fc5 FG |
2489 | if (!op_tracker.dump_historic_ops(f, true, filters)) { |
2490 | ss << error_str; | |
9f95a23c TL |
2491 | ret = -EINVAL; |
2492 | goto out; | |
c07f9fc5 FG |
2493 | } |
2494 | } | |
9f95a23c | 2495 | if (prefix == "dump_historic_slow_ops") { |
c07f9fc5 FG |
2496 | if (!op_tracker.dump_historic_slow_ops(f, filters)) { |
2497 | ss << error_str; | |
9f95a23c TL |
2498 | ret = -EINVAL; |
2499 | goto out; | |
c07f9fc5 | 2500 | } |
7c673cae | 2501 | } |
9f95a23c | 2502 | } else if (prefix == "dump_op_pq_state") { |
7c673cae FG |
2503 | f->open_object_section("pq"); |
2504 | op_shardedwq.dump(f); | |
2505 | f->close_section(); | |
9f95a23c | 2506 | } else if (prefix == "dump_blacklist") { |
7c673cae FG |
2507 | list<pair<entity_addr_t,utime_t> > bl; |
2508 | OSDMapRef curmap = service.get_osdmap(); | |
2509 | ||
2510 | f->open_array_section("blacklist"); | |
2511 | curmap->get_blacklist(&bl); | |
2512 | for (list<pair<entity_addr_t,utime_t> >::iterator it = bl.begin(); | |
2513 | it != bl.end(); ++it) { | |
224ce89b | 2514 | f->open_object_section("entry"); |
7c673cae FG |
2515 | f->open_object_section("entity_addr_t"); |
2516 | it->first.dump(f); | |
2517 | f->close_section(); //entity_addr_t | |
2518 | it->second.localtime(f->dump_stream("expire_time")); | |
2519 | f->close_section(); //entry | |
2520 | } | |
2521 | f->close_section(); //blacklist | |
9f95a23c | 2522 | } else if (prefix == "dump_watchers") { |
7c673cae FG |
2523 | list<obj_watch_item_t> watchers; |
2524 | // scan pg's | |
11fdf7f2 TL |
2525 | vector<PGRef> pgs; |
2526 | _get_pgs(&pgs); | |
2527 | for (auto& pg : pgs) { | |
2528 | list<obj_watch_item_t> pg_watchers; | |
2529 | pg->get_watchers(&pg_watchers); | |
2530 | watchers.splice(watchers.end(), pg_watchers); | |
7c673cae FG |
2531 | } |
2532 | ||
2533 | f->open_array_section("watchers"); | |
2534 | for (list<obj_watch_item_t>::iterator it = watchers.begin(); | |
2535 | it != watchers.end(); ++it) { | |
2536 | ||
224ce89b | 2537 | f->open_object_section("watch"); |
7c673cae FG |
2538 | |
2539 | f->dump_string("namespace", it->obj.nspace); | |
2540 | f->dump_string("object", it->obj.oid.name); | |
2541 | ||
2542 | f->open_object_section("entity_name"); | |
2543 | it->wi.name.dump(f); | |
2544 | f->close_section(); //entity_name_t | |
2545 | ||
224ce89b WB |
2546 | f->dump_unsigned("cookie", it->wi.cookie); |
2547 | f->dump_unsigned("timeout", it->wi.timeout_seconds); | |
7c673cae FG |
2548 | |
2549 | f->open_object_section("entity_addr_t"); | |
2550 | it->wi.addr.dump(f); | |
2551 | f->close_section(); //entity_addr_t | |
2552 | ||
2553 | f->close_section(); //watch | |
2554 | } | |
2555 | ||
2556 | f->close_section(); //watchers | |
9f95a23c | 2557 | } else if (prefix == "dump_recovery_reservations") { |
7c673cae FG |
2558 | f->open_object_section("reservations"); |
2559 | f->open_object_section("local_reservations"); | |
2560 | service.local_reserver.dump(f); | |
2561 | f->close_section(); | |
2562 | f->open_object_section("remote_reservations"); | |
2563 | service.remote_reserver.dump(f); | |
2564 | f->close_section(); | |
2565 | f->close_section(); | |
9f95a23c | 2566 | } else if (prefix == "dump_scrub_reservations") { |
eafe8130 TL |
2567 | f->open_object_section("scrub_reservations"); |
2568 | service.dump_scrub_reservations(f); | |
2569 | f->close_section(); | |
9f95a23c | 2570 | } else if (prefix == "get_latest_osdmap") { |
7c673cae | 2571 | get_latest_osdmap(); |
9f95a23c | 2572 | } else if (prefix == "set_heap_property") { |
7c673cae FG |
2573 | string property; |
2574 | int64_t value = 0; | |
2575 | string error; | |
2576 | bool success = false; | |
9f95a23c | 2577 | if (!cmd_getval(cmdmap, "property", property)) { |
7c673cae FG |
2578 | error = "unable to get property"; |
2579 | success = false; | |
9f95a23c | 2580 | } else if (!cmd_getval(cmdmap, "value", value)) { |
7c673cae FG |
2581 | error = "unable to get value"; |
2582 | success = false; | |
2583 | } else if (value < 0) { | |
2584 | error = "negative value not allowed"; | |
2585 | success = false; | |
2586 | } else if (!ceph_heap_set_numeric_property(property.c_str(), (size_t)value)) { | |
2587 | error = "invalid property"; | |
2588 | success = false; | |
2589 | } else { | |
2590 | success = true; | |
2591 | } | |
2592 | f->open_object_section("result"); | |
2593 | f->dump_string("error", error); | |
2594 | f->dump_bool("success", success); | |
2595 | f->close_section(); | |
9f95a23c | 2596 | } else if (prefix == "get_heap_property") { |
7c673cae FG |
2597 | string property; |
2598 | size_t value = 0; | |
2599 | string error; | |
2600 | bool success = false; | |
9f95a23c | 2601 | if (!cmd_getval(cmdmap, "property", property)) { |
7c673cae FG |
2602 | error = "unable to get property"; |
2603 | success = false; | |
2604 | } else if (!ceph_heap_get_numeric_property(property.c_str(), &value)) { | |
2605 | error = "invalid property"; | |
2606 | success = false; | |
2607 | } else { | |
2608 | success = true; | |
2609 | } | |
2610 | f->open_object_section("result"); | |
2611 | f->dump_string("error", error); | |
2612 | f->dump_bool("success", success); | |
2613 | f->dump_int("value", value); | |
2614 | f->close_section(); | |
9f95a23c | 2615 | } else if (prefix == "dump_objectstore_kv_stats") { |
7c673cae | 2616 | store->get_db_statistics(f); |
9f95a23c | 2617 | } else if (prefix == "dump_scrubs") { |
7c673cae | 2618 | service.dumps_scrub(f); |
9f95a23c | 2619 | } else if (prefix == "calc_objectstore_db_histogram") { |
7c673cae | 2620 | store->generate_db_histogram(f); |
9f95a23c | 2621 | } else if (prefix == "flush_store_cache") { |
11fdf7f2 | 2622 | store->flush_cache(&ss); |
9f95a23c | 2623 | } else if (prefix == "dump_pgstate_history") { |
7c673cae | 2624 | f->open_object_section("pgstate_history"); |
9f95a23c | 2625 | f->open_array_section("pgs"); |
11fdf7f2 TL |
2626 | vector<PGRef> pgs; |
2627 | _get_pgs(&pgs); | |
2628 | for (auto& pg : pgs) { | |
9f95a23c | 2629 | f->open_object_section("pg"); |
11fdf7f2 | 2630 | f->dump_stream("pg") << pg->pg_id; |
9f95a23c | 2631 | f->dump_string("currently", pg->get_current_state()); |
11fdf7f2 | 2632 | pg->dump_pgstate_history(f); |
9f95a23c | 2633 | f->close_section(); |
7c673cae FG |
2634 | } |
2635 | f->close_section(); | |
9f95a23c TL |
2636 | f->close_section(); |
2637 | } else if (prefix == "compact") { | |
224ce89b WB |
2638 | dout(1) << "triggering manual compaction" << dendl; |
2639 | auto start = ceph::coarse_mono_clock::now(); | |
2640 | store->compact(); | |
2641 | auto end = ceph::coarse_mono_clock::now(); | |
11fdf7f2 | 2642 | double duration = std::chrono::duration<double>(end-start).count(); |
224ce89b | 2643 | dout(1) << "finished manual compaction in " |
11fdf7f2 | 2644 | << duration |
224ce89b WB |
2645 | << " seconds" << dendl; |
2646 | f->open_object_section("compact_result"); | |
11fdf7f2 TL |
2647 | f->dump_float("elapsed_time", duration); |
2648 | f->close_section(); | |
9f95a23c | 2649 | } else if (prefix == "get_mapped_pools") { |
11fdf7f2 TL |
2650 | f->open_array_section("mapped_pools"); |
2651 | set<int64_t> poollist = get_mapped_pools(); | |
2652 | for (auto pool : poollist) { | |
2653 | f->dump_int("pool_id", pool); | |
2654 | } | |
2655 | f->close_section(); | |
9f95a23c | 2656 | } else if (prefix == "smart") { |
11fdf7f2 | 2657 | string devid; |
9f95a23c TL |
2658 | cmd_getval(cmdmap, "devid", devid); |
2659 | ostringstream out; | |
2660 | probe_smart(devid, out); | |
2661 | outbl.append(out.str()); | |
2662 | } else if (prefix == "list_devices") { | |
11fdf7f2 TL |
2663 | set<string> devnames; |
2664 | store->get_devices(&devnames); | |
9f95a23c | 2665 | f->open_array_section("list_devices"); |
11fdf7f2 TL |
2666 | for (auto dev : devnames) { |
2667 | if (dev.find("dm-") == 0) { | |
2668 | continue; | |
2669 | } | |
9f95a23c TL |
2670 | string err; |
2671 | f->open_object_section("device"); | |
11fdf7f2 | 2672 | f->dump_string("device", "/dev/" + dev); |
9f95a23c TL |
2673 | f->dump_string("device_id", get_device_id(dev, &err)); |
2674 | f->close_section(); | |
11fdf7f2 | 2675 | } |
224ce89b | 2676 | f->close_section(); |
9f95a23c TL |
2677 | } else if (prefix == "send_beacon") { |
2678 | lock_guard l(osd_lock); | |
11fdf7f2 TL |
2679 | if (is_active()) { |
2680 | send_beacon(ceph::coarse_mono_clock::now()); | |
2681 | } | |
9f95a23c TL |
2682 | } |
2683 | ||
2684 | else if (prefix == "cluster_log") { | |
2685 | vector<string> msg; | |
2686 | cmd_getval(cmdmap, "message", msg); | |
2687 | if (msg.empty()) { | |
2688 | ret = -EINVAL; | |
2689 | ss << "ignoring empty log message"; | |
2690 | goto out; | |
2691 | } | |
2692 | string message = msg.front(); | |
2693 | for (vector<string>::iterator a = ++msg.begin(); a != msg.end(); ++a) | |
2694 | message += " " + *a; | |
2695 | string lvl; | |
2696 | cmd_getval(cmdmap, "level", lvl); | |
2697 | clog_type level = string_to_clog_type(lvl); | |
2698 | if (level < 0) { | |
2699 | ret = -EINVAL; | |
2700 | ss << "unknown level '" << lvl << "'"; | |
2701 | goto out; | |
2702 | } | |
2703 | clog->do_log(level, message); | |
2704 | } | |
2705 | ||
2706 | else if (prefix == "bench") { | |
2707 | lock_guard l(osd_lock); | |
2708 | int64_t count; | |
2709 | int64_t bsize; | |
2710 | int64_t osize, onum; | |
2711 | // default count 1G, size 4MB | |
2712 | cmd_getval(cmdmap, "count", count, (int64_t)1 << 30); | |
2713 | cmd_getval(cmdmap, "size", bsize, (int64_t)4 << 20); | |
2714 | cmd_getval(cmdmap, "object_size", osize, (int64_t)0); | |
2715 | cmd_getval(cmdmap, "object_num", onum, (int64_t)0); | |
2716 | ||
2717 | uint32_t duration = cct->_conf->osd_bench_duration; | |
2718 | ||
2719 | if (bsize > (int64_t) cct->_conf->osd_bench_max_block_size) { | |
2720 | // let us limit the block size because the next checks rely on it | |
2721 | // having a sane value. If we allow any block size to be set things | |
2722 | // can still go sideways. | |
2723 | ss << "block 'size' values are capped at " | |
2724 | << byte_u_t(cct->_conf->osd_bench_max_block_size) << ". If you wish to use" | |
2725 | << " a higher value, please adjust 'osd_bench_max_block_size'"; | |
2726 | ret = -EINVAL; | |
2727 | goto out; | |
2728 | } else if (bsize < (int64_t) (1 << 20)) { | |
2729 | // entering the realm of small block sizes. | |
2730 | // limit the count to a sane value, assuming a configurable amount of | |
2731 | // IOPS and duration, so that the OSD doesn't get hung up on this, | |
2732 | // preventing timeouts from going off | |
2733 | int64_t max_count = | |
2734 | bsize * duration * cct->_conf->osd_bench_small_size_max_iops; | |
2735 | if (count > max_count) { | |
2736 | ss << "'count' values greater than " << max_count | |
2737 | << " for a block size of " << byte_u_t(bsize) << ", assuming " | |
2738 | << cct->_conf->osd_bench_small_size_max_iops << " IOPS," | |
2739 | << " for " << duration << " seconds," | |
2740 | << " can cause ill effects on osd. " | |
2741 | << " Please adjust 'osd_bench_small_size_max_iops' with a higher" | |
2742 | << " value if you wish to use a higher 'count'."; | |
2743 | ret = -EINVAL; | |
2744 | goto out; | |
eafe8130 TL |
2745 | } |
2746 | } else { | |
9f95a23c TL |
2747 | // 1MB block sizes are big enough so that we get more stuff done. |
2748 | // However, to avoid the osd from getting hung on this and having | |
2749 | // timers being triggered, we are going to limit the count assuming | |
2750 | // a configurable throughput and duration. | |
2751 | // NOTE: max_count is the total amount of bytes that we believe we | |
2752 | // will be able to write during 'duration' for the given | |
2753 | // throughput. The block size hardly impacts this unless it's | |
2754 | // way too big. Given we already check how big the block size | |
2755 | // is, it's safe to assume everything will check out. | |
2756 | int64_t max_count = | |
2757 | cct->_conf->osd_bench_large_size_max_throughput * duration; | |
2758 | if (count > max_count) { | |
2759 | ss << "'count' values greater than " << max_count | |
2760 | << " for a block size of " << byte_u_t(bsize) << ", assuming " | |
2761 | << byte_u_t(cct->_conf->osd_bench_large_size_max_throughput) << "/s," | |
2762 | << " for " << duration << " seconds," | |
2763 | << " can cause ill effects on osd. " | |
2764 | << " Please adjust 'osd_bench_large_size_max_throughput'" | |
2765 | << " with a higher value if you wish to use a higher 'count'."; | |
2766 | ret = -EINVAL; | |
2767 | goto out; | |
2768 | } | |
eafe8130 | 2769 | } |
eafe8130 | 2770 | |
9f95a23c TL |
2771 | if (osize && bsize > osize) |
2772 | bsize = osize; | |
eafe8130 | 2773 | |
9f95a23c TL |
2774 | dout(1) << " bench count " << count |
2775 | << " bsize " << byte_u_t(bsize) << dendl; | |
eafe8130 | 2776 | |
9f95a23c TL |
2777 | ObjectStore::Transaction cleanupt; |
2778 | ||
2779 | if (osize && onum) { | |
2780 | bufferlist bl; | |
2781 | bufferptr bp(osize); | |
2782 | bp.zero(); | |
2783 | bl.push_back(std::move(bp)); | |
2784 | bl.rebuild_page_aligned(); | |
2785 | for (int i=0; i<onum; ++i) { | |
2786 | char nm[30]; | |
2787 | snprintf(nm, sizeof(nm), "disk_bw_test_%d", i); | |
2788 | object_t oid(nm); | |
2789 | hobject_t soid(sobject_t(oid, 0)); | |
2790 | ObjectStore::Transaction t; | |
2791 | t.write(coll_t(), ghobject_t(soid), 0, osize, bl); | |
2792 | store->queue_transaction(service.meta_ch, std::move(t), NULL); | |
2793 | cleanupt.remove(coll_t(), ghobject_t(soid)); | |
2794 | } | |
2795 | } | |
2796 | ||
2797 | bufferlist bl; | |
2798 | bufferptr bp(bsize); | |
2799 | bp.zero(); | |
2800 | bl.push_back(std::move(bp)); | |
2801 | bl.rebuild_page_aligned(); | |
2802 | ||
2803 | { | |
2804 | C_SaferCond waiter; | |
2805 | if (!service.meta_ch->flush_commit(&waiter)) { | |
2806 | waiter.wait(); | |
2807 | } | |
2808 | } | |
2809 | ||
2810 | utime_t start = ceph_clock_now(); | |
2811 | for (int64_t pos = 0; pos < count; pos += bsize) { | |
2812 | char nm[30]; | |
2813 | unsigned offset = 0; | |
2814 | if (onum && osize) { | |
2815 | snprintf(nm, sizeof(nm), "disk_bw_test_%d", (int)(rand() % onum)); | |
2816 | offset = rand() % (osize / bsize) * bsize; | |
2817 | } else { | |
2818 | snprintf(nm, sizeof(nm), "disk_bw_test_%lld", (long long)pos); | |
2819 | } | |
2820 | object_t oid(nm); | |
2821 | hobject_t soid(sobject_t(oid, 0)); | |
2822 | ObjectStore::Transaction t; | |
2823 | t.write(coll_t::meta(), ghobject_t(soid), offset, bsize, bl); | |
2824 | store->queue_transaction(service.meta_ch, std::move(t), NULL); | |
2825 | if (!onum || !osize) | |
2826 | cleanupt.remove(coll_t::meta(), ghobject_t(soid)); | |
2827 | } | |
2828 | ||
2829 | { | |
2830 | C_SaferCond waiter; | |
2831 | if (!service.meta_ch->flush_commit(&waiter)) { | |
2832 | waiter.wait(); | |
2833 | } | |
2834 | } | |
2835 | utime_t end = ceph_clock_now(); | |
2836 | ||
2837 | // clean up | |
2838 | store->queue_transaction(service.meta_ch, std::move(cleanupt), NULL); | |
2839 | { | |
2840 | C_SaferCond waiter; | |
2841 | if (!service.meta_ch->flush_commit(&waiter)) { | |
2842 | waiter.wait(); | |
2843 | } | |
2844 | } | |
2845 | ||
2846 | double elapsed = end - start; | |
2847 | double rate = count / elapsed; | |
2848 | double iops = rate / bsize; | |
2849 | f->open_object_section("osd_bench_results"); | |
2850 | f->dump_int("bytes_written", count); | |
2851 | f->dump_int("blocksize", bsize); | |
2852 | f->dump_float("elapsed_sec", elapsed); | |
2853 | f->dump_float("bytes_per_sec", rate); | |
2854 | f->dump_float("iops", iops); | |
2855 | f->close_section(); | |
2856 | } | |
2857 | ||
2858 | else if (prefix == "flush_pg_stats") { | |
2859 | mgrc.send_pgstats(); | |
2860 | f->dump_unsigned("stat_seq", service.get_osd_stat_seq()); | |
2861 | } | |
2862 | ||
2863 | else if (prefix == "heap") { | |
2864 | ret = ceph::osd_cmds::heap(*cct, cmdmap, *f, ss); | |
2865 | } | |
2866 | ||
2867 | else if (prefix == "debug dump_missing") { | |
2868 | f->open_array_section("pgs"); | |
2869 | vector<PGRef> pgs; | |
2870 | _get_pgs(&pgs); | |
2871 | for (auto& pg : pgs) { | |
2872 | string s = stringify(pg->pg_id); | |
2873 | f->open_array_section(s.c_str()); | |
2874 | pg->lock(); | |
2875 | pg->dump_missing(f); | |
2876 | pg->unlock(); | |
2877 | f->close_section(); | |
2878 | } | |
2879 | f->close_section(); | |
2880 | } | |
2881 | ||
2882 | else if (prefix == "debug kick_recovery_wq") { | |
2883 | int64_t delay; | |
2884 | cmd_getval(cmdmap, "delay", delay); | |
2885 | ostringstream oss; | |
2886 | oss << delay; | |
2887 | ret = cct->_conf.set_val("osd_recovery_delay_start", oss.str().c_str()); | |
2888 | if (ret != 0) { | |
2889 | ss << "kick_recovery_wq: error setting " | |
2890 | << "osd_recovery_delay_start to '" << delay << "': error " | |
2891 | << ret; | |
2892 | goto out; | |
2893 | } | |
2894 | cct->_conf.apply_changes(nullptr); | |
2895 | ss << "kicking recovery queue. set osd_recovery_delay_start " | |
2896 | << "to " << cct->_conf->osd_recovery_delay_start; | |
2897 | } | |
2898 | ||
2899 | else if (prefix == "cpu_profiler") { | |
2900 | ostringstream ds; | |
2901 | string arg; | |
2902 | cmd_getval(cmdmap, "arg", arg); | |
2903 | vector<string> argvec; | |
2904 | get_str_vec(arg, argvec); | |
2905 | cpu_profiler_handle_command(argvec, ds); | |
2906 | outbl.append(ds.str()); | |
2907 | } | |
2908 | ||
2909 | else if (prefix == "dump_pg_recovery_stats") { | |
2910 | lock_guard l(osd_lock); | |
2911 | pg_recovery_stats.dump_formatted(f); | |
2912 | } | |
2913 | ||
2914 | else if (prefix == "reset_pg_recovery_stats") { | |
2915 | lock_guard l(osd_lock); | |
2916 | pg_recovery_stats.reset(); | |
2917 | } | |
2918 | ||
2919 | else if (prefix == "perf histogram dump") { | |
2920 | std::string logger; | |
2921 | std::string counter; | |
2922 | cmd_getval(cmdmap, "logger", logger); | |
2923 | cmd_getval(cmdmap, "counter", counter); | |
2924 | cct->get_perfcounters_collection()->dump_formatted_histograms( | |
2925 | f, false, logger, counter); | |
2926 | } | |
2927 | ||
2928 | else if (prefix == "cache drop") { | |
2929 | lock_guard l(osd_lock); | |
2930 | dout(20) << "clearing all caches" << dendl; | |
2931 | // Clear the objectstore's cache - onode and buffer for Bluestore, | |
2932 | // system's pagecache for Filestore | |
2933 | ret = store->flush_cache(&ss); | |
2934 | if (ret < 0) { | |
2935 | ss << "Error flushing objectstore cache: " << cpp_strerror(ret); | |
2936 | goto out; | |
2937 | } | |
2938 | // Clear the objectcontext cache (per PG) | |
2939 | vector<PGRef> pgs; | |
2940 | _get_pgs(&pgs); | |
2941 | for (auto& pg: pgs) { | |
2942 | pg->clear_cache(); | |
2943 | } | |
2944 | } | |
2945 | ||
2946 | else if (prefix == "cache status") { | |
2947 | lock_guard l(osd_lock); | |
2948 | int obj_ctx_count = 0; | |
2949 | vector<PGRef> pgs; | |
2950 | _get_pgs(&pgs); | |
2951 | for (auto& pg: pgs) { | |
2952 | obj_ctx_count += pg->get_cache_obj_count(); | |
2953 | } | |
2954 | f->open_object_section("cache_status"); | |
2955 | f->dump_int("object_ctx", obj_ctx_count); | |
2956 | store->dump_cache_stats(f); | |
2957 | f->close_section(); | |
2958 | } | |
2959 | ||
2960 | else if (prefix == "scrub_purged_snaps") { | |
2961 | lock_guard l(osd_lock); | |
2962 | scrub_purged_snaps(); | |
2963 | } | |
2964 | ||
2965 | else if (prefix == "dump_osd_network") { | |
2966 | lock_guard l(osd_lock); | |
2967 | int64_t value = 0; | |
2968 | if (!(cmd_getval(cmdmap, "value", value))) { | |
2969 | // Convert milliseconds to microseconds | |
2970 | value = static_cast<double>(g_conf().get_val<double>( | |
2971 | "mon_warn_on_slow_ping_time")) * 1000; | |
2972 | if (value == 0) { | |
2973 | double ratio = g_conf().get_val<double>("mon_warn_on_slow_ping_ratio"); | |
2974 | value = g_conf().get_val<int64_t>("osd_heartbeat_grace"); | |
2975 | value *= 1000000 * ratio; // Seconds of grace to microseconds at ratio | |
2976 | } | |
2977 | } else { | |
2978 | // Convert user input to microseconds | |
2979 | value *= 1000; | |
2980 | } | |
2981 | if (value < 0) value = 0; | |
2982 | ||
2983 | struct osd_ping_time_t { | |
2984 | uint32_t pingtime; | |
2985 | int to; | |
2986 | bool back; | |
2987 | std::array<uint32_t,3> times; | |
2988 | std::array<uint32_t,3> min; | |
2989 | std::array<uint32_t,3> max; | |
2990 | uint32_t last; | |
2991 | uint32_t last_update; | |
2992 | ||
2993 | bool operator<(const osd_ping_time_t& rhs) const { | |
2994 | if (pingtime < rhs.pingtime) | |
2995 | return true; | |
2996 | if (pingtime > rhs.pingtime) | |
2997 | return false; | |
2998 | if (to < rhs.to) | |
2999 | return true; | |
3000 | if (to > rhs.to) | |
3001 | return false; | |
3002 | return back; | |
3003 | } | |
3004 | }; | |
3005 | ||
3006 | set<osd_ping_time_t> sorted; | |
3007 | // Get pingtimes under lock and not on the stack | |
eafe8130 TL |
3008 | map<int, osd_stat_t::Interfaces> *pingtimes = new map<int, osd_stat_t::Interfaces>; |
3009 | service.get_hb_pingtime(pingtimes); | |
3010 | for (auto j : *pingtimes) { | |
3011 | if (j.second.last_update == 0) | |
3012 | continue; | |
3013 | osd_ping_time_t item; | |
3014 | item.pingtime = std::max(j.second.back_pingtime[0], j.second.back_pingtime[1]); | |
3015 | item.pingtime = std::max(item.pingtime, j.second.back_pingtime[2]); | |
3016 | if (item.pingtime >= value) { | |
3017 | item.to = j.first; | |
3018 | item.times[0] = j.second.back_pingtime[0]; | |
3019 | item.times[1] = j.second.back_pingtime[1]; | |
3020 | item.times[2] = j.second.back_pingtime[2]; | |
3021 | item.min[0] = j.second.back_min[0]; | |
3022 | item.min[1] = j.second.back_min[1]; | |
3023 | item.min[2] = j.second.back_min[2]; | |
3024 | item.max[0] = j.second.back_max[0]; | |
3025 | item.max[1] = j.second.back_max[1]; | |
3026 | item.max[2] = j.second.back_max[2]; | |
3027 | item.last = j.second.back_last; | |
3028 | item.back = true; | |
3029 | item.last_update = j.second.last_update; | |
3030 | sorted.emplace(item); | |
3031 | } | |
3032 | if (j.second.front_last == 0) | |
3033 | continue; | |
3034 | item.pingtime = std::max(j.second.front_pingtime[0], j.second.front_pingtime[1]); | |
3035 | item.pingtime = std::max(item.pingtime, j.second.front_pingtime[2]); | |
3036 | if (item.pingtime >= value) { | |
3037 | item.to = j.first; | |
3038 | item.times[0] = j.second.front_pingtime[0]; | |
3039 | item.times[1] = j.second.front_pingtime[1]; | |
3040 | item.times[2] = j.second.front_pingtime[2]; | |
3041 | item.min[0] = j.second.front_min[0]; | |
3042 | item.min[1] = j.second.front_min[1]; | |
3043 | item.min[2] = j.second.front_min[2]; | |
3044 | item.max[0] = j.second.front_max[0]; | |
3045 | item.max[1] = j.second.front_max[1]; | |
3046 | item.max[2] = j.second.front_max[2]; | |
3047 | item.last = j.second.front_last; | |
3048 | item.last_update = j.second.last_update; | |
3049 | item.back = false; | |
3050 | sorted.emplace(item); | |
3051 | } | |
3052 | } | |
3053 | delete pingtimes; | |
3054 | // | |
3055 | // Network ping times (1min 5min 15min) | |
3056 | f->open_object_section("network_ping_times"); | |
3057 | f->dump_int("threshold", value / 1000); | |
3058 | f->open_array_section("entries"); | |
3059 | for (auto &sitem : boost::adaptors::reverse(sorted)) { | |
3060 | ceph_assert(sitem.pingtime >= value); | |
3061 | f->open_object_section("entry"); | |
3062 | ||
3063 | const time_t lu(sitem.last_update); | |
3064 | char buffer[26]; | |
3065 | string lustr(ctime_r(&lu, buffer)); | |
3066 | lustr.pop_back(); // Remove trailing \n | |
3067 | auto stale = cct->_conf.get_val<int64_t>("osd_heartbeat_stale"); | |
3068 | f->dump_string("last update", lustr); | |
3069 | f->dump_bool("stale", ceph_clock_now().sec() - sitem.last_update > stale); | |
3070 | f->dump_int("from osd", whoami); | |
3071 | f->dump_int("to osd", sitem.to); | |
3072 | f->dump_string("interface", (sitem.back ? "back" : "front")); | |
3073 | f->open_object_section("average"); | |
3074 | f->dump_format_unquoted("1min", "%s", fixed_u_to_string(sitem.times[0],3).c_str()); | |
3075 | f->dump_format_unquoted("5min", "%s", fixed_u_to_string(sitem.times[1],3).c_str()); | |
3076 | f->dump_format_unquoted("15min", "%s", fixed_u_to_string(sitem.times[2],3).c_str()); | |
3077 | f->close_section(); // average | |
3078 | f->open_object_section("min"); | |
3079 | f->dump_format_unquoted("1min", "%s", fixed_u_to_string(sitem.max[0],3).c_str()); | |
3080 | f->dump_format_unquoted("5min", "%s", fixed_u_to_string(sitem.max[1],3).c_str()); | |
3081 | f->dump_format_unquoted("15min", "%s", fixed_u_to_string(sitem.max[2],3).c_str()); | |
3082 | f->close_section(); // min | |
3083 | f->open_object_section("max"); | |
3084 | f->dump_format_unquoted("1min", "%s", fixed_u_to_string(sitem.max[0],3).c_str()); | |
3085 | f->dump_format_unquoted("5min", "%s", fixed_u_to_string(sitem.max[1],3).c_str()); | |
3086 | f->dump_format_unquoted("15min", "%s", fixed_u_to_string(sitem.max[2],3).c_str()); | |
3087 | f->close_section(); // max | |
3088 | f->dump_format_unquoted("last", "%s", fixed_u_to_string(sitem.last,3).c_str()); | |
3089 | f->close_section(); // entry | |
3090 | } | |
3091 | f->close_section(); // entries | |
3092 | f->close_section(); // network_ping_times | |
7c673cae | 3093 | } else { |
11fdf7f2 | 3094 | ceph_abort_msg("broken asok registration"); |
7c673cae | 3095 | } |
9f95a23c TL |
3096 | |
3097 | out: | |
3098 | on_finish(ret, ss.str(), outbl); | |
7c673cae FG |
3099 | } |
3100 | ||
3101 | class TestOpsSocketHook : public AdminSocketHook { | |
3102 | OSDService *service; | |
3103 | ObjectStore *store; | |
3104 | public: | |
3105 | TestOpsSocketHook(OSDService *s, ObjectStore *st) : service(s), store(st) {} | |
9f95a23c TL |
3106 | int call(std::string_view command, const cmdmap_t& cmdmap, |
3107 | Formatter *f, | |
3108 | std::ostream& errss, | |
3109 | bufferlist& out) override { | |
3110 | int r = 0; | |
3111 | stringstream outss; | |
11fdf7f2 | 3112 | try { |
9f95a23c TL |
3113 | test_ops(service, store, command, cmdmap, outss); |
3114 | out.append(outss); | |
3115 | } catch (const TOPNSPC::common::bad_cmd_get& e) { | |
3116 | errss << e.what(); | |
3117 | r = -EINVAL; | |
11fdf7f2 | 3118 | } |
9f95a23c | 3119 | return r; |
7c673cae FG |
3120 | } |
3121 | void test_ops(OSDService *service, ObjectStore *store, | |
11fdf7f2 | 3122 | std::string_view command, const cmdmap_t& cmdmap, ostream &ss); |
7c673cae FG |
3123 | |
3124 | }; | |
3125 | ||
3126 | class OSD::C_Tick : public Context { | |
3127 | OSD *osd; | |
3128 | public: | |
3129 | explicit C_Tick(OSD *o) : osd(o) {} | |
3130 | void finish(int r) override { | |
3131 | osd->tick(); | |
3132 | } | |
3133 | }; | |
3134 | ||
3135 | class OSD::C_Tick_WithoutOSDLock : public Context { | |
3136 | OSD *osd; | |
3137 | public: | |
3138 | explicit C_Tick_WithoutOSDLock(OSD *o) : osd(o) {} | |
3139 | void finish(int r) override { | |
3140 | osd->tick_without_osd_lock(); | |
3141 | } | |
3142 | }; | |
3143 | ||
3144 | int OSD::enable_disable_fuse(bool stop) | |
3145 | { | |
3146 | #ifdef HAVE_LIBFUSE | |
3147 | int r; | |
3148 | string mntpath = cct->_conf->osd_data + "/fuse"; | |
3149 | if (fuse_store && (stop || !cct->_conf->osd_objectstore_fuse)) { | |
3150 | dout(1) << __func__ << " disabling" << dendl; | |
3151 | fuse_store->stop(); | |
3152 | delete fuse_store; | |
3153 | fuse_store = NULL; | |
3154 | r = ::rmdir(mntpath.c_str()); | |
7c673cae | 3155 | if (r < 0) { |
c07f9fc5 FG |
3156 | r = -errno; |
3157 | derr << __func__ << " failed to rmdir " << mntpath << ": " | |
3158 | << cpp_strerror(r) << dendl; | |
7c673cae FG |
3159 | return r; |
3160 | } | |
3161 | return 0; | |
3162 | } | |
3163 | if (!fuse_store && cct->_conf->osd_objectstore_fuse) { | |
3164 | dout(1) << __func__ << " enabling" << dendl; | |
3165 | r = ::mkdir(mntpath.c_str(), 0700); | |
3166 | if (r < 0) | |
3167 | r = -errno; | |
3168 | if (r < 0 && r != -EEXIST) { | |
3169 | derr << __func__ << " unable to create " << mntpath << ": " | |
3170 | << cpp_strerror(r) << dendl; | |
3171 | return r; | |
3172 | } | |
3173 | fuse_store = new FuseStore(store, mntpath); | |
3174 | r = fuse_store->start(); | |
3175 | if (r < 0) { | |
3176 | derr << __func__ << " unable to start fuse: " << cpp_strerror(r) << dendl; | |
3177 | delete fuse_store; | |
3178 | fuse_store = NULL; | |
3179 | return r; | |
3180 | } | |
3181 | } | |
3182 | #endif // HAVE_LIBFUSE | |
3183 | return 0; | |
3184 | } | |
3185 | ||
9f95a23c TL |
3186 | size_t OSD::get_num_cache_shards() |
3187 | { | |
3188 | return cct->_conf.get_val<Option::size_t>("osd_num_cache_shards"); | |
3189 | } | |
3190 | ||
31f18b77 FG |
3191 | int OSD::get_num_op_shards() |
3192 | { | |
3193 | if (cct->_conf->osd_op_num_shards) | |
3194 | return cct->_conf->osd_op_num_shards; | |
3195 | if (store_is_rotational) | |
3196 | return cct->_conf->osd_op_num_shards_hdd; | |
3197 | else | |
3198 | return cct->_conf->osd_op_num_shards_ssd; | |
3199 | } | |
3200 | ||
3201 | int OSD::get_num_op_threads() | |
3202 | { | |
3203 | if (cct->_conf->osd_op_num_threads_per_shard) | |
3204 | return get_num_op_shards() * cct->_conf->osd_op_num_threads_per_shard; | |
3205 | if (store_is_rotational) | |
3206 | return get_num_op_shards() * cct->_conf->osd_op_num_threads_per_shard_hdd; | |
3207 | else | |
3208 | return get_num_op_shards() * cct->_conf->osd_op_num_threads_per_shard_ssd; | |
3209 | } | |
3210 | ||
c07f9fc5 FG |
3211 | float OSD::get_osd_recovery_sleep() |
3212 | { | |
3213 | if (cct->_conf->osd_recovery_sleep) | |
3214 | return cct->_conf->osd_recovery_sleep; | |
d2e6a577 | 3215 | if (!store_is_rotational && !journal_is_rotational) |
c07f9fc5 | 3216 | return cct->_conf->osd_recovery_sleep_ssd; |
d2e6a577 | 3217 | else if (store_is_rotational && !journal_is_rotational) |
11fdf7f2 | 3218 | return cct->_conf.get_val<double>("osd_recovery_sleep_hybrid"); |
d2e6a577 FG |
3219 | else |
3220 | return cct->_conf->osd_recovery_sleep_hdd; | |
c07f9fc5 FG |
3221 | } |
3222 | ||
11fdf7f2 TL |
3223 | float OSD::get_osd_delete_sleep() |
3224 | { | |
3225 | float osd_delete_sleep = cct->_conf.get_val<double>("osd_delete_sleep"); | |
3226 | if (osd_delete_sleep > 0) | |
3227 | return osd_delete_sleep; | |
3228 | if (!store_is_rotational && !journal_is_rotational) | |
3229 | return cct->_conf.get_val<double>("osd_delete_sleep_ssd"); | |
3230 | if (store_is_rotational && !journal_is_rotational) | |
3231 | return cct->_conf.get_val<double>("osd_delete_sleep_hybrid"); | |
3232 | return cct->_conf.get_val<double>("osd_delete_sleep_hdd"); | |
3233 | } | |
3234 | ||
9f95a23c TL |
3235 | int OSD::get_recovery_max_active() |
3236 | { | |
3237 | if (cct->_conf->osd_recovery_max_active) | |
3238 | return cct->_conf->osd_recovery_max_active; | |
3239 | if (store_is_rotational) | |
3240 | return cct->_conf->osd_recovery_max_active_hdd; | |
3241 | else | |
3242 | return cct->_conf->osd_recovery_max_active_ssd; | |
3243 | } | |
3244 | ||
494da23a TL |
3245 | float OSD::get_osd_snap_trim_sleep() |
3246 | { | |
3247 | float osd_snap_trim_sleep = cct->_conf.get_val<double>("osd_snap_trim_sleep"); | |
3248 | if (osd_snap_trim_sleep > 0) | |
3249 | return osd_snap_trim_sleep; | |
3250 | if (!store_is_rotational && !journal_is_rotational) | |
3251 | return cct->_conf.get_val<double>("osd_snap_trim_sleep_ssd"); | |
3252 | if (store_is_rotational && !journal_is_rotational) | |
3253 | return cct->_conf.get_val<double>("osd_snap_trim_sleep_hybrid"); | |
3254 | return cct->_conf.get_val<double>("osd_snap_trim_sleep_hdd"); | |
3255 | } | |
3256 | ||
7c673cae FG |
3257 | int OSD::init() |
3258 | { | |
9f95a23c | 3259 | OSDMapRef osdmap; |
7c673cae | 3260 | CompatSet initial, diff; |
11fdf7f2 | 3261 | std::lock_guard lock(osd_lock); |
7c673cae FG |
3262 | if (is_stopping()) |
3263 | return 0; | |
3264 | ||
3265 | tick_timer.init(); | |
3266 | tick_timer_without_osd_lock.init(); | |
3267 | service.recovery_request_timer.init(); | |
11fdf7f2 TL |
3268 | service.sleep_timer.init(); |
3269 | ||
3270 | boot_finisher.start(); | |
3271 | ||
3272 | { | |
3273 | string val; | |
3274 | store->read_meta("require_osd_release", &val); | |
9f95a23c | 3275 | last_require_osd_release = ceph_release_from_name(val); |
11fdf7f2 | 3276 | } |
7c673cae FG |
3277 | |
3278 | // mount. | |
31f18b77 FG |
3279 | dout(2) << "init " << dev_path |
3280 | << " (looks like " << (store_is_rotational ? "hdd" : "ssd") << ")" | |
3281 | << dendl; | |
d2e6a577 | 3282 | dout(2) << "journal " << journal_path << dendl; |
11fdf7f2 | 3283 | ceph_assert(store); // call pre_init() first! |
7c673cae | 3284 | |
9f95a23c | 3285 | store->set_cache_shards(get_num_cache_shards()); |
7c673cae FG |
3286 | |
3287 | int r = store->mount(); | |
3288 | if (r < 0) { | |
3289 | derr << "OSD:init: unable to mount object store" << dendl; | |
3290 | return r; | |
3291 | } | |
d2e6a577 FG |
3292 | journal_is_rotational = store->is_journal_rotational(); |
3293 | dout(2) << "journal looks like " << (journal_is_rotational ? "hdd" : "ssd") | |
3294 | << dendl; | |
7c673cae FG |
3295 | |
3296 | enable_disable_fuse(false); | |
3297 | ||
3298 | dout(2) << "boot" << dendl; | |
3299 | ||
11fdf7f2 TL |
3300 | service.meta_ch = store->open_collection(coll_t::meta()); |
3301 | ||
7c673cae FG |
3302 | // initialize the daily loadavg with current 15min loadavg |
3303 | double loadavgs[3]; | |
3304 | if (getloadavg(loadavgs, 3) == 3) { | |
3305 | daily_loadavg = loadavgs[2]; | |
3306 | } else { | |
3307 | derr << "OSD::init() : couldn't read loadavgs\n" << dendl; | |
3308 | daily_loadavg = 1.0; | |
3309 | } | |
3310 | ||
3311 | int rotating_auth_attempts = 0; | |
11fdf7f2 TL |
3312 | auto rotating_auth_timeout = |
3313 | g_conf().get_val<int64_t>("rotating_keys_bootstrap_timeout"); | |
7c673cae FG |
3314 | |
3315 | // sanity check long object name handling | |
3316 | { | |
3317 | hobject_t l; | |
3318 | l.oid.name = string(cct->_conf->osd_max_object_name_len, 'n'); | |
3319 | l.set_key(string(cct->_conf->osd_max_object_name_len, 'k')); | |
3320 | l.nspace = string(cct->_conf->osd_max_object_namespace_len, 's'); | |
3321 | r = store->validate_hobject_key(l); | |
3322 | if (r < 0) { | |
3323 | derr << "backend (" << store->get_type() << ") is unable to support max " | |
3324 | << "object name[space] len" << dendl; | |
3325 | derr << " osd max object name len = " | |
3326 | << cct->_conf->osd_max_object_name_len << dendl; | |
3327 | derr << " osd max object namespace len = " | |
3328 | << cct->_conf->osd_max_object_namespace_len << dendl; | |
3329 | derr << cpp_strerror(r) << dendl; | |
3330 | if (cct->_conf->osd_check_max_object_name_len_on_startup) { | |
3331 | goto out; | |
3332 | } | |
3333 | derr << "osd_check_max_object_name_len_on_startup = false, starting anyway" | |
3334 | << dendl; | |
3335 | } else { | |
3336 | dout(20) << "configured osd_max_object_name[space]_len looks ok" << dendl; | |
3337 | } | |
3338 | } | |
3339 | ||
3340 | // read superblock | |
3341 | r = read_superblock(); | |
3342 | if (r < 0) { | |
3343 | derr << "OSD::init() : unable to read osd superblock" << dendl; | |
3344 | r = -EINVAL; | |
3345 | goto out; | |
3346 | } | |
3347 | ||
3348 | if (osd_compat.compare(superblock.compat_features) < 0) { | |
3349 | derr << "The disk uses features unsupported by the executable." << dendl; | |
3350 | derr << " ondisk features " << superblock.compat_features << dendl; | |
3351 | derr << " daemon features " << osd_compat << dendl; | |
3352 | ||
3353 | if (osd_compat.writeable(superblock.compat_features)) { | |
3354 | CompatSet diff = osd_compat.unsupported(superblock.compat_features); | |
3355 | derr << "it is still writeable, though. Missing features: " << diff << dendl; | |
3356 | r = -EOPNOTSUPP; | |
3357 | goto out; | |
3358 | } | |
3359 | else { | |
3360 | CompatSet diff = osd_compat.unsupported(superblock.compat_features); | |
3361 | derr << "Cannot write to disk! Missing features: " << diff << dendl; | |
3362 | r = -EOPNOTSUPP; | |
3363 | goto out; | |
3364 | } | |
3365 | } | |
3366 | ||
3367 | assert_warn(whoami == superblock.whoami); | |
3368 | if (whoami != superblock.whoami) { | |
3369 | derr << "OSD::init: superblock says osd" | |
3370 | << superblock.whoami << " but I am osd." << whoami << dendl; | |
3371 | r = -EINVAL; | |
3372 | goto out; | |
3373 | } | |
3374 | ||
9f95a23c TL |
3375 | startup_time = ceph::mono_clock::now(); |
3376 | ||
11fdf7f2 | 3377 | // load up "current" osdmap |
9f95a23c TL |
3378 | assert_warn(!get_osdmap()); |
3379 | if (get_osdmap()) { | |
11fdf7f2 TL |
3380 | derr << "OSD::init: unable to read current osdmap" << dendl; |
3381 | r = -EINVAL; | |
3382 | goto out; | |
3383 | } | |
3384 | osdmap = get_map(superblock.current_epoch); | |
9f95a23c | 3385 | set_osdmap(osdmap); |
11fdf7f2 TL |
3386 | |
3387 | // make sure we don't have legacy pgs deleting | |
3388 | { | |
3389 | vector<coll_t> ls; | |
3390 | int r = store->list_collections(ls); | |
3391 | ceph_assert(r >= 0); | |
3392 | for (auto c : ls) { | |
3393 | spg_t pgid; | |
3394 | if (c.is_pg(&pgid) && | |
3395 | !osdmap->have_pg_pool(pgid.pool())) { | |
3396 | ghobject_t oid = make_final_pool_info_oid(pgid.pool()); | |
3397 | if (!store->exists(service.meta_ch, oid)) { | |
3398 | derr << __func__ << " missing pg_pool_t for deleted pool " | |
3399 | << pgid.pool() << " for pg " << pgid | |
3400 | << "; please downgrade to luminous and allow " | |
3401 | << "pg deletion to complete before upgrading" << dendl; | |
3402 | ceph_abort(); | |
3403 | } | |
3404 | } | |
3405 | } | |
3406 | } | |
3407 | ||
7c673cae FG |
3408 | initial = get_osd_initial_compat_set(); |
3409 | diff = superblock.compat_features.unsupported(initial); | |
3410 | if (superblock.compat_features.merge(initial)) { | |
9f95a23c TL |
3411 | // Are we adding SNAPMAPPER2? |
3412 | if (diff.incompat.contains(CEPH_OSD_FEATURE_INCOMPAT_SNAPMAPPER2)) { | |
3413 | dout(1) << __func__ << " upgrade snap_mapper (first start as octopus)" | |
3414 | << dendl; | |
3415 | auto ch = service.meta_ch; | |
3416 | auto hoid = make_snapmapper_oid(); | |
3417 | unsigned max = cct->_conf->osd_target_transaction_size; | |
3418 | r = SnapMapper::convert_legacy(cct, store, ch, hoid, max); | |
3419 | if (r < 0) | |
3420 | goto out; | |
3421 | } | |
7c673cae FG |
3422 | // We need to persist the new compat_set before we |
3423 | // do anything else | |
3424 | dout(5) << "Upgrading superblock adding: " << diff << dendl; | |
3425 | ObjectStore::Transaction t; | |
3426 | write_superblock(t); | |
11fdf7f2 | 3427 | r = store->queue_transaction(service.meta_ch, std::move(t)); |
7c673cae FG |
3428 | if (r < 0) |
3429 | goto out; | |
3430 | } | |
3431 | ||
3432 | // make sure snap mapper object exists | |
11fdf7f2 | 3433 | if (!store->exists(service.meta_ch, OSD::make_snapmapper_oid())) { |
7c673cae FG |
3434 | dout(10) << "init creating/touching snapmapper object" << dendl; |
3435 | ObjectStore::Transaction t; | |
3436 | t.touch(coll_t::meta(), OSD::make_snapmapper_oid()); | |
11fdf7f2 | 3437 | r = store->queue_transaction(service.meta_ch, std::move(t)); |
7c673cae FG |
3438 | if (r < 0) |
3439 | goto out; | |
3440 | } | |
9f95a23c TL |
3441 | if (!store->exists(service.meta_ch, OSD::make_purged_snaps_oid())) { |
3442 | dout(10) << "init creating/touching purged_snaps object" << dendl; | |
3443 | ObjectStore::Transaction t; | |
3444 | t.touch(coll_t::meta(), OSD::make_purged_snaps_oid()); | |
3445 | r = store->queue_transaction(service.meta_ch, std::move(t)); | |
3446 | if (r < 0) | |
3447 | goto out; | |
3448 | } | |
7c673cae FG |
3449 | |
3450 | if (cct->_conf->osd_open_classes_on_start) { | |
9f95a23c | 3451 | int r = ClassHandler::get_instance().open_all_classes(); |
7c673cae FG |
3452 | if (r) |
3453 | dout(1) << "warning: got an error loading one or more classes: " << cpp_strerror(r) << dendl; | |
3454 | } | |
3455 | ||
11fdf7f2 | 3456 | check_osdmap_features(); |
7c673cae FG |
3457 | |
3458 | create_recoverystate_perf(); | |
3459 | ||
3460 | { | |
3461 | epoch_t bind_epoch = osdmap->get_epoch(); | |
3462 | service.set_epochs(NULL, NULL, &bind_epoch); | |
3463 | } | |
3464 | ||
3465 | clear_temp_objects(); | |
3466 | ||
d2e6a577 | 3467 | // initialize osdmap references in sharded wq |
11fdf7f2 TL |
3468 | for (auto& shard : shards) { |
3469 | std::lock_guard l(shard->osdmap_lock); | |
3470 | shard->shard_osdmap = osdmap; | |
3471 | } | |
d2e6a577 | 3472 | |
7c673cae FG |
3473 | // load up pgs (as they previously existed) |
3474 | load_pgs(); | |
3475 | ||
3476 | dout(2) << "superblock: I am osd." << superblock.whoami << dendl; | |
7c673cae FG |
3477 | |
3478 | create_logger(); | |
3479 | ||
11fdf7f2 TL |
3480 | // prime osd stats |
3481 | { | |
3482 | struct store_statfs_t stbuf; | |
3483 | osd_alert_list_t alerts; | |
3484 | int r = store->statfs(&stbuf, &alerts); | |
3485 | ceph_assert(r == 0); | |
3486 | service.set_statfs(stbuf, alerts); | |
3487 | } | |
3488 | ||
3489 | // client_messenger auth_client is already set up by monc. | |
3490 | for (auto m : { cluster_messenger, | |
3491 | objecter_messenger, | |
3492 | hb_front_client_messenger, | |
3493 | hb_back_client_messenger, | |
3494 | hb_front_server_messenger, | |
3495 | hb_back_server_messenger } ) { | |
3496 | m->set_auth_client(monc); | |
3497 | } | |
3498 | for (auto m : { client_messenger, | |
3499 | cluster_messenger, | |
3500 | hb_front_server_messenger, | |
3501 | hb_back_server_messenger }) { | |
3502 | m->set_auth_server(monc); | |
3503 | } | |
3504 | monc->set_handle_authentication_dispatcher(this); | |
7c673cae FG |
3505 | |
3506 | monc->set_want_keys(CEPH_ENTITY_TYPE_MON | CEPH_ENTITY_TYPE_OSD | |
3507 | | CEPH_ENTITY_TYPE_MGR); | |
3508 | r = monc->init(); | |
3509 | if (r < 0) | |
3510 | goto out; | |
3511 | ||
11fdf7f2 TL |
3512 | mgrc.set_pgstats_cb([this](){ return collect_pg_stats(); }); |
3513 | mgrc.set_perf_metric_query_cb( | |
9f95a23c TL |
3514 | [this](const ConfigPayload &config_payload) { |
3515 | set_perf_queries(config_payload); | |
11fdf7f2 | 3516 | }, |
9f95a23c TL |
3517 | [this] { |
3518 | return get_perf_reports(); | |
11fdf7f2 | 3519 | }); |
7c673cae | 3520 | mgrc.init(); |
7c673cae FG |
3521 | |
3522 | // tell monc about log_client so it will know about mon session resets | |
3523 | monc->set_log_client(&log_client); | |
3524 | update_log_config(); | |
3525 | ||
11fdf7f2 TL |
3526 | // i'm ready! |
3527 | client_messenger->add_dispatcher_tail(&mgrc); | |
3528 | client_messenger->add_dispatcher_tail(this); | |
3529 | cluster_messenger->add_dispatcher_head(this); | |
3530 | ||
3531 | hb_front_client_messenger->add_dispatcher_head(&heartbeat_dispatcher); | |
3532 | hb_back_client_messenger->add_dispatcher_head(&heartbeat_dispatcher); | |
3533 | hb_front_server_messenger->add_dispatcher_head(&heartbeat_dispatcher); | |
3534 | hb_back_server_messenger->add_dispatcher_head(&heartbeat_dispatcher); | |
3535 | ||
9f95a23c | 3536 | objecter_messenger->add_dispatcher_head(service.objecter.get()); |
11fdf7f2 | 3537 | |
28e407b8 AA |
3538 | service.init(); |
3539 | service.publish_map(osdmap); | |
3540 | service.publish_superblock(superblock); | |
3541 | service.max_oldest_map = superblock.oldest_map; | |
3542 | ||
11fdf7f2 TL |
3543 | for (auto& shard : shards) { |
3544 | // put PGs in a temporary set because we may modify pg_slots | |
3545 | // unordered_map below. | |
3546 | set<PGRef> pgs; | |
3547 | for (auto& i : shard->pg_slots) { | |
3548 | PGRef pg = i.second->pg; | |
3549 | if (!pg) { | |
3550 | continue; | |
3551 | } | |
3552 | pgs.insert(pg); | |
3553 | } | |
3554 | for (auto pg : pgs) { | |
9f95a23c | 3555 | std::scoped_lock l{*pg}; |
11fdf7f2 TL |
3556 | set<pair<spg_t,epoch_t>> new_children; |
3557 | set<pair<spg_t,epoch_t>> merge_pgs; | |
3558 | service.identify_splits_and_merges(pg->get_osdmap(), osdmap, pg->pg_id, | |
3559 | &new_children, &merge_pgs); | |
3560 | if (!new_children.empty()) { | |
3561 | for (auto shard : shards) { | |
3562 | shard->prime_splits(osdmap, &new_children); | |
3563 | } | |
3564 | assert(new_children.empty()); | |
3565 | } | |
3566 | if (!merge_pgs.empty()) { | |
3567 | for (auto shard : shards) { | |
3568 | shard->prime_merges(osdmap, &merge_pgs); | |
3569 | } | |
3570 | assert(merge_pgs.empty()); | |
3571 | } | |
11fdf7f2 TL |
3572 | } |
3573 | } | |
3574 | ||
7c673cae | 3575 | osd_op_tp.start(); |
7c673cae | 3576 | |
7c673cae FG |
3577 | // start the heartbeat |
3578 | heartbeat_thread.create("osd_srv_heartbt"); | |
3579 | ||
3580 | // tick | |
91327a77 AA |
3581 | tick_timer.add_event_after(get_tick_interval(), |
3582 | new C_Tick(this)); | |
7c673cae | 3583 | { |
11fdf7f2 | 3584 | std::lock_guard l(tick_timer_lock); |
91327a77 AA |
3585 | tick_timer_without_osd_lock.add_event_after(get_tick_interval(), |
3586 | new C_Tick_WithoutOSDLock(this)); | |
7c673cae FG |
3587 | } |
3588 | ||
9f95a23c | 3589 | osd_lock.unlock(); |
7c673cae FG |
3590 | |
3591 | r = monc->authenticate(); | |
3592 | if (r < 0) { | |
c07f9fc5 FG |
3593 | derr << __func__ << " authentication failed: " << cpp_strerror(r) |
3594 | << dendl; | |
11fdf7f2 | 3595 | exit(1); |
7c673cae FG |
3596 | } |
3597 | ||
11fdf7f2 | 3598 | while (monc->wait_auth_rotating(rotating_auth_timeout) < 0) { |
7c673cae FG |
3599 | derr << "unable to obtain rotating service keys; retrying" << dendl; |
3600 | ++rotating_auth_attempts; | |
11fdf7f2 | 3601 | if (rotating_auth_attempts > g_conf()->max_rotating_auth_attempts) { |
c07f9fc5 | 3602 | derr << __func__ << " wait_auth_rotating timed out" << dendl; |
11fdf7f2 | 3603 | exit(1); |
7c673cae FG |
3604 | } |
3605 | } | |
3606 | ||
3607 | r = update_crush_device_class(); | |
3608 | if (r < 0) { | |
d2e6a577 FG |
3609 | derr << __func__ << " unable to update_crush_device_class: " |
3610 | << cpp_strerror(r) << dendl; | |
11fdf7f2 | 3611 | exit(1); |
7c673cae FG |
3612 | } |
3613 | ||
3614 | r = update_crush_location(); | |
3615 | if (r < 0) { | |
d2e6a577 | 3616 | derr << __func__ << " unable to update_crush_location: " |
c07f9fc5 | 3617 | << cpp_strerror(r) << dendl; |
11fdf7f2 | 3618 | exit(1); |
7c673cae FG |
3619 | } |
3620 | ||
9f95a23c | 3621 | osd_lock.lock(); |
7c673cae FG |
3622 | if (is_stopping()) |
3623 | return 0; | |
3624 | ||
3625 | // start objecter *after* we have authenticated, so that we don't ignore | |
3626 | // the OSDMaps it requests. | |
3627 | service.final_init(); | |
3628 | ||
3629 | check_config(); | |
3630 | ||
3631 | dout(10) << "ensuring pgs have consumed prior maps" << dendl; | |
3632 | consume_map(); | |
7c673cae FG |
3633 | |
3634 | dout(0) << "done with init, starting boot process" << dendl; | |
3635 | ||
3636 | // subscribe to any pg creations | |
3637 | monc->sub_want("osd_pg_creates", last_pg_create_epoch, 0); | |
3638 | ||
3639 | // MgrClient needs this (it doesn't have MonClient reference itself) | |
3640 | monc->sub_want("mgrmap", 0, 0); | |
3641 | ||
3642 | // we don't need to ask for an osdmap here; objecter will | |
3643 | //monc->sub_want("osdmap", osdmap->get_epoch(), CEPH_SUBSCRIBE_ONETIME); | |
3644 | ||
3645 | monc->renew_subs(); | |
3646 | ||
3647 | start_boot(); | |
3648 | ||
3649 | return 0; | |
7c673cae FG |
3650 | |
3651 | out: | |
3652 | enable_disable_fuse(true); | |
3653 | store->umount(); | |
3654 | delete store; | |
3655 | store = NULL; | |
3656 | return r; | |
3657 | } | |
3658 | ||
3659 | void OSD::final_init() | |
3660 | { | |
3661 | AdminSocket *admin_socket = cct->get_admin_socket(); | |
3662 | asok_hook = new OSDSocketHook(this); | |
9f95a23c | 3663 | int r = admin_socket->register_command("status", asok_hook, |
7c673cae | 3664 | "high-level status of OSD"); |
11fdf7f2 | 3665 | ceph_assert(r == 0); |
9f95a23c | 3666 | r = admin_socket->register_command("flush_journal", |
7c673cae FG |
3667 | asok_hook, |
3668 | "flush the journal to permanent store"); | |
11fdf7f2 | 3669 | ceph_assert(r == 0); |
9f95a23c | 3670 | r = admin_socket->register_command("dump_ops_in_flight " \ |
c07f9fc5 FG |
3671 | "name=filterstr,type=CephString,n=N,req=false", |
3672 | asok_hook, | |
7c673cae | 3673 | "show the ops currently in flight"); |
11fdf7f2 | 3674 | ceph_assert(r == 0); |
9f95a23c | 3675 | r = admin_socket->register_command("ops " \ |
c07f9fc5 FG |
3676 | "name=filterstr,type=CephString,n=N,req=false", |
3677 | asok_hook, | |
7c673cae | 3678 | "show the ops currently in flight"); |
11fdf7f2 | 3679 | ceph_assert(r == 0); |
9f95a23c | 3680 | r = admin_socket->register_command("dump_blocked_ops " \ |
c07f9fc5 FG |
3681 | "name=filterstr,type=CephString,n=N,req=false", |
3682 | asok_hook, | |
7c673cae | 3683 | "show the blocked ops currently in flight"); |
11fdf7f2 | 3684 | ceph_assert(r == 0); |
9f95a23c | 3685 | r = admin_socket->register_command("dump_historic_ops " \ |
c07f9fc5 | 3686 | "name=filterstr,type=CephString,n=N,req=false", |
7c673cae FG |
3687 | asok_hook, |
3688 | "show recent ops"); | |
11fdf7f2 | 3689 | ceph_assert(r == 0); |
9f95a23c | 3690 | r = admin_socket->register_command("dump_historic_slow_ops " \ |
c07f9fc5 | 3691 | "name=filterstr,type=CephString,n=N,req=false", |
7c673cae FG |
3692 | asok_hook, |
3693 | "show slowest recent ops"); | |
11fdf7f2 | 3694 | ceph_assert(r == 0); |
9f95a23c | 3695 | r = admin_socket->register_command("dump_historic_ops_by_duration " \ |
c07f9fc5 | 3696 | "name=filterstr,type=CephString,n=N,req=false", |
7c673cae FG |
3697 | asok_hook, |
3698 | "show slowest recent ops, sorted by duration"); | |
11fdf7f2 | 3699 | ceph_assert(r == 0); |
9f95a23c | 3700 | r = admin_socket->register_command("dump_op_pq_state", |
7c673cae FG |
3701 | asok_hook, |
3702 | "dump op priority queue state"); | |
11fdf7f2 | 3703 | ceph_assert(r == 0); |
9f95a23c | 3704 | r = admin_socket->register_command("dump_blacklist", |
7c673cae FG |
3705 | asok_hook, |
3706 | "dump blacklisted clients and times"); | |
11fdf7f2 | 3707 | ceph_assert(r == 0); |
9f95a23c | 3708 | r = admin_socket->register_command("dump_watchers", |
7c673cae FG |
3709 | asok_hook, |
3710 | "show clients which have active watches," | |
3711 | " and on which objects"); | |
11fdf7f2 | 3712 | ceph_assert(r == 0); |
9f95a23c | 3713 | r = admin_socket->register_command("dump_recovery_reservations", |
7c673cae FG |
3714 | asok_hook, |
3715 | "show recovery reservations"); | |
11fdf7f2 | 3716 | ceph_assert(r == 0); |
9f95a23c | 3717 | r = admin_socket->register_command("dump_scrub_reservations", |
eafe8130 | 3718 | asok_hook, |
9f95a23c | 3719 | "show recovery reservations"); |
eafe8130 | 3720 | ceph_assert(r == 0); |
9f95a23c | 3721 | r = admin_socket->register_command("get_latest_osdmap", |
7c673cae FG |
3722 | asok_hook, |
3723 | "force osd to update the latest map from " | |
3724 | "the mon"); | |
11fdf7f2 | 3725 | ceph_assert(r == 0); |
7c673cae | 3726 | |
9f95a23c | 3727 | r = admin_socket->register_command("set_heap_property " \ |
7c673cae FG |
3728 | "name=property,type=CephString " \ |
3729 | "name=value,type=CephInt", | |
3730 | asok_hook, | |
3731 | "update malloc extension heap property"); | |
11fdf7f2 | 3732 | ceph_assert(r == 0); |
7c673cae | 3733 | |
9f95a23c | 3734 | r = admin_socket->register_command("get_heap_property " \ |
7c673cae FG |
3735 | "name=property,type=CephString", |
3736 | asok_hook, | |
3737 | "get malloc extension heap property"); | |
11fdf7f2 | 3738 | ceph_assert(r == 0); |
7c673cae FG |
3739 | |
3740 | r = admin_socket->register_command("dump_objectstore_kv_stats", | |
7c673cae FG |
3741 | asok_hook, |
3742 | "print statistics of kvdb which used by bluestore"); | |
11fdf7f2 | 3743 | ceph_assert(r == 0); |
7c673cae FG |
3744 | |
3745 | r = admin_socket->register_command("dump_scrubs", | |
7c673cae FG |
3746 | asok_hook, |
3747 | "print scheduled scrubs"); | |
11fdf7f2 | 3748 | ceph_assert(r == 0); |
7c673cae FG |
3749 | |
3750 | r = admin_socket->register_command("calc_objectstore_db_histogram", | |
7c673cae FG |
3751 | asok_hook, |
3752 | "Generate key value histogram of kvdb(rocksdb) which used by bluestore"); | |
11fdf7f2 | 3753 | ceph_assert(r == 0); |
7c673cae FG |
3754 | |
3755 | r = admin_socket->register_command("flush_store_cache", | |
7c673cae FG |
3756 | asok_hook, |
3757 | "Flush bluestore internal cache"); | |
11fdf7f2 | 3758 | ceph_assert(r == 0); |
9f95a23c | 3759 | r = admin_socket->register_command("dump_pgstate_history", |
7c673cae FG |
3760 | asok_hook, |
3761 | "show recent state history"); | |
11fdf7f2 | 3762 | ceph_assert(r == 0); |
7c673cae | 3763 | |
9f95a23c | 3764 | r = admin_socket->register_command("compact", |
224ce89b WB |
3765 | asok_hook, |
3766 | "Commpact object store's omap." | |
3767 | " WARNING: Compaction probably slows your requests"); | |
11fdf7f2 TL |
3768 | ceph_assert(r == 0); |
3769 | ||
9f95a23c | 3770 | r = admin_socket->register_command("get_mapped_pools", |
11fdf7f2 TL |
3771 | asok_hook, |
3772 | "dump pools whose PG(s) are mapped to this OSD."); | |
3773 | ||
3774 | ceph_assert(r == 0); | |
3775 | ||
9f95a23c | 3776 | r = admin_socket->register_command("smart name=devid,type=CephString,req=false", |
11fdf7f2 TL |
3777 | asok_hook, |
3778 | "probe OSD devices for SMART data."); | |
3779 | ||
3780 | ceph_assert(r == 0); | |
3781 | ||
9f95a23c | 3782 | r = admin_socket->register_command("list_devices", |
11fdf7f2 TL |
3783 | asok_hook, |
3784 | "list OSD devices."); | |
9f95a23c | 3785 | r = admin_socket->register_command("send_beacon", |
11fdf7f2 TL |
3786 | asok_hook, |
3787 | "send OSD beacon to mon immediately"); | |
224ce89b | 3788 | |
9f95a23c TL |
3789 | r = admin_socket->register_command( |
3790 | "dump_osd_network name=value,type=CephInt,req=false", asok_hook, | |
3791 | "Dump osd heartbeat network ping times"); | |
eafe8130 TL |
3792 | ceph_assert(r == 0); |
3793 | ||
7c673cae FG |
3794 | test_ops_hook = new TestOpsSocketHook(&(this->service), this->store); |
3795 | // Note: pools are CephString instead of CephPoolname because | |
3796 | // these commands traditionally support both pool names and numbers | |
3797 | r = admin_socket->register_command( | |
7c673cae FG |
3798 | "setomapval " \ |
3799 | "name=pool,type=CephString " \ | |
3800 | "name=objname,type=CephObjectname " \ | |
3801 | "name=key,type=CephString "\ | |
3802 | "name=val,type=CephString", | |
3803 | test_ops_hook, | |
3804 | "set omap key"); | |
11fdf7f2 | 3805 | ceph_assert(r == 0); |
7c673cae | 3806 | r = admin_socket->register_command( |
7c673cae FG |
3807 | "rmomapkey " \ |
3808 | "name=pool,type=CephString " \ | |
3809 | "name=objname,type=CephObjectname " \ | |
3810 | "name=key,type=CephString", | |
3811 | test_ops_hook, | |
3812 | "remove omap key"); | |
11fdf7f2 | 3813 | ceph_assert(r == 0); |
7c673cae | 3814 | r = admin_socket->register_command( |
7c673cae FG |
3815 | "setomapheader " \ |
3816 | "name=pool,type=CephString " \ | |
3817 | "name=objname,type=CephObjectname " \ | |
3818 | "name=header,type=CephString", | |
3819 | test_ops_hook, | |
3820 | "set omap header"); | |
11fdf7f2 | 3821 | ceph_assert(r == 0); |
7c673cae FG |
3822 | |
3823 | r = admin_socket->register_command( | |
7c673cae FG |
3824 | "getomap " \ |
3825 | "name=pool,type=CephString " \ | |
3826 | "name=objname,type=CephObjectname", | |
3827 | test_ops_hook, | |
3828 | "output entire object map"); | |
11fdf7f2 | 3829 | ceph_assert(r == 0); |
7c673cae FG |
3830 | |
3831 | r = admin_socket->register_command( | |
7c673cae FG |
3832 | "truncobj " \ |
3833 | "name=pool,type=CephString " \ | |
3834 | "name=objname,type=CephObjectname " \ | |
3835 | "name=len,type=CephInt", | |
3836 | test_ops_hook, | |
3837 | "truncate object to length"); | |
11fdf7f2 | 3838 | ceph_assert(r == 0); |
7c673cae FG |
3839 | |
3840 | r = admin_socket->register_command( | |
7c673cae FG |
3841 | "injectdataerr " \ |
3842 | "name=pool,type=CephString " \ | |
3843 | "name=objname,type=CephObjectname " \ | |
3844 | "name=shardid,type=CephInt,req=false,range=0|255", | |
3845 | test_ops_hook, | |
3846 | "inject data error to an object"); | |
11fdf7f2 | 3847 | ceph_assert(r == 0); |
7c673cae FG |
3848 | |
3849 | r = admin_socket->register_command( | |
7c673cae FG |
3850 | "injectmdataerr " \ |
3851 | "name=pool,type=CephString " \ | |
3852 | "name=objname,type=CephObjectname " \ | |
3853 | "name=shardid,type=CephInt,req=false,range=0|255", | |
3854 | test_ops_hook, | |
3855 | "inject metadata error to an object"); | |
11fdf7f2 | 3856 | ceph_assert(r == 0); |
7c673cae | 3857 | r = admin_socket->register_command( |
7c673cae FG |
3858 | "set_recovery_delay " \ |
3859 | "name=utime,type=CephInt,req=false", | |
3860 | test_ops_hook, | |
3861 | "Delay osd recovery by specified seconds"); | |
11fdf7f2 | 3862 | ceph_assert(r == 0); |
7c673cae | 3863 | r = admin_socket->register_command( |
7c673cae FG |
3864 | "injectfull " \ |
3865 | "name=type,type=CephString,req=false " \ | |
3866 | "name=count,type=CephInt,req=false ", | |
3867 | test_ops_hook, | |
3868 | "Inject a full disk (optional count times)"); | |
11fdf7f2 | 3869 | ceph_assert(r == 0); |
9f95a23c TL |
3870 | r = admin_socket->register_command( |
3871 | "bench " \ | |
3872 | "name=count,type=CephInt,req=false " \ | |
3873 | "name=size,type=CephInt,req=false " \ | |
3874 | "name=object_size,type=CephInt,req=false " \ | |
3875 | "name=object_num,type=CephInt,req=false ", | |
3876 | asok_hook, | |
3877 | "OSD benchmark: write <count> <size>-byte objects(with <obj_size> <obj_num>), " \ | |
3878 | "(default count=1G default size=4MB). Results in log."); | |
3879 | ceph_assert(r == 0); | |
3880 | r = admin_socket->register_command( | |
3881 | "cluster_log " \ | |
3882 | "name=level,type=CephChoices,strings=error,warning,info,debug " \ | |
3883 | "name=message,type=CephString,n=N", | |
3884 | asok_hook, | |
3885 | "log a message to the cluster log"); | |
3886 | ceph_assert(r == 0); | |
3887 | r = admin_socket->register_command( | |
3888 | "flush_pg_stats", | |
3889 | asok_hook, | |
3890 | "flush pg stats"); | |
3891 | ceph_assert(r == 0); | |
3892 | r = admin_socket->register_command( | |
3893 | "heap " \ | |
3894 | "name=heapcmd,type=CephChoices,strings=" \ | |
3895 | "dump|start_profiler|stop_profiler|release|get_release_rate|set_release_rate|stats " \ | |
3896 | "name=value,type=CephString,req=false", | |
3897 | asok_hook, | |
3898 | "show heap usage info (available only if compiled with tcmalloc)"); | |
3899 | ceph_assert(r == 0); | |
3900 | r = admin_socket->register_command( | |
3901 | "debug dump_missing " \ | |
3902 | "name=filename,type=CephFilepath", | |
3903 | asok_hook, | |
3904 | "dump missing objects to a named file"); | |
3905 | ceph_assert(r == 0); | |
3906 | r = admin_socket->register_command( | |
3907 | "debug kick_recovery_wq " \ | |
3908 | "name=delay,type=CephInt,range=0", | |
3909 | asok_hook, | |
3910 | "set osd_recovery_delay_start to <val>"); | |
3911 | ceph_assert(r == 0); | |
3912 | r = admin_socket->register_command( | |
3913 | "cpu_profiler " \ | |
3914 | "name=arg,type=CephChoices,strings=status|flush", | |
3915 | asok_hook, | |
3916 | "run cpu profiling on daemon"); | |
3917 | ceph_assert(r == 0); | |
3918 | r = admin_socket->register_command( | |
3919 | "dump_pg_recovery_stats", | |
3920 | asok_hook, | |
3921 | "dump pg recovery statistics"); | |
3922 | ceph_assert(r == 0); | |
3923 | r = admin_socket->register_command( | |
3924 | "reset_pg_recovery_stats", | |
3925 | asok_hook, | |
3926 | "reset pg recovery statistics"); | |
3927 | ceph_assert(r == 0); | |
3928 | r = admin_socket->register_command( | |
3929 | "cache drop", | |
3930 | asok_hook, | |
3931 | "Drop all OSD caches"); | |
3932 | ceph_assert(r == 0); | |
3933 | r = admin_socket->register_command( | |
3934 | "cache status", | |
3935 | asok_hook, | |
3936 | "Get OSD caches statistics"); | |
3937 | ceph_assert(r == 0); | |
3938 | r = admin_socket->register_command( | |
3939 | "scrub_purged_snaps", | |
3940 | asok_hook, | |
3941 | "Scrub purged_snaps vs snapmapper index"); | |
3942 | ceph_assert(r == 0); | |
7c673cae | 3943 | |
9f95a23c TL |
3944 | // -- pg commands -- |
3945 | // old form: ceph pg <pgid> command ... | |
3946 | r = admin_socket->register_command( | |
3947 | "pg " \ | |
3948 | "name=pgid,type=CephPgid " \ | |
3949 | "name=cmd,type=CephChoices,strings=query", | |
3950 | asok_hook, | |
3951 | ""); | |
3952 | ceph_assert(r == 0); | |
3953 | r = admin_socket->register_command( | |
3954 | "pg " \ | |
3955 | "name=pgid,type=CephPgid " \ | |
3956 | "name=cmd,type=CephChoices,strings=mark_unfound_lost " \ | |
3957 | "name=mulcmd,type=CephChoices,strings=revert|delete", | |
3958 | asok_hook, | |
3959 | ""); | |
3960 | ceph_assert(r == 0); | |
3961 | r = admin_socket->register_command( | |
3962 | "pg " \ | |
3963 | "name=pgid,type=CephPgid " \ | |
3964 | "name=cmd,type=CephChoices,strings=list_unfound " \ | |
3965 | "name=offset,type=CephString,req=false", | |
3966 | asok_hook, | |
3967 | ""); | |
3968 | ceph_assert(r == 0); | |
3969 | r = admin_socket->register_command( | |
3970 | "pg " \ | |
3971 | "name=pgid,type=CephPgid " \ | |
3972 | "name=cmd,type=CephChoices,strings=scrub " \ | |
3973 | "name=time,type=CephInt,req=false", | |
3974 | asok_hook, | |
3975 | ""); | |
3976 | ceph_assert(r == 0); | |
3977 | r = admin_socket->register_command( | |
3978 | "pg " \ | |
3979 | "name=pgid,type=CephPgid " \ | |
3980 | "name=cmd,type=CephChoices,strings=deep_scrub " \ | |
3981 | "name=time,type=CephInt,req=false", | |
3982 | asok_hook, | |
3983 | ""); | |
3984 | ceph_assert(r == 0); | |
3985 | // new form: tell <pgid> <cmd> for both cli and rest | |
3986 | r = admin_socket->register_command( | |
3987 | "query", | |
3988 | asok_hook, | |
3989 | "show details of a specific pg"); | |
3990 | ceph_assert(r == 0); | |
3991 | r = admin_socket->register_command( | |
3992 | "mark_unfound_lost " \ | |
3993 | "name=pgid,type=CephPgid,req=false " \ | |
3994 | "name=mulcmd,type=CephChoices,strings=revert|delete", | |
3995 | asok_hook, | |
3996 | "mark all unfound objects in this pg as lost, either removing or reverting to a prior version if one is available"); | |
3997 | ceph_assert(r == 0); | |
3998 | r = admin_socket->register_command( | |
3999 | "list_unfound " \ | |
4000 | "name=pgid,type=CephPgid,req=false " \ | |
4001 | "name=offset,type=CephString,req=false", | |
4002 | asok_hook, | |
4003 | "list unfound objects on this pg, perhaps starting at an offset given in JSON"); | |
4004 | ceph_assert(r == 0); | |
4005 | r = admin_socket->register_command( | |
4006 | "scrub " \ | |
4007 | "name=pgid,type=CephPgid,req=false " \ | |
4008 | "name=time,type=CephInt,req=false", | |
4009 | asok_hook, | |
4010 | "Trigger a scheduled scrub "); | |
4011 | ceph_assert(r == 0); | |
4012 | r = admin_socket->register_command( | |
4013 | "deep_scrub " \ | |
4014 | "name=pgid,type=CephPgid,req=false " \ | |
4015 | "name=time,type=CephInt,req=false", | |
4016 | asok_hook, | |
4017 | "Trigger a scheduled deep scrub "); | |
4018 | ceph_assert(r == 0); | |
4019 | } | |
7c673cae | 4020 | |
9f95a23c TL |
4021 | void OSD::create_logger() |
4022 | { | |
4023 | dout(10) << "create_logger" << dendl; | |
7c673cae | 4024 | |
9f95a23c | 4025 | logger = build_osd_logger(cct); |
7c673cae FG |
4026 | cct->get_perfcounters_collection()->add(logger); |
4027 | } | |
4028 | ||
4029 | void OSD::create_recoverystate_perf() | |
4030 | { | |
4031 | dout(10) << "create_recoverystate_perf" << dendl; | |
4032 | ||
9f95a23c | 4033 | recoverystate_perf = build_recoverystate_perf(cct); |
7c673cae FG |
4034 | cct->get_perfcounters_collection()->add(recoverystate_perf); |
4035 | } | |
4036 | ||
4037 | int OSD::shutdown() | |
4038 | { | |
92f5a8d4 TL |
4039 | if (cct->_conf->osd_fast_shutdown) { |
4040 | derr << "*** Immediate shutdown (osd_fast_shutdown=true) ***" << dendl; | |
4041 | cct->_log->flush(); | |
4042 | _exit(0); | |
4043 | } | |
4044 | ||
7c673cae FG |
4045 | if (!service.prepare_to_stop()) |
4046 | return 0; // already shutting down | |
9f95a23c | 4047 | osd_lock.lock(); |
7c673cae | 4048 | if (is_stopping()) { |
9f95a23c | 4049 | osd_lock.unlock(); |
7c673cae FG |
4050 | return 0; |
4051 | } | |
11fdf7f2 | 4052 | dout(0) << "shutdown" << dendl; |
7c673cae FG |
4053 | |
4054 | set_state(STATE_STOPPING); | |
4055 | ||
4056 | // Debugging | |
11fdf7f2 TL |
4057 | if (cct->_conf.get_val<bool>("osd_debug_shutdown")) { |
4058 | cct->_conf.set_val("debug_osd", "100"); | |
4059 | cct->_conf.set_val("debug_journal", "100"); | |
4060 | cct->_conf.set_val("debug_filestore", "100"); | |
4061 | cct->_conf.set_val("debug_bluestore", "100"); | |
4062 | cct->_conf.set_val("debug_ms", "100"); | |
4063 | cct->_conf.apply_changes(nullptr); | |
3efd9988 | 4064 | } |
7c673cae FG |
4065 | |
4066 | // stop MgrClient earlier as it's more like an internal consumer of OSD | |
4067 | mgrc.shutdown(); | |
4068 | ||
4069 | service.start_shutdown(); | |
4070 | ||
4071 | // stop sending work to pgs. this just prevents any new work in _process | |
4072 | // from racing with on_shutdown and potentially entering the pg after. | |
4073 | op_shardedwq.drain(); | |
4074 | ||
4075 | // Shutdown PGs | |
4076 | { | |
11fdf7f2 TL |
4077 | vector<PGRef> pgs; |
4078 | _get_pgs(&pgs); | |
4079 | for (auto pg : pgs) { | |
4080 | pg->shutdown(); | |
7c673cae FG |
4081 | } |
4082 | } | |
7c673cae FG |
4083 | |
4084 | // drain op queue again (in case PGs requeued something) | |
4085 | op_shardedwq.drain(); | |
4086 | { | |
4087 | finished.clear(); // zap waiters (bleh, this is messy) | |
11fdf7f2 | 4088 | waiting_for_osdmap.clear(); |
7c673cae FG |
4089 | } |
4090 | ||
7c673cae | 4091 | // unregister commands |
11fdf7f2 | 4092 | cct->get_admin_socket()->unregister_commands(asok_hook); |
7c673cae FG |
4093 | delete asok_hook; |
4094 | asok_hook = NULL; | |
4095 | ||
11fdf7f2 | 4096 | cct->get_admin_socket()->unregister_commands(test_ops_hook); |
7c673cae FG |
4097 | delete test_ops_hook; |
4098 | test_ops_hook = NULL; | |
4099 | ||
9f95a23c | 4100 | osd_lock.unlock(); |
7c673cae | 4101 | |
9f95a23c TL |
4102 | { |
4103 | std::lock_guard l{heartbeat_lock}; | |
4104 | heartbeat_stop = true; | |
4105 | heartbeat_cond.notify_all(); | |
4106 | heartbeat_peers.clear(); | |
4107 | } | |
7c673cae FG |
4108 | heartbeat_thread.join(); |
4109 | ||
9f95a23c TL |
4110 | hb_back_server_messenger->mark_down_all(); |
4111 | hb_front_server_messenger->mark_down_all(); | |
4112 | hb_front_client_messenger->mark_down_all(); | |
4113 | hb_back_client_messenger->mark_down_all(); | |
4114 | ||
7c673cae FG |
4115 | osd_op_tp.drain(); |
4116 | osd_op_tp.stop(); | |
4117 | dout(10) << "op sharded tp stopped" << dendl; | |
4118 | ||
7c673cae FG |
4119 | dout(10) << "stopping agent" << dendl; |
4120 | service.agent_stop(); | |
4121 | ||
11fdf7f2 TL |
4122 | boot_finisher.wait_for_empty(); |
4123 | ||
9f95a23c | 4124 | osd_lock.lock(); |
7c673cae | 4125 | |
11fdf7f2 | 4126 | boot_finisher.stop(); |
494da23a | 4127 | reset_heartbeat_peers(true); |
7c673cae FG |
4128 | |
4129 | tick_timer.shutdown(); | |
4130 | ||
4131 | { | |
11fdf7f2 | 4132 | std::lock_guard l(tick_timer_lock); |
7c673cae FG |
4133 | tick_timer_without_osd_lock.shutdown(); |
4134 | } | |
4135 | ||
4136 | // note unmount epoch | |
9f95a23c | 4137 | dout(10) << "noting clean unmount in epoch " << get_osdmap_epoch() << dendl; |
7c673cae | 4138 | superblock.mounted = service.get_boot_epoch(); |
9f95a23c | 4139 | superblock.clean_thru = get_osdmap_epoch(); |
7c673cae FG |
4140 | ObjectStore::Transaction t; |
4141 | write_superblock(t); | |
11fdf7f2 | 4142 | int r = store->queue_transaction(service.meta_ch, std::move(t)); |
7c673cae FG |
4143 | if (r) { |
4144 | derr << "OSD::shutdown: error writing superblock: " | |
4145 | << cpp_strerror(r) << dendl; | |
4146 | } | |
4147 | ||
4148 | ||
31f18b77 FG |
4149 | service.shutdown_reserver(); |
4150 | ||
7c673cae FG |
4151 | // Remove PGs |
4152 | #ifdef PG_DEBUG_REFS | |
4153 | service.dump_live_pgids(); | |
4154 | #endif | |
11fdf7f2 TL |
4155 | while (true) { |
4156 | vector<PGRef> pgs; | |
4157 | _get_pgs(&pgs, true); | |
4158 | if (pgs.empty()) { | |
4159 | break; | |
4160 | } | |
4161 | for (auto& pg : pgs) { | |
4162 | if (pg->is_deleted()) { | |
4163 | continue; | |
4164 | } | |
4165 | dout(20) << " kicking pg " << pg << dendl; | |
4166 | pg->lock(); | |
4167 | if (pg->get_num_ref() != 1) { | |
4168 | derr << "pgid " << pg->get_pgid() << " has ref count of " | |
4169 | << pg->get_num_ref() << dendl; | |
7c673cae | 4170 | #ifdef PG_DEBUG_REFS |
11fdf7f2 | 4171 | pg->dump_live_ids(); |
7c673cae | 4172 | #endif |
31f18b77 FG |
4173 | if (cct->_conf->osd_shutdown_pgref_assert) { |
4174 | ceph_abort(); | |
4175 | } | |
7c673cae | 4176 | } |
11fdf7f2 TL |
4177 | pg->ch.reset(); |
4178 | pg->unlock(); | |
7c673cae | 4179 | } |
7c673cae FG |
4180 | } |
4181 | #ifdef PG_DEBUG_REFS | |
4182 | service.dump_live_pgids(); | |
4183 | #endif | |
f64942e4 | 4184 | |
9f95a23c | 4185 | osd_lock.unlock(); |
11fdf7f2 | 4186 | cct->_conf.remove_observer(this); |
9f95a23c | 4187 | osd_lock.lock(); |
7c673cae | 4188 | |
11fdf7f2 TL |
4189 | service.meta_ch.reset(); |
4190 | ||
7c673cae FG |
4191 | dout(10) << "syncing store" << dendl; |
4192 | enable_disable_fuse(true); | |
4193 | ||
4194 | if (cct->_conf->osd_journal_flush_on_shutdown) { | |
4195 | dout(10) << "flushing journal" << dendl; | |
4196 | store->flush_journal(); | |
4197 | } | |
4198 | ||
7c673cae | 4199 | monc->shutdown(); |
9f95a23c TL |
4200 | osd_lock.unlock(); |
4201 | { | |
4202 | std::unique_lock l{map_lock}; | |
4203 | set_osdmap(OSDMapRef()); | |
4204 | } | |
11fdf7f2 TL |
4205 | for (auto s : shards) { |
4206 | std::lock_guard l(s->osdmap_lock); | |
4207 | s->shard_osdmap = OSDMapRef(); | |
4208 | } | |
7c673cae | 4209 | service.shutdown(); |
11fdf7f2 TL |
4210 | |
4211 | std::lock_guard lock(osd_lock); | |
4212 | store->umount(); | |
4213 | delete store; | |
4214 | store = nullptr; | |
4215 | dout(10) << "Store synced" << dendl; | |
4216 | ||
7c673cae FG |
4217 | op_tracker.on_shutdown(); |
4218 | ||
9f95a23c | 4219 | ClassHandler::get_instance().shutdown(); |
7c673cae FG |
4220 | client_messenger->shutdown(); |
4221 | cluster_messenger->shutdown(); | |
4222 | hb_front_client_messenger->shutdown(); | |
4223 | hb_back_client_messenger->shutdown(); | |
4224 | objecter_messenger->shutdown(); | |
4225 | hb_front_server_messenger->shutdown(); | |
4226 | hb_back_server_messenger->shutdown(); | |
4227 | ||
7c673cae FG |
4228 | return r; |
4229 | } | |
4230 | ||
4231 | int OSD::mon_cmd_maybe_osd_create(string &cmd) | |
4232 | { | |
4233 | bool created = false; | |
4234 | while (true) { | |
4235 | dout(10) << __func__ << " cmd: " << cmd << dendl; | |
4236 | vector<string> vcmd{cmd}; | |
4237 | bufferlist inbl; | |
4238 | C_SaferCond w; | |
4239 | string outs; | |
4240 | monc->start_mon_command(vcmd, inbl, NULL, &outs, &w); | |
4241 | int r = w.wait(); | |
4242 | if (r < 0) { | |
4243 | if (r == -ENOENT && !created) { | |
4244 | string newcmd = "{\"prefix\": \"osd create\", \"id\": " + stringify(whoami) | |
4245 | + ", \"uuid\": \"" + stringify(superblock.osd_fsid) + "\"}"; | |
4246 | vector<string> vnewcmd{newcmd}; | |
4247 | bufferlist inbl; | |
4248 | C_SaferCond w; | |
4249 | string outs; | |
4250 | monc->start_mon_command(vnewcmd, inbl, NULL, &outs, &w); | |
4251 | int r = w.wait(); | |
4252 | if (r < 0) { | |
4253 | derr << __func__ << " fail: osd does not exist and created failed: " | |
4254 | << cpp_strerror(r) << dendl; | |
4255 | return r; | |
4256 | } | |
4257 | created = true; | |
4258 | continue; | |
4259 | } | |
4260 | derr << __func__ << " fail: '" << outs << "': " << cpp_strerror(r) << dendl; | |
4261 | return r; | |
4262 | } | |
4263 | break; | |
4264 | } | |
4265 | ||
4266 | return 0; | |
4267 | } | |
4268 | ||
4269 | int OSD::update_crush_location() | |
4270 | { | |
4271 | if (!cct->_conf->osd_crush_update_on_start) { | |
4272 | dout(10) << __func__ << " osd_crush_update_on_start = false" << dendl; | |
4273 | return 0; | |
4274 | } | |
4275 | ||
4276 | char weight[32]; | |
4277 | if (cct->_conf->osd_crush_initial_weight >= 0) { | |
4278 | snprintf(weight, sizeof(weight), "%.4lf", cct->_conf->osd_crush_initial_weight); | |
4279 | } else { | |
4280 | struct store_statfs_t st; | |
11fdf7f2 TL |
4281 | osd_alert_list_t alerts; |
4282 | int r = store->statfs(&st, &alerts); | |
7c673cae FG |
4283 | if (r < 0) { |
4284 | derr << "statfs: " << cpp_strerror(r) << dendl; | |
4285 | return r; | |
4286 | } | |
4287 | snprintf(weight, sizeof(weight), "%.4lf", | |
11fdf7f2 TL |
4288 | std::max(.00001, |
4289 | double(st.total) / | |
4290 | double(1ull << 40 /* TB */))); | |
7c673cae FG |
4291 | } |
4292 | ||
9f95a23c | 4293 | dout(10) << __func__ << " crush location is " << cct->crush_location << dendl; |
7c673cae FG |
4294 | |
4295 | string cmd = | |
4296 | string("{\"prefix\": \"osd crush create-or-move\", ") + | |
9f95a23c TL |
4297 | string("\"id\": ") + stringify(whoami) + ", " + |
4298 | string("\"weight\":") + weight + ", " + | |
4299 | string("\"args\": [") + stringify(cct->crush_location) + "]}"; | |
7c673cae FG |
4300 | return mon_cmd_maybe_osd_create(cmd); |
4301 | } | |
4302 | ||
4303 | int OSD::update_crush_device_class() | |
4304 | { | |
224ce89b WB |
4305 | if (!cct->_conf->osd_class_update_on_start) { |
4306 | dout(10) << __func__ << " osd_class_update_on_start = false" << dendl; | |
4307 | return 0; | |
4308 | } | |
4309 | ||
7c673cae FG |
4310 | string device_class; |
4311 | int r = store->read_meta("crush_device_class", &device_class); | |
224ce89b WB |
4312 | if (r < 0 || device_class.empty()) { |
4313 | device_class = store->get_default_device_class(); | |
4314 | } | |
4315 | ||
4316 | if (device_class.empty()) { | |
d2e6a577 | 4317 | dout(20) << __func__ << " no device class stored locally" << dendl; |
7c673cae | 4318 | return 0; |
224ce89b | 4319 | } |
7c673cae FG |
4320 | |
4321 | string cmd = | |
4322 | string("{\"prefix\": \"osd crush set-device-class\", ") + | |
224ce89b WB |
4323 | string("\"class\": \"") + device_class + string("\", ") + |
4324 | string("\"ids\": [\"") + stringify(whoami) + string("\"]}"); | |
7c673cae | 4325 | |
224ce89b | 4326 | r = mon_cmd_maybe_osd_create(cmd); |
11fdf7f2 TL |
4327 | if (r == -EBUSY) { |
4328 | // good, already bound to a device-class | |
4329 | return 0; | |
4330 | } else { | |
4331 | return r; | |
4332 | } | |
7c673cae FG |
4333 | } |
4334 | ||
4335 | void OSD::write_superblock(ObjectStore::Transaction& t) | |
4336 | { | |
4337 | dout(10) << "write_superblock " << superblock << dendl; | |
4338 | ||
4339 | //hack: at minimum it's using the baseline feature set | |
4340 | if (!superblock.compat_features.incompat.contains(CEPH_OSD_FEATURE_INCOMPAT_BASE)) | |
4341 | superblock.compat_features.incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_BASE); | |
4342 | ||
4343 | bufferlist bl; | |
11fdf7f2 | 4344 | encode(superblock, bl); |
7c673cae FG |
4345 | t.write(coll_t::meta(), OSD_SUPERBLOCK_GOBJECT, 0, bl.length(), bl); |
4346 | } | |
4347 | ||
4348 | int OSD::read_superblock() | |
4349 | { | |
4350 | bufferlist bl; | |
11fdf7f2 | 4351 | int r = store->read(service.meta_ch, OSD_SUPERBLOCK_GOBJECT, 0, 0, bl); |
7c673cae FG |
4352 | if (r < 0) |
4353 | return r; | |
4354 | ||
11fdf7f2 TL |
4355 | auto p = bl.cbegin(); |
4356 | decode(superblock, p); | |
7c673cae FG |
4357 | |
4358 | dout(10) << "read_superblock " << superblock << dendl; | |
4359 | ||
4360 | return 0; | |
4361 | } | |
4362 | ||
4363 | void OSD::clear_temp_objects() | |
4364 | { | |
4365 | dout(10) << __func__ << dendl; | |
4366 | vector<coll_t> ls; | |
4367 | store->list_collections(ls); | |
4368 | for (vector<coll_t>::iterator p = ls.begin(); p != ls.end(); ++p) { | |
4369 | spg_t pgid; | |
4370 | if (!p->is_pg(&pgid)) | |
4371 | continue; | |
4372 | ||
4373 | // list temp objects | |
4374 | dout(20) << " clearing temps in " << *p << " pgid " << pgid << dendl; | |
4375 | ||
4376 | vector<ghobject_t> temps; | |
4377 | ghobject_t next; | |
4378 | while (1) { | |
4379 | vector<ghobject_t> objects; | |
11fdf7f2 TL |
4380 | auto ch = store->open_collection(*p); |
4381 | ceph_assert(ch); | |
4382 | store->collection_list(ch, next, ghobject_t::get_max(), | |
7c673cae FG |
4383 | store->get_ideal_list_max(), |
4384 | &objects, &next); | |
4385 | if (objects.empty()) | |
4386 | break; | |
4387 | vector<ghobject_t>::iterator q; | |
4388 | for (q = objects.begin(); q != objects.end(); ++q) { | |
4389 | // Hammer set pool for temps to -1, so check for clean-up | |
4390 | if (q->hobj.is_temp() || (q->hobj.pool == -1)) { | |
4391 | temps.push_back(*q); | |
4392 | } else { | |
4393 | break; | |
4394 | } | |
4395 | } | |
4396 | // If we saw a non-temp object and hit the break above we can | |
4397 | // break out of the while loop too. | |
4398 | if (q != objects.end()) | |
4399 | break; | |
4400 | } | |
4401 | if (!temps.empty()) { | |
4402 | ObjectStore::Transaction t; | |
4403 | int removed = 0; | |
4404 | for (vector<ghobject_t>::iterator q = temps.begin(); q != temps.end(); ++q) { | |
4405 | dout(20) << " removing " << *p << " object " << *q << dendl; | |
4406 | t.remove(*p, *q); | |
4407 | if (++removed > cct->_conf->osd_target_transaction_size) { | |
11fdf7f2 | 4408 | store->queue_transaction(service.meta_ch, std::move(t)); |
7c673cae FG |
4409 | t = ObjectStore::Transaction(); |
4410 | removed = 0; | |
4411 | } | |
4412 | } | |
4413 | if (removed) { | |
11fdf7f2 | 4414 | store->queue_transaction(service.meta_ch, std::move(t)); |
7c673cae FG |
4415 | } |
4416 | } | |
4417 | } | |
4418 | } | |
4419 | ||
4420 | void OSD::recursive_remove_collection(CephContext* cct, | |
4421 | ObjectStore *store, spg_t pgid, | |
4422 | coll_t tmp) | |
4423 | { | |
4424 | OSDriver driver( | |
4425 | store, | |
4426 | coll_t(), | |
4427 | make_snapmapper_oid()); | |
4428 | ||
11fdf7f2 | 4429 | ObjectStore::CollectionHandle ch = store->open_collection(tmp); |
7c673cae FG |
4430 | ObjectStore::Transaction t; |
4431 | SnapMapper mapper(cct, &driver, 0, 0, 0, pgid.shard); | |
4432 | ||
11fdf7f2 TL |
4433 | ghobject_t next; |
4434 | int max = cct->_conf->osd_target_transaction_size; | |
7c673cae | 4435 | vector<ghobject_t> objects; |
11fdf7f2 TL |
4436 | objects.reserve(max); |
4437 | while (true) { | |
4438 | objects.clear(); | |
4439 | store->collection_list(ch, next, ghobject_t::get_max(), | |
4440 | max, &objects, &next); | |
4441 | generic_dout(10) << __func__ << " " << objects << dendl; | |
4442 | if (objects.empty()) | |
4443 | break; | |
4444 | for (auto& p: objects) { | |
4445 | OSDriver::OSTransaction _t(driver.get_transaction(&t)); | |
4446 | int r = mapper.remove_oid(p.hobj, &_t); | |
4447 | if (r != 0 && r != -ENOENT) | |
4448 | ceph_abort(); | |
4449 | t.remove(tmp, p); | |
7c673cae | 4450 | } |
11fdf7f2 TL |
4451 | int r = store->queue_transaction(ch, std::move(t)); |
4452 | ceph_assert(r == 0); | |
4453 | t = ObjectStore::Transaction(); | |
7c673cae FG |
4454 | } |
4455 | t.remove_collection(tmp); | |
11fdf7f2 TL |
4456 | int r = store->queue_transaction(ch, std::move(t)); |
4457 | ceph_assert(r == 0); | |
7c673cae FG |
4458 | |
4459 | C_SaferCond waiter; | |
11fdf7f2 | 4460 | if (!ch->flush_commit(&waiter)) { |
7c673cae FG |
4461 | waiter.wait(); |
4462 | } | |
4463 | } | |
4464 | ||
4465 | ||
4466 | // ====================================================== | |
4467 | // PG's | |
4468 | ||
7c673cae FG |
4469 | PG* OSD::_make_pg( |
4470 | OSDMapRef createmap, | |
4471 | spg_t pgid) | |
4472 | { | |
11fdf7f2 TL |
4473 | dout(10) << __func__ << " " << pgid << dendl; |
4474 | pg_pool_t pi; | |
4475 | map<string,string> ec_profile; | |
4476 | string name; | |
4477 | if (createmap->have_pg_pool(pgid.pool())) { | |
4478 | pi = *createmap->get_pg_pool(pgid.pool()); | |
4479 | name = createmap->get_pool_name(pgid.pool()); | |
4480 | if (pi.is_erasure()) { | |
4481 | ec_profile = createmap->get_erasure_code_profile(pi.erasure_code_profile); | |
4482 | } | |
4483 | } else { | |
4484 | // pool was deleted; grab final pg_pool_t off disk. | |
4485 | ghobject_t oid = make_final_pool_info_oid(pgid.pool()); | |
4486 | bufferlist bl; | |
4487 | int r = store->read(service.meta_ch, oid, 0, 0, bl); | |
4488 | if (r < 0) { | |
4489 | derr << __func__ << " missing pool " << pgid.pool() << " tombstone" | |
4490 | << dendl; | |
4491 | return nullptr; | |
4492 | } | |
4493 | ceph_assert(r >= 0); | |
4494 | auto p = bl.cbegin(); | |
4495 | decode(pi, p); | |
4496 | decode(name, p); | |
4497 | if (p.end()) { // dev release v13.0.2 did not include ec_profile | |
4498 | derr << __func__ << " missing ec_profile from pool " << pgid.pool() | |
4499 | << " tombstone" << dendl; | |
4500 | return nullptr; | |
4501 | } | |
4502 | decode(ec_profile, p); | |
4503 | } | |
4504 | PGPool pool(cct, createmap, pgid.pool(), pi, name); | |
7c673cae | 4505 | PG *pg; |
11fdf7f2 TL |
4506 | if (pi.type == pg_pool_t::TYPE_REPLICATED || |
4507 | pi.type == pg_pool_t::TYPE_ERASURE) | |
4508 | pg = new PrimaryLogPG(&service, createmap, pool, ec_profile, pgid); | |
7c673cae FG |
4509 | else |
4510 | ceph_abort(); | |
7c673cae FG |
4511 | return pg; |
4512 | } | |
4513 | ||
11fdf7f2 | 4514 | void OSD::_get_pgs(vector<PGRef> *v, bool clear_too) |
7c673cae | 4515 | { |
11fdf7f2 TL |
4516 | v->clear(); |
4517 | v->reserve(get_num_pgs()); | |
4518 | for (auto& s : shards) { | |
4519 | std::lock_guard l(s->shard_lock); | |
4520 | for (auto& j : s->pg_slots) { | |
4521 | if (j.second->pg && | |
4522 | !j.second->pg->is_deleted()) { | |
4523 | v->push_back(j.second->pg); | |
4524 | if (clear_too) { | |
4525 | s->_detach_pg(j.second.get()); | |
4526 | } | |
4527 | } | |
7c673cae | 4528 | } |
7c673cae | 4529 | } |
7c673cae FG |
4530 | } |
4531 | ||
11fdf7f2 | 4532 | void OSD::_get_pgids(vector<spg_t> *v) |
7c673cae | 4533 | { |
11fdf7f2 TL |
4534 | v->clear(); |
4535 | v->reserve(get_num_pgs()); | |
4536 | for (auto& s : shards) { | |
4537 | std::lock_guard l(s->shard_lock); | |
4538 | for (auto& j : s->pg_slots) { | |
4539 | if (j.second->pg && | |
4540 | !j.second->pg->is_deleted()) { | |
4541 | v->push_back(j.first); | |
4542 | } | |
7c673cae FG |
4543 | } |
4544 | } | |
7c673cae FG |
4545 | } |
4546 | ||
11fdf7f2 | 4547 | void OSD::register_pg(PGRef pg) |
7c673cae | 4548 | { |
11fdf7f2 TL |
4549 | spg_t pgid = pg->get_pgid(); |
4550 | uint32_t shard_index = pgid.hash_to_shard(num_shards); | |
4551 | auto sdata = shards[shard_index]; | |
4552 | std::lock_guard l(sdata->shard_lock); | |
4553 | auto r = sdata->pg_slots.emplace(pgid, make_unique<OSDShardPGSlot>()); | |
4554 | ceph_assert(r.second); | |
4555 | auto *slot = r.first->second.get(); | |
4556 | dout(20) << __func__ << " " << pgid << " " << pg << dendl; | |
4557 | sdata->_attach_pg(slot, pg.get()); | |
4558 | } | |
7c673cae | 4559 | |
11fdf7f2 TL |
4560 | bool OSD::try_finish_pg_delete(PG *pg, unsigned old_pg_num) |
4561 | { | |
4562 | auto sdata = pg->osd_shard; | |
4563 | ceph_assert(sdata); | |
4564 | { | |
4565 | std::lock_guard l(sdata->shard_lock); | |
4566 | auto p = sdata->pg_slots.find(pg->pg_id); | |
4567 | if (p == sdata->pg_slots.end() || | |
4568 | !p->second->pg) { | |
4569 | dout(20) << __func__ << " " << pg->pg_id << " not found" << dendl; | |
4570 | return false; | |
4571 | } | |
4572 | if (p->second->waiting_for_merge_epoch) { | |
4573 | dout(20) << __func__ << " " << pg->pg_id << " waiting for merge" << dendl; | |
4574 | return false; | |
4575 | } | |
4576 | dout(20) << __func__ << " " << pg->pg_id << " " << pg << dendl; | |
4577 | sdata->_detach_pg(p->second.get()); | |
4578 | } | |
7c673cae | 4579 | |
11fdf7f2 TL |
4580 | for (auto shard : shards) { |
4581 | shard->unprime_split_children(pg->pg_id, old_pg_num); | |
4582 | } | |
7c673cae | 4583 | |
11fdf7f2 TL |
4584 | // update pg count now since we might not get an osdmap any time soon. |
4585 | if (pg->is_primary()) | |
4586 | service.logger->dec(l_osd_pg_primary); | |
9f95a23c TL |
4587 | else if (pg->is_nonprimary()) |
4588 | service.logger->dec(l_osd_pg_replica); // misnomver | |
11fdf7f2 TL |
4589 | else |
4590 | service.logger->dec(l_osd_pg_stray); | |
7c673cae | 4591 | |
11fdf7f2 | 4592 | return true; |
7c673cae FG |
4593 | } |
4594 | ||
11fdf7f2 | 4595 | PGRef OSD::_lookup_pg(spg_t pgid) |
7c673cae | 4596 | { |
11fdf7f2 TL |
4597 | uint32_t shard_index = pgid.hash_to_shard(num_shards); |
4598 | auto sdata = shards[shard_index]; | |
4599 | std::lock_guard l(sdata->shard_lock); | |
4600 | auto p = sdata->pg_slots.find(pgid); | |
4601 | if (p == sdata->pg_slots.end()) { | |
7c673cae | 4602 | return nullptr; |
11fdf7f2 TL |
4603 | } |
4604 | return p->second->pg; | |
7c673cae FG |
4605 | } |
4606 | ||
11fdf7f2 | 4607 | PGRef OSD::_lookup_lock_pg(spg_t pgid) |
31f18b77 | 4608 | { |
11fdf7f2 TL |
4609 | PGRef pg = _lookup_pg(pgid); |
4610 | if (!pg) { | |
4611 | return nullptr; | |
4612 | } | |
4613 | pg->lock(); | |
4614 | if (!pg->is_deleted()) { | |
4615 | return pg; | |
4616 | } | |
4617 | pg->unlock(); | |
4618 | return nullptr; | |
31f18b77 FG |
4619 | } |
4620 | ||
11fdf7f2 | 4621 | PGRef OSD::lookup_lock_pg(spg_t pgid) |
7c673cae | 4622 | { |
11fdf7f2 | 4623 | return _lookup_lock_pg(pgid); |
7c673cae FG |
4624 | } |
4625 | ||
4626 | void OSD::load_pgs() | |
4627 | { | |
9f95a23c | 4628 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae | 4629 | dout(0) << "load_pgs" << dendl; |
11fdf7f2 | 4630 | |
7c673cae | 4631 | { |
11fdf7f2 TL |
4632 | auto pghist = make_pg_num_history_oid(); |
4633 | bufferlist bl; | |
4634 | int r = store->read(service.meta_ch, pghist, 0, 0, bl, 0); | |
4635 | if (r >= 0 && bl.length() > 0) { | |
4636 | auto p = bl.cbegin(); | |
4637 | decode(pg_num_history, p); | |
4638 | } | |
4639 | dout(20) << __func__ << " pg_num_history " << pg_num_history << dendl; | |
7c673cae FG |
4640 | } |
4641 | ||
4642 | vector<coll_t> ls; | |
4643 | int r = store->list_collections(ls); | |
4644 | if (r < 0) { | |
4645 | derr << "failed to list pgs: " << cpp_strerror(-r) << dendl; | |
4646 | } | |
4647 | ||
11fdf7f2 | 4648 | int num = 0; |
7c673cae FG |
4649 | for (vector<coll_t>::iterator it = ls.begin(); |
4650 | it != ls.end(); | |
4651 | ++it) { | |
4652 | spg_t pgid; | |
4653 | if (it->is_temp(&pgid) || | |
4654 | (it->is_pg(&pgid) && PG::_has_removal_flag(store, pgid))) { | |
11fdf7f2 TL |
4655 | dout(10) << "load_pgs " << *it |
4656 | << " removing, legacy or flagged for removal pg" << dendl; | |
7c673cae FG |
4657 | recursive_remove_collection(cct, store, pgid, *it); |
4658 | continue; | |
4659 | } | |
4660 | ||
4661 | if (!it->is_pg(&pgid)) { | |
4662 | dout(10) << "load_pgs ignoring unrecognized " << *it << dendl; | |
4663 | continue; | |
4664 | } | |
4665 | ||
7c673cae | 4666 | dout(10) << "pgid " << pgid << " coll " << coll_t(pgid) << dendl; |
7c673cae | 4667 | epoch_t map_epoch = 0; |
11fdf7f2 | 4668 | int r = PG::peek_map_epoch(store, pgid, &map_epoch); |
7c673cae FG |
4669 | if (r < 0) { |
4670 | derr << __func__ << " unable to peek at " << pgid << " metadata, skipping" | |
4671 | << dendl; | |
4672 | continue; | |
4673 | } | |
4674 | ||
11fdf7f2 | 4675 | PGRef pg; |
7c673cae FG |
4676 | if (map_epoch > 0) { |
4677 | OSDMapRef pgosdmap = service.try_get_map(map_epoch); | |
4678 | if (!pgosdmap) { | |
9f95a23c | 4679 | if (!get_osdmap()->have_pg_pool(pgid.pool())) { |
7c673cae FG |
4680 | derr << __func__ << ": could not find map for epoch " << map_epoch |
4681 | << " on pg " << pgid << ", but the pool is not present in the " | |
4682 | << "current map, so this is probably a result of bug 10617. " | |
4683 | << "Skipping the pg for now, you can use ceph-objectstore-tool " | |
4684 | << "to clean it up later." << dendl; | |
4685 | continue; | |
4686 | } else { | |
4687 | derr << __func__ << ": have pgid " << pgid << " at epoch " | |
4688 | << map_epoch << ", but missing map. Crashing." | |
4689 | << dendl; | |
11fdf7f2 | 4690 | ceph_abort_msg("Missing map in load_pgs"); |
7c673cae FG |
4691 | } |
4692 | } | |
11fdf7f2 | 4693 | pg = _make_pg(pgosdmap, pgid); |
7c673cae | 4694 | } else { |
9f95a23c | 4695 | pg = _make_pg(get_osdmap(), pgid); |
7c673cae | 4696 | } |
11fdf7f2 TL |
4697 | if (!pg) { |
4698 | recursive_remove_collection(cct, store, pgid, *it); | |
4699 | continue; | |
4700 | } | |
4701 | ||
4702 | // there can be no waiters here, so we don't call _wake_pg_slot | |
7c673cae | 4703 | |
11fdf7f2 | 4704 | pg->lock(); |
7c673cae FG |
4705 | pg->ch = store->open_collection(pg->coll); |
4706 | ||
4707 | // read pg state, log | |
11fdf7f2 | 4708 | pg->read_state(store); |
7c673cae | 4709 | |
94b18763 FG |
4710 | if (pg->dne()) { |
4711 | dout(10) << "load_pgs " << *it << " deleting dne" << dendl; | |
4712 | pg->ch = nullptr; | |
94b18763 | 4713 | pg->unlock(); |
94b18763 FG |
4714 | recursive_remove_collection(cct, store, pgid, *it); |
4715 | continue; | |
4716 | } | |
11fdf7f2 TL |
4717 | { |
4718 | uint32_t shard_index = pgid.hash_to_shard(shards.size()); | |
4719 | assert(NULL != shards[shard_index]); | |
4720 | store->set_collection_commit_queue(pg->coll, &(shards[shard_index]->context_queue)); | |
4721 | } | |
7c673cae FG |
4722 | |
4723 | pg->reg_next_scrub(); | |
4724 | ||
11fdf7f2 | 4725 | dout(10) << __func__ << " loaded " << *pg << dendl; |
7c673cae | 4726 | pg->unlock(); |
7c673cae | 4727 | |
11fdf7f2 TL |
4728 | register_pg(pg); |
4729 | ++num; | |
7c673cae | 4730 | } |
11fdf7f2 | 4731 | dout(0) << __func__ << " opened " << num << " pgs" << dendl; |
7c673cae FG |
4732 | } |
4733 | ||
4734 | ||
11fdf7f2 TL |
4735 | PGRef OSD::handle_pg_create_info(const OSDMapRef& osdmap, |
4736 | const PGCreateInfo *info) | |
4737 | { | |
4738 | spg_t pgid = info->pgid; | |
7c673cae | 4739 | |
11fdf7f2 TL |
4740 | if (maybe_wait_for_max_pg(osdmap, pgid, info->by_mon)) { |
4741 | dout(10) << __func__ << " hit max pg, dropping" << dendl; | |
4742 | return nullptr; | |
4743 | } | |
3efd9988 | 4744 | |
9f95a23c | 4745 | PeeringCtx rctx = create_context(); |
7c673cae | 4746 | |
11fdf7f2 | 4747 | OSDMapRef startmap = get_map(info->epoch); |
7c673cae | 4748 | |
11fdf7f2 TL |
4749 | if (info->by_mon) { |
4750 | int64_t pool_id = pgid.pgid.pool(); | |
4751 | const pg_pool_t *pool = osdmap->get_pg_pool(pool_id); | |
4752 | if (!pool) { | |
4753 | dout(10) << __func__ << " ignoring " << pgid << ", pool dne" << dendl; | |
4754 | return nullptr; | |
4755 | } | |
9f95a23c | 4756 | if (osdmap->require_osd_release >= ceph_release_t::nautilus && |
11fdf7f2 TL |
4757 | !pool->has_flag(pg_pool_t::FLAG_CREATING)) { |
4758 | // this ensures we do not process old creating messages after the | |
4759 | // pool's initial pgs have been created (and pg are subsequently | |
4760 | // allowed to split or merge). | |
4761 | dout(20) << __func__ << " dropping " << pgid | |
4762 | << "create, pool does not have CREATING flag set" << dendl; | |
4763 | return nullptr; | |
7c673cae FG |
4764 | } |
4765 | } | |
7c673cae | 4766 | |
11fdf7f2 TL |
4767 | int up_primary, acting_primary; |
4768 | vector<int> up, acting; | |
4769 | startmap->pg_to_up_acting_osds( | |
4770 | pgid.pgid, &up, &up_primary, &acting, &acting_primary); | |
7c673cae | 4771 | |
11fdf7f2 TL |
4772 | const pg_pool_t* pp = startmap->get_pg_pool(pgid.pool()); |
4773 | if (pp->has_flag(pg_pool_t::FLAG_EC_OVERWRITES) && | |
4774 | store->get_type() != "bluestore") { | |
4775 | clog->warn() << "pg " << pgid | |
4776 | << " is at risk of silent data corruption: " | |
4777 | << "the pool allows ec overwrites but is not stored in " | |
4778 | << "bluestore, so deep scrubbing will not detect bitrot"; | |
7c673cae | 4779 | } |
9f95a23c TL |
4780 | create_pg_collection( |
4781 | rctx.transaction, pgid, pgid.get_split_bits(pp->get_pg_num())); | |
4782 | init_pg_ondisk(rctx.transaction, pgid, pp); | |
7c673cae | 4783 | |
9f95a23c | 4784 | int role = startmap->calc_pg_role(pg_shard_t(whoami, pgid.shard), acting); |
7c673cae | 4785 | |
11fdf7f2 TL |
4786 | PGRef pg = _make_pg(startmap, pgid); |
4787 | pg->ch = store->create_new_collection(pg->coll); | |
7c673cae | 4788 | |
11fdf7f2 TL |
4789 | { |
4790 | uint32_t shard_index = pgid.hash_to_shard(shards.size()); | |
4791 | assert(NULL != shards[shard_index]); | |
4792 | store->set_collection_commit_queue(pg->coll, &(shards[shard_index]->context_queue)); | |
7c673cae | 4793 | } |
7c673cae | 4794 | |
11fdf7f2 | 4795 | pg->lock(true); |
7c673cae | 4796 | |
11fdf7f2 TL |
4797 | // we are holding the shard lock |
4798 | ceph_assert(!pg->is_deleted()); | |
4799 | ||
4800 | pg->init( | |
4801 | role, | |
4802 | up, | |
4803 | up_primary, | |
4804 | acting, | |
4805 | acting_primary, | |
4806 | info->history, | |
4807 | info->past_intervals, | |
4808 | false, | |
4809 | rctx.transaction); | |
7c673cae | 4810 | |
92f5a8d4 TL |
4811 | pg->init_collection_pool_opts(); |
4812 | ||
11fdf7f2 | 4813 | if (pg->is_primary()) { |
9f95a23c | 4814 | std::lock_guard locker{m_perf_queries_lock}; |
11fdf7f2 TL |
4815 | pg->set_dynamic_perf_stats_queries(m_perf_queries); |
4816 | } | |
7c673cae | 4817 | |
9f95a23c TL |
4818 | pg->handle_initialize(rctx); |
4819 | pg->handle_activate_map(rctx); | |
7c673cae | 4820 | |
11fdf7f2 | 4821 | dispatch_context(rctx, pg.get(), osdmap, nullptr); |
7c673cae | 4822 | |
11fdf7f2 TL |
4823 | dout(10) << __func__ << " new pg " << *pg << dendl; |
4824 | return pg; | |
7c673cae FG |
4825 | } |
4826 | ||
11fdf7f2 TL |
4827 | bool OSD::maybe_wait_for_max_pg(const OSDMapRef& osdmap, |
4828 | spg_t pgid, | |
4829 | bool is_mon_create) | |
3efd9988 FG |
4830 | { |
4831 | const auto max_pgs_per_osd = | |
11fdf7f2 TL |
4832 | (cct->_conf.get_val<uint64_t>("mon_max_pg_per_osd") * |
4833 | cct->_conf.get_val<double>("osd_max_pg_per_osd_hard_ratio")); | |
3efd9988 | 4834 | |
11fdf7f2 | 4835 | if (num_pgs < max_pgs_per_osd) { |
3efd9988 FG |
4836 | return false; |
4837 | } | |
11fdf7f2 TL |
4838 | |
4839 | std::lock_guard l(pending_creates_lock); | |
3efd9988 FG |
4840 | if (is_mon_create) { |
4841 | pending_creates_from_mon++; | |
4842 | } else { | |
9f95a23c TL |
4843 | bool is_primary = osdmap->get_pg_acting_role(pgid, whoami) == 0; |
4844 | pending_creates_from_osd.emplace(pgid, is_primary); | |
3efd9988 | 4845 | } |
1adf2230 | 4846 | dout(1) << __func__ << " withhold creation of pg " << pgid |
11fdf7f2 | 4847 | << ": " << num_pgs << " >= "<< max_pgs_per_osd << dendl; |
3efd9988 FG |
4848 | return true; |
4849 | } | |
4850 | ||
4851 | // to re-trigger a peering, we have to twiddle the pg mapping a little bit, | |
4852 | // see PG::should_restart_peering(). OSDMap::pg_to_up_acting_osds() will turn | |
4853 | // to up set if pg_temp is empty. so an empty pg_temp won't work. | |
4854 | static vector<int32_t> twiddle(const vector<int>& acting) { | |
4855 | if (acting.size() > 1) { | |
4856 | return {acting[0]}; | |
4857 | } else { | |
4858 | vector<int32_t> twiddled(acting.begin(), acting.end()); | |
4859 | twiddled.push_back(-1); | |
4860 | return twiddled; | |
4861 | } | |
4862 | } | |
4863 | ||
4864 | void OSD::resume_creating_pg() | |
4865 | { | |
4866 | bool do_sub_pg_creates = false; | |
b32b8144 | 4867 | bool have_pending_creates = false; |
3efd9988 FG |
4868 | { |
4869 | const auto max_pgs_per_osd = | |
11fdf7f2 TL |
4870 | (cct->_conf.get_val<uint64_t>("mon_max_pg_per_osd") * |
4871 | cct->_conf.get_val<double>("osd_max_pg_per_osd_hard_ratio")); | |
4872 | if (max_pgs_per_osd <= num_pgs) { | |
3efd9988 FG |
4873 | // this could happen if admin decreases this setting before a PG is removed |
4874 | return; | |
4875 | } | |
11fdf7f2 TL |
4876 | unsigned spare_pgs = max_pgs_per_osd - num_pgs; |
4877 | std::lock_guard l(pending_creates_lock); | |
3efd9988 | 4878 | if (pending_creates_from_mon > 0) { |
11fdf7f2 TL |
4879 | dout(20) << __func__ << " pending_creates_from_mon " |
4880 | << pending_creates_from_mon << dendl; | |
3efd9988 FG |
4881 | do_sub_pg_creates = true; |
4882 | if (pending_creates_from_mon >= spare_pgs) { | |
4883 | spare_pgs = pending_creates_from_mon = 0; | |
4884 | } else { | |
4885 | spare_pgs -= pending_creates_from_mon; | |
4886 | pending_creates_from_mon = 0; | |
4887 | } | |
4888 | } | |
4889 | auto pg = pending_creates_from_osd.cbegin(); | |
4890 | while (spare_pgs > 0 && pg != pending_creates_from_osd.cend()) { | |
94b18763 | 4891 | dout(20) << __func__ << " pg " << pg->first << dendl; |
3efd9988 | 4892 | vector<int> acting; |
9f95a23c TL |
4893 | get_osdmap()->pg_to_up_acting_osds(pg->first.pgid, nullptr, nullptr, &acting, nullptr); |
4894 | service.queue_want_pg_temp(pg->first.pgid, twiddle(acting), true); | |
3efd9988 | 4895 | pg = pending_creates_from_osd.erase(pg); |
94b18763 | 4896 | do_sub_pg_creates = true; |
3efd9988 FG |
4897 | spare_pgs--; |
4898 | } | |
b32b8144 FG |
4899 | have_pending_creates = (pending_creates_from_mon > 0 || |
4900 | !pending_creates_from_osd.empty()); | |
3efd9988 | 4901 | } |
b32b8144 FG |
4902 | |
4903 | bool do_renew_subs = false; | |
3efd9988 FG |
4904 | if (do_sub_pg_creates) { |
4905 | if (monc->sub_want("osd_pg_creates", last_pg_create_epoch, 0)) { | |
4906 | dout(4) << __func__ << ": resolicit pg creates from mon since " | |
4907 | << last_pg_create_epoch << dendl; | |
b32b8144 | 4908 | do_renew_subs = true; |
3efd9988 FG |
4909 | } |
4910 | } | |
9f95a23c | 4911 | version_t start = get_osdmap_epoch() + 1; |
b32b8144 FG |
4912 | if (have_pending_creates) { |
4913 | // don't miss any new osdmap deleting PGs | |
4914 | if (monc->sub_want("osdmap", start, 0)) { | |
4915 | dout(4) << __func__ << ": resolicit osdmap from mon since " | |
4916 | << start << dendl; | |
4917 | do_renew_subs = true; | |
4918 | } | |
94b18763 | 4919 | } else if (do_sub_pg_creates) { |
b32b8144 FG |
4920 | // no need to subscribe the osdmap continuously anymore |
4921 | // once the pgtemp and/or mon_subscribe(pg_creates) is sent | |
4922 | if (monc->sub_want_increment("osdmap", start, CEPH_SUBSCRIBE_ONETIME)) { | |
11fdf7f2 | 4923 | dout(4) << __func__ << ": re-subscribe osdmap(onetime) since " |
b32b8144 FG |
4924 | << start << dendl; |
4925 | do_renew_subs = true; | |
4926 | } | |
4927 | } | |
4928 | ||
4929 | if (do_renew_subs) { | |
4930 | monc->renew_subs(); | |
4931 | } | |
4932 | ||
94b18763 | 4933 | service.send_pg_temp(); |
3efd9988 | 4934 | } |
7c673cae FG |
4935 | |
4936 | void OSD::build_initial_pg_history( | |
4937 | spg_t pgid, | |
4938 | epoch_t created, | |
4939 | utime_t created_stamp, | |
4940 | pg_history_t *h, | |
4941 | PastIntervals *pi) | |
4942 | { | |
4943 | dout(10) << __func__ << " " << pgid << " created " << created << dendl; | |
9f95a23c | 4944 | *h = pg_history_t(created, created_stamp); |
7c673cae FG |
4945 | |
4946 | OSDMapRef lastmap = service.get_map(created); | |
4947 | int up_primary, acting_primary; | |
4948 | vector<int> up, acting; | |
4949 | lastmap->pg_to_up_acting_osds( | |
4950 | pgid.pgid, &up, &up_primary, &acting, &acting_primary); | |
4951 | ||
4952 | ostringstream debug; | |
9f95a23c | 4953 | for (epoch_t e = created + 1; e <= get_osdmap_epoch(); ++e) { |
7c673cae FG |
4954 | OSDMapRef osdmap = service.get_map(e); |
4955 | int new_up_primary, new_acting_primary; | |
4956 | vector<int> new_up, new_acting; | |
4957 | osdmap->pg_to_up_acting_osds( | |
4958 | pgid.pgid, &new_up, &new_up_primary, &new_acting, &new_acting_primary); | |
4959 | ||
4960 | // this is a bit imprecise, but sufficient? | |
4961 | struct min_size_predicate_t : public IsPGRecoverablePredicate { | |
4962 | const pg_pool_t *pi; | |
4963 | bool operator()(const set<pg_shard_t> &have) const { | |
4964 | return have.size() >= pi->min_size; | |
4965 | } | |
11fdf7f2 | 4966 | explicit min_size_predicate_t(const pg_pool_t *i) : pi(i) {} |
7c673cae FG |
4967 | } min_size_predicate(osdmap->get_pg_pool(pgid.pgid.pool())); |
4968 | ||
4969 | bool new_interval = PastIntervals::check_new_interval( | |
4970 | acting_primary, | |
4971 | new_acting_primary, | |
4972 | acting, new_acting, | |
4973 | up_primary, | |
4974 | new_up_primary, | |
4975 | up, new_up, | |
4976 | h->same_interval_since, | |
4977 | h->last_epoch_clean, | |
9f95a23c TL |
4978 | osdmap.get(), |
4979 | lastmap.get(), | |
7c673cae | 4980 | pgid.pgid, |
9f95a23c | 4981 | min_size_predicate, |
7c673cae FG |
4982 | pi, |
4983 | &debug); | |
4984 | if (new_interval) { | |
4985 | h->same_interval_since = e; | |
181888fb FG |
4986 | if (up != new_up) { |
4987 | h->same_up_since = e; | |
4988 | } | |
4989 | if (acting_primary != new_acting_primary) { | |
4990 | h->same_primary_since = e; | |
4991 | } | |
4992 | if (pgid.pgid.is_split(lastmap->get_pg_num(pgid.pgid.pool()), | |
4993 | osdmap->get_pg_num(pgid.pgid.pool()), | |
4994 | nullptr)) { | |
4995 | h->last_epoch_split = e; | |
4996 | } | |
4997 | up = new_up; | |
4998 | acting = new_acting; | |
4999 | up_primary = new_up_primary; | |
5000 | acting_primary = new_acting_primary; | |
c07f9fc5 | 5001 | } |
7c673cae FG |
5002 | lastmap = osdmap; |
5003 | } | |
5004 | dout(20) << __func__ << " " << debug.str() << dendl; | |
5005 | dout(10) << __func__ << " " << *h << " " << *pi | |
5006 | << " [" << (pi->empty() ? pair<epoch_t,epoch_t>(0,0) : | |
5007 | pi->get_bounds()) << ")" | |
5008 | << dendl; | |
5009 | } | |
5010 | ||
7c673cae FG |
5011 | void OSD::_add_heartbeat_peer(int p) |
5012 | { | |
5013 | if (p == whoami) | |
5014 | return; | |
5015 | HeartbeatInfo *hi; | |
5016 | ||
5017 | map<int,HeartbeatInfo>::iterator i = heartbeat_peers.find(p); | |
5018 | if (i == heartbeat_peers.end()) { | |
9f95a23c | 5019 | pair<ConnectionRef,ConnectionRef> cons = service.get_con_osd_hb(p, get_osdmap_epoch()); |
7c673cae FG |
5020 | if (!cons.first) |
5021 | return; | |
9f95a23c TL |
5022 | assert(cons.second); |
5023 | ||
7c673cae FG |
5024 | hi = &heartbeat_peers[p]; |
5025 | hi->peer = p; | |
9f95a23c TL |
5026 | |
5027 | auto stamps = service.get_hb_stamps(p); | |
5028 | ||
5029 | auto sb = ceph::make_ref<Session>(cct, cons.first.get()); | |
5030 | sb->peer = p; | |
5031 | sb->stamps = stamps; | |
eafe8130 | 5032 | hi->hb_interval_start = ceph_clock_now(); |
7c673cae | 5033 | hi->con_back = cons.first.get(); |
9f95a23c TL |
5034 | hi->con_back->set_priv(sb); |
5035 | ||
5036 | auto sf = ceph::make_ref<Session>(cct, cons.second.get()); | |
5037 | sf->peer = p; | |
5038 | sf->stamps = stamps; | |
5039 | hi->con_front = cons.second.get(); | |
5040 | hi->con_front->set_priv(sf); | |
5041 | ||
5042 | dout(10) << "_add_heartbeat_peer: new peer osd." << p | |
5043 | << " " << hi->con_back->get_peer_addr() | |
5044 | << " " << hi->con_front->get_peer_addr() | |
5045 | << dendl; | |
7c673cae FG |
5046 | } else { |
5047 | hi = &i->second; | |
5048 | } | |
9f95a23c | 5049 | hi->epoch = get_osdmap_epoch(); |
7c673cae FG |
5050 | } |
5051 | ||
5052 | void OSD::_remove_heartbeat_peer(int n) | |
5053 | { | |
5054 | map<int,HeartbeatInfo>::iterator q = heartbeat_peers.find(n); | |
11fdf7f2 | 5055 | ceph_assert(q != heartbeat_peers.end()); |
7c673cae FG |
5056 | dout(20) << " removing heartbeat peer osd." << n |
5057 | << " " << q->second.con_back->get_peer_addr() | |
5058 | << " " << (q->second.con_front ? q->second.con_front->get_peer_addr() : entity_addr_t()) | |
5059 | << dendl; | |
9f95a23c | 5060 | q->second.clear_mark_down(); |
7c673cae FG |
5061 | heartbeat_peers.erase(q); |
5062 | } | |
5063 | ||
5064 | void OSD::need_heartbeat_peer_update() | |
5065 | { | |
5066 | if (is_stopping()) | |
5067 | return; | |
5068 | dout(20) << "need_heartbeat_peer_update" << dendl; | |
5069 | heartbeat_set_peers_need_update(); | |
5070 | } | |
5071 | ||
5072 | void OSD::maybe_update_heartbeat_peers() | |
5073 | { | |
9f95a23c | 5074 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae | 5075 | |
11fdf7f2 | 5076 | if (is_waiting_for_healthy() || is_active()) { |
7c673cae FG |
5077 | utime_t now = ceph_clock_now(); |
5078 | if (last_heartbeat_resample == utime_t()) { | |
5079 | last_heartbeat_resample = now; | |
5080 | heartbeat_set_peers_need_update(); | |
5081 | } else if (!heartbeat_peers_need_update()) { | |
5082 | utime_t dur = now - last_heartbeat_resample; | |
5083 | if (dur > cct->_conf->osd_heartbeat_grace) { | |
5084 | dout(10) << "maybe_update_heartbeat_peers forcing update after " << dur << " seconds" << dendl; | |
5085 | heartbeat_set_peers_need_update(); | |
5086 | last_heartbeat_resample = now; | |
494da23a TL |
5087 | // automatically clean up any stale heartbeat peers |
5088 | // if we are unhealthy, then clean all | |
5089 | reset_heartbeat_peers(is_waiting_for_healthy()); | |
7c673cae FG |
5090 | } |
5091 | } | |
5092 | } | |
5093 | ||
5094 | if (!heartbeat_peers_need_update()) | |
5095 | return; | |
5096 | heartbeat_clear_peers_need_update(); | |
5097 | ||
11fdf7f2 | 5098 | std::lock_guard l(heartbeat_lock); |
7c673cae FG |
5099 | |
5100 | dout(10) << "maybe_update_heartbeat_peers updating" << dendl; | |
5101 | ||
5102 | ||
5103 | // build heartbeat from set | |
5104 | if (is_active()) { | |
11fdf7f2 TL |
5105 | vector<PGRef> pgs; |
5106 | _get_pgs(&pgs); | |
5107 | for (auto& pg : pgs) { | |
5108 | pg->with_heartbeat_peers([&](int peer) { | |
9f95a23c | 5109 | if (get_osdmap()->is_up(peer)) { |
11fdf7f2 TL |
5110 | _add_heartbeat_peer(peer); |
5111 | } | |
5112 | }); | |
7c673cae FG |
5113 | } |
5114 | } | |
5115 | ||
5116 | // include next and previous up osds to ensure we have a fully-connected set | |
5117 | set<int> want, extras; | |
9f95a23c | 5118 | const int next = get_osdmap()->get_next_up_osd_after(whoami); |
7c673cae FG |
5119 | if (next >= 0) |
5120 | want.insert(next); | |
9f95a23c | 5121 | int prev = get_osdmap()->get_previous_up_osd_before(whoami); |
7c673cae FG |
5122 | if (prev >= 0 && prev != next) |
5123 | want.insert(prev); | |
5124 | ||
11fdf7f2 TL |
5125 | // make sure we have at least **min_down** osds coming from different |
5126 | // subtree level (e.g., hosts) for fast failure detection. | |
5127 | auto min_down = cct->_conf.get_val<uint64_t>("mon_osd_min_down_reporters"); | |
5128 | auto subtree = cct->_conf.get_val<string>("mon_osd_reporter_subtree_level"); | |
9f95a23c TL |
5129 | auto limit = std::max(min_down, (uint64_t)cct->_conf->osd_heartbeat_min_peers); |
5130 | get_osdmap()->get_random_up_osds_by_subtree( | |
5131 | whoami, subtree, limit, want, &want); | |
11fdf7f2 | 5132 | |
7c673cae FG |
5133 | for (set<int>::iterator p = want.begin(); p != want.end(); ++p) { |
5134 | dout(10) << " adding neighbor peer osd." << *p << dendl; | |
5135 | extras.insert(*p); | |
5136 | _add_heartbeat_peer(*p); | |
5137 | } | |
5138 | ||
5139 | // remove down peers; enumerate extras | |
5140 | map<int,HeartbeatInfo>::iterator p = heartbeat_peers.begin(); | |
5141 | while (p != heartbeat_peers.end()) { | |
9f95a23c | 5142 | if (!get_osdmap()->is_up(p->first)) { |
7c673cae FG |
5143 | int o = p->first; |
5144 | ++p; | |
5145 | _remove_heartbeat_peer(o); | |
5146 | continue; | |
5147 | } | |
9f95a23c | 5148 | if (p->second.epoch < get_osdmap_epoch()) { |
7c673cae FG |
5149 | extras.insert(p->first); |
5150 | } | |
5151 | ++p; | |
5152 | } | |
5153 | ||
5154 | // too few? | |
11fdf7f2 | 5155 | for (int n = next; n >= 0; ) { |
7c673cae FG |
5156 | if ((int)heartbeat_peers.size() >= cct->_conf->osd_heartbeat_min_peers) |
5157 | break; | |
5158 | if (!extras.count(n) && !want.count(n) && n != whoami) { | |
5159 | dout(10) << " adding random peer osd." << n << dendl; | |
5160 | extras.insert(n); | |
5161 | _add_heartbeat_peer(n); | |
5162 | } | |
9f95a23c | 5163 | n = get_osdmap()->get_next_up_osd_after(n); |
11fdf7f2 | 5164 | if (n == next) |
7c673cae FG |
5165 | break; // came full circle; stop |
5166 | } | |
5167 | ||
5168 | // too many? | |
5169 | for (set<int>::iterator p = extras.begin(); | |
5170 | (int)heartbeat_peers.size() > cct->_conf->osd_heartbeat_min_peers && p != extras.end(); | |
5171 | ++p) { | |
5172 | if (want.count(*p)) | |
5173 | continue; | |
5174 | _remove_heartbeat_peer(*p); | |
5175 | } | |
5176 | ||
5177 | dout(10) << "maybe_update_heartbeat_peers " << heartbeat_peers.size() << " peers, extras " << extras << dendl; | |
9f95a23c TL |
5178 | |
5179 | // clean up stale failure pending | |
5180 | for (auto it = failure_pending.begin(); it != failure_pending.end();) { | |
5181 | if (heartbeat_peers.count(it->first) == 0) { | |
5182 | send_still_alive(get_osdmap_epoch(), it->first, it->second.second); | |
5183 | failure_pending.erase(it++); | |
5184 | } else { | |
5185 | it++; | |
5186 | } | |
5187 | } | |
7c673cae FG |
5188 | } |
5189 | ||
494da23a | 5190 | void OSD::reset_heartbeat_peers(bool all) |
7c673cae | 5191 | { |
9f95a23c | 5192 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae | 5193 | dout(10) << "reset_heartbeat_peers" << dendl; |
494da23a TL |
5194 | utime_t stale = ceph_clock_now(); |
5195 | stale -= cct->_conf.get_val<int64_t>("osd_heartbeat_stale"); | |
11fdf7f2 | 5196 | std::lock_guard l(heartbeat_lock); |
494da23a TL |
5197 | for (auto it = heartbeat_peers.begin(); it != heartbeat_peers.end();) { |
5198 | HeartbeatInfo& hi = it->second; | |
5199 | if (all || hi.is_stale(stale)) { | |
9f95a23c | 5200 | hi.clear_mark_down(); |
494da23a TL |
5201 | // stop sending failure_report to mon too |
5202 | failure_queue.erase(it->first); | |
5203 | heartbeat_peers.erase(it++); | |
5204 | } else { | |
5205 | it++; | |
7c673cae | 5206 | } |
7c673cae | 5207 | } |
7c673cae FG |
5208 | } |
5209 | ||
5210 | void OSD::handle_osd_ping(MOSDPing *m) | |
5211 | { | |
5212 | if (superblock.cluster_fsid != m->fsid) { | |
5213 | dout(20) << "handle_osd_ping from " << m->get_source_inst() | |
9f95a23c TL |
5214 | << " bad fsid " << m->fsid << " != " << superblock.cluster_fsid |
5215 | << dendl; | |
7c673cae FG |
5216 | m->put(); |
5217 | return; | |
5218 | } | |
5219 | ||
5220 | int from = m->get_source().num(); | |
5221 | ||
9f95a23c | 5222 | heartbeat_lock.lock(); |
7c673cae | 5223 | if (is_stopping()) { |
9f95a23c | 5224 | heartbeat_lock.unlock(); |
7c673cae FG |
5225 | m->put(); |
5226 | return; | |
5227 | } | |
5228 | ||
9f95a23c TL |
5229 | utime_t now = ceph_clock_now(); |
5230 | auto mnow = service.get_mnow(); | |
5231 | ConnectionRef con(m->get_connection()); | |
7c673cae | 5232 | OSDMapRef curmap = service.get_osdmap(); |
c07f9fc5 | 5233 | if (!curmap) { |
9f95a23c | 5234 | heartbeat_lock.unlock(); |
c07f9fc5 FG |
5235 | m->put(); |
5236 | return; | |
5237 | } | |
7c673cae | 5238 | |
9f95a23c TL |
5239 | auto sref = con->get_priv(); |
5240 | Session *s = static_cast<Session*>(sref.get()); | |
5241 | if (!s) { | |
5242 | heartbeat_lock.unlock(); | |
5243 | m->put(); | |
5244 | return; | |
5245 | } | |
5246 | if (!s->stamps) { | |
5247 | s->peer = from; | |
5248 | s->stamps = service.get_hb_stamps(from); | |
5249 | } | |
5250 | ||
7c673cae FG |
5251 | switch (m->op) { |
5252 | ||
5253 | case MOSDPing::PING: | |
5254 | { | |
5255 | if (cct->_conf->osd_debug_drop_ping_probability > 0) { | |
5256 | auto heartbeat_drop = debug_heartbeat_drops_remaining.find(from); | |
5257 | if (heartbeat_drop != debug_heartbeat_drops_remaining.end()) { | |
5258 | if (heartbeat_drop->second == 0) { | |
5259 | debug_heartbeat_drops_remaining.erase(heartbeat_drop); | |
5260 | } else { | |
5261 | --heartbeat_drop->second; | |
5262 | dout(5) << "Dropping heartbeat from " << from | |
5263 | << ", " << heartbeat_drop->second | |
5264 | << " remaining to drop" << dendl; | |
5265 | break; | |
5266 | } | |
5267 | } else if (cct->_conf->osd_debug_drop_ping_probability > | |
5268 | ((((double)(rand()%100))/100.0))) { | |
5269 | heartbeat_drop = | |
5270 | debug_heartbeat_drops_remaining.insert(std::make_pair(from, | |
5271 | cct->_conf->osd_debug_drop_ping_duration)).first; | |
5272 | dout(5) << "Dropping heartbeat from " << from | |
5273 | << ", " << heartbeat_drop->second | |
5274 | << " remaining to drop" << dendl; | |
5275 | break; | |
5276 | } | |
5277 | } | |
5278 | ||
9f95a23c TL |
5279 | ceph::signedspan sender_delta_ub{}; |
5280 | s->stamps->got_ping( | |
5281 | m->up_from, | |
5282 | mnow, | |
5283 | m->mono_send_stamp, | |
5284 | m->delta_ub, | |
5285 | &sender_delta_ub); | |
5286 | dout(20) << __func__ << " new stamps " << *s->stamps << dendl; | |
5287 | ||
7c673cae | 5288 | if (!cct->get_heartbeat_map()->is_healthy()) { |
9f95a23c TL |
5289 | dout(10) << "internal heartbeat not healthy, dropping ping request" |
5290 | << dendl; | |
7c673cae FG |
5291 | break; |
5292 | } | |
5293 | ||
5294 | Message *r = new MOSDPing(monc->get_fsid(), | |
5295 | curmap->get_epoch(), | |
9f95a23c TL |
5296 | MOSDPing::PING_REPLY, |
5297 | m->ping_stamp, | |
5298 | m->mono_ping_stamp, | |
5299 | mnow, | |
5300 | service.get_up_epoch(), | |
5301 | cct->_conf->osd_heartbeat_min_size, | |
5302 | sender_delta_ub); | |
5303 | con->send_message(r); | |
7c673cae FG |
5304 | |
5305 | if (curmap->is_up(from)) { | |
7c673cae | 5306 | if (is_active()) { |
9f95a23c TL |
5307 | ConnectionRef cluster_con = service.get_con_osd_cluster( |
5308 | from, curmap->get_epoch()); | |
5309 | if (cluster_con) { | |
5310 | service.maybe_share_map(cluster_con.get(), curmap, m->map_epoch); | |
7c673cae FG |
5311 | } |
5312 | } | |
5313 | } else if (!curmap->exists(from) || | |
5314 | curmap->get_down_at(from) > m->map_epoch) { | |
5315 | // tell them they have died | |
5316 | Message *r = new MOSDPing(monc->get_fsid(), | |
5317 | curmap->get_epoch(), | |
5318 | MOSDPing::YOU_DIED, | |
9f95a23c TL |
5319 | m->ping_stamp, |
5320 | m->mono_ping_stamp, | |
5321 | mnow, | |
5322 | service.get_up_epoch(), | |
31f18b77 | 5323 | cct->_conf->osd_heartbeat_min_size); |
9f95a23c | 5324 | con->send_message(r); |
7c673cae FG |
5325 | } |
5326 | } | |
5327 | break; | |
5328 | ||
5329 | case MOSDPing::PING_REPLY: | |
5330 | { | |
5331 | map<int,HeartbeatInfo>::iterator i = heartbeat_peers.find(from); | |
5332 | if (i != heartbeat_peers.end()) { | |
9f95a23c | 5333 | auto acked = i->second.ping_history.find(m->ping_stamp); |
11fdf7f2 | 5334 | if (acked != i->second.ping_history.end()) { |
11fdf7f2 | 5335 | int &unacknowledged = acked->second.second; |
9f95a23c | 5336 | if (con == i->second.con_back) { |
11fdf7f2 TL |
5337 | dout(25) << "handle_osd_ping got reply from osd." << from |
5338 | << " first_tx " << i->second.first_tx | |
5339 | << " last_tx " << i->second.last_tx | |
9f95a23c TL |
5340 | << " last_rx_back " << i->second.last_rx_back |
5341 | << " -> " << now | |
11fdf7f2 TL |
5342 | << " last_rx_front " << i->second.last_rx_front |
5343 | << dendl; | |
5344 | i->second.last_rx_back = now; | |
5345 | ceph_assert(unacknowledged > 0); | |
5346 | --unacknowledged; | |
5347 | // if there is no front con, set both stamps. | |
5348 | if (i->second.con_front == NULL) { | |
5349 | i->second.last_rx_front = now; | |
5350 | ceph_assert(unacknowledged > 0); | |
5351 | --unacknowledged; | |
5352 | } | |
9f95a23c | 5353 | } else if (con == i->second.con_front) { |
11fdf7f2 TL |
5354 | dout(25) << "handle_osd_ping got reply from osd." << from |
5355 | << " first_tx " << i->second.first_tx | |
5356 | << " last_tx " << i->second.last_tx | |
5357 | << " last_rx_back " << i->second.last_rx_back | |
9f95a23c TL |
5358 | << " last_rx_front " << i->second.last_rx_front |
5359 | << " -> " << now | |
11fdf7f2 TL |
5360 | << dendl; |
5361 | i->second.last_rx_front = now; | |
5362 | ceph_assert(unacknowledged > 0); | |
5363 | --unacknowledged; | |
5364 | } | |
7c673cae | 5365 | |
11fdf7f2 TL |
5366 | if (unacknowledged == 0) { |
5367 | // succeeded in getting all replies | |
5368 | dout(25) << "handle_osd_ping got all replies from osd." << from | |
9f95a23c | 5369 | << " , erase pending ping(sent at " << m->ping_stamp << ")" |
11fdf7f2 TL |
5370 | << " and older pending ping(s)" |
5371 | << dendl; | |
eafe8130 TL |
5372 | |
5373 | #define ROUND_S_TO_USEC(sec) (uint32_t)((sec) * 1000 * 1000 + 0.5) | |
5374 | ++i->second.hb_average_count; | |
9f95a23c | 5375 | uint32_t back_pingtime = ROUND_S_TO_USEC(i->second.last_rx_back - m->ping_stamp); |
eafe8130 TL |
5376 | i->second.hb_total_back += back_pingtime; |
5377 | if (back_pingtime < i->second.hb_min_back) | |
5378 | i->second.hb_min_back = back_pingtime; | |
5379 | if (back_pingtime > i->second.hb_max_back) | |
5380 | i->second.hb_max_back = back_pingtime; | |
9f95a23c | 5381 | uint32_t front_pingtime = ROUND_S_TO_USEC(i->second.last_rx_front - m->ping_stamp); |
eafe8130 TL |
5382 | i->second.hb_total_front += front_pingtime; |
5383 | if (front_pingtime < i->second.hb_min_front) | |
5384 | i->second.hb_min_front = front_pingtime; | |
5385 | if (front_pingtime > i->second.hb_max_front) | |
5386 | i->second.hb_max_front = front_pingtime; | |
5387 | ||
5388 | ceph_assert(i->second.hb_interval_start != utime_t()); | |
5389 | if (i->second.hb_interval_start == utime_t()) | |
5390 | i->second.hb_interval_start = now; | |
5391 | int64_t hb_avg_time_period = 60; | |
5392 | if (cct->_conf.get_val<int64_t>("debug_heartbeat_testing_span")) { | |
5393 | hb_avg_time_period = cct->_conf.get_val<int64_t>("debug_heartbeat_testing_span"); | |
5394 | } | |
5395 | if (now - i->second.hb_interval_start >= utime_t(hb_avg_time_period, 0)) { | |
5396 | uint32_t back_avg = i->second.hb_total_back / i->second.hb_average_count; | |
5397 | uint32_t back_min = i->second.hb_min_back; | |
5398 | uint32_t back_max = i->second.hb_max_back; | |
5399 | uint32_t front_avg = i->second.hb_total_front / i->second.hb_average_count; | |
5400 | uint32_t front_min = i->second.hb_min_front; | |
5401 | uint32_t front_max = i->second.hb_max_front; | |
5402 | ||
5403 | // Reset for new interval | |
5404 | i->second.hb_average_count = 0; | |
5405 | i->second.hb_interval_start = now; | |
5406 | i->second.hb_total_back = i->second.hb_max_back = 0; | |
5407 | i->second.hb_min_back = UINT_MAX; | |
5408 | i->second.hb_total_front = i->second.hb_max_front = 0; | |
5409 | i->second.hb_min_front = UINT_MAX; | |
5410 | ||
5411 | // Record per osd interace ping times | |
5412 | // Based on osd_heartbeat_interval ignoring that it is randomly short than this interval | |
5413 | if (i->second.hb_back_pingtime.size() == 0) { | |
5414 | ceph_assert(i->second.hb_front_pingtime.size() == 0); | |
5415 | for (unsigned k = 0 ; k < hb_vector_size; ++k) { | |
5416 | i->second.hb_back_pingtime.push_back(back_avg); | |
5417 | i->second.hb_back_min.push_back(back_min); | |
5418 | i->second.hb_back_max.push_back(back_max); | |
5419 | i->second.hb_front_pingtime.push_back(front_avg); | |
5420 | i->second.hb_front_min.push_back(front_min); | |
5421 | i->second.hb_front_max.push_back(front_max); | |
5422 | ++i->second.hb_index; | |
5423 | } | |
5424 | } else { | |
5425 | int index = i->second.hb_index & (hb_vector_size - 1); | |
5426 | i->second.hb_back_pingtime[index] = back_avg; | |
5427 | i->second.hb_back_min[index] = back_min; | |
5428 | i->second.hb_back_max[index] = back_max; | |
5429 | i->second.hb_front_pingtime[index] = front_avg; | |
5430 | i->second.hb_front_min[index] = front_min; | |
5431 | i->second.hb_front_max[index] = front_max; | |
5432 | ++i->second.hb_index; | |
5433 | } | |
5434 | ||
5435 | { | |
5436 | std::lock_guard l(service.stat_lock); | |
5437 | service.osd_stat.hb_pingtime[from].last_update = now.sec(); | |
5438 | service.osd_stat.hb_pingtime[from].back_last = back_pingtime; | |
5439 | ||
5440 | uint32_t total = 0; | |
5441 | uint32_t min = UINT_MAX; | |
5442 | uint32_t max = 0; | |
5443 | uint32_t count = 0; | |
5444 | uint32_t which = 0; | |
5445 | uint32_t size = (uint32_t)i->second.hb_back_pingtime.size(); | |
5446 | for (int32_t k = size - 1 ; k >= 0; --k) { | |
5447 | ++count; | |
5448 | int index = (i->second.hb_index + k) % size; | |
5449 | total += i->second.hb_back_pingtime[index]; | |
5450 | if (i->second.hb_back_min[index] < min) | |
5451 | min = i->second.hb_back_min[index]; | |
5452 | if (i->second.hb_back_max[index] > max) | |
5453 | max = i->second.hb_back_max[index]; | |
5454 | if (count == 1 || count == 5 || count == 15) { | |
5455 | service.osd_stat.hb_pingtime[from].back_pingtime[which] = total / count; | |
5456 | service.osd_stat.hb_pingtime[from].back_min[which] = min; | |
5457 | service.osd_stat.hb_pingtime[from].back_max[which] = max; | |
5458 | which++; | |
5459 | if (count == 15) | |
5460 | break; | |
5461 | } | |
5462 | } | |
5463 | ||
5464 | if (i->second.con_front != NULL) { | |
5465 | service.osd_stat.hb_pingtime[from].front_last = front_pingtime; | |
5466 | ||
5467 | total = 0; | |
5468 | min = UINT_MAX; | |
5469 | max = 0; | |
5470 | count = 0; | |
5471 | which = 0; | |
5472 | for (int32_t k = size - 1 ; k >= 0; --k) { | |
5473 | ++count; | |
5474 | int index = (i->second.hb_index + k) % size; | |
5475 | total += i->second.hb_front_pingtime[index]; | |
5476 | if (i->second.hb_front_min[index] < min) | |
5477 | min = i->second.hb_front_min[index]; | |
5478 | if (i->second.hb_front_max[index] > max) | |
5479 | max = i->second.hb_front_max[index]; | |
5480 | if (count == 1 || count == 5 || count == 15) { | |
5481 | service.osd_stat.hb_pingtime[from].front_pingtime[which] = total / count; | |
5482 | service.osd_stat.hb_pingtime[from].front_min[which] = min; | |
5483 | service.osd_stat.hb_pingtime[from].front_max[which] = max; | |
5484 | which++; | |
5485 | if (count == 15) | |
5486 | break; | |
5487 | } | |
5488 | } | |
5489 | } | |
5490 | } | |
5491 | } else { | |
5492 | std::lock_guard l(service.stat_lock); | |
5493 | service.osd_stat.hb_pingtime[from].back_last = back_pingtime; | |
5494 | if (i->second.con_front != NULL) | |
5495 | service.osd_stat.hb_pingtime[from].front_last = front_pingtime; | |
5496 | } | |
11fdf7f2 | 5497 | i->second.ping_history.erase(i->second.ping_history.begin(), ++acked); |
7c673cae FG |
5498 | } |
5499 | ||
11fdf7f2 TL |
5500 | if (i->second.is_healthy(now)) { |
5501 | // Cancel false reports | |
5502 | auto failure_queue_entry = failure_queue.find(from); | |
5503 | if (failure_queue_entry != failure_queue.end()) { | |
5504 | dout(10) << "handle_osd_ping canceling queued " | |
5505 | << "failure report for osd." << from << dendl; | |
5506 | failure_queue.erase(failure_queue_entry); | |
5507 | } | |
5508 | ||
5509 | auto failure_pending_entry = failure_pending.find(from); | |
5510 | if (failure_pending_entry != failure_pending.end()) { | |
5511 | dout(10) << "handle_osd_ping canceling in-flight " | |
5512 | << "failure report for osd." << from << dendl; | |
5513 | send_still_alive(curmap->get_epoch(), | |
5514 | from, | |
5515 | failure_pending_entry->second.second); | |
5516 | failure_pending.erase(failure_pending_entry); | |
5517 | } | |
7c673cae | 5518 | } |
11fdf7f2 TL |
5519 | } else { |
5520 | // old replies, deprecated by newly sent pings. | |
9f95a23c | 5521 | dout(10) << "handle_osd_ping no pending ping(sent at " << m->ping_stamp |
11fdf7f2 TL |
5522 | << ") is found, treat as covered by newly sent pings " |
5523 | << "and ignore" | |
5524 | << dendl; | |
7c673cae FG |
5525 | } |
5526 | } | |
5527 | ||
5528 | if (m->map_epoch && | |
5529 | curmap->is_up(from)) { | |
7c673cae | 5530 | if (is_active()) { |
9f95a23c TL |
5531 | ConnectionRef cluster_con = service.get_con_osd_cluster( |
5532 | from, curmap->get_epoch()); | |
5533 | if (cluster_con) { | |
5534 | service.maybe_share_map(cluster_con.get(), curmap, m->map_epoch); | |
7c673cae FG |
5535 | } |
5536 | } | |
5537 | } | |
9f95a23c TL |
5538 | |
5539 | s->stamps->got_ping_reply( | |
5540 | mnow, | |
5541 | m->mono_send_stamp, | |
5542 | m->delta_ub); | |
5543 | dout(20) << __func__ << " new stamps " << *s->stamps << dendl; | |
7c673cae FG |
5544 | } |
5545 | break; | |
5546 | ||
5547 | case MOSDPing::YOU_DIED: | |
5548 | dout(10) << "handle_osd_ping " << m->get_source_inst() | |
5549 | << " says i am down in " << m->map_epoch << dendl; | |
5550 | osdmap_subscribe(curmap->get_epoch()+1, false); | |
5551 | break; | |
5552 | } | |
5553 | ||
9f95a23c | 5554 | heartbeat_lock.unlock(); |
7c673cae FG |
5555 | m->put(); |
5556 | } | |
5557 | ||
5558 | void OSD::heartbeat_entry() | |
5559 | { | |
9f95a23c | 5560 | std::unique_lock l(heartbeat_lock); |
7c673cae FG |
5561 | if (is_stopping()) |
5562 | return; | |
5563 | while (!heartbeat_stop) { | |
5564 | heartbeat(); | |
5565 | ||
eafe8130 TL |
5566 | double wait; |
5567 | if (cct->_conf.get_val<bool>("debug_disable_randomized_ping")) { | |
5568 | wait = (float)cct->_conf->osd_heartbeat_interval; | |
5569 | } else { | |
5570 | wait = .5 + ((float)(rand() % 10)/10.0) * (float)cct->_conf->osd_heartbeat_interval; | |
5571 | } | |
9f95a23c | 5572 | auto w = ceph::make_timespan(wait); |
7c673cae | 5573 | dout(30) << "heartbeat_entry sleeping for " << wait << dendl; |
9f95a23c | 5574 | heartbeat_cond.wait_for(l, w); |
7c673cae FG |
5575 | if (is_stopping()) |
5576 | return; | |
5577 | dout(30) << "heartbeat_entry woke up" << dendl; | |
5578 | } | |
5579 | } | |
5580 | ||
5581 | void OSD::heartbeat_check() | |
5582 | { | |
9f95a23c | 5583 | ceph_assert(ceph_mutex_is_locked(heartbeat_lock)); |
7c673cae FG |
5584 | utime_t now = ceph_clock_now(); |
5585 | ||
11fdf7f2 | 5586 | // check for incoming heartbeats (move me elsewhere?) |
7c673cae FG |
5587 | for (map<int,HeartbeatInfo>::iterator p = heartbeat_peers.begin(); |
5588 | p != heartbeat_peers.end(); | |
5589 | ++p) { | |
5590 | ||
5591 | if (p->second.first_tx == utime_t()) { | |
5592 | dout(25) << "heartbeat_check we haven't sent ping to osd." << p->first | |
11fdf7f2 | 5593 | << " yet, skipping" << dendl; |
7c673cae FG |
5594 | continue; |
5595 | } | |
5596 | ||
5597 | dout(25) << "heartbeat_check osd." << p->first | |
5598 | << " first_tx " << p->second.first_tx | |
5599 | << " last_tx " << p->second.last_tx | |
5600 | << " last_rx_back " << p->second.last_rx_back | |
5601 | << " last_rx_front " << p->second.last_rx_front | |
5602 | << dendl; | |
11fdf7f2 TL |
5603 | if (p->second.is_unhealthy(now)) { |
5604 | utime_t oldest_deadline = p->second.ping_history.begin()->second.first; | |
7c673cae FG |
5605 | if (p->second.last_rx_back == utime_t() || |
5606 | p->second.last_rx_front == utime_t()) { | |
11fdf7f2 TL |
5607 | derr << "heartbeat_check: no reply from " |
5608 | << p->second.con_front->get_peer_addr().get_sockaddr() | |
5609 | << " osd." << p->first | |
5610 | << " ever on either front or back, first ping sent " | |
5611 | << p->second.first_tx | |
5612 | << " (oldest deadline " << oldest_deadline << ")" | |
5613 | << dendl; | |
7c673cae | 5614 | // fail |
11fdf7f2 | 5615 | failure_queue[p->first] = p->second.first_tx; |
7c673cae | 5616 | } else { |
11fdf7f2 TL |
5617 | derr << "heartbeat_check: no reply from " |
5618 | << p->second.con_front->get_peer_addr().get_sockaddr() | |
7c673cae FG |
5619 | << " osd." << p->first << " since back " << p->second.last_rx_back |
5620 | << " front " << p->second.last_rx_front | |
11fdf7f2 TL |
5621 | << " (oldest deadline " << oldest_deadline << ")" |
5622 | << dendl; | |
7c673cae | 5623 | // fail |
11fdf7f2 | 5624 | failure_queue[p->first] = std::min(p->second.last_rx_back, p->second.last_rx_front); |
7c673cae FG |
5625 | } |
5626 | } | |
5627 | } | |
5628 | } | |
5629 | ||
5630 | void OSD::heartbeat() | |
5631 | { | |
9f95a23c | 5632 | ceph_assert(ceph_mutex_is_locked_by_me(heartbeat_lock)); |
7c673cae FG |
5633 | dout(30) << "heartbeat" << dendl; |
5634 | ||
5635 | // get CPU load avg | |
5636 | double loadavgs[1]; | |
11fdf7f2 TL |
5637 | int hb_interval = cct->_conf->osd_heartbeat_interval; |
5638 | int n_samples = 86400; | |
5639 | if (hb_interval > 1) { | |
5640 | n_samples /= hb_interval; | |
5641 | if (n_samples < 1) | |
5642 | n_samples = 1; | |
5643 | } | |
5644 | ||
7c673cae FG |
5645 | if (getloadavg(loadavgs, 1) == 1) { |
5646 | logger->set(l_osd_loadavg, 100 * loadavgs[0]); | |
5647 | daily_loadavg = (daily_loadavg * (n_samples - 1) + loadavgs[0]) / n_samples; | |
5648 | dout(30) << "heartbeat: daily_loadavg " << daily_loadavg << dendl; | |
5649 | } | |
5650 | ||
5651 | dout(30) << "heartbeat checking stats" << dendl; | |
5652 | ||
11fdf7f2 | 5653 | // refresh peer list and osd stats |
7c673cae FG |
5654 | vector<int> hb_peers; |
5655 | for (map<int,HeartbeatInfo>::iterator p = heartbeat_peers.begin(); | |
5656 | p != heartbeat_peers.end(); | |
5657 | ++p) | |
5658 | hb_peers.push_back(p->first); | |
7c673cae | 5659 | |
11fdf7f2 TL |
5660 | auto new_stat = service.set_osd_stat(hb_peers, get_num_pgs()); |
5661 | dout(5) << __func__ << " " << new_stat << dendl; | |
5662 | ceph_assert(new_stat.statfs.total); | |
5663 | ||
5664 | float pratio; | |
5665 | float ratio = service.compute_adjusted_ratio(new_stat, &pratio); | |
5666 | ||
5667 | service.check_full_status(ratio, pratio); | |
7c673cae FG |
5668 | |
5669 | utime_t now = ceph_clock_now(); | |
9f95a23c | 5670 | auto mnow = service.get_mnow(); |
11fdf7f2 TL |
5671 | utime_t deadline = now; |
5672 | deadline += cct->_conf->osd_heartbeat_grace; | |
7c673cae FG |
5673 | |
5674 | // send heartbeats | |
5675 | for (map<int,HeartbeatInfo>::iterator i = heartbeat_peers.begin(); | |
5676 | i != heartbeat_peers.end(); | |
5677 | ++i) { | |
5678 | int peer = i->first; | |
9f95a23c TL |
5679 | dout(30) << "heartbeat sending ping to osd." << peer << dendl; |
5680 | ||
7c673cae FG |
5681 | i->second.last_tx = now; |
5682 | if (i->second.first_tx == utime_t()) | |
5683 | i->second.first_tx = now; | |
11fdf7f2 TL |
5684 | i->second.ping_history[now] = make_pair(deadline, |
5685 | HeartbeatInfo::HEARTBEAT_MAX_CONN); | |
eafe8130 TL |
5686 | if (i->second.hb_interval_start == utime_t()) |
5687 | i->second.hb_interval_start = now; | |
9f95a23c TL |
5688 | |
5689 | Session *s = static_cast<Session*>(i->second.con_back->get_priv().get()); | |
5690 | std::optional<ceph::signedspan> delta_ub; | |
5691 | s->stamps->sent_ping(&delta_ub); | |
5692 | ||
5693 | i->second.con_back->send_message( | |
5694 | new MOSDPing(monc->get_fsid(), | |
5695 | service.get_osdmap_epoch(), | |
5696 | MOSDPing::PING, | |
5697 | now, | |
5698 | mnow, | |
5699 | mnow, | |
5700 | service.get_up_epoch(), | |
5701 | cct->_conf->osd_heartbeat_min_size, | |
5702 | delta_ub)); | |
7c673cae FG |
5703 | |
5704 | if (i->second.con_front) | |
9f95a23c TL |
5705 | i->second.con_front->send_message( |
5706 | new MOSDPing(monc->get_fsid(), | |
5707 | service.get_osdmap_epoch(), | |
5708 | MOSDPing::PING, | |
5709 | now, | |
5710 | mnow, | |
5711 | mnow, | |
5712 | service.get_up_epoch(), | |
5713 | cct->_conf->osd_heartbeat_min_size, | |
5714 | delta_ub)); | |
7c673cae FG |
5715 | } |
5716 | ||
5717 | logger->set(l_osd_hb_to, heartbeat_peers.size()); | |
5718 | ||
5719 | // hmm.. am i all alone? | |
5720 | dout(30) << "heartbeat lonely?" << dendl; | |
5721 | if (heartbeat_peers.empty()) { | |
5722 | if (now - last_mon_heartbeat > cct->_conf->osd_mon_heartbeat_interval && is_active()) { | |
5723 | last_mon_heartbeat = now; | |
5724 | dout(10) << "i have no heartbeat peers; checking mon for new map" << dendl; | |
9f95a23c | 5725 | osdmap_subscribe(get_osdmap_epoch() + 1, false); |
7c673cae FG |
5726 | } |
5727 | } | |
5728 | ||
5729 | dout(30) << "heartbeat done" << dendl; | |
5730 | } | |
5731 | ||
5732 | bool OSD::heartbeat_reset(Connection *con) | |
5733 | { | |
11fdf7f2 TL |
5734 | std::lock_guard l(heartbeat_lock); |
5735 | auto s = con->get_priv(); | |
9f95a23c | 5736 | dout(20) << __func__ << " con " << con << " s " << s.get() << dendl; |
11fdf7f2 | 5737 | con->set_priv(nullptr); |
7c673cae | 5738 | if (s) { |
7c673cae | 5739 | if (is_stopping()) { |
7c673cae FG |
5740 | return true; |
5741 | } | |
9f95a23c TL |
5742 | auto session = static_cast<Session*>(s.get()); |
5743 | auto p = heartbeat_peers.find(session->peer); | |
7c673cae FG |
5744 | if (p != heartbeat_peers.end() && |
5745 | (p->second.con_back == con || | |
5746 | p->second.con_front == con)) { | |
5747 | dout(10) << "heartbeat_reset failed hb con " << con << " for osd." << p->second.peer | |
5748 | << ", reopening" << dendl; | |
9f95a23c | 5749 | p->second.clear_mark_down(con); |
7c673cae FG |
5750 | pair<ConnectionRef,ConnectionRef> newcon = service.get_con_osd_hb(p->second.peer, p->second.epoch); |
5751 | if (newcon.first) { | |
5752 | p->second.con_back = newcon.first.get(); | |
11fdf7f2 | 5753 | p->second.con_back->set_priv(s); |
7c673cae FG |
5754 | if (newcon.second) { |
5755 | p->second.con_front = newcon.second.get(); | |
11fdf7f2 | 5756 | p->second.con_front->set_priv(s); |
7c673cae | 5757 | } |
11fdf7f2 | 5758 | p->second.ping_history.clear(); |
7c673cae FG |
5759 | } else { |
5760 | dout(10) << "heartbeat_reset failed hb con " << con << " for osd." << p->second.peer | |
5761 | << ", raced with osdmap update, closing out peer" << dendl; | |
5762 | heartbeat_peers.erase(p); | |
5763 | } | |
5764 | } else { | |
5765 | dout(10) << "heartbeat_reset closing (old) failed hb con " << con << dendl; | |
5766 | } | |
7c673cae FG |
5767 | } |
5768 | return true; | |
5769 | } | |
5770 | ||
5771 | ||
5772 | ||
5773 | // ========================================= | |
5774 | ||
5775 | void OSD::tick() | |
5776 | { | |
9f95a23c | 5777 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae FG |
5778 | dout(10) << "tick" << dendl; |
5779 | ||
9f95a23c TL |
5780 | utime_t now = ceph_clock_now(); |
5781 | // throw out any obsolete markdown log | |
5782 | utime_t grace = utime_t(cct->_conf->osd_max_markdown_period, 0); | |
5783 | while (!osd_markdown_log.empty() && | |
5784 | osd_markdown_log.front() + grace < now) | |
5785 | osd_markdown_log.pop_front(); | |
5786 | ||
7c673cae FG |
5787 | if (is_active() || is_waiting_for_healthy()) { |
5788 | maybe_update_heartbeat_peers(); | |
5789 | } | |
5790 | ||
5791 | if (is_waiting_for_healthy()) { | |
5792 | start_boot(); | |
494da23a TL |
5793 | } |
5794 | ||
5795 | if (is_waiting_for_healthy() || is_booting()) { | |
5796 | std::lock_guard l(heartbeat_lock); | |
494da23a TL |
5797 | if (now - last_mon_heartbeat > cct->_conf->osd_mon_heartbeat_interval) { |
5798 | last_mon_heartbeat = now; | |
5799 | dout(1) << __func__ << " checking mon for new map" << dendl; | |
9f95a23c | 5800 | osdmap_subscribe(get_osdmap_epoch() + 1, false); |
11fdf7f2 | 5801 | } |
7c673cae FG |
5802 | } |
5803 | ||
5804 | do_waiters(); | |
5805 | ||
9f95a23c TL |
5806 | // scrub purged_snaps every deep scrub interval |
5807 | { | |
5808 | const utime_t last = superblock.last_purged_snaps_scrub; | |
5809 | utime_t next = last; | |
5810 | next += cct->_conf->osd_scrub_min_interval; | |
5811 | std::mt19937 rng; | |
5812 | // use a seed that is stable for each scrub interval, but varies | |
5813 | // by OSD to avoid any herds. | |
5814 | rng.seed(whoami + superblock.last_purged_snaps_scrub.sec()); | |
5815 | double r = (rng() % 1024) / 1024; | |
5816 | next += | |
5817 | cct->_conf->osd_scrub_min_interval * | |
5818 | cct->_conf->osd_scrub_interval_randomize_ratio * r; | |
5819 | if (next < ceph_clock_now()) { | |
5820 | dout(20) << __func__ << " last_purged_snaps_scrub " << last | |
5821 | << " next " << next << " ... now" << dendl; | |
5822 | scrub_purged_snaps(); | |
5823 | } else { | |
5824 | dout(20) << __func__ << " last_purged_snaps_scrub " << last | |
5825 | << " next " << next << dendl; | |
5826 | } | |
5827 | } | |
5828 | ||
91327a77 | 5829 | tick_timer.add_event_after(get_tick_interval(), new C_Tick(this)); |
7c673cae FG |
5830 | } |
5831 | ||
5832 | void OSD::tick_without_osd_lock() | |
5833 | { | |
9f95a23c | 5834 | ceph_assert(ceph_mutex_is_locked(tick_timer_lock)); |
7c673cae FG |
5835 | dout(10) << "tick_without_osd_lock" << dendl; |
5836 | ||
7c673cae FG |
5837 | logger->set(l_osd_cached_crc, buffer::get_cached_crc()); |
5838 | logger->set(l_osd_cached_crc_adjusted, buffer::get_cached_crc_adjusted()); | |
5839 | logger->set(l_osd_missed_crc, buffer::get_missed_crc()); | |
11fdf7f2 TL |
5840 | |
5841 | // refresh osd stats | |
5842 | struct store_statfs_t stbuf; | |
5843 | osd_alert_list_t alerts; | |
5844 | int r = store->statfs(&stbuf, &alerts); | |
5845 | ceph_assert(r == 0); | |
5846 | service.set_statfs(stbuf, alerts); | |
7c673cae FG |
5847 | |
5848 | // osd_lock is not being held, which means the OSD state | |
5849 | // might change when doing the monitor report | |
5850 | if (is_active() || is_waiting_for_healthy()) { | |
9f95a23c TL |
5851 | { |
5852 | std::lock_guard l{heartbeat_lock}; | |
5853 | heartbeat_check(); | |
5854 | } | |
5855 | map_lock.lock_shared(); | |
11fdf7f2 | 5856 | std::lock_guard l(mon_report_lock); |
7c673cae FG |
5857 | |
5858 | // mon report? | |
7c673cae | 5859 | utime_t now = ceph_clock_now(); |
11fdf7f2 TL |
5860 | if (service.need_fullness_update() || |
5861 | now - last_mon_report > cct->_conf->osd_mon_report_interval) { | |
7c673cae | 5862 | last_mon_report = now; |
7c673cae FG |
5863 | send_full_update(); |
5864 | send_failures(); | |
7c673cae | 5865 | } |
9f95a23c | 5866 | map_lock.unlock_shared(); |
11fdf7f2 TL |
5867 | |
5868 | epoch_t max_waiting_epoch = 0; | |
5869 | for (auto s : shards) { | |
5870 | max_waiting_epoch = std::max(max_waiting_epoch, | |
5871 | s->get_max_waiting_epoch()); | |
5872 | } | |
5873 | if (max_waiting_epoch > get_osdmap()->get_epoch()) { | |
5874 | dout(20) << __func__ << " max_waiting_epoch " << max_waiting_epoch | |
5875 | << ", requesting new map" << dendl; | |
5876 | osdmap_subscribe(superblock.newest_map + 1, false); | |
5877 | } | |
7c673cae FG |
5878 | } |
5879 | ||
5880 | if (is_active()) { | |
5881 | if (!scrub_random_backoff()) { | |
5882 | sched_scrub(); | |
5883 | } | |
5884 | service.promote_throttle_recalibrate(); | |
3efd9988 | 5885 | resume_creating_pg(); |
224ce89b WB |
5886 | bool need_send_beacon = false; |
5887 | const auto now = ceph::coarse_mono_clock::now(); | |
5888 | { | |
5889 | // borrow lec lock to pretect last_sent_beacon from changing | |
11fdf7f2 | 5890 | std::lock_guard l{min_last_epoch_clean_lock}; |
224ce89b WB |
5891 | const auto elapsed = now - last_sent_beacon; |
5892 | if (chrono::duration_cast<chrono::seconds>(elapsed).count() > | |
5893 | cct->_conf->osd_beacon_report_interval) { | |
5894 | need_send_beacon = true; | |
5895 | } | |
5896 | } | |
5897 | if (need_send_beacon) { | |
5898 | send_beacon(now); | |
5899 | } | |
7c673cae FG |
5900 | } |
5901 | ||
11fdf7f2 | 5902 | mgrc.update_daemon_health(get_health_metrics()); |
7c673cae | 5903 | service.kick_recovery_queue(); |
91327a77 AA |
5904 | tick_timer_without_osd_lock.add_event_after(get_tick_interval(), |
5905 | new C_Tick_WithoutOSDLock(this)); | |
7c673cae FG |
5906 | } |
5907 | ||
7c673cae FG |
5908 | // Usage: |
5909 | // setomapval <pool-id> [namespace/]<obj-name> <key> <val> | |
5910 | // rmomapkey <pool-id> [namespace/]<obj-name> <key> | |
5911 | // setomapheader <pool-id> [namespace/]<obj-name> <header> | |
5912 | // getomap <pool> [namespace/]<obj-name> | |
5913 | // truncobj <pool-id> [namespace/]<obj-name> <newlen> | |
5914 | // injectmdataerr [namespace/]<obj-name> [shardid] | |
5915 | // injectdataerr [namespace/]<obj-name> [shardid] | |
5916 | // | |
5917 | // set_recovery_delay [utime] | |
5918 | void TestOpsSocketHook::test_ops(OSDService *service, ObjectStore *store, | |
11fdf7f2 TL |
5919 | std::string_view command, |
5920 | const cmdmap_t& cmdmap, ostream &ss) | |
7c673cae FG |
5921 | { |
5922 | //Test support | |
5923 | //Support changing the omap on a single osd by using the Admin Socket to | |
5924 | //directly request the osd make a change. | |
5925 | if (command == "setomapval" || command == "rmomapkey" || | |
5926 | command == "setomapheader" || command == "getomap" || | |
5927 | command == "truncobj" || command == "injectmdataerr" || | |
5928 | command == "injectdataerr" | |
5929 | ) { | |
5930 | pg_t rawpg; | |
5931 | int64_t pool; | |
5932 | OSDMapRef curmap = service->get_osdmap(); | |
5933 | int r = -1; | |
5934 | ||
5935 | string poolstr; | |
5936 | ||
9f95a23c | 5937 | cmd_getval(cmdmap, "pool", poolstr); |
7c673cae FG |
5938 | pool = curmap->lookup_pg_pool_name(poolstr); |
5939 | //If we can't find it by name then maybe id specified | |
5940 | if (pool < 0 && isdigit(poolstr[0])) | |
5941 | pool = atoll(poolstr.c_str()); | |
5942 | if (pool < 0) { | |
b5b8bbf5 | 5943 | ss << "Invalid pool '" << poolstr << "''"; |
7c673cae FG |
5944 | return; |
5945 | } | |
5946 | ||
5947 | string objname, nspace; | |
9f95a23c | 5948 | cmd_getval(cmdmap, "objname", objname); |
7c673cae FG |
5949 | std::size_t found = objname.find_first_of('/'); |
5950 | if (found != string::npos) { | |
5951 | nspace = objname.substr(0, found); | |
5952 | objname = objname.substr(found+1); | |
5953 | } | |
5954 | object_locator_t oloc(pool, nspace); | |
5955 | r = curmap->object_locator_to_pg(object_t(objname), oloc, rawpg); | |
5956 | ||
5957 | if (r < 0) { | |
5958 | ss << "Invalid namespace/objname"; | |
5959 | return; | |
5960 | } | |
5961 | ||
5962 | int64_t shardid; | |
9f95a23c | 5963 | cmd_getval(cmdmap, "shardid", shardid, int64_t(shard_id_t::NO_SHARD)); |
7c673cae FG |
5964 | hobject_t obj(object_t(objname), string(""), CEPH_NOSNAP, rawpg.ps(), pool, nspace); |
5965 | ghobject_t gobj(obj, ghobject_t::NO_GEN, shard_id_t(uint8_t(shardid))); | |
5966 | spg_t pgid(curmap->raw_pg_to_pg(rawpg), shard_id_t(shardid)); | |
5967 | if (curmap->pg_is_ec(rawpg)) { | |
5968 | if ((command != "injectdataerr") && (command != "injectmdataerr")) { | |
5969 | ss << "Must not call on ec pool, except injectdataerr or injectmdataerr"; | |
5970 | return; | |
5971 | } | |
5972 | } | |
5973 | ||
5974 | ObjectStore::Transaction t; | |
5975 | ||
5976 | if (command == "setomapval") { | |
5977 | map<string, bufferlist> newattrs; | |
5978 | bufferlist val; | |
5979 | string key, valstr; | |
9f95a23c TL |
5980 | cmd_getval(cmdmap, "key", key); |
5981 | cmd_getval(cmdmap, "val", valstr); | |
7c673cae FG |
5982 | |
5983 | val.append(valstr); | |
5984 | newattrs[key] = val; | |
5985 | t.omap_setkeys(coll_t(pgid), ghobject_t(obj), newattrs); | |
11fdf7f2 | 5986 | r = store->queue_transaction(service->meta_ch, std::move(t)); |
7c673cae FG |
5987 | if (r < 0) |
5988 | ss << "error=" << r; | |
5989 | else | |
5990 | ss << "ok"; | |
5991 | } else if (command == "rmomapkey") { | |
5992 | string key; | |
9f95a23c | 5993 | cmd_getval(cmdmap, "key", key); |
7c673cae | 5994 | |
9f95a23c | 5995 | t.omap_rmkey(coll_t(pgid), ghobject_t(obj), key); |
11fdf7f2 | 5996 | r = store->queue_transaction(service->meta_ch, std::move(t)); |
7c673cae FG |
5997 | if (r < 0) |
5998 | ss << "error=" << r; | |
5999 | else | |
6000 | ss << "ok"; | |
6001 | } else if (command == "setomapheader") { | |
6002 | bufferlist newheader; | |
6003 | string headerstr; | |
6004 | ||
9f95a23c | 6005 | cmd_getval(cmdmap, "header", headerstr); |
7c673cae FG |
6006 | newheader.append(headerstr); |
6007 | t.omap_setheader(coll_t(pgid), ghobject_t(obj), newheader); | |
11fdf7f2 | 6008 | r = store->queue_transaction(service->meta_ch, std::move(t)); |
7c673cae FG |
6009 | if (r < 0) |
6010 | ss << "error=" << r; | |
6011 | else | |
6012 | ss << "ok"; | |
6013 | } else if (command == "getomap") { | |
6014 | //Debug: Output entire omap | |
6015 | bufferlist hdrbl; | |
6016 | map<string, bufferlist> keyvals; | |
11fdf7f2 TL |
6017 | auto ch = store->open_collection(coll_t(pgid)); |
6018 | if (!ch) { | |
6019 | ss << "unable to open collection for " << pgid; | |
6020 | r = -ENOENT; | |
6021 | } else { | |
6022 | r = store->omap_get(ch, ghobject_t(obj), &hdrbl, &keyvals); | |
6023 | if (r >= 0) { | |
7c673cae FG |
6024 | ss << "header=" << string(hdrbl.c_str(), hdrbl.length()); |
6025 | for (map<string, bufferlist>::iterator it = keyvals.begin(); | |
11fdf7f2 | 6026 | it != keyvals.end(); ++it) |
7c673cae FG |
6027 | ss << " key=" << (*it).first << " val=" |
6028 | << string((*it).second.c_str(), (*it).second.length()); | |
11fdf7f2 | 6029 | } else { |
7c673cae | 6030 | ss << "error=" << r; |
11fdf7f2 | 6031 | } |
7c673cae FG |
6032 | } |
6033 | } else if (command == "truncobj") { | |
6034 | int64_t trunclen; | |
9f95a23c | 6035 | cmd_getval(cmdmap, "len", trunclen); |
7c673cae | 6036 | t.truncate(coll_t(pgid), ghobject_t(obj), trunclen); |
11fdf7f2 | 6037 | r = store->queue_transaction(service->meta_ch, std::move(t)); |
7c673cae FG |
6038 | if (r < 0) |
6039 | ss << "error=" << r; | |
6040 | else | |
6041 | ss << "ok"; | |
6042 | } else if (command == "injectdataerr") { | |
6043 | store->inject_data_error(gobj); | |
6044 | ss << "ok"; | |
6045 | } else if (command == "injectmdataerr") { | |
6046 | store->inject_mdata_error(gobj); | |
6047 | ss << "ok"; | |
6048 | } | |
6049 | return; | |
6050 | } | |
6051 | if (command == "set_recovery_delay") { | |
6052 | int64_t delay; | |
9f95a23c | 6053 | cmd_getval(cmdmap, "utime", delay, (int64_t)0); |
7c673cae FG |
6054 | ostringstream oss; |
6055 | oss << delay; | |
11fdf7f2 | 6056 | int r = service->cct->_conf.set_val("osd_recovery_delay_start", |
7c673cae FG |
6057 | oss.str().c_str()); |
6058 | if (r != 0) { | |
6059 | ss << "set_recovery_delay: error setting " | |
6060 | << "osd_recovery_delay_start to '" << delay << "': error " | |
6061 | << r; | |
6062 | return; | |
6063 | } | |
11fdf7f2 | 6064 | service->cct->_conf.apply_changes(nullptr); |
7c673cae FG |
6065 | ss << "set_recovery_delay: set osd_recovery_delay_start " |
6066 | << "to " << service->cct->_conf->osd_recovery_delay_start; | |
6067 | return; | |
6068 | } | |
7c673cae FG |
6069 | if (command == "injectfull") { |
6070 | int64_t count; | |
6071 | string type; | |
6072 | OSDService::s_names state; | |
9f95a23c TL |
6073 | cmd_getval(cmdmap, "type", type, string("full")); |
6074 | cmd_getval(cmdmap, "count", count, (int64_t)-1); | |
7c673cae FG |
6075 | if (type == "none" || count == 0) { |
6076 | type = "none"; | |
6077 | count = 0; | |
6078 | } | |
6079 | state = service->get_full_state(type); | |
6080 | if (state == OSDService::s_names::INVALID) { | |
6081 | ss << "Invalid type use (none, nearfull, backfillfull, full, failsafe)"; | |
6082 | return; | |
6083 | } | |
6084 | service->set_injectfull(state, count); | |
6085 | return; | |
6086 | } | |
6087 | ss << "Internal error - command=" << command; | |
6088 | } | |
6089 | ||
7c673cae FG |
6090 | // ========================================= |
6091 | ||
6092 | void OSD::ms_handle_connect(Connection *con) | |
6093 | { | |
6094 | dout(10) << __func__ << " con " << con << dendl; | |
6095 | if (con->get_peer_type() == CEPH_ENTITY_TYPE_MON) { | |
11fdf7f2 | 6096 | std::lock_guard l(osd_lock); |
7c673cae FG |
6097 | if (is_stopping()) |
6098 | return; | |
6099 | dout(10) << __func__ << " on mon" << dendl; | |
6100 | ||
6101 | if (is_preboot()) { | |
6102 | start_boot(); | |
6103 | } else if (is_booting()) { | |
6104 | _send_boot(); // resend boot message | |
6105 | } else { | |
9f95a23c | 6106 | map_lock.lock_shared(); |
11fdf7f2 | 6107 | std::lock_guard l2(mon_report_lock); |
7c673cae FG |
6108 | |
6109 | utime_t now = ceph_clock_now(); | |
6110 | last_mon_report = now; | |
6111 | ||
6112 | // resend everything, it's a new session | |
6113 | send_full_update(); | |
6114 | send_alive(); | |
6115 | service.requeue_pg_temp(); | |
11fdf7f2 | 6116 | service.clear_sent_ready_to_merge(); |
7c673cae | 6117 | service.send_pg_temp(); |
11fdf7f2 TL |
6118 | service.send_ready_to_merge(); |
6119 | service.send_pg_created(); | |
7c673cae FG |
6120 | requeue_failures(); |
6121 | send_failures(); | |
7c673cae | 6122 | |
9f95a23c | 6123 | map_lock.unlock_shared(); |
7c673cae FG |
6124 | if (is_active()) { |
6125 | send_beacon(ceph::coarse_mono_clock::now()); | |
6126 | } | |
6127 | } | |
6128 | ||
6129 | // full map requests may happen while active or pre-boot | |
6130 | if (requested_full_first) { | |
6131 | rerequest_full_maps(); | |
6132 | } | |
6133 | } | |
6134 | } | |
6135 | ||
6136 | void OSD::ms_handle_fast_connect(Connection *con) | |
6137 | { | |
6138 | if (con->get_peer_type() != CEPH_ENTITY_TYPE_MON && | |
6139 | con->get_peer_type() != CEPH_ENTITY_TYPE_MGR) { | |
9f95a23c TL |
6140 | if (auto s = ceph::ref_cast<Session>(con->get_priv()); !s) { |
6141 | s = ceph::make_ref<Session>(cct, con); | |
6142 | con->set_priv(s); | |
7c673cae FG |
6143 | dout(10) << " new session (outgoing) " << s << " con=" << s->con |
6144 | << " addr=" << s->con->get_peer_addr() << dendl; | |
6145 | // we don't connect to clients | |
11fdf7f2 | 6146 | ceph_assert(con->get_peer_type() == CEPH_ENTITY_TYPE_OSD); |
7c673cae FG |
6147 | s->entity_name.set_type(CEPH_ENTITY_TYPE_OSD); |
6148 | } | |
7c673cae FG |
6149 | } |
6150 | } | |
6151 | ||
6152 | void OSD::ms_handle_fast_accept(Connection *con) | |
6153 | { | |
6154 | if (con->get_peer_type() != CEPH_ENTITY_TYPE_MON && | |
6155 | con->get_peer_type() != CEPH_ENTITY_TYPE_MGR) { | |
9f95a23c TL |
6156 | if (auto s = ceph::ref_cast<Session>(con->get_priv()); !s) { |
6157 | s = ceph::make_ref<Session>(cct, con); | |
6158 | con->set_priv(s); | |
7c673cae FG |
6159 | dout(10) << "new session (incoming)" << s << " con=" << con |
6160 | << " addr=" << con->get_peer_addr() | |
6161 | << " must have raced with connect" << dendl; | |
11fdf7f2 | 6162 | ceph_assert(con->get_peer_type() == CEPH_ENTITY_TYPE_OSD); |
7c673cae FG |
6163 | s->entity_name.set_type(CEPH_ENTITY_TYPE_OSD); |
6164 | } | |
7c673cae FG |
6165 | } |
6166 | } | |
6167 | ||
6168 | bool OSD::ms_handle_reset(Connection *con) | |
6169 | { | |
9f95a23c TL |
6170 | auto session = ceph::ref_cast<Session>(con->get_priv()); |
6171 | dout(2) << "ms_handle_reset con " << con << " session " << session.get() << dendl; | |
7c673cae FG |
6172 | if (!session) |
6173 | return false; | |
6174 | session->wstate.reset(con); | |
11fdf7f2 TL |
6175 | session->con->set_priv(nullptr); |
6176 | session->con.reset(); // break con <-> session ref cycle | |
7c673cae FG |
6177 | // note that we break session->con *before* the session_handle_reset |
6178 | // cleanup below. this avoids a race between us and | |
6179 | // PG::add_backoff, Session::check_backoff, etc. | |
9f95a23c | 6180 | session_handle_reset(session); |
7c673cae FG |
6181 | return true; |
6182 | } | |
6183 | ||
6184 | bool OSD::ms_handle_refused(Connection *con) | |
6185 | { | |
6186 | if (!cct->_conf->osd_fast_fail_on_connection_refused) | |
6187 | return false; | |
6188 | ||
9f95a23c TL |
6189 | auto session = ceph::ref_cast<Session>(con->get_priv()); |
6190 | dout(2) << "ms_handle_refused con " << con << " session " << session.get() << dendl; | |
7c673cae FG |
6191 | if (!session) |
6192 | return false; | |
6193 | int type = con->get_peer_type(); | |
6194 | // handle only OSD failures here | |
6195 | if (monc && (type == CEPH_ENTITY_TYPE_OSD)) { | |
6196 | OSDMapRef osdmap = get_osdmap(); | |
6197 | if (osdmap) { | |
6198 | int id = osdmap->identify_osd_on_all_channels(con->get_peer_addr()); | |
6199 | if (id >= 0 && osdmap->is_up(id)) { | |
6200 | // I'm cheating mon heartbeat grace logic, because we know it's not going | |
6201 | // to respawn alone. +1 so we won't hit any boundary case. | |
11fdf7f2 TL |
6202 | monc->send_mon_message( |
6203 | new MOSDFailure( | |
6204 | monc->get_fsid(), | |
6205 | id, | |
6206 | osdmap->get_addrs(id), | |
6207 | cct->_conf->osd_heartbeat_grace + 1, | |
6208 | osdmap->get_epoch(), | |
6209 | MOSDFailure::FLAG_IMMEDIATE | MOSDFailure::FLAG_FAILED | |
6210 | )); | |
7c673cae FG |
6211 | } |
6212 | } | |
6213 | } | |
7c673cae FG |
6214 | return true; |
6215 | } | |
6216 | ||
6217 | struct C_OSD_GetVersion : public Context { | |
6218 | OSD *osd; | |
6219 | uint64_t oldest, newest; | |
6220 | explicit C_OSD_GetVersion(OSD *o) : osd(o), oldest(0), newest(0) {} | |
6221 | void finish(int r) override { | |
6222 | if (r >= 0) | |
6223 | osd->_got_mon_epochs(oldest, newest); | |
6224 | } | |
6225 | }; | |
6226 | ||
6227 | void OSD::start_boot() | |
6228 | { | |
6229 | if (!_is_healthy()) { | |
6230 | // if we are not healthy, do not mark ourselves up (yet) | |
6231 | dout(1) << "not healthy; waiting to boot" << dendl; | |
6232 | if (!is_waiting_for_healthy()) | |
6233 | start_waiting_for_healthy(); | |
6234 | // send pings sooner rather than later | |
6235 | heartbeat_kick(); | |
6236 | return; | |
6237 | } | |
6238 | dout(1) << __func__ << dendl; | |
6239 | set_state(STATE_PREBOOT); | |
6240 | dout(10) << "start_boot - have maps " << superblock.oldest_map | |
6241 | << ".." << superblock.newest_map << dendl; | |
6242 | C_OSD_GetVersion *c = new C_OSD_GetVersion(this); | |
6243 | monc->get_version("osdmap", &c->newest, &c->oldest, c); | |
6244 | } | |
6245 | ||
6246 | void OSD::_got_mon_epochs(epoch_t oldest, epoch_t newest) | |
6247 | { | |
11fdf7f2 | 6248 | std::lock_guard l(osd_lock); |
7c673cae FG |
6249 | if (is_preboot()) { |
6250 | _preboot(oldest, newest); | |
6251 | } | |
6252 | } | |
6253 | ||
6254 | void OSD::_preboot(epoch_t oldest, epoch_t newest) | |
6255 | { | |
11fdf7f2 | 6256 | ceph_assert(is_preboot()); |
7c673cae FG |
6257 | dout(10) << __func__ << " _preboot mon has osdmaps " |
6258 | << oldest << ".." << newest << dendl; | |
6259 | ||
6260 | // ensure our local fullness awareness is accurate | |
81eedcae TL |
6261 | { |
6262 | std::lock_guard l(heartbeat_lock); | |
6263 | heartbeat(); | |
6264 | } | |
7c673cae | 6265 | |
9f95a23c TL |
6266 | const auto& monmap = monc->monmap; |
6267 | const auto osdmap = get_osdmap(); | |
7c673cae | 6268 | // if our map within recent history, try to add ourselves to the osdmap. |
31f18b77 FG |
6269 | if (osdmap->get_epoch() == 0) { |
6270 | derr << "waiting for initial osdmap" << dendl; | |
c07f9fc5 | 6271 | } else if (osdmap->is_destroyed(whoami)) { |
b32b8144 FG |
6272 | derr << "osdmap says I am destroyed" << dendl; |
6273 | // provide a small margin so we don't livelock seeing if we | |
6274 | // un-destroyed ourselves. | |
6275 | if (osdmap->get_epoch() > newest - 1) { | |
6276 | exit(0); | |
6277 | } | |
81eedcae | 6278 | } else if (osdmap->is_noup(whoami)) { |
7c673cae FG |
6279 | derr << "osdmap NOUP flag is set, waiting for it to clear" << dendl; |
6280 | } else if (!osdmap->test_flag(CEPH_OSDMAP_SORTBITWISE)) { | |
6281 | derr << "osdmap SORTBITWISE OSDMap flag is NOT set; please set it" | |
6282 | << dendl; | |
9f95a23c | 6283 | } else if (osdmap->require_osd_release < ceph_release_t::luminous) { |
11fdf7f2 | 6284 | derr << "osdmap require_osd_release < luminous; please upgrade to luminous" |
7c673cae | 6285 | << dendl; |
7c673cae FG |
6286 | } else if (service.need_fullness_update()) { |
6287 | derr << "osdmap fullness state needs update" << dendl; | |
6288 | send_full_update(); | |
9f95a23c TL |
6289 | } else if (monmap.min_mon_release >= ceph_release_t::octopus && |
6290 | superblock.purged_snaps_last < superblock.current_epoch) { | |
6291 | dout(10) << __func__ << " purged_snaps_last " << superblock.purged_snaps_last | |
6292 | << " < newest_map " << superblock.current_epoch << dendl; | |
6293 | _get_purged_snaps(); | |
7c673cae FG |
6294 | } else if (osdmap->get_epoch() >= oldest - 1 && |
6295 | osdmap->get_epoch() + cct->_conf->osd_map_message_max > newest) { | |
11fdf7f2 TL |
6296 | |
6297 | // wait for pgs to fully catch up in a different thread, since | |
6298 | // this thread might be required for splitting and merging PGs to | |
6299 | // make progress. | |
6300 | boot_finisher.queue( | |
9f95a23c | 6301 | new LambdaContext( |
11fdf7f2 | 6302 | [this](int r) { |
9f95a23c | 6303 | std::unique_lock l(osd_lock); |
11fdf7f2 TL |
6304 | if (is_preboot()) { |
6305 | dout(10) << __func__ << " waiting for peering work to drain" | |
6306 | << dendl; | |
9f95a23c | 6307 | l.unlock(); |
11fdf7f2 | 6308 | for (auto shard : shards) { |
9f95a23c | 6309 | shard->wait_min_pg_epoch(get_osdmap_epoch()); |
11fdf7f2 | 6310 | } |
9f95a23c | 6311 | l.lock(); |
11fdf7f2 TL |
6312 | } |
6313 | if (is_preboot()) { | |
6314 | _send_boot(); | |
6315 | } | |
6316 | })); | |
6317 | return; | |
7c673cae FG |
6318 | } |
6319 | ||
6320 | // get all the latest maps | |
6321 | if (osdmap->get_epoch() + 1 >= oldest) | |
6322 | osdmap_subscribe(osdmap->get_epoch() + 1, false); | |
6323 | else | |
6324 | osdmap_subscribe(oldest - 1, true); | |
6325 | } | |
6326 | ||
9f95a23c TL |
6327 | void OSD::_get_purged_snaps() |
6328 | { | |
6329 | // NOTE: this is a naive, stateless implementaiton. it may send multiple | |
6330 | // overlapping requests to the mon, which will be somewhat inefficient, but | |
6331 | // it should be reliable. | |
6332 | dout(10) << __func__ << " purged_snaps_last " << superblock.purged_snaps_last | |
6333 | << ", newest_map " << superblock.current_epoch << dendl; | |
6334 | MMonGetPurgedSnaps *m = new MMonGetPurgedSnaps( | |
6335 | superblock.purged_snaps_last + 1, | |
6336 | superblock.current_epoch + 1); | |
6337 | monc->send_mon_message(m); | |
6338 | } | |
6339 | ||
6340 | void OSD::handle_get_purged_snaps_reply(MMonGetPurgedSnapsReply *m) | |
6341 | { | |
6342 | dout(10) << __func__ << " " << *m << dendl; | |
6343 | ObjectStore::Transaction t; | |
6344 | if (!is_preboot() || | |
6345 | m->last < superblock.purged_snaps_last) { | |
6346 | goto out; | |
6347 | } | |
6348 | SnapMapper::record_purged_snaps(cct, store, service.meta_ch, | |
6349 | make_purged_snaps_oid(), &t, | |
6350 | m->purged_snaps); | |
6351 | superblock.purged_snaps_last = m->last; | |
6352 | write_superblock(t); | |
6353 | store->queue_transaction( | |
6354 | service.meta_ch, | |
6355 | std::move(t)); | |
6356 | service.publish_superblock(superblock); | |
6357 | if (m->last < superblock.current_epoch) { | |
6358 | _get_purged_snaps(); | |
6359 | } else { | |
6360 | start_boot(); | |
6361 | } | |
6362 | out: | |
6363 | m->put(); | |
6364 | } | |
6365 | ||
7c673cae FG |
6366 | void OSD::send_full_update() |
6367 | { | |
6368 | if (!service.need_fullness_update()) | |
6369 | return; | |
6370 | unsigned state = 0; | |
6371 | if (service.is_full()) { | |
6372 | state = CEPH_OSD_FULL; | |
6373 | } else if (service.is_backfillfull()) { | |
6374 | state = CEPH_OSD_BACKFILLFULL; | |
6375 | } else if (service.is_nearfull()) { | |
6376 | state = CEPH_OSD_NEARFULL; | |
6377 | } | |
6378 | set<string> s; | |
6379 | OSDMap::calc_state_set(state, s); | |
6380 | dout(10) << __func__ << " want state " << s << dendl; | |
9f95a23c | 6381 | monc->send_mon_message(new MOSDFull(get_osdmap_epoch(), state)); |
7c673cae FG |
6382 | } |
6383 | ||
6384 | void OSD::start_waiting_for_healthy() | |
6385 | { | |
6386 | dout(1) << "start_waiting_for_healthy" << dendl; | |
6387 | set_state(STATE_WAITING_FOR_HEALTHY); | |
6388 | last_heartbeat_resample = utime_t(); | |
181888fb FG |
6389 | |
6390 | // subscribe to osdmap updates, in case our peers really are known to be dead | |
9f95a23c | 6391 | osdmap_subscribe(get_osdmap_epoch() + 1, false); |
7c673cae FG |
6392 | } |
6393 | ||
6394 | bool OSD::_is_healthy() | |
6395 | { | |
6396 | if (!cct->get_heartbeat_map()->is_healthy()) { | |
6397 | dout(1) << "is_healthy false -- internal heartbeat failed" << dendl; | |
6398 | return false; | |
6399 | } | |
6400 | ||
6401 | if (is_waiting_for_healthy()) { | |
11fdf7f2 | 6402 | utime_t now = ceph_clock_now(); |
9f95a23c TL |
6403 | if (osd_markdown_log.empty()) { |
6404 | dout(5) << __func__ << " force returning true since last markdown" | |
6405 | << " was " << cct->_conf->osd_max_markdown_period | |
6406 | << "s ago" << dendl; | |
11fdf7f2 TL |
6407 | return true; |
6408 | } | |
6409 | std::lock_guard l(heartbeat_lock); | |
7c673cae FG |
6410 | int num = 0, up = 0; |
6411 | for (map<int,HeartbeatInfo>::iterator p = heartbeat_peers.begin(); | |
6412 | p != heartbeat_peers.end(); | |
6413 | ++p) { | |
11fdf7f2 | 6414 | if (p->second.is_healthy(now)) |
7c673cae FG |
6415 | ++up; |
6416 | ++num; | |
6417 | } | |
6418 | if ((float)up < (float)num * cct->_conf->osd_heartbeat_min_healthy_ratio) { | |
6419 | dout(1) << "is_healthy false -- only " << up << "/" << num << " up peers (less than " | |
6420 | << int(cct->_conf->osd_heartbeat_min_healthy_ratio * 100.0) << "%)" << dendl; | |
6421 | return false; | |
6422 | } | |
6423 | } | |
6424 | ||
6425 | return true; | |
6426 | } | |
6427 | ||
6428 | void OSD::_send_boot() | |
6429 | { | |
6430 | dout(10) << "_send_boot" << dendl; | |
11fdf7f2 TL |
6431 | Connection *local_connection = |
6432 | cluster_messenger->get_loopback_connection().get(); | |
6433 | entity_addrvec_t client_addrs = client_messenger->get_myaddrs(); | |
6434 | entity_addrvec_t cluster_addrs = cluster_messenger->get_myaddrs(); | |
6435 | entity_addrvec_t hb_back_addrs = hb_back_server_messenger->get_myaddrs(); | |
6436 | entity_addrvec_t hb_front_addrs = hb_front_server_messenger->get_myaddrs(); | |
6437 | ||
6438 | dout(20) << " initial client_addrs " << client_addrs | |
6439 | << ", cluster_addrs " << cluster_addrs | |
6440 | << ", hb_back_addrs " << hb_back_addrs | |
6441 | << ", hb_front_addrs " << hb_front_addrs | |
6442 | << dendl; | |
6443 | if (cluster_messenger->set_addr_unknowns(client_addrs)) { | |
6444 | dout(10) << " assuming cluster_addrs match client_addrs " | |
6445 | << client_addrs << dendl; | |
6446 | cluster_addrs = cluster_messenger->get_myaddrs(); | |
6447 | } | |
6448 | if (auto session = local_connection->get_priv(); !session) { | |
6449 | cluster_messenger->ms_deliver_handle_fast_connect(local_connection); | |
7c673cae FG |
6450 | } |
6451 | ||
7c673cae | 6452 | local_connection = hb_back_server_messenger->get_loopback_connection().get(); |
11fdf7f2 TL |
6453 | if (hb_back_server_messenger->set_addr_unknowns(cluster_addrs)) { |
6454 | dout(10) << " assuming hb_back_addrs match cluster_addrs " | |
6455 | << cluster_addrs << dendl; | |
6456 | hb_back_addrs = hb_back_server_messenger->get_myaddrs(); | |
7c673cae | 6457 | } |
11fdf7f2 TL |
6458 | if (auto session = local_connection->get_priv(); !session) { |
6459 | hb_back_server_messenger->ms_deliver_handle_fast_connect(local_connection); | |
7c673cae FG |
6460 | } |
6461 | ||
11fdf7f2 TL |
6462 | local_connection = hb_front_server_messenger->get_loopback_connection().get(); |
6463 | if (hb_front_server_messenger->set_addr_unknowns(client_addrs)) { | |
6464 | dout(10) << " assuming hb_front_addrs match client_addrs " | |
6465 | << client_addrs << dendl; | |
6466 | hb_front_addrs = hb_front_server_messenger->get_myaddrs(); | |
6467 | } | |
6468 | if (auto session = local_connection->get_priv(); !session) { | |
6469 | hb_front_server_messenger->ms_deliver_handle_fast_connect(local_connection); | |
6470 | } | |
6471 | ||
6472 | // we now know what our front and back addrs will be, and we are | |
6473 | // about to tell the mon what our metadata (including numa bindings) | |
6474 | // are, so now is a good time! | |
6475 | set_numa_affinity(); | |
6476 | ||
6477 | MOSDBoot *mboot = new MOSDBoot( | |
6478 | superblock, get_osdmap_epoch(), service.get_boot_epoch(), | |
6479 | hb_back_addrs, hb_front_addrs, cluster_addrs, | |
6480 | CEPH_FEATURES_ALL); | |
6481 | dout(10) << " final client_addrs " << client_addrs | |
6482 | << ", cluster_addrs " << cluster_addrs | |
6483 | << ", hb_back_addrs " << hb_back_addrs | |
6484 | << ", hb_front_addrs " << hb_front_addrs | |
7c673cae FG |
6485 | << dendl; |
6486 | _collect_metadata(&mboot->metadata); | |
6487 | monc->send_mon_message(mboot); | |
6488 | set_state(STATE_BOOTING); | |
6489 | } | |
6490 | ||
6491 | void OSD::_collect_metadata(map<string,string> *pm) | |
6492 | { | |
6493 | // config info | |
6494 | (*pm)["osd_data"] = dev_path; | |
c07f9fc5 FG |
6495 | if (store->get_type() == "filestore") { |
6496 | // not applicable for bluestore | |
6497 | (*pm)["osd_journal"] = journal_path; | |
6498 | } | |
11fdf7f2 TL |
6499 | (*pm)["front_addr"] = stringify(client_messenger->get_myaddrs()); |
6500 | (*pm)["back_addr"] = stringify(cluster_messenger->get_myaddrs()); | |
6501 | (*pm)["hb_front_addr"] = stringify(hb_front_server_messenger->get_myaddrs()); | |
6502 | (*pm)["hb_back_addr"] = stringify(hb_back_server_messenger->get_myaddrs()); | |
7c673cae FG |
6503 | |
6504 | // backend | |
6505 | (*pm)["osd_objectstore"] = store->get_type(); | |
31f18b77 | 6506 | (*pm)["rotational"] = store_is_rotational ? "1" : "0"; |
d2e6a577 | 6507 | (*pm)["journal_rotational"] = journal_is_rotational ? "1" : "0"; |
c07f9fc5 | 6508 | (*pm)["default_device_class"] = store->get_default_device_class(); |
7c673cae FG |
6509 | store->collect_metadata(pm); |
6510 | ||
6511 | collect_sys_info(pm, cct); | |
6512 | ||
11fdf7f2 TL |
6513 | (*pm)["front_iface"] = pick_iface( |
6514 | cct, | |
6515 | client_messenger->get_myaddrs().front().get_sockaddr_storage()); | |
6516 | (*pm)["back_iface"] = pick_iface( | |
6517 | cct, | |
6518 | cluster_messenger->get_myaddrs().front().get_sockaddr_storage()); | |
6519 | ||
6520 | // network numa | |
6521 | { | |
6522 | int node = -1; | |
6523 | set<int> nodes; | |
6524 | set<string> unknown; | |
6525 | for (auto nm : { "front_iface", "back_iface" }) { | |
6526 | if (!(*pm)[nm].size()) { | |
6527 | unknown.insert(nm); | |
6528 | continue; | |
6529 | } | |
6530 | int n = -1; | |
6531 | int r = get_iface_numa_node((*pm)[nm], &n); | |
6532 | if (r < 0) { | |
6533 | unknown.insert((*pm)[nm]); | |
6534 | continue; | |
6535 | } | |
6536 | nodes.insert(n); | |
6537 | if (node < 0) { | |
6538 | node = n; | |
6539 | } | |
6540 | } | |
6541 | if (unknown.size()) { | |
6542 | (*pm)["network_numa_unknown_ifaces"] = stringify(unknown); | |
6543 | } | |
6544 | if (!nodes.empty()) { | |
6545 | (*pm)["network_numa_nodes"] = stringify(nodes); | |
6546 | } | |
6547 | if (node >= 0 && nodes.size() == 1 && unknown.empty()) { | |
6548 | (*pm)["network_numa_node"] = stringify(node); | |
6549 | } | |
6550 | } | |
6551 | ||
6552 | if (numa_node >= 0) { | |
6553 | (*pm)["numa_node"] = stringify(numa_node); | |
6554 | (*pm)["numa_node_cpus"] = cpu_set_to_str_list(numa_cpu_set_size, | |
6555 | &numa_cpu_set); | |
6556 | } | |
6557 | ||
6558 | set<string> devnames; | |
6559 | store->get_devices(&devnames); | |
9f95a23c TL |
6560 | map<string,string> errs; |
6561 | get_device_metadata(devnames, pm, &errs); | |
6562 | for (auto& i : errs) { | |
6563 | dout(1) << __func__ << " " << i.first << ": " << i.second << dendl; | |
11fdf7f2 | 6564 | } |
7c673cae FG |
6565 | dout(10) << __func__ << " " << *pm << dendl; |
6566 | } | |
6567 | ||
6568 | void OSD::queue_want_up_thru(epoch_t want) | |
6569 | { | |
9f95a23c TL |
6570 | std::shared_lock map_locker{map_lock}; |
6571 | epoch_t cur = get_osdmap()->get_up_thru(whoami); | |
6572 | std::lock_guard report_locker(mon_report_lock); | |
7c673cae FG |
6573 | if (want > up_thru_wanted) { |
6574 | dout(10) << "queue_want_up_thru now " << want << " (was " << up_thru_wanted << ")" | |
6575 | << ", currently " << cur | |
6576 | << dendl; | |
6577 | up_thru_wanted = want; | |
6578 | send_alive(); | |
6579 | } else { | |
6580 | dout(10) << "queue_want_up_thru want " << want << " <= queued " << up_thru_wanted | |
6581 | << ", currently " << cur | |
6582 | << dendl; | |
6583 | } | |
7c673cae FG |
6584 | } |
6585 | ||
6586 | void OSD::send_alive() | |
6587 | { | |
9f95a23c TL |
6588 | ceph_assert(ceph_mutex_is_locked(mon_report_lock)); |
6589 | const auto osdmap = get_osdmap(); | |
7c673cae FG |
6590 | if (!osdmap->exists(whoami)) |
6591 | return; | |
6592 | epoch_t up_thru = osdmap->get_up_thru(whoami); | |
6593 | dout(10) << "send_alive up_thru currently " << up_thru << " want " << up_thru_wanted << dendl; | |
6594 | if (up_thru_wanted > up_thru) { | |
6595 | dout(10) << "send_alive want " << up_thru_wanted << dendl; | |
6596 | monc->send_mon_message(new MOSDAlive(osdmap->get_epoch(), up_thru_wanted)); | |
6597 | } | |
6598 | } | |
6599 | ||
6600 | void OSD::request_full_map(epoch_t first, epoch_t last) | |
6601 | { | |
6602 | dout(10) << __func__ << " " << first << ".." << last | |
6603 | << ", previously requested " | |
6604 | << requested_full_first << ".." << requested_full_last << dendl; | |
9f95a23c | 6605 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
11fdf7f2 TL |
6606 | ceph_assert(first > 0 && last > 0); |
6607 | ceph_assert(first <= last); | |
6608 | ceph_assert(first >= requested_full_first); // we shouldn't ever ask for older maps | |
7c673cae FG |
6609 | if (requested_full_first == 0) { |
6610 | // first request | |
6611 | requested_full_first = first; | |
6612 | requested_full_last = last; | |
6613 | } else if (last <= requested_full_last) { | |
6614 | // dup | |
6615 | return; | |
6616 | } else { | |
6617 | // additional request | |
6618 | first = requested_full_last + 1; | |
6619 | requested_full_last = last; | |
6620 | } | |
6621 | MMonGetOSDMap *req = new MMonGetOSDMap; | |
6622 | req->request_full(first, last); | |
6623 | monc->send_mon_message(req); | |
6624 | } | |
6625 | ||
6626 | void OSD::got_full_map(epoch_t e) | |
6627 | { | |
11fdf7f2 | 6628 | ceph_assert(requested_full_first <= requested_full_last); |
9f95a23c | 6629 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae FG |
6630 | if (requested_full_first == 0) { |
6631 | dout(20) << __func__ << " " << e << ", nothing requested" << dendl; | |
6632 | return; | |
6633 | } | |
6634 | if (e < requested_full_first) { | |
6635 | dout(10) << __func__ << " " << e << ", requested " << requested_full_first | |
6636 | << ".." << requested_full_last | |
6637 | << ", ignoring" << dendl; | |
6638 | return; | |
6639 | } | |
6640 | if (e >= requested_full_last) { | |
6641 | dout(10) << __func__ << " " << e << ", requested " << requested_full_first | |
6642 | << ".." << requested_full_last << ", resetting" << dendl; | |
6643 | requested_full_first = requested_full_last = 0; | |
6644 | return; | |
6645 | } | |
6646 | ||
6647 | requested_full_first = e + 1; | |
6648 | ||
6649 | dout(10) << __func__ << " " << e << ", requested " << requested_full_first | |
6650 | << ".." << requested_full_last | |
6651 | << ", still need more" << dendl; | |
6652 | } | |
6653 | ||
6654 | void OSD::requeue_failures() | |
6655 | { | |
11fdf7f2 | 6656 | std::lock_guard l(heartbeat_lock); |
7c673cae FG |
6657 | unsigned old_queue = failure_queue.size(); |
6658 | unsigned old_pending = failure_pending.size(); | |
11fdf7f2 | 6659 | for (auto p = failure_pending.begin(); p != failure_pending.end(); ) { |
7c673cae FG |
6660 | failure_queue[p->first] = p->second.first; |
6661 | failure_pending.erase(p++); | |
6662 | } | |
6663 | dout(10) << __func__ << " " << old_queue << " + " << old_pending << " -> " | |
6664 | << failure_queue.size() << dendl; | |
6665 | } | |
6666 | ||
6667 | void OSD::send_failures() | |
6668 | { | |
9f95a23c TL |
6669 | ceph_assert(ceph_mutex_is_locked(map_lock)); |
6670 | ceph_assert(ceph_mutex_is_locked(mon_report_lock)); | |
11fdf7f2 | 6671 | std::lock_guard l(heartbeat_lock); |
7c673cae | 6672 | utime_t now = ceph_clock_now(); |
9f95a23c | 6673 | const auto osdmap = get_osdmap(); |
7c673cae FG |
6674 | while (!failure_queue.empty()) { |
6675 | int osd = failure_queue.begin()->first; | |
7c673cae FG |
6676 | if (!failure_pending.count(osd)) { |
6677 | int failed_for = (int)(double)(now - failure_queue.begin()->second); | |
11fdf7f2 TL |
6678 | monc->send_mon_message( |
6679 | new MOSDFailure( | |
6680 | monc->get_fsid(), | |
6681 | osd, | |
6682 | osdmap->get_addrs(osd), | |
6683 | failed_for, | |
6684 | osdmap->get_epoch())); | |
6685 | failure_pending[osd] = make_pair(failure_queue.begin()->second, | |
6686 | osdmap->get_addrs(osd)); | |
7c673cae FG |
6687 | } |
6688 | failure_queue.erase(osd); | |
6689 | } | |
6690 | } | |
6691 | ||
11fdf7f2 | 6692 | void OSD::send_still_alive(epoch_t epoch, int osd, const entity_addrvec_t &addrs) |
7c673cae | 6693 | { |
11fdf7f2 TL |
6694 | MOSDFailure *m = new MOSDFailure(monc->get_fsid(), osd, addrs, 0, epoch, |
6695 | MOSDFailure::FLAG_ALIVE); | |
7c673cae FG |
6696 | monc->send_mon_message(m); |
6697 | } | |
6698 | ||
11fdf7f2 | 6699 | void OSD::cancel_pending_failures() |
7c673cae | 6700 | { |
11fdf7f2 TL |
6701 | std::lock_guard l(heartbeat_lock); |
6702 | auto it = failure_pending.begin(); | |
6703 | while (it != failure_pending.end()) { | |
6704 | dout(10) << __func__ << " canceling in-flight failure report for osd." | |
6705 | << it->first << dendl; | |
9f95a23c | 6706 | send_still_alive(get_osdmap_epoch(), it->first, it->second.second); |
11fdf7f2 | 6707 | failure_pending.erase(it++); |
7c673cae | 6708 | } |
7c673cae FG |
6709 | } |
6710 | ||
6711 | void OSD::send_beacon(const ceph::coarse_mono_clock::time_point& now) | |
6712 | { | |
6713 | const auto& monmap = monc->monmap; | |
6714 | // send beacon to mon even if we are just connected, and the monmap is not | |
6715 | // initialized yet by then. | |
6716 | if (monmap.epoch > 0 && | |
6717 | monmap.get_required_features().contains_all( | |
6718 | ceph::features::mon::FEATURE_LUMINOUS)) { | |
6719 | dout(20) << __func__ << " sending" << dendl; | |
7c673cae FG |
6720 | MOSDBeacon* beacon = nullptr; |
6721 | { | |
11fdf7f2 | 6722 | std::lock_guard l{min_last_epoch_clean_lock}; |
9f95a23c TL |
6723 | beacon = new MOSDBeacon(get_osdmap_epoch(), |
6724 | min_last_epoch_clean, | |
6725 | superblock.last_purged_snaps_scrub); | |
494da23a | 6726 | beacon->pgs = min_last_epoch_clean_pgs; |
224ce89b | 6727 | last_sent_beacon = now; |
7c673cae FG |
6728 | } |
6729 | monc->send_mon_message(beacon); | |
6730 | } else { | |
6731 | dout(20) << __func__ << " not sending" << dendl; | |
6732 | } | |
6733 | } | |
6734 | ||
7c673cae FG |
6735 | void OSD::handle_command(MCommand *m) |
6736 | { | |
6737 | ConnectionRef con = m->get_connection(); | |
9f95a23c | 6738 | auto session = ceph::ref_cast<Session>(con->get_priv()); |
7c673cae | 6739 | if (!session) { |
9f95a23c | 6740 | con->send_message(new MCommandReply(m, -EACCES)); |
7c673cae FG |
6741 | m->put(); |
6742 | return; | |
6743 | } | |
9f95a23c TL |
6744 | if (!session->caps.allow_all()) { |
6745 | con->send_message(new MCommandReply(m, -EACCES)); | |
7c673cae FG |
6746 | m->put(); |
6747 | return; | |
6748 | } | |
9f95a23c | 6749 | cct->get_admin_socket()->queue_tell_command(m); |
7c673cae FG |
6750 | m->put(); |
6751 | } | |
6752 | ||
f64942e4 AA |
6753 | namespace { |
6754 | class unlock_guard { | |
9f95a23c | 6755 | ceph::mutex& m; |
f64942e4 | 6756 | public: |
9f95a23c | 6757 | explicit unlock_guard(ceph::mutex& mutex) |
f64942e4 AA |
6758 | : m(mutex) |
6759 | { | |
11fdf7f2 | 6760 | m.unlock(); |
f64942e4 AA |
6761 | } |
6762 | unlock_guard(unlock_guard&) = delete; | |
6763 | ~unlock_guard() { | |
11fdf7f2 | 6764 | m.lock(); |
f64942e4 AA |
6765 | } |
6766 | }; | |
6767 | } | |
6768 | ||
9f95a23c | 6769 | void OSD::scrub_purged_snaps() |
7c673cae | 6770 | { |
9f95a23c TL |
6771 | dout(10) << __func__ << dendl; |
6772 | ceph_assert(ceph_mutex_is_locked(osd_lock)); | |
6773 | SnapMapper::Scrubber s(cct, store, service.meta_ch, | |
6774 | make_snapmapper_oid(), | |
6775 | make_purged_snaps_oid()); | |
6776 | clog->debug() << "purged_snaps scrub starts"; | |
6777 | osd_lock.unlock(); | |
6778 | s.run(); | |
6779 | if (s.stray.size()) { | |
6780 | clog->debug() << "purged_snaps scrub found " << s.stray.size() << " strays"; | |
6781 | } else { | |
6782 | clog->debug() << "purged_snaps scrub ok"; | |
224ce89b | 6783 | } |
9f95a23c TL |
6784 | set<pair<spg_t,snapid_t>> queued; |
6785 | for (auto& [pool, snap, hash, shard] : s.stray) { | |
6786 | const pg_pool_t *pi = get_osdmap()->get_pg_pool(pool); | |
6787 | if (!pi) { | |
6788 | dout(20) << __func__ << " pool " << pool << " dne" << dendl; | |
6789 | continue; | |
11fdf7f2 | 6790 | } |
9f95a23c TL |
6791 | pg_t pgid(pi->raw_hash_to_pg(hash), pool); |
6792 | spg_t spgid(pgid, shard); | |
6793 | pair<spg_t,snapid_t> p(spgid, snap); | |
6794 | if (queued.count(p)) { | |
6795 | dout(20) << __func__ << " pg " << spgid << " snap " << snap | |
6796 | << " already queued" << dendl; | |
6797 | continue; | |
11fdf7f2 | 6798 | } |
9f95a23c TL |
6799 | PGRef pg = lookup_lock_pg(spgid); |
6800 | if (!pg) { | |
6801 | dout(20) << __func__ << " pg " << spgid << " not found" << dendl; | |
6802 | continue; | |
11fdf7f2 | 6803 | } |
9f95a23c TL |
6804 | queued.insert(p); |
6805 | dout(10) << __func__ << " requeue pg " << spgid << " " << pg << " snap " | |
6806 | << snap << dendl; | |
6807 | pg->queue_snap_retrim(snap); | |
6808 | pg->unlock(); | |
7c673cae | 6809 | } |
9f95a23c TL |
6810 | osd_lock.lock(); |
6811 | if (is_stopping()) { | |
6812 | return; | |
6813 | } | |
6814 | dout(10) << __func__ << " done queueing pgs, updating superblock" << dendl; | |
6815 | ObjectStore::Transaction t; | |
6816 | superblock.last_purged_snaps_scrub = ceph_clock_now(); | |
6817 | write_superblock(t); | |
6818 | int tr = store->queue_transaction(service.meta_ch, std::move(t), nullptr); | |
6819 | ceph_assert(tr == 0); | |
6820 | if (is_active()) { | |
6821 | send_beacon(ceph::coarse_mono_clock::now()); | |
6822 | } | |
6823 | dout(10) << __func__ << " done" << dendl; | |
11fdf7f2 TL |
6824 | } |
6825 | ||
6826 | void OSD::probe_smart(const string& only_devid, ostream& ss) | |
6827 | { | |
6828 | set<string> devnames; | |
6829 | store->get_devices(&devnames); | |
6830 | uint64_t smart_timeout = cct->_conf.get_val<uint64_t>( | |
6831 | "osd_smart_report_timeout"); | |
6832 | ||
6833 | // == typedef std::map<std::string, mValue> mObject; | |
6834 | json_spirit::mObject json_map; | |
6835 | ||
6836 | for (auto dev : devnames) { | |
6837 | // smartctl works only on physical devices; filter out any logical device | |
6838 | if (dev.find("dm-") == 0) { | |
6839 | continue; | |
6840 | } | |
6841 | ||
6842 | string err; | |
6843 | string devid = get_device_id(dev, &err); | |
6844 | if (devid.size() == 0) { | |
6845 | dout(10) << __func__ << " no unique id for dev " << dev << " (" | |
6846 | << err << "), skipping" << dendl; | |
6847 | continue; | |
6848 | } | |
6849 | if (only_devid.size() && devid != only_devid) { | |
6850 | continue; | |
6851 | } | |
6852 | ||
6853 | json_spirit::mValue smart_json; | |
6854 | if (block_device_get_metrics(dev, smart_timeout, | |
6855 | &smart_json)) { | |
6856 | dout(10) << "block_device_get_metrics failed for /dev/" << dev << dendl; | |
6857 | continue; | |
6858 | } | |
6859 | json_map[devid] = smart_json; | |
7c673cae | 6860 | } |
11fdf7f2 | 6861 | json_spirit::write(json_map, ss, json_spirit::pretty_print); |
7c673cae FG |
6862 | } |
6863 | ||
6864 | bool OSD::heartbeat_dispatch(Message *m) | |
6865 | { | |
6866 | dout(30) << "heartbeat_dispatch " << m << dendl; | |
6867 | switch (m->get_type()) { | |
6868 | ||
6869 | case CEPH_MSG_PING: | |
6870 | dout(10) << "ping from " << m->get_source_inst() << dendl; | |
6871 | m->put(); | |
6872 | break; | |
6873 | ||
6874 | case MSG_OSD_PING: | |
6875 | handle_osd_ping(static_cast<MOSDPing*>(m)); | |
6876 | break; | |
6877 | ||
6878 | default: | |
6879 | dout(0) << "dropping unexpected message " << *m << " from " << m->get_source_inst() << dendl; | |
6880 | m->put(); | |
6881 | } | |
6882 | ||
6883 | return true; | |
6884 | } | |
6885 | ||
6886 | bool OSD::ms_dispatch(Message *m) | |
6887 | { | |
6888 | dout(20) << "OSD::ms_dispatch: " << *m << dendl; | |
6889 | if (m->get_type() == MSG_OSD_MARK_ME_DOWN) { | |
6890 | service.got_stop_ack(); | |
6891 | m->put(); | |
6892 | return true; | |
6893 | } | |
6894 | ||
6895 | // lock! | |
6896 | ||
9f95a23c | 6897 | osd_lock.lock(); |
7c673cae | 6898 | if (is_stopping()) { |
9f95a23c | 6899 | osd_lock.unlock(); |
7c673cae FG |
6900 | m->put(); |
6901 | return true; | |
6902 | } | |
6903 | ||
6904 | do_waiters(); | |
6905 | _dispatch(m); | |
6906 | ||
9f95a23c | 6907 | osd_lock.unlock(); |
7c673cae FG |
6908 | |
6909 | return true; | |
6910 | } | |
6911 | ||
9f95a23c TL |
6912 | void OSDService::maybe_share_map( |
6913 | Connection *con, | |
6914 | const OSDMapRef& osdmap, | |
6915 | epoch_t peer_epoch_lb) | |
7c673cae | 6916 | { |
9f95a23c TL |
6917 | // NOTE: we assume caller hold something that keeps the Connection itself |
6918 | // pinned (e.g., an OpRequest's MessageRef). | |
6919 | auto session = ceph::ref_cast<Session>(con->get_priv()); | |
6920 | if (!session) { | |
7c673cae FG |
6921 | return; |
6922 | } | |
7c673cae | 6923 | |
9f95a23c TL |
6924 | // assume the peer has the newer of the op's sent_epoch and what |
6925 | // we think we sent them. | |
7c673cae | 6926 | session->sent_epoch_lock.lock(); |
9f95a23c TL |
6927 | if (peer_epoch_lb > session->last_sent_epoch) { |
6928 | dout(10) << __func__ << " con " << con | |
6929 | << " " << con->get_peer_addr() | |
6930 | << " map epoch " << session->last_sent_epoch | |
6931 | << " -> " << peer_epoch_lb << " (as per caller)" << dendl; | |
6932 | session->last_sent_epoch = peer_epoch_lb; | |
6933 | } | |
6934 | epoch_t last_sent_epoch = session->last_sent_epoch; | |
7c673cae FG |
6935 | session->sent_epoch_lock.unlock(); |
6936 | ||
9f95a23c TL |
6937 | if (osdmap->get_epoch() <= last_sent_epoch) { |
6938 | return; | |
6939 | } | |
11fdf7f2 | 6940 | |
9f95a23c TL |
6941 | send_incremental_map(last_sent_epoch, con, osdmap); |
6942 | last_sent_epoch = osdmap->get_epoch(); | |
7c673cae FG |
6943 | |
6944 | session->sent_epoch_lock.lock(); | |
6945 | if (session->last_sent_epoch < last_sent_epoch) { | |
9f95a23c TL |
6946 | dout(10) << __func__ << " con " << con |
6947 | << " " << con->get_peer_addr() | |
6948 | << " map epoch " << session->last_sent_epoch | |
6949 | << " -> " << last_sent_epoch << " (shared)" << dendl; | |
7c673cae FG |
6950 | session->last_sent_epoch = last_sent_epoch; |
6951 | } | |
6952 | session->sent_epoch_lock.unlock(); | |
7c673cae FG |
6953 | } |
6954 | ||
9f95a23c | 6955 | void OSD::dispatch_session_waiting(const ceph::ref_t<Session>& session, OSDMapRef osdmap) |
7c673cae | 6956 | { |
9f95a23c | 6957 | ceph_assert(ceph_mutex_is_locked(session->session_dispatch_lock)); |
7c673cae FG |
6958 | |
6959 | auto i = session->waiting_on_map.begin(); | |
6960 | while (i != session->waiting_on_map.end()) { | |
6961 | OpRequestRef op = &(*i); | |
11fdf7f2 | 6962 | ceph_assert(ms_can_fast_dispatch(op->get_req())); |
9f95a23c | 6963 | auto m = op->get_req<MOSDFastDispatchOp>(); |
7c673cae FG |
6964 | if (m->get_min_epoch() > osdmap->get_epoch()) { |
6965 | break; | |
6966 | } | |
6967 | session->waiting_on_map.erase(i++); | |
6968 | op->put(); | |
6969 | ||
6970 | spg_t pgid; | |
6971 | if (m->get_type() == CEPH_MSG_OSD_OP) { | |
6972 | pg_t actual_pgid = osdmap->raw_pg_to_pg( | |
6973 | static_cast<const MOSDOp*>(m)->get_pg()); | |
6974 | if (!osdmap->get_primary_shard(actual_pgid, &pgid)) { | |
6975 | continue; | |
6976 | } | |
6977 | } else { | |
6978 | pgid = m->get_spg(); | |
6979 | } | |
11fdf7f2 | 6980 | enqueue_op(pgid, std::move(op), m->get_map_epoch()); |
7c673cae FG |
6981 | } |
6982 | ||
6983 | if (session->waiting_on_map.empty()) { | |
6984 | clear_session_waiting_on_map(session); | |
6985 | } else { | |
6986 | register_session_waiting_on_map(session); | |
6987 | } | |
6988 | } | |
6989 | ||
6990 | void OSD::ms_fast_dispatch(Message *m) | |
6991 | { | |
11fdf7f2 | 6992 | FUNCTRACE(cct); |
7c673cae FG |
6993 | if (service.is_stopping()) { |
6994 | m->put(); | |
6995 | return; | |
6996 | } | |
11fdf7f2 TL |
6997 | |
6998 | // peering event? | |
6999 | switch (m->get_type()) { | |
7000 | case CEPH_MSG_PING: | |
7001 | dout(10) << "ping from " << m->get_source() << dendl; | |
7002 | m->put(); | |
7003 | return; | |
11fdf7f2 TL |
7004 | case MSG_OSD_FORCE_RECOVERY: |
7005 | handle_fast_force_recovery(static_cast<MOSDForceRecovery*>(m)); | |
7006 | return; | |
7007 | case MSG_OSD_SCRUB2: | |
7008 | handle_fast_scrub(static_cast<MOSDScrub2*>(m)); | |
7009 | return; | |
7010 | ||
7011 | case MSG_OSD_PG_CREATE2: | |
7012 | return handle_fast_pg_create(static_cast<MOSDPGCreate2*>(m)); | |
7013 | case MSG_OSD_PG_QUERY: | |
7014 | return handle_fast_pg_query(static_cast<MOSDPGQuery*>(m)); | |
7015 | case MSG_OSD_PG_NOTIFY: | |
7016 | return handle_fast_pg_notify(static_cast<MOSDPGNotify*>(m)); | |
7017 | case MSG_OSD_PG_INFO: | |
7018 | return handle_fast_pg_info(static_cast<MOSDPGInfo*>(m)); | |
7019 | case MSG_OSD_PG_REMOVE: | |
7020 | return handle_fast_pg_remove(static_cast<MOSDPGRemove*>(m)); | |
7021 | ||
7022 | // these are single-pg messages that handle themselves | |
7023 | case MSG_OSD_PG_LOG: | |
7024 | case MSG_OSD_PG_TRIM: | |
9f95a23c TL |
7025 | case MSG_OSD_PG_NOTIFY2: |
7026 | case MSG_OSD_PG_QUERY2: | |
7027 | case MSG_OSD_PG_INFO2: | |
11fdf7f2 TL |
7028 | case MSG_OSD_BACKFILL_RESERVE: |
7029 | case MSG_OSD_RECOVERY_RESERVE: | |
9f95a23c TL |
7030 | case MSG_OSD_PG_LEASE: |
7031 | case MSG_OSD_PG_LEASE_ACK: | |
11fdf7f2 TL |
7032 | { |
7033 | MOSDPeeringOp *pm = static_cast<MOSDPeeringOp*>(m); | |
7034 | if (require_osd_peer(pm)) { | |
7035 | enqueue_peering_evt( | |
7036 | pm->get_spg(), | |
7037 | PGPeeringEventRef(pm->get_event())); | |
7038 | } | |
7039 | pm->put(); | |
7040 | return; | |
7041 | } | |
7042 | } | |
7043 | ||
7c673cae FG |
7044 | OpRequestRef op = op_tracker.create_request<OpRequest, Message*>(m); |
7045 | { | |
7046 | #ifdef WITH_LTTNG | |
7047 | osd_reqid_t reqid = op->get_reqid(); | |
7048 | #endif | |
7049 | tracepoint(osd, ms_fast_dispatch, reqid.name._type, | |
7050 | reqid.name._num, reqid.tid, reqid.inc); | |
7051 | } | |
7052 | ||
7053 | if (m->trace) | |
7054 | op->osd_trace.init("osd op", &trace_endpoint, &m->trace); | |
7055 | ||
11fdf7f2 | 7056 | // note sender epoch, min req's epoch |
7c673cae FG |
7057 | op->sent_epoch = static_cast<MOSDFastDispatchOp*>(m)->get_map_epoch(); |
7058 | op->min_epoch = static_cast<MOSDFastDispatchOp*>(m)->get_min_epoch(); | |
11fdf7f2 | 7059 | ceph_assert(op->min_epoch <= op->sent_epoch); // sanity check! |
7c673cae FG |
7060 | |
7061 | service.maybe_inject_dispatch_delay(); | |
7062 | ||
7063 | if (m->get_connection()->has_features(CEPH_FEATUREMASK_RESEND_ON_SPLIT) || | |
7064 | m->get_type() != CEPH_MSG_OSD_OP) { | |
7065 | // queue it directly | |
7066 | enqueue_op( | |
7067 | static_cast<MOSDFastDispatchOp*>(m)->get_spg(), | |
11fdf7f2 | 7068 | std::move(op), |
7c673cae FG |
7069 | static_cast<MOSDFastDispatchOp*>(m)->get_map_epoch()); |
7070 | } else { | |
7071 | // legacy client, and this is an MOSDOp (the *only* fast dispatch | |
7072 | // message that didn't have an explicit spg_t); we need to map | |
7073 | // them to an spg_t while preserving delivery order. | |
11fdf7f2 TL |
7074 | auto priv = m->get_connection()->get_priv(); |
7075 | if (auto session = static_cast<Session*>(priv.get()); session) { | |
7076 | std::lock_guard l{session->session_dispatch_lock}; | |
7077 | op->get(); | |
7078 | session->waiting_on_map.push_back(*op); | |
7079 | OSDMapRef nextmap = service.get_nextmap_reserved(); | |
7080 | dispatch_session_waiting(session, nextmap); | |
7081 | service.release_map(nextmap); | |
7c673cae FG |
7082 | } |
7083 | } | |
7084 | OID_EVENT_TRACE_WITH_MSG(m, "MS_FAST_DISPATCH_END", false); | |
7085 | } | |
7086 | ||
11fdf7f2 | 7087 | int OSD::ms_handle_authentication(Connection *con) |
7c673cae | 7088 | { |
11fdf7f2 | 7089 | int ret = 0; |
9f95a23c | 7090 | auto s = ceph::ref_cast<Session>(con->get_priv()); |
11fdf7f2 | 7091 | if (!s) { |
9f95a23c TL |
7092 | s = ceph::make_ref<Session>(cct, con); |
7093 | con->set_priv(s); | |
11fdf7f2 TL |
7094 | s->entity_name = con->get_peer_entity_name(); |
7095 | dout(10) << __func__ << " new session " << s << " con " << s->con | |
7096 | << " entity " << s->entity_name | |
7097 | << " addr " << con->get_peer_addrs() << dendl; | |
7098 | } else { | |
7099 | dout(10) << __func__ << " existing session " << s << " con " << s->con | |
7100 | << " entity " << s->entity_name | |
7101 | << " addr " << con->get_peer_addrs() << dendl; | |
7c673cae FG |
7102 | } |
7103 | ||
11fdf7f2 | 7104 | AuthCapsInfo &caps_info = con->get_peer_caps_info(); |
9f95a23c | 7105 | if (caps_info.allow_all) { |
11fdf7f2 | 7106 | s->caps.set_allow_all(); |
9f95a23c | 7107 | } else if (caps_info.caps.length() > 0) { |
11fdf7f2 TL |
7108 | bufferlist::const_iterator p = caps_info.caps.cbegin(); |
7109 | string str; | |
7110 | try { | |
7111 | decode(str, p); | |
7112 | } | |
7113 | catch (buffer::error& e) { | |
7114 | dout(10) << __func__ << " session " << s << " " << s->entity_name | |
7115 | << " failed to decode caps string" << dendl; | |
9f95a23c | 7116 | ret = -EACCES; |
11fdf7f2 TL |
7117 | } |
7118 | if (!ret) { | |
7c673cae | 7119 | bool success = s->caps.parse(str); |
11fdf7f2 TL |
7120 | if (success) { |
7121 | dout(10) << __func__ << " session " << s | |
7122 | << " " << s->entity_name | |
7123 | << " has caps " << s->caps << " '" << str << "'" << dendl; | |
7124 | ret = 1; | |
7125 | } else { | |
7126 | dout(10) << __func__ << " session " << s << " " << s->entity_name | |
7127 | << " failed to parse caps '" << str << "'" << dendl; | |
9f95a23c | 7128 | ret = -EACCES; |
11fdf7f2 | 7129 | } |
7c673cae | 7130 | } |
7c673cae | 7131 | } |
11fdf7f2 | 7132 | return ret; |
7c673cae FG |
7133 | } |
7134 | ||
7135 | void OSD::do_waiters() | |
7136 | { | |
9f95a23c | 7137 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae FG |
7138 | |
7139 | dout(10) << "do_waiters -- start" << dendl; | |
7140 | while (!finished.empty()) { | |
7141 | OpRequestRef next = finished.front(); | |
7142 | finished.pop_front(); | |
7143 | dispatch_op(next); | |
7144 | } | |
7145 | dout(10) << "do_waiters -- finish" << dendl; | |
7146 | } | |
7147 | ||
7148 | void OSD::dispatch_op(OpRequestRef op) | |
7149 | { | |
7150 | switch (op->get_req()->get_type()) { | |
7151 | ||
7152 | case MSG_OSD_PG_CREATE: | |
7153 | handle_pg_create(op); | |
7154 | break; | |
7c673cae FG |
7155 | } |
7156 | } | |
7157 | ||
7158 | void OSD::_dispatch(Message *m) | |
7159 | { | |
9f95a23c | 7160 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae FG |
7161 | dout(20) << "_dispatch " << m << " " << *m << dendl; |
7162 | ||
7163 | switch (m->get_type()) { | |
7c673cae FG |
7164 | // -- don't need OSDMap -- |
7165 | ||
7166 | // map and replication | |
7167 | case CEPH_MSG_OSD_MAP: | |
7168 | handle_osd_map(static_cast<MOSDMap*>(m)); | |
7169 | break; | |
9f95a23c TL |
7170 | case MSG_MON_GET_PURGED_SNAPS_REPLY: |
7171 | handle_get_purged_snaps_reply(static_cast<MMonGetPurgedSnapsReply*>(m)); | |
7172 | break; | |
7c673cae FG |
7173 | |
7174 | // osd | |
7c673cae FG |
7175 | case MSG_OSD_SCRUB: |
7176 | handle_scrub(static_cast<MOSDScrub*>(m)); | |
7177 | break; | |
7178 | ||
11fdf7f2 TL |
7179 | case MSG_COMMAND: |
7180 | handle_command(static_cast<MCommand*>(m)); | |
7181 | return; | |
c07f9fc5 | 7182 | |
7c673cae FG |
7183 | // -- need OSDMap -- |
7184 | ||
7185 | case MSG_OSD_PG_CREATE: | |
7c673cae FG |
7186 | { |
7187 | OpRequestRef op = op_tracker.create_request<OpRequest, Message*>(m); | |
7188 | if (m->trace) | |
7189 | op->osd_trace.init("osd op", &trace_endpoint, &m->trace); | |
7190 | // no map? starting up? | |
9f95a23c | 7191 | if (!get_osdmap()) { |
7c673cae FG |
7192 | dout(7) << "no OSDMap, not booted" << dendl; |
7193 | logger->inc(l_osd_waiting_for_map); | |
7194 | waiting_for_osdmap.push_back(op); | |
7195 | op->mark_delayed("no osdmap"); | |
7196 | break; | |
7197 | } | |
7198 | ||
7199 | // need OSDMap | |
7200 | dispatch_op(op); | |
7201 | } | |
7202 | } | |
7203 | } | |
7204 | ||
11fdf7f2 | 7205 | // remove me post-nautilus |
7c673cae FG |
7206 | void OSD::handle_scrub(MOSDScrub *m) |
7207 | { | |
7208 | dout(10) << "handle_scrub " << *m << dendl; | |
7209 | if (!require_mon_or_mgr_peer(m)) { | |
7210 | m->put(); | |
7211 | return; | |
7212 | } | |
7213 | if (m->fsid != monc->get_fsid()) { | |
11fdf7f2 TL |
7214 | dout(0) << "handle_scrub fsid " << m->fsid << " != " << monc->get_fsid() |
7215 | << dendl; | |
7c673cae FG |
7216 | m->put(); |
7217 | return; | |
7218 | } | |
7219 | ||
11fdf7f2 TL |
7220 | vector<spg_t> spgs; |
7221 | _get_pgids(&spgs); | |
7222 | ||
7223 | if (!m->scrub_pgs.empty()) { | |
7224 | vector<spg_t> v; | |
7225 | for (auto pgid : m->scrub_pgs) { | |
7c673cae | 7226 | spg_t pcand; |
9f95a23c | 7227 | if (get_osdmap()->get_primary_shard(pgid, &pcand) && |
11fdf7f2 TL |
7228 | std::find(spgs.begin(), spgs.end(), pcand) != spgs.end()) { |
7229 | v.push_back(pcand); | |
7c673cae FG |
7230 | } |
7231 | } | |
11fdf7f2 TL |
7232 | spgs.swap(v); |
7233 | } | |
7234 | ||
7235 | for (auto pgid : spgs) { | |
7236 | enqueue_peering_evt( | |
7237 | pgid, | |
7238 | PGPeeringEventRef( | |
7239 | std::make_shared<PGPeeringEvent>( | |
7240 | get_osdmap_epoch(), | |
7241 | get_osdmap_epoch(), | |
9f95a23c | 7242 | PeeringState::RequestScrub(m->deep, m->repair)))); |
7c673cae FG |
7243 | } |
7244 | ||
7245 | m->put(); | |
7246 | } | |
7247 | ||
11fdf7f2 TL |
7248 | void OSD::handle_fast_scrub(MOSDScrub2 *m) |
7249 | { | |
7250 | dout(10) << __func__ << " " << *m << dendl; | |
7251 | if (!require_mon_or_mgr_peer(m)) { | |
7252 | m->put(); | |
7253 | return; | |
7254 | } | |
7255 | if (m->fsid != monc->get_fsid()) { | |
7256 | dout(0) << __func__ << " fsid " << m->fsid << " != " << monc->get_fsid() | |
7257 | << dendl; | |
7258 | m->put(); | |
7259 | return; | |
7260 | } | |
7261 | for (auto pgid : m->scrub_pgs) { | |
7262 | enqueue_peering_evt( | |
7263 | pgid, | |
7264 | PGPeeringEventRef( | |
7265 | std::make_shared<PGPeeringEvent>( | |
7266 | m->epoch, | |
7267 | m->epoch, | |
9f95a23c | 7268 | PeeringState::RequestScrub(m->deep, m->repair)))); |
11fdf7f2 TL |
7269 | } |
7270 | m->put(); | |
7271 | } | |
7272 | ||
7c673cae FG |
7273 | bool OSD::scrub_random_backoff() |
7274 | { | |
7275 | bool coin_flip = (rand() / (double)RAND_MAX >= | |
7276 | cct->_conf->osd_scrub_backoff_ratio); | |
7277 | if (!coin_flip) { | |
7278 | dout(20) << "scrub_random_backoff lost coin flip, randomly backing off" << dendl; | |
7279 | return true; | |
7280 | } | |
7281 | return false; | |
7282 | } | |
7283 | ||
7284 | OSDService::ScrubJob::ScrubJob(CephContext* cct, | |
7285 | const spg_t& pg, const utime_t& timestamp, | |
7286 | double pool_scrub_min_interval, | |
7287 | double pool_scrub_max_interval, bool must) | |
7288 | : cct(cct), | |
7289 | pgid(pg), | |
7290 | sched_time(timestamp), | |
7291 | deadline(timestamp) | |
7292 | { | |
7293 | // if not explicitly requested, postpone the scrub with a random delay | |
7294 | if (!must) { | |
7295 | double scrub_min_interval = pool_scrub_min_interval > 0 ? | |
7296 | pool_scrub_min_interval : cct->_conf->osd_scrub_min_interval; | |
7297 | double scrub_max_interval = pool_scrub_max_interval > 0 ? | |
7298 | pool_scrub_max_interval : cct->_conf->osd_scrub_max_interval; | |
7299 | ||
7300 | sched_time += scrub_min_interval; | |
7301 | double r = rand() / (double)RAND_MAX; | |
7302 | sched_time += | |
7303 | scrub_min_interval * cct->_conf->osd_scrub_interval_randomize_ratio * r; | |
11fdf7f2 TL |
7304 | if (scrub_max_interval == 0) { |
7305 | deadline = utime_t(); | |
7306 | } else { | |
7307 | deadline += scrub_max_interval; | |
7308 | } | |
7309 | ||
7c673cae FG |
7310 | } |
7311 | } | |
7312 | ||
7313 | bool OSDService::ScrubJob::ScrubJob::operator<(const OSDService::ScrubJob& rhs) const { | |
7314 | if (sched_time < rhs.sched_time) | |
7315 | return true; | |
7316 | if (sched_time > rhs.sched_time) | |
7317 | return false; | |
7318 | return pgid < rhs.pgid; | |
7319 | } | |
7320 | ||
9f95a23c TL |
7321 | double OSD::scrub_sleep_time(bool must_scrub) |
7322 | { | |
7323 | if (must_scrub) { | |
7324 | return cct->_conf->osd_scrub_sleep; | |
7325 | } | |
7326 | utime_t now = ceph_clock_now(); | |
7327 | if (scrub_time_permit(now)) { | |
7328 | return cct->_conf->osd_scrub_sleep; | |
7329 | } | |
7330 | double normal_sleep = cct->_conf->osd_scrub_sleep; | |
7331 | double extended_sleep = cct->_conf->osd_scrub_extended_sleep; | |
7332 | return std::max(extended_sleep, normal_sleep); | |
7333 | } | |
7334 | ||
7c673cae FG |
7335 | bool OSD::scrub_time_permit(utime_t now) |
7336 | { | |
7337 | struct tm bdt; | |
7338 | time_t tt = now.sec(); | |
7339 | localtime_r(&tt, &bdt); | |
28e407b8 AA |
7340 | |
7341 | bool day_permit = false; | |
7342 | if (cct->_conf->osd_scrub_begin_week_day < cct->_conf->osd_scrub_end_week_day) { | |
7343 | if (bdt.tm_wday >= cct->_conf->osd_scrub_begin_week_day && bdt.tm_wday < cct->_conf->osd_scrub_end_week_day) { | |
7344 | day_permit = true; | |
7345 | } | |
7346 | } else { | |
7347 | if (bdt.tm_wday >= cct->_conf->osd_scrub_begin_week_day || bdt.tm_wday < cct->_conf->osd_scrub_end_week_day) { | |
7348 | day_permit = true; | |
7349 | } | |
7350 | } | |
7351 | ||
7352 | if (!day_permit) { | |
7353 | dout(20) << __func__ << " should run between week day " << cct->_conf->osd_scrub_begin_week_day | |
7354 | << " - " << cct->_conf->osd_scrub_end_week_day | |
7355 | << " now " << bdt.tm_wday << " = no" << dendl; | |
7356 | return false; | |
7357 | } | |
7358 | ||
7c673cae FG |
7359 | bool time_permit = false; |
7360 | if (cct->_conf->osd_scrub_begin_hour < cct->_conf->osd_scrub_end_hour) { | |
7361 | if (bdt.tm_hour >= cct->_conf->osd_scrub_begin_hour && bdt.tm_hour < cct->_conf->osd_scrub_end_hour) { | |
7362 | time_permit = true; | |
7363 | } | |
7364 | } else { | |
7365 | if (bdt.tm_hour >= cct->_conf->osd_scrub_begin_hour || bdt.tm_hour < cct->_conf->osd_scrub_end_hour) { | |
7366 | time_permit = true; | |
7367 | } | |
7368 | } | |
7369 | if (!time_permit) { | |
7370 | dout(20) << __func__ << " should run between " << cct->_conf->osd_scrub_begin_hour | |
7371 | << " - " << cct->_conf->osd_scrub_end_hour | |
7372 | << " now " << bdt.tm_hour << " = no" << dendl; | |
7373 | } else { | |
7374 | dout(20) << __func__ << " should run between " << cct->_conf->osd_scrub_begin_hour | |
7375 | << " - " << cct->_conf->osd_scrub_end_hour | |
7376 | << " now " << bdt.tm_hour << " = yes" << dendl; | |
7377 | } | |
7378 | return time_permit; | |
7379 | } | |
7380 | ||
7381 | bool OSD::scrub_load_below_threshold() | |
7382 | { | |
7383 | double loadavgs[3]; | |
7384 | if (getloadavg(loadavgs, 3) != 3) { | |
7385 | dout(10) << __func__ << " couldn't read loadavgs\n" << dendl; | |
7386 | return false; | |
7387 | } | |
7388 | ||
7389 | // allow scrub if below configured threshold | |
91327a77 AA |
7390 | long cpus = sysconf(_SC_NPROCESSORS_ONLN); |
7391 | double loadavg_per_cpu = cpus > 0 ? loadavgs[0] / cpus : loadavgs[0]; | |
7392 | if (loadavg_per_cpu < cct->_conf->osd_scrub_load_threshold) { | |
7393 | dout(20) << __func__ << " loadavg per cpu " << loadavg_per_cpu | |
7c673cae FG |
7394 | << " < max " << cct->_conf->osd_scrub_load_threshold |
7395 | << " = yes" << dendl; | |
7396 | return true; | |
7397 | } | |
7398 | ||
7399 | // allow scrub if below daily avg and currently decreasing | |
7400 | if (loadavgs[0] < daily_loadavg && loadavgs[0] < loadavgs[2]) { | |
7401 | dout(20) << __func__ << " loadavg " << loadavgs[0] | |
7402 | << " < daily_loadavg " << daily_loadavg | |
7403 | << " and < 15m avg " << loadavgs[2] | |
7404 | << " = yes" << dendl; | |
7405 | return true; | |
7406 | } | |
7407 | ||
7408 | dout(20) << __func__ << " loadavg " << loadavgs[0] | |
7409 | << " >= max " << cct->_conf->osd_scrub_load_threshold | |
7410 | << " and ( >= daily_loadavg " << daily_loadavg | |
7411 | << " or >= 15m avg " << loadavgs[2] | |
7412 | << ") = no" << dendl; | |
7413 | return false; | |
7414 | } | |
7415 | ||
7416 | void OSD::sched_scrub() | |
7417 | { | |
7418 | // if not permitted, fail fast | |
eafe8130 | 7419 | if (!service.can_inc_scrubs()) { |
7c673cae FG |
7420 | return; |
7421 | } | |
eafe8130 TL |
7422 | bool allow_requested_repair_only = false; |
7423 | if (service.is_recovery_active()) { | |
7424 | if (!cct->_conf->osd_scrub_during_recovery && cct->_conf->osd_repair_during_recovery) { | |
7425 | dout(10) << __func__ | |
7426 | << " will only schedule explicitly requested repair due to active recovery" | |
7427 | << dendl; | |
7428 | allow_requested_repair_only = true; | |
7429 | } else if (!cct->_conf->osd_scrub_during_recovery && !cct->_conf->osd_repair_during_recovery) { | |
7430 | dout(20) << __func__ << " not scheduling scrubs due to active recovery" << dendl; | |
7431 | return; | |
7432 | } | |
b5b8bbf5 FG |
7433 | } |
7434 | ||
7c673cae FG |
7435 | utime_t now = ceph_clock_now(); |
7436 | bool time_permit = scrub_time_permit(now); | |
7437 | bool load_is_low = scrub_load_below_threshold(); | |
7438 | dout(20) << "sched_scrub load_is_low=" << (int)load_is_low << dendl; | |
7439 | ||
7440 | OSDService::ScrubJob scrub; | |
7441 | if (service.first_scrub_stamp(&scrub)) { | |
7442 | do { | |
7443 | dout(30) << "sched_scrub examine " << scrub.pgid << " at " << scrub.sched_time << dendl; | |
7444 | ||
7445 | if (scrub.sched_time > now) { | |
7446 | // save ourselves some effort | |
7447 | dout(10) << "sched_scrub " << scrub.pgid << " scheduled at " << scrub.sched_time | |
7448 | << " > " << now << dendl; | |
7449 | break; | |
7450 | } | |
7451 | ||
11fdf7f2 | 7452 | if ((scrub.deadline.is_zero() || scrub.deadline >= now) && !(time_permit && load_is_low)) { |
7c673cae FG |
7453 | dout(10) << __func__ << " not scheduling scrub for " << scrub.pgid << " due to " |
7454 | << (!time_permit ? "time not permit" : "high load") << dendl; | |
7455 | continue; | |
7456 | } | |
7457 | ||
11fdf7f2 | 7458 | PGRef pg = _lookup_lock_pg(scrub.pgid); |
7c673cae FG |
7459 | if (!pg) |
7460 | continue; | |
494da23a TL |
7461 | // This has already started, so go on to the next scrub job |
7462 | if (pg->scrubber.active) { | |
7463 | pg->unlock(); | |
7464 | dout(30) << __func__ << ": already in progress pgid " << scrub.pgid << dendl; | |
7465 | continue; | |
7466 | } | |
eafe8130 TL |
7467 | // Skip other kinds of scrubing if only explicitly requested repairing is allowed |
7468 | if (allow_requested_repair_only && !pg->scrubber.must_repair) { | |
7469 | pg->unlock(); | |
7470 | dout(10) << __func__ << " skip " << scrub.pgid | |
7471 | << " because repairing is not explicitly requested on it" | |
7472 | << dendl; | |
7473 | continue; | |
7474 | } | |
494da23a | 7475 | // If it is reserving, let it resolve before going to the next scrub job |
eafe8130 | 7476 | if (pg->scrubber.local_reserved && !pg->scrubber.active) { |
494da23a TL |
7477 | pg->unlock(); |
7478 | dout(30) << __func__ << ": reserve in progress pgid " << scrub.pgid << dendl; | |
7479 | break; | |
7480 | } | |
11fdf7f2 TL |
7481 | dout(10) << "sched_scrub scrubbing " << scrub.pgid << " at " << scrub.sched_time |
7482 | << (pg->get_must_scrub() ? ", explicitly requested" : | |
7483 | (load_is_low ? ", load_is_low" : " deadline < now")) | |
7484 | << dendl; | |
7485 | if (pg->sched_scrub()) { | |
7486 | pg->unlock(); | |
7487 | break; | |
7c673cae FG |
7488 | } |
7489 | pg->unlock(); | |
7490 | } while (service.next_scrub_stamp(scrub, &scrub)); | |
7491 | } | |
7492 | dout(20) << "sched_scrub done" << dendl; | |
7493 | } | |
7494 | ||
494da23a TL |
7495 | void OSD::resched_all_scrubs() |
7496 | { | |
7497 | dout(10) << __func__ << ": start" << dendl; | |
7498 | OSDService::ScrubJob scrub; | |
7499 | if (service.first_scrub_stamp(&scrub)) { | |
7500 | do { | |
7501 | dout(20) << __func__ << ": examine " << scrub.pgid << dendl; | |
7502 | ||
7503 | PGRef pg = _lookup_lock_pg(scrub.pgid); | |
7504 | if (!pg) | |
7505 | continue; | |
7506 | if (!pg->scrubber.must_scrub && !pg->scrubber.need_auto) { | |
7507 | dout(20) << __func__ << ": reschedule " << scrub.pgid << dendl; | |
7508 | pg->on_info_history_change(); | |
7509 | } | |
7510 | pg->unlock(); | |
7511 | } while (service.next_scrub_stamp(scrub, &scrub)); | |
7512 | } | |
7513 | dout(10) << __func__ << ": done" << dendl; | |
7514 | } | |
7515 | ||
11fdf7f2 TL |
7516 | MPGStats* OSD::collect_pg_stats() |
7517 | { | |
7518 | // This implementation unconditionally sends every is_primary PG's | |
7519 | // stats every time we're called. This has equivalent cost to the | |
7520 | // previous implementation's worst case where all PGs are busy and | |
7521 | // their stats are always enqueued for sending. | |
9f95a23c | 7522 | std::shared_lock l{map_lock}; |
11fdf7f2 | 7523 | |
11fdf7f2 TL |
7524 | osd_stat_t cur_stat = service.get_osd_stat(); |
7525 | cur_stat.os_perf_stat = store->get_cur_stats(); | |
7526 | ||
9f95a23c | 7527 | auto m = new MPGStats(monc->get_fsid(), get_osdmap_epoch()); |
11fdf7f2 TL |
7528 | m->osd_stat = cur_stat; |
7529 | ||
7530 | std::lock_guard lec{min_last_epoch_clean_lock}; | |
9f95a23c | 7531 | min_last_epoch_clean = get_osdmap_epoch(); |
11fdf7f2 TL |
7532 | min_last_epoch_clean_pgs.clear(); |
7533 | ||
7534 | std::set<int64_t> pool_set; | |
7535 | vector<PGRef> pgs; | |
7536 | _get_pgs(&pgs); | |
7537 | for (auto& pg : pgs) { | |
7538 | auto pool = pg->pg_id.pgid.pool(); | |
7539 | pool_set.emplace((int64_t)pool); | |
7540 | if (!pg->is_primary()) { | |
7541 | continue; | |
7542 | } | |
7543 | pg->get_pg_stats([&](const pg_stat_t& s, epoch_t lec) { | |
7544 | m->pg_stat[pg->pg_id.pgid] = s; | |
7545 | min_last_epoch_clean = min(min_last_epoch_clean, lec); | |
7546 | min_last_epoch_clean_pgs.push_back(pg->pg_id.pgid); | |
7547 | }); | |
7548 | } | |
7549 | store_statfs_t st; | |
81eedcae | 7550 | bool per_pool_stats = false; |
9f95a23c | 7551 | bool per_pool_omap_stats = false; |
11fdf7f2 | 7552 | for (auto p : pool_set) { |
9f95a23c | 7553 | int r = store->pool_statfs(p, &st, &per_pool_omap_stats); |
11fdf7f2 TL |
7554 | if (r == -ENOTSUP) { |
7555 | break; | |
7556 | } else { | |
7557 | assert(r >= 0); | |
7558 | m->pool_stat[p] = st; | |
81eedcae | 7559 | per_pool_stats = true; |
11fdf7f2 TL |
7560 | } |
7561 | } | |
7c673cae | 7562 | |
81eedcae TL |
7563 | // indicate whether we are reporting per-pool stats |
7564 | m->osd_stat.num_osds = 1; | |
7565 | m->osd_stat.num_per_pool_osds = per_pool_stats ? 1 : 0; | |
9f95a23c | 7566 | m->osd_stat.num_per_pool_omap_osds = per_pool_omap_stats ? 1 : 0; |
81eedcae | 7567 | |
11fdf7f2 TL |
7568 | return m; |
7569 | } | |
7c673cae | 7570 | |
11fdf7f2 | 7571 | vector<DaemonHealthMetric> OSD::get_health_metrics() |
b32b8144 | 7572 | { |
11fdf7f2 TL |
7573 | vector<DaemonHealthMetric> metrics; |
7574 | { | |
7575 | utime_t oldest_secs; | |
7576 | const utime_t now = ceph_clock_now(); | |
7577 | auto too_old = now; | |
7578 | too_old -= cct->_conf.get_val<double>("osd_op_complaint_time"); | |
7579 | int slow = 0; | |
7580 | TrackedOpRef oldest_op; | |
7581 | auto count_slow_ops = [&](TrackedOp& op) { | |
7582 | if (op.get_initiated() < too_old) { | |
9f95a23c TL |
7583 | stringstream ss; |
7584 | ss << "slow request " << op.get_desc() | |
7585 | << " initiated " | |
7586 | << op.get_initiated() | |
7587 | << " currently " | |
7588 | << op.state_string(); | |
7589 | lgeneric_subdout(cct,osd,20) << ss.str() << dendl; | |
7590 | clog->warn() << ss.str(); | |
11fdf7f2 TL |
7591 | slow++; |
7592 | if (!oldest_op || op.get_initiated() < oldest_op->get_initiated()) { | |
7593 | oldest_op = &op; | |
7594 | } | |
7595 | return true; | |
7596 | } else { | |
7597 | return false; | |
7598 | } | |
7599 | }; | |
7600 | if (op_tracker.visit_ops_in_flight(&oldest_secs, count_slow_ops)) { | |
7601 | if (slow) { | |
7602 | derr << __func__ << " reporting " << slow << " slow ops, oldest is " | |
7603 | << oldest_op->get_desc() << dendl; | |
7604 | } | |
7605 | metrics.emplace_back(daemon_metric::SLOW_OPS, slow, oldest_secs); | |
7606 | } else { | |
7607 | // no news is not good news. | |
7608 | metrics.emplace_back(daemon_metric::SLOW_OPS, 0, 0); | |
7609 | } | |
7610 | } | |
7611 | { | |
7612 | std::lock_guard l(pending_creates_lock); | |
7613 | auto n_primaries = pending_creates_from_mon; | |
7614 | for (const auto& create : pending_creates_from_osd) { | |
7615 | if (create.second) { | |
7616 | n_primaries++; | |
7617 | } | |
b32b8144 | 7618 | } |
11fdf7f2 | 7619 | metrics.emplace_back(daemon_metric::PENDING_CREATING_PGS, n_primaries); |
b32b8144 | 7620 | } |
b32b8144 FG |
7621 | return metrics; |
7622 | } | |
7623 | ||
7c673cae FG |
7624 | // ===================================================== |
7625 | // MAP | |
7626 | ||
7627 | void OSD::wait_for_new_map(OpRequestRef op) | |
7628 | { | |
7629 | // ask? | |
7630 | if (waiting_for_osdmap.empty()) { | |
9f95a23c | 7631 | osdmap_subscribe(get_osdmap_epoch() + 1, false); |
7c673cae FG |
7632 | } |
7633 | ||
7634 | logger->inc(l_osd_waiting_for_map); | |
7635 | waiting_for_osdmap.push_back(op); | |
7636 | op->mark_delayed("wait for new map"); | |
7637 | } | |
7638 | ||
7639 | ||
7640 | /** update_map | |
7641 | * assimilate new OSDMap(s). scan pgs, etc. | |
7642 | */ | |
7643 | ||
7644 | void OSD::note_down_osd(int peer) | |
7645 | { | |
9f95a23c TL |
7646 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7647 | cluster_messenger->mark_down_addrs(get_osdmap()->get_cluster_addrs(peer)); | |
7c673cae | 7648 | |
9f95a23c | 7649 | std::lock_guard l{heartbeat_lock}; |
7c673cae FG |
7650 | failure_queue.erase(peer); |
7651 | failure_pending.erase(peer); | |
7652 | map<int,HeartbeatInfo>::iterator p = heartbeat_peers.find(peer); | |
7653 | if (p != heartbeat_peers.end()) { | |
9f95a23c | 7654 | p->second.clear_mark_down(); |
7c673cae FG |
7655 | heartbeat_peers.erase(p); |
7656 | } | |
7c673cae FG |
7657 | } |
7658 | ||
7659 | void OSD::note_up_osd(int peer) | |
7660 | { | |
7c673cae FG |
7661 | heartbeat_set_peers_need_update(); |
7662 | } | |
7663 | ||
7664 | struct C_OnMapCommit : public Context { | |
7665 | OSD *osd; | |
7666 | epoch_t first, last; | |
7667 | MOSDMap *msg; | |
7668 | C_OnMapCommit(OSD *o, epoch_t f, epoch_t l, MOSDMap *m) | |
7669 | : osd(o), first(f), last(l), msg(m) {} | |
7670 | void finish(int r) override { | |
7671 | osd->_committed_osd_maps(first, last, msg); | |
7672 | msg->put(); | |
7673 | } | |
7674 | }; | |
7675 | ||
7c673cae FG |
7676 | void OSD::osdmap_subscribe(version_t epoch, bool force_request) |
7677 | { | |
11fdf7f2 | 7678 | std::lock_guard l(osdmap_subscribe_lock); |
181888fb | 7679 | if (latest_subscribed_epoch >= epoch && !force_request) |
7c673cae FG |
7680 | return; |
7681 | ||
11fdf7f2 | 7682 | latest_subscribed_epoch = std::max<uint64_t>(epoch, latest_subscribed_epoch); |
181888fb | 7683 | |
7c673cae FG |
7684 | if (monc->sub_want_increment("osdmap", epoch, CEPH_SUBSCRIBE_ONETIME) || |
7685 | force_request) { | |
7686 | monc->renew_subs(); | |
7687 | } | |
7688 | } | |
7689 | ||
7690 | void OSD::trim_maps(epoch_t oldest, int nreceived, bool skip_maps) | |
7691 | { | |
7692 | epoch_t min = std::min(oldest, service.map_cache.cached_key_lower_bound()); | |
7693 | if (min <= superblock.oldest_map) | |
7694 | return; | |
7695 | ||
7696 | int num = 0; | |
7697 | ObjectStore::Transaction t; | |
7698 | for (epoch_t e = superblock.oldest_map; e < min; ++e) { | |
7699 | dout(20) << " removing old osdmap epoch " << e << dendl; | |
7700 | t.remove(coll_t::meta(), get_osdmap_pobject_name(e)); | |
7701 | t.remove(coll_t::meta(), get_inc_osdmap_pobject_name(e)); | |
7702 | superblock.oldest_map = e + 1; | |
7703 | num++; | |
7704 | if (num >= cct->_conf->osd_target_transaction_size && num >= nreceived) { | |
7705 | service.publish_superblock(superblock); | |
7706 | write_superblock(t); | |
11fdf7f2 TL |
7707 | int tr = store->queue_transaction(service.meta_ch, std::move(t), nullptr); |
7708 | ceph_assert(tr == 0); | |
7c673cae FG |
7709 | num = 0; |
7710 | if (!skip_maps) { | |
7711 | // skip_maps leaves us with a range of old maps if we fail to remove all | |
7712 | // of them before moving superblock.oldest_map forward to the first map | |
7713 | // in the incoming MOSDMap msg. so we should continue removing them in | |
7714 | // this case, even we could do huge series of delete transactions all at | |
7715 | // once. | |
7716 | break; | |
7717 | } | |
7718 | } | |
7719 | } | |
7720 | if (num > 0) { | |
7721 | service.publish_superblock(superblock); | |
7722 | write_superblock(t); | |
11fdf7f2 TL |
7723 | int tr = store->queue_transaction(service.meta_ch, std::move(t), nullptr); |
7724 | ceph_assert(tr == 0); | |
7c673cae FG |
7725 | } |
7726 | // we should not remove the cached maps | |
11fdf7f2 | 7727 | ceph_assert(min <= service.map_cache.cached_key_lower_bound()); |
7c673cae FG |
7728 | } |
7729 | ||
7730 | void OSD::handle_osd_map(MOSDMap *m) | |
7731 | { | |
11fdf7f2 TL |
7732 | // wait for pgs to catch up |
7733 | { | |
7734 | // we extend the map cache pins to accomodate pgs slow to consume maps | |
7735 | // for some period, until we hit the max_lag_factor bound, at which point | |
7736 | // we block here to stop injesting more maps than they are able to keep | |
7737 | // up with. | |
7738 | epoch_t max_lag = cct->_conf->osd_map_cache_size * | |
7739 | m_osd_pg_epoch_max_lag_factor; | |
7740 | ceph_assert(max_lag > 0); | |
7741 | epoch_t osd_min = 0; | |
7742 | for (auto shard : shards) { | |
7743 | epoch_t min = shard->get_min_pg_epoch(); | |
7744 | if (osd_min == 0 || min < osd_min) { | |
7745 | osd_min = min; | |
7746 | } | |
7747 | } | |
9f95a23c | 7748 | epoch_t osdmap_epoch = get_osdmap_epoch(); |
11fdf7f2 | 7749 | if (osd_min > 0 && |
9f95a23c TL |
7750 | osdmap_epoch > max_lag && |
7751 | osdmap_epoch - max_lag > osd_min) { | |
7752 | epoch_t need = osdmap_epoch - max_lag; | |
11fdf7f2 TL |
7753 | dout(10) << __func__ << " waiting for pgs to catch up (need " << need |
7754 | << " max_lag " << max_lag << ")" << dendl; | |
7755 | for (auto shard : shards) { | |
7756 | epoch_t min = shard->get_min_pg_epoch(); | |
7757 | if (need > min) { | |
7758 | dout(10) << __func__ << " waiting for pgs to consume " << need | |
7759 | << " (shard " << shard->shard_id << " min " << min | |
7760 | << ", map cache is " << cct->_conf->osd_map_cache_size | |
7761 | << ", max_lag_factor " << m_osd_pg_epoch_max_lag_factor | |
7762 | << ")" << dendl; | |
7763 | unlock_guard unlock{osd_lock}; | |
7764 | shard->wait_min_pg_epoch(need); | |
7765 | } | |
7766 | } | |
7767 | } | |
7768 | } | |
7769 | ||
9f95a23c | 7770 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
11fdf7f2 TL |
7771 | map<epoch_t,OSDMapRef> added_maps; |
7772 | map<epoch_t,bufferlist> added_maps_bl; | |
7c673cae FG |
7773 | if (m->fsid != monc->get_fsid()) { |
7774 | dout(0) << "handle_osd_map fsid " << m->fsid << " != " | |
7775 | << monc->get_fsid() << dendl; | |
7776 | m->put(); | |
7777 | return; | |
7778 | } | |
7779 | if (is_initializing()) { | |
7780 | dout(0) << "ignoring osdmap until we have initialized" << dendl; | |
7781 | m->put(); | |
7782 | return; | |
7783 | } | |
7784 | ||
9f95a23c TL |
7785 | auto session = ceph::ref_cast<Session>(m->get_connection()->get_priv()); |
7786 | if (session && !(session->entity_name.is_mon() || | |
7c673cae FG |
7787 | session->entity_name.is_osd())) { |
7788 | //not enough perms! | |
7789 | dout(10) << "got osd map from Session " << session | |
7790 | << " which we can't take maps from (not a mon or osd)" << dendl; | |
7791 | m->put(); | |
7c673cae FG |
7792 | return; |
7793 | } | |
7c673cae FG |
7794 | |
7795 | // share with the objecter | |
7796 | if (!is_preboot()) | |
7797 | service.objecter->handle_osd_map(m); | |
7798 | ||
7799 | epoch_t first = m->get_first(); | |
7800 | epoch_t last = m->get_last(); | |
7801 | dout(3) << "handle_osd_map epochs [" << first << "," << last << "], i have " | |
7802 | << superblock.newest_map | |
7803 | << ", src has [" << m->oldest_map << "," << m->newest_map << "]" | |
7804 | << dendl; | |
7805 | ||
7806 | logger->inc(l_osd_map); | |
7807 | logger->inc(l_osd_mape, last - first + 1); | |
7808 | if (first <= superblock.newest_map) | |
7809 | logger->inc(l_osd_mape_dup, superblock.newest_map - first + 1); | |
7810 | if (service.max_oldest_map < m->oldest_map) { | |
7811 | service.max_oldest_map = m->oldest_map; | |
11fdf7f2 | 7812 | ceph_assert(service.max_oldest_map >= superblock.oldest_map); |
7c673cae FG |
7813 | } |
7814 | ||
7815 | // make sure there is something new, here, before we bother flushing | |
7816 | // the queues and such | |
7817 | if (last <= superblock.newest_map) { | |
7818 | dout(10) << " no new maps here, dropping" << dendl; | |
7819 | m->put(); | |
7820 | return; | |
7821 | } | |
7822 | ||
7823 | // missing some? | |
7824 | bool skip_maps = false; | |
7825 | if (first > superblock.newest_map + 1) { | |
7826 | dout(10) << "handle_osd_map message skips epochs " | |
7827 | << superblock.newest_map + 1 << ".." << (first-1) << dendl; | |
7828 | if (m->oldest_map <= superblock.newest_map + 1) { | |
7829 | osdmap_subscribe(superblock.newest_map + 1, false); | |
7830 | m->put(); | |
7831 | return; | |
7832 | } | |
7833 | // always try to get the full range of maps--as many as we can. this | |
7834 | // 1- is good to have | |
7835 | // 2- is at present the only way to ensure that we get a *full* map as | |
7836 | // the first map! | |
7837 | if (m->oldest_map < first) { | |
7838 | osdmap_subscribe(m->oldest_map - 1, true); | |
7839 | m->put(); | |
7840 | return; | |
7841 | } | |
7842 | skip_maps = true; | |
7843 | } | |
7844 | ||
7845 | ObjectStore::Transaction t; | |
7846 | uint64_t txn_size = 0; | |
7847 | ||
9f95a23c TL |
7848 | map<epoch_t,mempool::osdmap::map<int64_t,snap_interval_set_t>> purged_snaps; |
7849 | ||
7c673cae | 7850 | // store new maps: queue for disk and put in the osdmap cache |
11fdf7f2 | 7851 | epoch_t start = std::max(superblock.newest_map + 1, first); |
7c673cae FG |
7852 | for (epoch_t e = start; e <= last; e++) { |
7853 | if (txn_size >= t.get_num_bytes()) { | |
7854 | derr << __func__ << " transaction size overflowed" << dendl; | |
11fdf7f2 | 7855 | ceph_assert(txn_size < t.get_num_bytes()); |
7c673cae FG |
7856 | } |
7857 | txn_size = t.get_num_bytes(); | |
7858 | map<epoch_t,bufferlist>::iterator p; | |
7859 | p = m->maps.find(e); | |
7860 | if (p != m->maps.end()) { | |
7861 | dout(10) << "handle_osd_map got full map for epoch " << e << dendl; | |
7862 | OSDMap *o = new OSDMap; | |
7863 | bufferlist& bl = p->second; | |
7864 | ||
7865 | o->decode(bl); | |
7866 | ||
9f95a23c TL |
7867 | purged_snaps[e] = o->get_new_purged_snaps(); |
7868 | ||
7c673cae FG |
7869 | ghobject_t fulloid = get_osdmap_pobject_name(e); |
7870 | t.write(coll_t::meta(), fulloid, 0, bl.length(), bl); | |
11fdf7f2 TL |
7871 | added_maps[e] = add_map(o); |
7872 | added_maps_bl[e] = bl; | |
7c673cae FG |
7873 | got_full_map(e); |
7874 | continue; | |
7875 | } | |
7876 | ||
7877 | p = m->incremental_maps.find(e); | |
7878 | if (p != m->incremental_maps.end()) { | |
7879 | dout(10) << "handle_osd_map got inc map for epoch " << e << dendl; | |
7880 | bufferlist& bl = p->second; | |
7881 | ghobject_t oid = get_inc_osdmap_pobject_name(e); | |
7882 | t.write(coll_t::meta(), oid, 0, bl.length(), bl); | |
7c673cae FG |
7883 | |
7884 | OSDMap *o = new OSDMap; | |
7885 | if (e > 1) { | |
7886 | bufferlist obl; | |
7887 | bool got = get_map_bl(e - 1, obl); | |
11fdf7f2 TL |
7888 | if (!got) { |
7889 | auto p = added_maps_bl.find(e - 1); | |
7890 | ceph_assert(p != added_maps_bl.end()); | |
7891 | obl = p->second; | |
7892 | } | |
7c673cae FG |
7893 | o->decode(obl); |
7894 | } | |
7895 | ||
7896 | OSDMap::Incremental inc; | |
11fdf7f2 | 7897 | auto p = bl.cbegin(); |
7c673cae | 7898 | inc.decode(p); |
494da23a | 7899 | |
7c673cae | 7900 | if (o->apply_incremental(inc) < 0) { |
9f95a23c | 7901 | derr << "ERROR: bad fsid? i have " << get_osdmap()->get_fsid() << " and inc has " << inc.fsid << dendl; |
11fdf7f2 | 7902 | ceph_abort_msg("bad fsid"); |
7c673cae FG |
7903 | } |
7904 | ||
7905 | bufferlist fbl; | |
7906 | o->encode(fbl, inc.encode_features | CEPH_FEATURE_RESERVED); | |
7907 | ||
7908 | bool injected_failure = false; | |
7909 | if (cct->_conf->osd_inject_bad_map_crc_probability > 0 && | |
7910 | (rand() % 10000) < cct->_conf->osd_inject_bad_map_crc_probability*10000.0) { | |
7911 | derr << __func__ << " injecting map crc failure" << dendl; | |
7912 | injected_failure = true; | |
7913 | } | |
7914 | ||
7915 | if ((inc.have_crc && o->get_crc() != inc.full_crc) || injected_failure) { | |
7916 | dout(2) << "got incremental " << e | |
7917 | << " but failed to encode full with correct crc; requesting" | |
7918 | << dendl; | |
7919 | clog->warn() << "failed to encode map e" << e << " with expected crc"; | |
7920 | dout(20) << "my encoded map was:\n"; | |
7921 | fbl.hexdump(*_dout); | |
7922 | *_dout << dendl; | |
7923 | delete o; | |
7924 | request_full_map(e, last); | |
7925 | last = e - 1; | |
7926 | break; | |
7927 | } | |
7928 | got_full_map(e); | |
9f95a23c | 7929 | purged_snaps[e] = o->get_new_purged_snaps(); |
7c673cae FG |
7930 | |
7931 | ghobject_t fulloid = get_osdmap_pobject_name(e); | |
7932 | t.write(coll_t::meta(), fulloid, 0, fbl.length(), fbl); | |
11fdf7f2 TL |
7933 | added_maps[e] = add_map(o); |
7934 | added_maps_bl[e] = fbl; | |
7c673cae FG |
7935 | continue; |
7936 | } | |
7937 | ||
11fdf7f2 | 7938 | ceph_abort_msg("MOSDMap lied about what maps it had?"); |
7c673cae FG |
7939 | } |
7940 | ||
7941 | // even if this map isn't from a mon, we may have satisfied our subscription | |
7942 | monc->sub_got("osdmap", last); | |
7943 | ||
7944 | if (!m->maps.empty() && requested_full_first) { | |
7945 | dout(10) << __func__ << " still missing full maps " << requested_full_first | |
7946 | << ".." << requested_full_last << dendl; | |
7947 | rerequest_full_maps(); | |
7948 | } | |
7949 | ||
7c673cae FG |
7950 | if (superblock.oldest_map) { |
7951 | // make sure we at least keep pace with incoming maps | |
7952 | trim_maps(m->oldest_map, last - first + 1, skip_maps); | |
11fdf7f2 | 7953 | pg_num_history.prune(superblock.oldest_map); |
7c673cae FG |
7954 | } |
7955 | ||
7956 | if (!superblock.oldest_map || skip_maps) | |
7957 | superblock.oldest_map = first; | |
7958 | superblock.newest_map = last; | |
7959 | superblock.current_epoch = last; | |
7960 | ||
7961 | // note in the superblock that we were clean thru the prior epoch | |
7962 | epoch_t boot_epoch = service.get_boot_epoch(); | |
7963 | if (boot_epoch && boot_epoch >= superblock.mounted) { | |
7964 | superblock.mounted = boot_epoch; | |
7965 | superblock.clean_thru = last; | |
7966 | } | |
7967 | ||
11fdf7f2 TL |
7968 | // check for pg_num changes and deleted pools |
7969 | OSDMapRef lastmap; | |
7970 | for (auto& i : added_maps) { | |
7971 | if (!lastmap) { | |
7972 | if (!(lastmap = service.try_get_map(i.first - 1))) { | |
7973 | dout(10) << __func__ << " can't get previous map " << i.first - 1 | |
7974 | << " probably first start of this osd" << dendl; | |
7975 | continue; | |
7976 | } | |
7977 | } | |
7978 | ceph_assert(lastmap->get_epoch() + 1 == i.second->get_epoch()); | |
7979 | for (auto& j : lastmap->get_pools()) { | |
7980 | if (!i.second->have_pg_pool(j.first)) { | |
7981 | pg_num_history.log_pool_delete(i.first, j.first); | |
7982 | dout(10) << __func__ << " recording final pg_pool_t for pool " | |
7983 | << j.first << dendl; | |
7984 | // this information is needed by _make_pg() if have to restart before | |
7985 | // the pool is deleted and need to instantiate a new (zombie) PG[Pool]. | |
7986 | ghobject_t obj = make_final_pool_info_oid(j.first); | |
7987 | bufferlist bl; | |
7988 | encode(j.second, bl, CEPH_FEATURES_ALL); | |
7989 | string name = lastmap->get_pool_name(j.first); | |
7990 | encode(name, bl); | |
7991 | map<string,string> profile; | |
7992 | if (lastmap->get_pg_pool(j.first)->is_erasure()) { | |
7993 | profile = lastmap->get_erasure_code_profile( | |
7994 | lastmap->get_pg_pool(j.first)->erasure_code_profile); | |
7995 | } | |
7996 | encode(profile, bl); | |
7997 | t.write(coll_t::meta(), obj, 0, bl.length(), bl); | |
11fdf7f2 TL |
7998 | } else if (unsigned new_pg_num = i.second->get_pg_num(j.first); |
7999 | new_pg_num != j.second.get_pg_num()) { | |
8000 | dout(10) << __func__ << " recording pool " << j.first << " pg_num " | |
8001 | << j.second.get_pg_num() << " -> " << new_pg_num << dendl; | |
8002 | pg_num_history.log_pg_num_change(i.first, j.first, new_pg_num); | |
8003 | } | |
8004 | } | |
8005 | for (auto& j : i.second->get_pools()) { | |
8006 | if (!lastmap->have_pg_pool(j.first)) { | |
8007 | dout(10) << __func__ << " recording new pool " << j.first << " pg_num " | |
8008 | << j.second.get_pg_num() << dendl; | |
8009 | pg_num_history.log_pg_num_change(i.first, j.first, | |
8010 | j.second.get_pg_num()); | |
8011 | } | |
8012 | } | |
8013 | lastmap = i.second; | |
8014 | } | |
8015 | pg_num_history.epoch = last; | |
8016 | { | |
8017 | bufferlist bl; | |
8018 | ::encode(pg_num_history, bl); | |
8019 | t.write(coll_t::meta(), make_pg_num_history_oid(), 0, bl.length(), bl); | |
8020 | dout(20) << __func__ << " pg_num_history " << pg_num_history << dendl; | |
8021 | } | |
8022 | ||
9f95a23c TL |
8023 | // record new purged_snaps |
8024 | if (superblock.purged_snaps_last == start - 1) { | |
8025 | SnapMapper::record_purged_snaps(cct, store, service.meta_ch, | |
8026 | make_purged_snaps_oid(), &t, | |
8027 | purged_snaps); | |
8028 | superblock.purged_snaps_last = last; | |
8029 | } else { | |
8030 | dout(10) << __func__ << " superblock purged_snaps_last is " | |
8031 | << superblock.purged_snaps_last | |
8032 | << ", not recording new purged_snaps" << dendl; | |
8033 | } | |
8034 | ||
7c673cae FG |
8035 | // superblock and commit |
8036 | write_superblock(t); | |
11fdf7f2 | 8037 | t.register_on_commit(new C_OnMapCommit(this, start, last, m)); |
7c673cae | 8038 | store->queue_transaction( |
11fdf7f2 TL |
8039 | service.meta_ch, |
8040 | std::move(t)); | |
7c673cae FG |
8041 | service.publish_superblock(superblock); |
8042 | } | |
8043 | ||
8044 | void OSD::_committed_osd_maps(epoch_t first, epoch_t last, MOSDMap *m) | |
8045 | { | |
8046 | dout(10) << __func__ << " " << first << ".." << last << dendl; | |
8047 | if (is_stopping()) { | |
8048 | dout(10) << __func__ << " bailing, we are shutting down" << dendl; | |
8049 | return; | |
8050 | } | |
11fdf7f2 | 8051 | std::lock_guard l(osd_lock); |
31f18b77 FG |
8052 | if (is_stopping()) { |
8053 | dout(10) << __func__ << " bailing, we are shutting down" << dendl; | |
8054 | return; | |
8055 | } | |
9f95a23c | 8056 | map_lock.lock(); |
7c673cae FG |
8057 | |
8058 | bool do_shutdown = false; | |
8059 | bool do_restart = false; | |
8060 | bool network_error = false; | |
9f95a23c | 8061 | OSDMapRef osdmap; |
7c673cae FG |
8062 | |
8063 | // advance through the new maps | |
8064 | for (epoch_t cur = first; cur <= last; cur++) { | |
8065 | dout(10) << " advance to epoch " << cur | |
8066 | << " (<= last " << last | |
8067 | << " <= newest_map " << superblock.newest_map | |
8068 | << ")" << dendl; | |
8069 | ||
8070 | OSDMapRef newmap = get_map(cur); | |
11fdf7f2 | 8071 | ceph_assert(newmap); // we just cached it above! |
7c673cae FG |
8072 | |
8073 | // start blacklisting messages sent to peers that go down. | |
8074 | service.pre_publish_map(newmap); | |
8075 | ||
8076 | // kill connections to newly down osds | |
8077 | bool waited_for_reservations = false; | |
8078 | set<int> old; | |
9f95a23c | 8079 | osdmap = get_osdmap(); |
7c673cae FG |
8080 | osdmap->get_all_osds(old); |
8081 | for (set<int>::iterator p = old.begin(); p != old.end(); ++p) { | |
8082 | if (*p != whoami && | |
8083 | osdmap->is_up(*p) && // in old map | |
8084 | newmap->is_down(*p)) { // but not the new one | |
8085 | if (!waited_for_reservations) { | |
8086 | service.await_reserved_maps(); | |
8087 | waited_for_reservations = true; | |
8088 | } | |
8089 | note_down_osd(*p); | |
8090 | } else if (*p != whoami && | |
8091 | osdmap->is_down(*p) && | |
8092 | newmap->is_up(*p)) { | |
8093 | note_up_osd(*p); | |
8094 | } | |
8095 | } | |
8096 | ||
81eedcae | 8097 | if (osdmap->is_noup(whoami) != newmap->is_noup(whoami)) { |
7c673cae FG |
8098 | dout(10) << __func__ << " NOUP flag changed in " << newmap->get_epoch() |
8099 | << dendl; | |
8100 | if (is_booting()) { | |
8101 | // this captures the case where we sent the boot message while | |
8102 | // NOUP was being set on the mon and our boot request was | |
8103 | // dropped, and then later it is cleared. it imperfectly | |
8104 | // handles the case where our original boot message was not | |
8105 | // dropped and we restart even though we might have booted, but | |
8106 | // that is harmless (boot will just take slightly longer). | |
8107 | do_restart = true; | |
8108 | } | |
8109 | } | |
8110 | ||
9f95a23c TL |
8111 | osdmap = std::move(newmap); |
8112 | set_osdmap(osdmap); | |
7c673cae FG |
8113 | epoch_t up_epoch; |
8114 | epoch_t boot_epoch; | |
8115 | service.retrieve_epochs(&boot_epoch, &up_epoch, NULL); | |
8116 | if (!up_epoch && | |
8117 | osdmap->is_up(whoami) && | |
11fdf7f2 | 8118 | osdmap->get_addrs(whoami) == client_messenger->get_myaddrs()) { |
7c673cae FG |
8119 | up_epoch = osdmap->get_epoch(); |
8120 | dout(10) << "up_epoch is " << up_epoch << dendl; | |
8121 | if (!boot_epoch) { | |
8122 | boot_epoch = osdmap->get_epoch(); | |
8123 | dout(10) << "boot_epoch is " << boot_epoch << dendl; | |
8124 | } | |
8125 | service.set_epochs(&boot_epoch, &up_epoch, NULL); | |
8126 | } | |
8127 | } | |
8128 | ||
7c673cae FG |
8129 | epoch_t _bind_epoch = service.get_bind_epoch(); |
8130 | if (osdmap->is_up(whoami) && | |
11fdf7f2 TL |
8131 | osdmap->get_addrs(whoami).legacy_equals( |
8132 | client_messenger->get_myaddrs()) && | |
7c673cae FG |
8133 | _bind_epoch < osdmap->get_up_from(whoami)) { |
8134 | ||
8135 | if (is_booting()) { | |
8136 | dout(1) << "state: booting -> active" << dendl; | |
8137 | set_state(STATE_ACTIVE); | |
11fdf7f2 | 8138 | do_restart = false; |
7c673cae FG |
8139 | |
8140 | // set incarnation so that osd_reqid_t's we generate for our | |
8141 | // objecter requests are unique across restarts. | |
8142 | service.objecter->set_client_incarnation(osdmap->get_epoch()); | |
11fdf7f2 | 8143 | cancel_pending_failures(); |
7c673cae FG |
8144 | } |
8145 | } | |
8146 | ||
8147 | if (osdmap->get_epoch() > 0 && | |
8148 | is_active()) { | |
8149 | if (!osdmap->exists(whoami)) { | |
9f95a23c | 8150 | derr << "map says i do not exist. shutting down." << dendl; |
7c673cae FG |
8151 | do_shutdown = true; // don't call shutdown() while we have |
8152 | // everything paused | |
9f95a23c TL |
8153 | } else if (osdmap->is_stop(whoami)) { |
8154 | derr << "map says i am stopped by admin. shutting down." << dendl; | |
8155 | do_shutdown = true; | |
7c673cae | 8156 | } else if (!osdmap->is_up(whoami) || |
11fdf7f2 TL |
8157 | !osdmap->get_addrs(whoami).legacy_equals( |
8158 | client_messenger->get_myaddrs()) || | |
8159 | !osdmap->get_cluster_addrs(whoami).legacy_equals( | |
8160 | cluster_messenger->get_myaddrs()) || | |
8161 | !osdmap->get_hb_back_addrs(whoami).legacy_equals( | |
8162 | hb_back_server_messenger->get_myaddrs()) || | |
8163 | !osdmap->get_hb_front_addrs(whoami).legacy_equals( | |
8164 | hb_front_server_messenger->get_myaddrs())) { | |
7c673cae FG |
8165 | if (!osdmap->is_up(whoami)) { |
8166 | if (service.is_preparing_to_stop() || service.is_stopping()) { | |
8167 | service.got_stop_ack(); | |
8168 | } else { | |
c07f9fc5 FG |
8169 | clog->warn() << "Monitor daemon marked osd." << whoami << " down, " |
8170 | "but it is still running"; | |
8171 | clog->debug() << "map e" << osdmap->get_epoch() | |
8172 | << " wrongly marked me down at e" | |
8173 | << osdmap->get_down_at(whoami); | |
7c673cae | 8174 | } |
9f95a23c TL |
8175 | if (monc->monmap.min_mon_release >= ceph_release_t::octopus) { |
8176 | // note that this is best-effort... | |
8177 | monc->send_mon_message( | |
8178 | new MOSDMarkMeDead( | |
8179 | monc->get_fsid(), | |
8180 | whoami, | |
8181 | osdmap->get_epoch())); | |
8182 | } | |
11fdf7f2 TL |
8183 | } else if (!osdmap->get_addrs(whoami).legacy_equals( |
8184 | client_messenger->get_myaddrs())) { | |
7c673cae | 8185 | clog->error() << "map e" << osdmap->get_epoch() |
11fdf7f2 TL |
8186 | << " had wrong client addr (" << osdmap->get_addrs(whoami) |
8187 | << " != my " << client_messenger->get_myaddrs() << ")"; | |
8188 | } else if (!osdmap->get_cluster_addrs(whoami).legacy_equals( | |
8189 | cluster_messenger->get_myaddrs())) { | |
7c673cae FG |
8190 | clog->error() << "map e" << osdmap->get_epoch() |
8191 | << " had wrong cluster addr (" | |
11fdf7f2 TL |
8192 | << osdmap->get_cluster_addrs(whoami) |
8193 | << " != my " << cluster_messenger->get_myaddrs() << ")"; | |
8194 | } else if (!osdmap->get_hb_back_addrs(whoami).legacy_equals( | |
8195 | hb_back_server_messenger->get_myaddrs())) { | |
7c673cae | 8196 | clog->error() << "map e" << osdmap->get_epoch() |
c07f9fc5 | 8197 | << " had wrong heartbeat back addr (" |
11fdf7f2 TL |
8198 | << osdmap->get_hb_back_addrs(whoami) |
8199 | << " != my " << hb_back_server_messenger->get_myaddrs() | |
7c673cae | 8200 | << ")"; |
11fdf7f2 TL |
8201 | } else if (!osdmap->get_hb_front_addrs(whoami).legacy_equals( |
8202 | hb_front_server_messenger->get_myaddrs())) { | |
7c673cae | 8203 | clog->error() << "map e" << osdmap->get_epoch() |
c07f9fc5 | 8204 | << " had wrong heartbeat front addr (" |
11fdf7f2 TL |
8205 | << osdmap->get_hb_front_addrs(whoami) |
8206 | << " != my " << hb_front_server_messenger->get_myaddrs() | |
7c673cae FG |
8207 | << ")"; |
8208 | } | |
8209 | ||
8210 | if (!service.is_stopping()) { | |
8211 | epoch_t up_epoch = 0; | |
8212 | epoch_t bind_epoch = osdmap->get_epoch(); | |
8213 | service.set_epochs(NULL,&up_epoch, &bind_epoch); | |
8214 | do_restart = true; | |
8215 | ||
8216 | //add markdown log | |
8217 | utime_t now = ceph_clock_now(); | |
8218 | utime_t grace = utime_t(cct->_conf->osd_max_markdown_period, 0); | |
8219 | osd_markdown_log.push_back(now); | |
7c673cae | 8220 | if ((int)osd_markdown_log.size() > cct->_conf->osd_max_markdown_count) { |
9f95a23c TL |
8221 | derr << __func__ << " marked down " |
8222 | << osd_markdown_log.size() | |
8223 | << " > osd_max_markdown_count " | |
8224 | << cct->_conf->osd_max_markdown_count | |
8225 | << " in last " << grace << " seconds, shutting down" | |
8226 | << dendl; | |
7c673cae FG |
8227 | do_restart = false; |
8228 | do_shutdown = true; | |
8229 | } | |
8230 | ||
8231 | start_waiting_for_healthy(); | |
8232 | ||
8233 | set<int> avoid_ports; | |
8234 | #if defined(__FreeBSD__) | |
8235 | // prevent FreeBSD from grabbing the client_messenger port during | |
8236 | // rebinding. In which case a cluster_meesneger will connect also | |
8237 | // to the same port | |
11fdf7f2 | 8238 | client_messenger->get_myaddrs().get_ports(&avoid_ports); |
7c673cae | 8239 | #endif |
11fdf7f2 | 8240 | cluster_messenger->get_myaddrs().get_ports(&avoid_ports); |
7c673cae FG |
8241 | |
8242 | int r = cluster_messenger->rebind(avoid_ports); | |
8243 | if (r != 0) { | |
8244 | do_shutdown = true; // FIXME: do_restart? | |
8245 | network_error = true; | |
9f95a23c TL |
8246 | derr << __func__ << " marked down:" |
8247 | << " rebind cluster_messenger failed" << dendl; | |
7c673cae FG |
8248 | } |
8249 | ||
9f95a23c TL |
8250 | hb_back_server_messenger->mark_down_all(); |
8251 | hb_front_server_messenger->mark_down_all(); | |
7c673cae FG |
8252 | hb_front_client_messenger->mark_down_all(); |
8253 | hb_back_client_messenger->mark_down_all(); | |
8254 | ||
494da23a | 8255 | reset_heartbeat_peers(true); |
7c673cae FG |
8256 | } |
8257 | } | |
8258 | } | |
8259 | ||
9f95a23c | 8260 | map_lock.unlock(); |
7c673cae | 8261 | |
11fdf7f2 | 8262 | check_osdmap_features(); |
7c673cae FG |
8263 | |
8264 | // yay! | |
8265 | consume_map(); | |
8266 | ||
8267 | if (is_active() || is_waiting_for_healthy()) | |
8268 | maybe_update_heartbeat_peers(); | |
8269 | ||
11fdf7f2 | 8270 | if (is_active()) { |
7c673cae FG |
8271 | activate_map(); |
8272 | } | |
8273 | ||
31f18b77 | 8274 | if (do_shutdown) { |
7c673cae | 8275 | if (network_error) { |
11fdf7f2 | 8276 | cancel_pending_failures(); |
7c673cae FG |
8277 | } |
8278 | // trigger shutdown in a different thread | |
8279 | dout(0) << __func__ << " shutdown OSD via async signal" << dendl; | |
8280 | queue_async_signal(SIGINT); | |
8281 | } | |
31f18b77 FG |
8282 | else if (m->newest_map && m->newest_map > last) { |
8283 | dout(10) << " msg say newest map is " << m->newest_map | |
8284 | << ", requesting more" << dendl; | |
8285 | osdmap_subscribe(osdmap->get_epoch()+1, false); | |
8286 | } | |
7c673cae FG |
8287 | else if (is_preboot()) { |
8288 | if (m->get_source().is_mon()) | |
8289 | _preboot(m->oldest_map, m->newest_map); | |
8290 | else | |
8291 | start_boot(); | |
8292 | } | |
8293 | else if (do_restart) | |
8294 | start_boot(); | |
8295 | ||
8296 | } | |
8297 | ||
11fdf7f2 | 8298 | void OSD::check_osdmap_features() |
7c673cae FG |
8299 | { |
8300 | // adjust required feature bits? | |
8301 | ||
8302 | // we have to be a bit careful here, because we are accessing the | |
8303 | // Policy structures without taking any lock. in particular, only | |
8304 | // modify integer values that can safely be read by a racing CPU. | |
8305 | // since we are only accessing existing Policy structures a their | |
8306 | // current memory location, and setting or clearing bits in integer | |
8307 | // fields, and we are the only writer, this is not a problem. | |
8308 | ||
9f95a23c | 8309 | const auto osdmap = get_osdmap(); |
7c673cae FG |
8310 | { |
8311 | Messenger::Policy p = client_messenger->get_default_policy(); | |
8312 | uint64_t mask; | |
8313 | uint64_t features = osdmap->get_features(entity_name_t::TYPE_CLIENT, &mask); | |
8314 | if ((p.features_required & mask) != features) { | |
8315 | dout(0) << "crush map has features " << features | |
8316 | << ", adjusting msgr requires for clients" << dendl; | |
8317 | p.features_required = (p.features_required & ~mask) | features; | |
8318 | client_messenger->set_default_policy(p); | |
8319 | } | |
8320 | } | |
8321 | { | |
8322 | Messenger::Policy p = client_messenger->get_policy(entity_name_t::TYPE_MON); | |
8323 | uint64_t mask; | |
8324 | uint64_t features = osdmap->get_features(entity_name_t::TYPE_MON, &mask); | |
8325 | if ((p.features_required & mask) != features) { | |
8326 | dout(0) << "crush map has features " << features | |
8327 | << " was " << p.features_required | |
8328 | << ", adjusting msgr requires for mons" << dendl; | |
8329 | p.features_required = (p.features_required & ~mask) | features; | |
8330 | client_messenger->set_policy(entity_name_t::TYPE_MON, p); | |
8331 | } | |
8332 | } | |
8333 | { | |
8334 | Messenger::Policy p = cluster_messenger->get_policy(entity_name_t::TYPE_OSD); | |
8335 | uint64_t mask; | |
8336 | uint64_t features = osdmap->get_features(entity_name_t::TYPE_OSD, &mask); | |
8337 | ||
8338 | if ((p.features_required & mask) != features) { | |
8339 | dout(0) << "crush map has features " << features | |
8340 | << ", adjusting msgr requires for osds" << dendl; | |
8341 | p.features_required = (p.features_required & ~mask) | features; | |
8342 | cluster_messenger->set_policy(entity_name_t::TYPE_OSD, p); | |
8343 | } | |
8344 | ||
11fdf7f2 | 8345 | if (!superblock.compat_features.incompat.contains(CEPH_OSD_FEATURE_INCOMPAT_SHARDS)) { |
7c673cae FG |
8346 | dout(0) << __func__ << " enabling on-disk ERASURE CODES compat feature" << dendl; |
8347 | superblock.compat_features.incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_SHARDS); | |
8348 | ObjectStore::Transaction t; | |
8349 | write_superblock(t); | |
11fdf7f2 TL |
8350 | int err = store->queue_transaction(service.meta_ch, std::move(t), NULL); |
8351 | ceph_assert(err == 0); | |
7c673cae FG |
8352 | } |
8353 | } | |
11fdf7f2 | 8354 | |
9f95a23c TL |
8355 | if (osdmap->require_osd_release < ceph_release_t::nautilus) { |
8356 | hb_front_server_messenger->set_require_authorizer(false); | |
8357 | hb_back_server_messenger->set_require_authorizer(false); | |
8358 | } else { | |
8359 | hb_front_server_messenger->set_require_authorizer(true); | |
8360 | hb_back_server_messenger->set_require_authorizer(true); | |
11fdf7f2 TL |
8361 | } |
8362 | ||
8363 | if (osdmap->require_osd_release != last_require_osd_release) { | |
8364 | dout(1) << __func__ << " require_osd_release " << last_require_osd_release | |
8365 | << " -> " << to_string(osdmap->require_osd_release) << dendl; | |
8366 | store->write_meta("require_osd_release", | |
8367 | stringify((int)osdmap->require_osd_release)); | |
8368 | last_require_osd_release = osdmap->require_osd_release; | |
8369 | } | |
7c673cae FG |
8370 | } |
8371 | ||
11fdf7f2 TL |
8372 | struct C_FinishSplits : public Context { |
8373 | OSD *osd; | |
8374 | set<PGRef> pgs; | |
8375 | C_FinishSplits(OSD *osd, const set<PGRef> &in) | |
8376 | : osd(osd), pgs(in) {} | |
8377 | void finish(int r) override { | |
8378 | osd->_finish_splits(pgs); | |
8379 | } | |
8380 | }; | |
8381 | ||
8382 | void OSD::_finish_splits(set<PGRef>& pgs) | |
7c673cae | 8383 | { |
11fdf7f2 TL |
8384 | dout(10) << __func__ << " " << pgs << dendl; |
8385 | if (is_stopping()) | |
8386 | return; | |
11fdf7f2 TL |
8387 | for (set<PGRef>::iterator i = pgs.begin(); |
8388 | i != pgs.end(); | |
8389 | ++i) { | |
8390 | PG *pg = i->get(); | |
7c673cae | 8391 | |
9f95a23c | 8392 | PeeringCtx rctx = create_context(); |
11fdf7f2 TL |
8393 | pg->lock(); |
8394 | dout(10) << __func__ << " " << *pg << dendl; | |
8395 | epoch_t e = pg->get_osdmap_epoch(); | |
9f95a23c | 8396 | pg->handle_initialize(rctx); |
11fdf7f2 | 8397 | pg->queue_null(e, e); |
9f95a23c | 8398 | dispatch_context(rctx, pg, service.get_osdmap()); |
11fdf7f2 | 8399 | pg->unlock(); |
7c673cae | 8400 | |
11fdf7f2 TL |
8401 | unsigned shard_index = pg->pg_id.hash_to_shard(num_shards); |
8402 | shards[shard_index]->register_and_wake_split_child(pg); | |
7c673cae | 8403 | } |
11fdf7f2 TL |
8404 | }; |
8405 | ||
8406 | bool OSD::add_merge_waiter(OSDMapRef nextmap, spg_t target, PGRef src, | |
8407 | unsigned need) | |
8408 | { | |
8409 | std::lock_guard l(merge_lock); | |
8410 | auto& p = merge_waiters[nextmap->get_epoch()][target]; | |
8411 | p[src->pg_id] = src; | |
8412 | dout(10) << __func__ << " added merge_waiter " << src->pg_id | |
8413 | << " for " << target << ", have " << p.size() << "/" << need | |
8414 | << dendl; | |
8415 | return p.size() == need; | |
8416 | } | |
8417 | ||
8418 | bool OSD::advance_pg( | |
8419 | epoch_t osd_epoch, | |
8420 | PG *pg, | |
8421 | ThreadPool::TPHandle &handle, | |
9f95a23c | 8422 | PeeringCtx &rctx) |
11fdf7f2 TL |
8423 | { |
8424 | if (osd_epoch <= pg->get_osdmap_epoch()) { | |
8425 | return true; | |
8426 | } | |
8427 | ceph_assert(pg->is_locked()); | |
8428 | OSDMapRef lastmap = pg->get_osdmap(); | |
8429 | ceph_assert(lastmap->get_epoch() < osd_epoch); | |
8430 | set<PGRef> new_pgs; // any split children | |
8431 | bool ret = true; | |
8432 | ||
8433 | unsigned old_pg_num = lastmap->have_pg_pool(pg->pg_id.pool()) ? | |
8434 | lastmap->get_pg_num(pg->pg_id.pool()) : 0; | |
8435 | for (epoch_t next_epoch = pg->get_osdmap_epoch() + 1; | |
8436 | next_epoch <= osd_epoch; | |
7c673cae FG |
8437 | ++next_epoch) { |
8438 | OSDMapRef nextmap = service.try_get_map(next_epoch); | |
8439 | if (!nextmap) { | |
8440 | dout(20) << __func__ << " missing map " << next_epoch << dendl; | |
7c673cae FG |
8441 | continue; |
8442 | } | |
8443 | ||
11fdf7f2 TL |
8444 | unsigned new_pg_num = |
8445 | (old_pg_num && nextmap->have_pg_pool(pg->pg_id.pool())) ? | |
8446 | nextmap->get_pg_num(pg->pg_id.pool()) : 0; | |
8447 | if (old_pg_num && new_pg_num && old_pg_num != new_pg_num) { | |
8448 | // check for merge | |
8449 | if (nextmap->have_pg_pool(pg->pg_id.pool())) { | |
8450 | spg_t parent; | |
8451 | if (pg->pg_id.is_merge_source( | |
8452 | old_pg_num, | |
8453 | new_pg_num, | |
8454 | &parent)) { | |
8455 | // we are merge source | |
8456 | PGRef spg = pg; // carry a ref | |
8457 | dout(1) << __func__ << " " << pg->pg_id | |
8458 | << " is merge source, target is " << parent | |
8459 | << dendl; | |
8460 | pg->write_if_dirty(rctx); | |
9f95a23c TL |
8461 | if (!new_pgs.empty()) { |
8462 | rctx.transaction.register_on_applied(new C_FinishSplits(this, | |
8463 | new_pgs)); | |
8464 | new_pgs.clear(); | |
8465 | } | |
8466 | dispatch_context(rctx, pg, pg->get_osdmap(), &handle); | |
11fdf7f2 | 8467 | pg->ch->flush(); |
eafe8130 TL |
8468 | // release backoffs explicitly, since the on_shutdown path |
8469 | // aggressively tears down backoff state. | |
8470 | if (pg->is_primary()) { | |
8471 | pg->release_pg_backoffs(); | |
8472 | } | |
11fdf7f2 TL |
8473 | pg->on_shutdown(); |
8474 | OSDShard *sdata = pg->osd_shard; | |
8475 | { | |
8476 | std::lock_guard l(sdata->shard_lock); | |
8477 | if (pg->pg_slot) { | |
8478 | sdata->_detach_pg(pg->pg_slot); | |
8479 | // update pg count now since we might not get an osdmap | |
8480 | // any time soon. | |
8481 | if (pg->is_primary()) | |
8482 | logger->dec(l_osd_pg_primary); | |
9f95a23c TL |
8483 | else if (pg->is_nonprimary()) |
8484 | logger->dec(l_osd_pg_replica); // misnomer | |
11fdf7f2 TL |
8485 | else |
8486 | logger->dec(l_osd_pg_stray); | |
8487 | } | |
8488 | } | |
8489 | pg->unlock(); | |
8490 | ||
8491 | set<spg_t> children; | |
8492 | parent.is_split(new_pg_num, old_pg_num, &children); | |
8493 | if (add_merge_waiter(nextmap, parent, pg, children.size())) { | |
8494 | enqueue_peering_evt( | |
8495 | parent, | |
8496 | PGPeeringEventRef( | |
8497 | std::make_shared<PGPeeringEvent>( | |
8498 | nextmap->get_epoch(), | |
8499 | nextmap->get_epoch(), | |
8500 | NullEvt()))); | |
8501 | } | |
8502 | ret = false; | |
8503 | goto out; | |
8504 | } else if (pg->pg_id.is_merge_target(old_pg_num, new_pg_num)) { | |
8505 | // we are merge target | |
8506 | set<spg_t> children; | |
8507 | pg->pg_id.is_split(new_pg_num, old_pg_num, &children); | |
8508 | dout(20) << __func__ << " " << pg->pg_id | |
8509 | << " is merge target, sources are " << children | |
8510 | << dendl; | |
8511 | map<spg_t,PGRef> sources; | |
8512 | { | |
8513 | std::lock_guard l(merge_lock); | |
8514 | auto& s = merge_waiters[nextmap->get_epoch()][pg->pg_id]; | |
8515 | unsigned need = children.size(); | |
8516 | dout(20) << __func__ << " have " << s.size() << "/" | |
8517 | << need << dendl; | |
8518 | if (s.size() == need) { | |
8519 | sources.swap(s); | |
8520 | merge_waiters[nextmap->get_epoch()].erase(pg->pg_id); | |
8521 | if (merge_waiters[nextmap->get_epoch()].empty()) { | |
8522 | merge_waiters.erase(nextmap->get_epoch()); | |
8523 | } | |
8524 | } | |
8525 | } | |
8526 | if (!sources.empty()) { | |
8527 | unsigned new_pg_num = nextmap->get_pg_num(pg->pg_id.pool()); | |
8528 | unsigned split_bits = pg->pg_id.get_split_bits(new_pg_num); | |
8529 | dout(1) << __func__ << " merging " << pg->pg_id << dendl; | |
8530 | pg->merge_from( | |
8531 | sources, rctx, split_bits, | |
8532 | nextmap->get_pg_pool( | |
8533 | pg->pg_id.pool())->last_pg_merge_meta); | |
8534 | pg->pg_slot->waiting_for_merge_epoch = 0; | |
8535 | } else { | |
8536 | dout(20) << __func__ << " not ready to merge yet" << dendl; | |
8537 | pg->write_if_dirty(rctx); | |
9f95a23c TL |
8538 | if (!new_pgs.empty()) { |
8539 | rctx.transaction.register_on_applied(new C_FinishSplits(this, | |
8540 | new_pgs)); | |
8541 | new_pgs.clear(); | |
8542 | } | |
8543 | dispatch_context(rctx, pg, pg->get_osdmap(), &handle); | |
11fdf7f2 TL |
8544 | pg->unlock(); |
8545 | // kick source(s) to get them ready | |
8546 | for (auto& i : children) { | |
8547 | dout(20) << __func__ << " kicking source " << i << dendl; | |
8548 | enqueue_peering_evt( | |
8549 | i, | |
8550 | PGPeeringEventRef( | |
8551 | std::make_shared<PGPeeringEvent>( | |
8552 | nextmap->get_epoch(), | |
8553 | nextmap->get_epoch(), | |
8554 | NullEvt()))); | |
8555 | } | |
8556 | ret = false; | |
8557 | goto out; | |
8558 | } | |
8559 | } | |
8560 | } | |
8561 | } | |
8562 | ||
7c673cae FG |
8563 | vector<int> newup, newacting; |
8564 | int up_primary, acting_primary; | |
8565 | nextmap->pg_to_up_acting_osds( | |
11fdf7f2 | 8566 | pg->pg_id.pgid, |
7c673cae FG |
8567 | &newup, &up_primary, |
8568 | &newacting, &acting_primary); | |
8569 | pg->handle_advance_map( | |
8570 | nextmap, lastmap, newup, up_primary, | |
8571 | newacting, acting_primary, rctx); | |
8572 | ||
494da23a TL |
8573 | auto oldpool = lastmap->get_pools().find(pg->pg_id.pool()); |
8574 | auto newpool = nextmap->get_pools().find(pg->pg_id.pool()); | |
8575 | if (oldpool != lastmap->get_pools().end() | |
8576 | && newpool != nextmap->get_pools().end()) { | |
8577 | dout(20) << __func__ | |
8578 | << " new pool opts " << newpool->second.opts | |
8579 | << " old pool opts " << oldpool->second.opts | |
8580 | << dendl; | |
8581 | ||
8582 | double old_min_interval = 0, new_min_interval = 0; | |
8583 | oldpool->second.opts.get(pool_opts_t::SCRUB_MIN_INTERVAL, &old_min_interval); | |
8584 | newpool->second.opts.get(pool_opts_t::SCRUB_MIN_INTERVAL, &new_min_interval); | |
8585 | ||
8586 | double old_max_interval = 0, new_max_interval = 0; | |
8587 | oldpool->second.opts.get(pool_opts_t::SCRUB_MAX_INTERVAL, &old_max_interval); | |
8588 | newpool->second.opts.get(pool_opts_t::SCRUB_MAX_INTERVAL, &new_max_interval); | |
8589 | ||
8590 | // Assume if an interval is change from set to unset or vice versa the actual config | |
8591 | // is different. Keep it simple even if it is possible to call resched_all_scrub() | |
8592 | // unnecessarily. | |
8593 | if (old_min_interval != new_min_interval || old_max_interval != new_max_interval) { | |
8594 | pg->on_info_history_change(); | |
8595 | } | |
8596 | } | |
8597 | ||
11fdf7f2 TL |
8598 | if (new_pg_num && old_pg_num != new_pg_num) { |
8599 | // check for split | |
8600 | set<spg_t> children; | |
8601 | if (pg->pg_id.is_split( | |
8602 | old_pg_num, | |
8603 | new_pg_num, | |
8604 | &children)) { | |
8605 | split_pgs( | |
8606 | pg, children, &new_pgs, lastmap, nextmap, | |
8607 | rctx); | |
8608 | } | |
7c673cae FG |
8609 | } |
8610 | ||
8611 | lastmap = nextmap; | |
11fdf7f2 | 8612 | old_pg_num = new_pg_num; |
7c673cae FG |
8613 | handle.reset_tp_timeout(); |
8614 | } | |
7c673cae | 8615 | pg->handle_activate_map(rctx); |
11fdf7f2 TL |
8616 | |
8617 | ret = true; | |
8618 | out: | |
8619 | if (!new_pgs.empty()) { | |
9f95a23c | 8620 | rctx.transaction.register_on_applied(new C_FinishSplits(this, new_pgs)); |
7c673cae | 8621 | } |
11fdf7f2 | 8622 | return ret; |
7c673cae FG |
8623 | } |
8624 | ||
8625 | void OSD::consume_map() | |
8626 | { | |
9f95a23c TL |
8627 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
8628 | auto osdmap = get_osdmap(); | |
7c673cae FG |
8629 | dout(7) << "consume_map version " << osdmap->get_epoch() << dendl; |
8630 | ||
3efd9988 FG |
8631 | /** make sure the cluster is speaking in SORTBITWISE, because we don't |
8632 | * speak the older sorting version any more. Be careful not to force | |
8633 | * a shutdown if we are merely processing old maps, though. | |
8634 | */ | |
8635 | if (!osdmap->test_flag(CEPH_OSDMAP_SORTBITWISE) && is_active()) { | |
8636 | derr << __func__ << " SORTBITWISE flag is not set" << dendl; | |
8637 | ceph_abort(); | |
8638 | } | |
8639 | ||
11fdf7f2 TL |
8640 | service.pre_publish_map(osdmap); |
8641 | service.await_reserved_maps(); | |
8642 | service.publish_map(osdmap); | |
7c673cae | 8643 | |
11fdf7f2 TL |
8644 | // prime splits and merges |
8645 | set<pair<spg_t,epoch_t>> newly_split; // splits, and when | |
8646 | set<pair<spg_t,epoch_t>> merge_pgs; // merge participants, and when | |
8647 | for (auto& shard : shards) { | |
8648 | shard->identify_splits_and_merges(osdmap, &newly_split, &merge_pgs); | |
8649 | } | |
8650 | if (!newly_split.empty()) { | |
8651 | for (auto& shard : shards) { | |
8652 | shard->prime_splits(osdmap, &newly_split); | |
8653 | } | |
8654 | ceph_assert(newly_split.empty()); | |
8655 | } | |
7c673cae | 8656 | |
11fdf7f2 TL |
8657 | // prune sent_ready_to_merge |
8658 | service.prune_sent_ready_to_merge(osdmap); | |
7c673cae | 8659 | |
11fdf7f2 TL |
8660 | // FIXME, maybe: We could race against an incoming peering message |
8661 | // that instantiates a merge PG after identify_merges() below and | |
8662 | // never set up its peer to complete the merge. An OSD restart | |
8663 | // would clear it up. This is a hard race to resolve, | |
8664 | // extraordinarily rare (we only merge PGs that are stable and | |
8665 | // clean, so it'd have to be an imported PG to an OSD with a | |
8666 | // slightly stale OSDMap...), so I'm ignoring it for now. We plan to | |
8667 | // replace all of this with a seastar-based code soon anyway. | |
8668 | if (!merge_pgs.empty()) { | |
8669 | // mark the pgs we already have, or create new and empty merge | |
8670 | // participants for those we are missing. do this all under the | |
8671 | // shard lock so we don't have to worry about racing pg creates | |
8672 | // via _process. | |
8673 | for (auto& shard : shards) { | |
8674 | shard->prime_merges(osdmap, &merge_pgs); | |
7c673cae | 8675 | } |
11fdf7f2 TL |
8676 | ceph_assert(merge_pgs.empty()); |
8677 | } | |
8678 | ||
8679 | service.prune_pg_created(); | |
8680 | ||
8681 | unsigned pushes_to_free = 0; | |
8682 | for (auto& shard : shards) { | |
8683 | shard->consume_map(osdmap, &pushes_to_free); | |
8684 | } | |
8685 | ||
8686 | vector<spg_t> pgids; | |
8687 | _get_pgids(&pgids); | |
8688 | ||
8689 | // count (FIXME, probably during seastar rewrite) | |
8690 | int num_pg_primary = 0, num_pg_replica = 0, num_pg_stray = 0; | |
8691 | vector<PGRef> pgs; | |
8692 | _get_pgs(&pgs); | |
8693 | for (auto& pg : pgs) { | |
8694 | // FIXME (probably during seastar rewrite): this is lockless and | |
8695 | // racy, but we don't want to take pg lock here. | |
8696 | if (pg->is_primary()) | |
8697 | num_pg_primary++; | |
9f95a23c TL |
8698 | else if (pg->is_nonprimary()) |
8699 | num_pg_replica++; // misnomer | |
11fdf7f2 TL |
8700 | else |
8701 | num_pg_stray++; | |
8702 | } | |
3efd9988 | 8703 | |
11fdf7f2 TL |
8704 | { |
8705 | // FIXME (as part of seastar rewrite): move to OSDShard | |
8706 | std::lock_guard l(pending_creates_lock); | |
8707 | for (auto pg = pending_creates_from_osd.begin(); | |
8708 | pg != pending_creates_from_osd.end();) { | |
9f95a23c | 8709 | if (osdmap->get_pg_acting_role(pg->first, whoami) < 0) { |
11fdf7f2 TL |
8710 | dout(10) << __func__ << " pg " << pg->first << " doesn't map here, " |
8711 | << "discarding pending_create_from_osd" << dendl; | |
3efd9988 FG |
8712 | pg = pending_creates_from_osd.erase(pg); |
8713 | } else { | |
8714 | ++pg; | |
8715 | } | |
8716 | } | |
7c673cae FG |
8717 | } |
8718 | ||
7c673cae FG |
8719 | service.maybe_inject_dispatch_delay(); |
8720 | ||
8721 | dispatch_sessions_waiting_on_map(); | |
8722 | ||
8723 | service.maybe_inject_dispatch_delay(); | |
8724 | ||
11fdf7f2 | 8725 | service.release_reserved_pushes(pushes_to_free); |
7c673cae | 8726 | |
11fdf7f2 TL |
8727 | // queue null events to push maps down to individual PGs |
8728 | for (auto pgid : pgids) { | |
8729 | enqueue_peering_evt( | |
8730 | pgid, | |
8731 | PGPeeringEventRef( | |
8732 | std::make_shared<PGPeeringEvent>( | |
8733 | osdmap->get_epoch(), | |
8734 | osdmap->get_epoch(), | |
8735 | NullEvt()))); | |
7c673cae | 8736 | } |
11fdf7f2 | 8737 | logger->set(l_osd_pg, pgids.size()); |
7c673cae FG |
8738 | logger->set(l_osd_pg_primary, num_pg_primary); |
8739 | logger->set(l_osd_pg_replica, num_pg_replica); | |
8740 | logger->set(l_osd_pg_stray, num_pg_stray); | |
8741 | } | |
8742 | ||
8743 | void OSD::activate_map() | |
8744 | { | |
9f95a23c TL |
8745 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
8746 | auto osdmap = get_osdmap(); | |
7c673cae FG |
8747 | |
8748 | dout(7) << "activate_map version " << osdmap->get_epoch() << dendl; | |
8749 | ||
7c673cae FG |
8750 | // norecover? |
8751 | if (osdmap->test_flag(CEPH_OSDMAP_NORECOVER)) { | |
8752 | if (!service.recovery_is_paused()) { | |
8753 | dout(1) << "pausing recovery (NORECOVER flag set)" << dendl; | |
8754 | service.pause_recovery(); | |
8755 | } | |
8756 | } else { | |
8757 | if (service.recovery_is_paused()) { | |
8758 | dout(1) << "unpausing recovery (NORECOVER flag unset)" << dendl; | |
8759 | service.unpause_recovery(); | |
8760 | } | |
8761 | } | |
8762 | ||
8763 | service.activate_map(); | |
8764 | ||
8765 | // process waiters | |
8766 | take_waiters(waiting_for_osdmap); | |
8767 | } | |
8768 | ||
8769 | bool OSD::require_mon_peer(const Message *m) | |
8770 | { | |
8771 | if (!m->get_connection()->peer_is_mon()) { | |
8772 | dout(0) << "require_mon_peer received from non-mon " | |
8773 | << m->get_connection()->get_peer_addr() | |
8774 | << " " << *m << dendl; | |
8775 | return false; | |
8776 | } | |
8777 | return true; | |
8778 | } | |
8779 | ||
8780 | bool OSD::require_mon_or_mgr_peer(const Message *m) | |
8781 | { | |
8782 | if (!m->get_connection()->peer_is_mon() && | |
8783 | !m->get_connection()->peer_is_mgr()) { | |
8784 | dout(0) << "require_mon_or_mgr_peer received from non-mon, non-mgr " | |
8785 | << m->get_connection()->get_peer_addr() | |
8786 | << " " << *m << dendl; | |
8787 | return false; | |
8788 | } | |
8789 | return true; | |
8790 | } | |
8791 | ||
8792 | bool OSD::require_osd_peer(const Message *m) | |
8793 | { | |
8794 | if (!m->get_connection()->peer_is_osd()) { | |
8795 | dout(0) << "require_osd_peer received from non-osd " | |
8796 | << m->get_connection()->get_peer_addr() | |
8797 | << " " << *m << dendl; | |
8798 | return false; | |
8799 | } | |
8800 | return true; | |
8801 | } | |
8802 | ||
8803 | bool OSD::require_self_aliveness(const Message *m, epoch_t epoch) | |
8804 | { | |
8805 | epoch_t up_epoch = service.get_up_epoch(); | |
8806 | if (epoch < up_epoch) { | |
8807 | dout(7) << "from pre-up epoch " << epoch << " < " << up_epoch << dendl; | |
8808 | return false; | |
8809 | } | |
8810 | ||
8811 | if (!is_active()) { | |
8812 | dout(7) << "still in boot state, dropping message " << *m << dendl; | |
8813 | return false; | |
8814 | } | |
8815 | ||
8816 | return true; | |
8817 | } | |
8818 | ||
9f95a23c | 8819 | bool OSD::require_same_peer_instance(const Message *m, const OSDMapRef& map, |
7c673cae FG |
8820 | bool is_fast_dispatch) |
8821 | { | |
8822 | int from = m->get_source().num(); | |
8823 | ||
8824 | if (map->is_down(from) || | |
11fdf7f2 | 8825 | (map->get_cluster_addrs(from) != m->get_source_addrs())) { |
7c673cae FG |
8826 | dout(5) << "from dead osd." << from << ", marking down, " |
8827 | << " msg was " << m->get_source_inst().addr | |
11fdf7f2 TL |
8828 | << " expected " |
8829 | << (map->is_up(from) ? | |
8830 | map->get_cluster_addrs(from) : entity_addrvec_t()) | |
7c673cae FG |
8831 | << dendl; |
8832 | ConnectionRef con = m->get_connection(); | |
8833 | con->mark_down(); | |
9f95a23c | 8834 | if (auto s = ceph::ref_cast<Session>(con->get_priv()); s) { |
7c673cae | 8835 | if (!is_fast_dispatch) |
9f95a23c | 8836 | s->session_dispatch_lock.lock(); |
7c673cae | 8837 | clear_session_waiting_on_map(s); |
11fdf7f2 TL |
8838 | con->set_priv(nullptr); // break ref <-> session cycle, if any |
8839 | s->con.reset(); | |
7c673cae | 8840 | if (!is_fast_dispatch) |
9f95a23c | 8841 | s->session_dispatch_lock.unlock(); |
7c673cae FG |
8842 | } |
8843 | return false; | |
8844 | } | |
8845 | return true; | |
8846 | } | |
8847 | ||
8848 | ||
8849 | /* | |
8850 | * require that we have same (or newer) map, and that | |
8851 | * the source is the pg primary. | |
8852 | */ | |
8853 | bool OSD::require_same_or_newer_map(OpRequestRef& op, epoch_t epoch, | |
8854 | bool is_fast_dispatch) | |
8855 | { | |
8856 | const Message *m = op->get_req(); | |
9f95a23c | 8857 | const auto osdmap = get_osdmap(); |
7c673cae FG |
8858 | dout(15) << "require_same_or_newer_map " << epoch |
8859 | << " (i am " << osdmap->get_epoch() << ") " << m << dendl; | |
8860 | ||
9f95a23c | 8861 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae FG |
8862 | |
8863 | // do they have a newer map? | |
8864 | if (epoch > osdmap->get_epoch()) { | |
8865 | dout(7) << "waiting for newer map epoch " << epoch | |
8866 | << " > my " << osdmap->get_epoch() << " with " << m << dendl; | |
8867 | wait_for_new_map(op); | |
8868 | return false; | |
8869 | } | |
8870 | ||
8871 | if (!require_self_aliveness(op->get_req(), epoch)) { | |
8872 | return false; | |
8873 | } | |
8874 | ||
8875 | // ok, our map is same or newer.. do they still exist? | |
8876 | if (m->get_connection()->get_messenger() == cluster_messenger && | |
8877 | !require_same_peer_instance(op->get_req(), osdmap, is_fast_dispatch)) { | |
8878 | return false; | |
8879 | } | |
8880 | ||
8881 | return true; | |
8882 | } | |
8883 | ||
8884 | ||
8885 | ||
8886 | ||
8887 | ||
8888 | // ---------------------------------------- | |
8889 | // pg creation | |
8890 | ||
8891 | void OSD::split_pgs( | |
8892 | PG *parent, | |
31f18b77 | 8893 | const set<spg_t> &childpgids, set<PGRef> *out_pgs, |
7c673cae FG |
8894 | OSDMapRef curmap, |
8895 | OSDMapRef nextmap, | |
9f95a23c | 8896 | PeeringCtx &rctx) |
7c673cae | 8897 | { |
11fdf7f2 TL |
8898 | unsigned pg_num = nextmap->get_pg_num(parent->pg_id.pool()); |
8899 | parent->update_snap_mapper_bits(parent->get_pgid().get_split_bits(pg_num)); | |
7c673cae | 8900 | |
11fdf7f2 TL |
8901 | vector<object_stat_sum_t> updated_stats; |
8902 | parent->start_split_stats(childpgids, &updated_stats); | |
7c673cae FG |
8903 | |
8904 | vector<object_stat_sum_t>::iterator stat_iter = updated_stats.begin(); | |
8905 | for (set<spg_t>::const_iterator i = childpgids.begin(); | |
8906 | i != childpgids.end(); | |
8907 | ++i, ++stat_iter) { | |
11fdf7f2 TL |
8908 | ceph_assert(stat_iter != updated_stats.end()); |
8909 | dout(10) << __func__ << " splitting " << *parent << " into " << *i << dendl; | |
7c673cae FG |
8910 | PG* child = _make_pg(nextmap, *i); |
8911 | child->lock(true); | |
8912 | out_pgs->insert(child); | |
11fdf7f2 | 8913 | child->ch = store->create_new_collection(child->coll); |
7c673cae | 8914 | |
11fdf7f2 TL |
8915 | { |
8916 | uint32_t shard_index = i->hash_to_shard(shards.size()); | |
8917 | assert(NULL != shards[shard_index]); | |
8918 | store->set_collection_commit_queue(child->coll, &(shards[shard_index]->context_queue)); | |
8919 | } | |
7c673cae | 8920 | |
11fdf7f2 TL |
8921 | unsigned split_bits = i->get_split_bits(pg_num); |
8922 | dout(10) << " pg_num is " << pg_num | |
8923 | << ", m_seed " << i->ps() | |
8924 | << ", split_bits is " << split_bits << dendl; | |
7c673cae FG |
8925 | parent->split_colls( |
8926 | *i, | |
8927 | split_bits, | |
8928 | i->ps(), | |
11fdf7f2 | 8929 | &child->get_pool().info, |
9f95a23c | 8930 | rctx.transaction); |
7c673cae FG |
8931 | parent->split_into( |
8932 | i->pgid, | |
8933 | child, | |
8934 | split_bits); | |
7c673cae | 8935 | |
92f5a8d4 TL |
8936 | child->init_collection_pool_opts(); |
8937 | ||
9f95a23c | 8938 | child->finish_split_stats(*stat_iter, rctx.transaction); |
7c673cae FG |
8939 | child->unlock(); |
8940 | } | |
11fdf7f2 | 8941 | ceph_assert(stat_iter != updated_stats.end()); |
9f95a23c | 8942 | parent->finish_split_stats(*stat_iter, rctx.transaction); |
7c673cae FG |
8943 | } |
8944 | ||
8945 | /* | |
8946 | * holding osd_lock | |
8947 | */ | |
8948 | void OSD::handle_pg_create(OpRequestRef op) | |
8949 | { | |
9f95a23c TL |
8950 | // NOTE: this can be removed in P release (mimic is the last version to |
8951 | // send MOSDPGCreate messages). | |
8952 | ||
8953 | auto m = op->get_req<MOSDPGCreate>(); | |
11fdf7f2 | 8954 | ceph_assert(m->get_type() == MSG_OSD_PG_CREATE); |
7c673cae FG |
8955 | |
8956 | dout(10) << "handle_pg_create " << *m << dendl; | |
8957 | ||
8958 | if (!require_mon_peer(op->get_req())) { | |
8959 | return; | |
8960 | } | |
8961 | ||
8962 | if (!require_same_or_newer_map(op, m->epoch, false)) | |
8963 | return; | |
8964 | ||
8965 | op->mark_started(); | |
8966 | ||
9f95a23c | 8967 | const auto osdmap = get_osdmap(); |
7c673cae FG |
8968 | map<pg_t,utime_t>::const_iterator ci = m->ctimes.begin(); |
8969 | for (map<pg_t,pg_create_t>::const_iterator p = m->mkpg.begin(); | |
8970 | p != m->mkpg.end(); | |
8971 | ++p, ++ci) { | |
11fdf7f2 | 8972 | ceph_assert(ci != m->ctimes.end() && ci->first == p->first); |
7c673cae FG |
8973 | epoch_t created = p->second.created; |
8974 | if (p->second.split_bits) // Skip split pgs | |
8975 | continue; | |
8976 | pg_t on = p->first; | |
8977 | ||
7c673cae FG |
8978 | if (!osdmap->have_pg_pool(on.pool())) { |
8979 | dout(20) << "ignoring pg on deleted pool " << on << dendl; | |
8980 | continue; | |
8981 | } | |
8982 | ||
8983 | dout(20) << "mkpg " << on << " e" << created << "@" << ci->second << dendl; | |
8984 | ||
9f95a23c TL |
8985 | spg_t pgid; |
8986 | bool mapped = osdmap->get_primary_shard(on, &pgid); | |
8987 | ceph_assert(mapped); | |
8988 | ||
7c673cae FG |
8989 | // is it still ours? |
8990 | vector<int> up, acting; | |
8991 | int up_primary = -1; | |
8992 | int acting_primary = -1; | |
8993 | osdmap->pg_to_up_acting_osds(on, &up, &up_primary, &acting, &acting_primary); | |
9f95a23c | 8994 | int role = osdmap->calc_pg_role(pg_shard_t(whoami, pgid.shard), acting); |
7c673cae FG |
8995 | |
8996 | if (acting_primary != whoami) { | |
8997 | dout(10) << "mkpg " << on << " not acting_primary (" << acting_primary | |
8998 | << "), my role=" << role << ", skipping" << dendl; | |
8999 | continue; | |
9000 | } | |
9001 | ||
7c673cae | 9002 | |
11fdf7f2 | 9003 | PastIntervals pi; |
7c673cae FG |
9004 | pg_history_t history; |
9005 | build_initial_pg_history(pgid, created, ci->second, &history, &pi); | |
9006 | ||
11fdf7f2 TL |
9007 | // The mon won't resend unless the primary changed, so we ignore |
9008 | // same_interval_since. We'll pass this history with the current | |
9009 | // epoch as the event. | |
7c673cae FG |
9010 | if (history.same_primary_since > m->epoch) { |
9011 | dout(10) << __func__ << ": got obsolete pg create on pgid " | |
9012 | << pgid << " from epoch " << m->epoch | |
9013 | << ", primary changed in " << history.same_primary_since | |
9014 | << dendl; | |
9015 | continue; | |
9016 | } | |
11fdf7f2 TL |
9017 | enqueue_peering_evt( |
9018 | pgid, | |
9019 | PGPeeringEventRef( | |
9020 | std::make_shared<PGPeeringEvent>( | |
9021 | osdmap->get_epoch(), | |
9022 | osdmap->get_epoch(), | |
9023 | NullEvt(), | |
9024 | true, | |
9025 | new PGCreateInfo( | |
9026 | pgid, | |
9027 | osdmap->get_epoch(), | |
9028 | history, | |
9029 | pi, | |
9030 | true) | |
9031 | ))); | |
7c673cae | 9032 | } |
7c673cae | 9033 | |
3efd9988 | 9034 | { |
11fdf7f2 | 9035 | std::lock_guard l(pending_creates_lock); |
3efd9988 FG |
9036 | if (pending_creates_from_mon == 0) { |
9037 | last_pg_create_epoch = m->epoch; | |
9038 | } | |
9039 | } | |
11fdf7f2 | 9040 | |
7c673cae FG |
9041 | maybe_update_heartbeat_peers(); |
9042 | } | |
9043 | ||
9044 | ||
9045 | // ---------------------------------------- | |
9046 | // peering and recovery | |
9047 | ||
9f95a23c | 9048 | PeeringCtx OSD::create_context() |
7c673cae | 9049 | { |
9f95a23c | 9050 | return PeeringCtx(get_osdmap()->require_osd_release); |
7c673cae FG |
9051 | } |
9052 | ||
9f95a23c | 9053 | void OSD::dispatch_context(PeeringCtx &ctx, PG *pg, OSDMapRef curmap, |
7c673cae FG |
9054 | ThreadPool::TPHandle *handle) |
9055 | { | |
11fdf7f2 TL |
9056 | if (!service.get_osdmap()->is_up(whoami)) { |
9057 | dout(20) << __func__ << " not up in osdmap" << dendl; | |
9058 | } else if (!is_active()) { | |
9059 | dout(20) << __func__ << " not active" << dendl; | |
9060 | } else { | |
9f95a23c TL |
9061 | for (auto& [osd, ls] : ctx.message_map) { |
9062 | if (!curmap->is_up(osd)) { | |
9063 | dout(20) << __func__ << " skipping down osd." << osd << dendl; | |
9064 | continue; | |
9065 | } | |
9066 | ConnectionRef con = service.get_con_osd_cluster( | |
9067 | osd, curmap->get_epoch()); | |
9068 | if (!con) { | |
9069 | dout(20) << __func__ << " skipping osd." << osd << " (NULL con)" | |
9070 | << dendl; | |
9071 | continue; | |
9072 | } | |
9073 | service.maybe_share_map(con.get(), curmap); | |
9074 | for (auto m : ls) { | |
9075 | con->send_message2(m); | |
9076 | } | |
9077 | ls.clear(); | |
9078 | } | |
7c673cae | 9079 | } |
9f95a23c | 9080 | if ((!ctx.transaction.empty() || ctx.transaction.has_contexts()) && pg) { |
7c673cae | 9081 | int tr = store->queue_transaction( |
11fdf7f2 | 9082 | pg->ch, |
9f95a23c | 9083 | std::move(ctx.transaction), TrackedOpRef(), |
7c673cae | 9084 | handle); |
11fdf7f2 | 9085 | ceph_assert(tr == 0); |
7c673cae | 9086 | } |
7c673cae FG |
9087 | } |
9088 | ||
11fdf7f2 | 9089 | void OSD::handle_fast_pg_create(MOSDPGCreate2 *m) |
7c673cae | 9090 | { |
11fdf7f2 TL |
9091 | dout(7) << __func__ << " " << *m << " from " << m->get_source() << dendl; |
9092 | if (!require_mon_peer(m)) { | |
9093 | m->put(); | |
7c673cae | 9094 | return; |
7c673cae | 9095 | } |
11fdf7f2 TL |
9096 | for (auto& p : m->pgs) { |
9097 | spg_t pgid = p.first; | |
9098 | epoch_t created = p.second.first; | |
9099 | utime_t created_stamp = p.second.second; | |
9f95a23c TL |
9100 | auto q = m->pg_extra.find(pgid); |
9101 | if (q == m->pg_extra.end()) { | |
9102 | dout(20) << __func__ << " " << pgid << " e" << created | |
9103 | << "@" << created_stamp | |
9104 | << " (no history or past_intervals)" << dendl; | |
9105 | // pre-octopus ... no pg history. this can be removed in Q release. | |
9106 | enqueue_peering_evt( | |
9107 | pgid, | |
9108 | PGPeeringEventRef( | |
9109 | std::make_shared<PGPeeringEvent>( | |
9110 | m->epoch, | |
9111 | m->epoch, | |
9112 | NullEvt(), | |
9113 | true, | |
9114 | new PGCreateInfo( | |
9115 | pgid, | |
9116 | created, | |
9117 | pg_history_t(created, created_stamp), | |
9118 | PastIntervals(), | |
9119 | true) | |
9120 | ))); | |
9121 | } else { | |
9122 | dout(20) << __func__ << " " << pgid << " e" << created | |
9123 | << "@" << created_stamp | |
9124 | << " history " << q->second.first | |
9125 | << " pi " << q->second.second << dendl; | |
9126 | if (!q->second.second.empty() && | |
9127 | m->epoch < q->second.second.get_bounds().second) { | |
9128 | clog->error() << "got pg_create on " << pgid << " epoch " << m->epoch | |
9129 | << " and unmatched past_intervals " << q->second.second | |
9130 | << " (history " << q->second.first << ")"; | |
9131 | } else { | |
9132 | enqueue_peering_evt( | |
9133 | pgid, | |
9134 | PGPeeringEventRef( | |
9135 | std::make_shared<PGPeeringEvent>( | |
9136 | m->epoch, | |
9137 | m->epoch, | |
9138 | NullEvt(), | |
9139 | true, | |
9140 | new PGCreateInfo( | |
9141 | pgid, | |
9142 | m->epoch, | |
9143 | q->second.first, | |
9144 | q->second.second, | |
9145 | true) | |
9146 | ))); | |
9147 | } | |
9148 | } | |
11fdf7f2 | 9149 | } |
7c673cae | 9150 | |
11fdf7f2 TL |
9151 | { |
9152 | std::lock_guard l(pending_creates_lock); | |
9153 | if (pending_creates_from_mon == 0) { | |
9154 | last_pg_create_epoch = m->epoch; | |
9155 | } | |
7c673cae FG |
9156 | } |
9157 | ||
11fdf7f2 | 9158 | m->put(); |
7c673cae FG |
9159 | } |
9160 | ||
11fdf7f2 | 9161 | void OSD::handle_fast_pg_query(MOSDPGQuery *m) |
7c673cae | 9162 | { |
11fdf7f2 TL |
9163 | dout(7) << __func__ << " " << *m << " from " << m->get_source() << dendl; |
9164 | if (!require_osd_peer(m)) { | |
9165 | m->put(); | |
7c673cae | 9166 | return; |
11fdf7f2 | 9167 | } |
7c673cae | 9168 | int from = m->get_source().num(); |
11fdf7f2 TL |
9169 | for (auto& p : m->pg_list) { |
9170 | enqueue_peering_evt( | |
9171 | p.first, | |
9172 | PGPeeringEventRef( | |
9173 | std::make_shared<PGPeeringEvent>( | |
9174 | p.second.epoch_sent, p.second.epoch_sent, | |
9175 | MQuery( | |
9176 | p.first, | |
9177 | pg_shard_t(from, p.second.from), | |
9178 | p.second, | |
9179 | p.second.epoch_sent), | |
9180 | false)) | |
7c673cae FG |
9181 | ); |
9182 | } | |
11fdf7f2 | 9183 | m->put(); |
7c673cae FG |
9184 | } |
9185 | ||
11fdf7f2 | 9186 | void OSD::handle_fast_pg_notify(MOSDPGNotify* m) |
7c673cae | 9187 | { |
11fdf7f2 TL |
9188 | dout(7) << __func__ << " " << *m << " from " << m->get_source() << dendl; |
9189 | if (!require_osd_peer(m)) { | |
9190 | m->put(); | |
7c673cae FG |
9191 | return; |
9192 | } | |
11fdf7f2 TL |
9193 | int from = m->get_source().num(); |
9194 | for (auto& p : m->get_pg_list()) { | |
9f95a23c | 9195 | spg_t pgid(p.info.pgid.pgid, p.to); |
11fdf7f2 TL |
9196 | enqueue_peering_evt( |
9197 | pgid, | |
9198 | PGPeeringEventRef( | |
9199 | std::make_shared<PGPeeringEvent>( | |
9f95a23c TL |
9200 | p.epoch_sent, |
9201 | p.query_epoch, | |
11fdf7f2 | 9202 | MNotifyRec( |
9f95a23c TL |
9203 | pgid, pg_shard_t(from, p.from), |
9204 | p, | |
9205 | m->get_connection()->get_features()), | |
11fdf7f2 TL |
9206 | true, |
9207 | new PGCreateInfo( | |
9208 | pgid, | |
9f95a23c TL |
9209 | p.query_epoch, |
9210 | p.info.history, | |
9211 | p.past_intervals, | |
11fdf7f2 TL |
9212 | false) |
9213 | ))); | |
7c673cae | 9214 | } |
11fdf7f2 | 9215 | m->put(); |
7c673cae FG |
9216 | } |
9217 | ||
11fdf7f2 | 9218 | void OSD::handle_fast_pg_info(MOSDPGInfo* m) |
7c673cae | 9219 | { |
11fdf7f2 TL |
9220 | dout(7) << __func__ << " " << *m << " from " << m->get_source() << dendl; |
9221 | if (!require_osd_peer(m)) { | |
9222 | m->put(); | |
7c673cae FG |
9223 | return; |
9224 | } | |
11fdf7f2 TL |
9225 | int from = m->get_source().num(); |
9226 | for (auto& p : m->pg_list) { | |
9227 | enqueue_peering_evt( | |
9f95a23c | 9228 | spg_t(p.info.pgid.pgid, p.to), |
11fdf7f2 TL |
9229 | PGPeeringEventRef( |
9230 | std::make_shared<PGPeeringEvent>( | |
9f95a23c | 9231 | p.epoch_sent, p.query_epoch, |
11fdf7f2 | 9232 | MInfoRec( |
9f95a23c TL |
9233 | pg_shard_t(from, p.from), |
9234 | p.info, | |
9235 | p.epoch_sent))) | |
11fdf7f2 | 9236 | ); |
7c673cae | 9237 | } |
11fdf7f2 | 9238 | m->put(); |
7c673cae FG |
9239 | } |
9240 | ||
11fdf7f2 | 9241 | void OSD::handle_fast_pg_remove(MOSDPGRemove *m) |
7c673cae | 9242 | { |
11fdf7f2 TL |
9243 | dout(7) << __func__ << " " << *m << " from " << m->get_source() << dendl; |
9244 | if (!require_osd_peer(m)) { | |
9245 | m->put(); | |
7c673cae FG |
9246 | return; |
9247 | } | |
11fdf7f2 TL |
9248 | for (auto& pgid : m->pg_list) { |
9249 | enqueue_peering_evt( | |
9250 | pgid, | |
9251 | PGPeeringEventRef( | |
9252 | std::make_shared<PGPeeringEvent>( | |
9253 | m->get_epoch(), m->get_epoch(), | |
9f95a23c | 9254 | PeeringState::DeleteStart()))); |
7c673cae | 9255 | } |
11fdf7f2 | 9256 | m->put(); |
7c673cae FG |
9257 | } |
9258 | ||
11fdf7f2 | 9259 | void OSD::handle_fast_force_recovery(MOSDForceRecovery *m) |
c07f9fc5 | 9260 | { |
11fdf7f2 TL |
9261 | dout(10) << __func__ << " " << *m << dendl; |
9262 | if (!require_mon_or_mgr_peer(m)) { | |
9263 | m->put(); | |
9264 | return; | |
9265 | } | |
9266 | epoch_t epoch = get_osdmap_epoch(); | |
9267 | for (auto pgid : m->forced_pgs) { | |
9268 | if (m->options & OFR_BACKFILL) { | |
9269 | if (m->options & OFR_CANCEL) { | |
9270 | enqueue_peering_evt( | |
9271 | pgid, | |
9272 | PGPeeringEventRef( | |
9273 | std::make_shared<PGPeeringEvent>( | |
9274 | epoch, epoch, | |
9f95a23c | 9275 | PeeringState::UnsetForceBackfill()))); |
11fdf7f2 TL |
9276 | } else { |
9277 | enqueue_peering_evt( | |
9278 | pgid, | |
9279 | PGPeeringEventRef( | |
9280 | std::make_shared<PGPeeringEvent>( | |
9281 | epoch, epoch, | |
9f95a23c | 9282 | PeeringState::SetForceBackfill()))); |
11fdf7f2 TL |
9283 | } |
9284 | } else if (m->options & OFR_RECOVERY) { | |
9285 | if (m->options & OFR_CANCEL) { | |
9286 | enqueue_peering_evt( | |
9287 | pgid, | |
9288 | PGPeeringEventRef( | |
9289 | std::make_shared<PGPeeringEvent>( | |
9290 | epoch, epoch, | |
9f95a23c | 9291 | PeeringState::UnsetForceRecovery()))); |
11fdf7f2 TL |
9292 | } else { |
9293 | enqueue_peering_evt( | |
9294 | pgid, | |
9295 | PGPeeringEventRef( | |
9296 | std::make_shared<PGPeeringEvent>( | |
9297 | epoch, epoch, | |
9f95a23c | 9298 | PeeringState::SetForceRecovery()))); |
c07f9fc5 FG |
9299 | } |
9300 | } | |
9301 | } | |
11fdf7f2 | 9302 | m->put(); |
c07f9fc5 | 9303 | } |
7c673cae | 9304 | |
11fdf7f2 | 9305 | void OSD::handle_pg_query_nopg(const MQuery& q) |
7c673cae | 9306 | { |
11fdf7f2 TL |
9307 | spg_t pgid = q.pgid; |
9308 | dout(10) << __func__ << " " << pgid << dendl; | |
7c673cae | 9309 | |
11fdf7f2 TL |
9310 | OSDMapRef osdmap = get_osdmap(); |
9311 | if (!osdmap->have_pg_pool(pgid.pool())) | |
7c673cae FG |
9312 | return; |
9313 | ||
11fdf7f2 TL |
9314 | dout(10) << " pg " << pgid << " dne" << dendl; |
9315 | pg_info_t empty(spg_t(pgid.pgid, q.query.to)); | |
9316 | ConnectionRef con = service.get_con_osd_cluster(q.from.osd, osdmap->get_epoch()); | |
9317 | if (con) { | |
9318 | Message *m; | |
9319 | if (q.query.type == pg_query_t::LOG || | |
9320 | q.query.type == pg_query_t::FULLLOG) { | |
9321 | m = new MOSDPGLog( | |
9322 | q.query.from, q.query.to, | |
9323 | osdmap->get_epoch(), empty, | |
9324 | q.query.epoch_sent); | |
7c673cae | 9325 | } else { |
9f95a23c | 9326 | vector<pg_notify_t> ls; |
11fdf7f2 | 9327 | ls.push_back( |
9f95a23c TL |
9328 | pg_notify_t( |
9329 | q.query.from, q.query.to, | |
9330 | q.query.epoch_sent, | |
9331 | osdmap->get_epoch(), | |
9332 | empty, | |
11fdf7f2 | 9333 | PastIntervals())); |
9f95a23c | 9334 | m = new MOSDPGNotify(osdmap->get_epoch(), std::move(ls)); |
7c673cae | 9335 | } |
9f95a23c | 9336 | service.maybe_share_map(con.get(), osdmap); |
11fdf7f2 | 9337 | con->send_message(m); |
7c673cae FG |
9338 | } |
9339 | } | |
9340 | ||
9f95a23c TL |
9341 | void OSDService::queue_check_readable(spg_t spgid, |
9342 | epoch_t lpr, | |
9343 | ceph::signedspan delay) | |
9344 | { | |
9345 | if (delay == ceph::signedspan::zero()) { | |
9346 | osd->enqueue_peering_evt( | |
9347 | spgid, | |
9348 | PGPeeringEventRef( | |
9349 | std::make_shared<PGPeeringEvent>( | |
9350 | lpr, lpr, | |
9351 | PeeringState::CheckReadable()))); | |
9352 | } else { | |
9353 | mono_timer.add_event( | |
9354 | delay, | |
9355 | [this, spgid, lpr]() { | |
9356 | queue_check_readable(spgid, lpr); | |
9357 | }); | |
9358 | } | |
9359 | } | |
9360 | ||
7c673cae | 9361 | |
7c673cae FG |
9362 | // ========================================================= |
9363 | // RECOVERY | |
9364 | ||
9365 | void OSDService::_maybe_queue_recovery() { | |
9f95a23c | 9366 | ceph_assert(ceph_mutex_is_locked_by_me(recovery_lock)); |
7c673cae FG |
9367 | uint64_t available_pushes; |
9368 | while (!awaiting_throttle.empty() && | |
9369 | _recover_now(&available_pushes)) { | |
11fdf7f2 | 9370 | uint64_t to_start = std::min( |
7c673cae FG |
9371 | available_pushes, |
9372 | cct->_conf->osd_recovery_max_single_start); | |
9373 | _queue_for_recovery(awaiting_throttle.front(), to_start); | |
9374 | awaiting_throttle.pop_front(); | |
11fdf7f2 TL |
9375 | dout(10) << __func__ << " starting " << to_start |
9376 | << ", recovery_ops_reserved " << recovery_ops_reserved | |
9377 | << " -> " << (recovery_ops_reserved + to_start) << dendl; | |
7c673cae FG |
9378 | recovery_ops_reserved += to_start; |
9379 | } | |
9380 | } | |
9381 | ||
9382 | bool OSDService::_recover_now(uint64_t *available_pushes) | |
9383 | { | |
9384 | if (available_pushes) | |
9385 | *available_pushes = 0; | |
9386 | ||
9387 | if (ceph_clock_now() < defer_recovery_until) { | |
9388 | dout(15) << __func__ << " defer until " << defer_recovery_until << dendl; | |
9389 | return false; | |
9390 | } | |
9391 | ||
9392 | if (recovery_paused) { | |
9393 | dout(15) << __func__ << " paused" << dendl; | |
9394 | return false; | |
9395 | } | |
9396 | ||
9f95a23c | 9397 | uint64_t max = osd->get_recovery_max_active(); |
7c673cae FG |
9398 | if (max <= recovery_ops_active + recovery_ops_reserved) { |
9399 | dout(15) << __func__ << " active " << recovery_ops_active | |
9400 | << " + reserved " << recovery_ops_reserved | |
9401 | << " >= max " << max << dendl; | |
9402 | return false; | |
9403 | } | |
9404 | ||
9405 | if (available_pushes) | |
9406 | *available_pushes = max - recovery_ops_active - recovery_ops_reserved; | |
9407 | ||
9408 | return true; | |
9409 | } | |
9410 | ||
9f95a23c TL |
9411 | unsigned OSDService::get_target_pg_log_entries() const |
9412 | { | |
9413 | auto num_pgs = osd->get_num_pgs(); | |
9414 | auto target = cct->_conf->osd_target_pg_log_entries_per_osd; | |
9415 | if (num_pgs > 0 && target > 0) { | |
9416 | // target an even spread of our budgeted log entries across all | |
9417 | // PGs. note that while we only get to control the entry count | |
9418 | // for primary PGs, we'll normally be responsible for a mix of | |
9419 | // primary and replica PGs (for the same pool(s) even), so this | |
9420 | // will work out. | |
9421 | return std::max<unsigned>( | |
9422 | std::min<unsigned>(target / num_pgs, | |
9423 | cct->_conf->osd_max_pg_log_entries), | |
9424 | cct->_conf->osd_min_pg_log_entries); | |
9425 | } else { | |
9426 | // fall back to a per-pg value. | |
9427 | return cct->_conf->osd_min_pg_log_entries; | |
9428 | } | |
9429 | } | |
9430 | ||
7c673cae FG |
9431 | void OSD::do_recovery( |
9432 | PG *pg, epoch_t queued, uint64_t reserved_pushes, | |
9433 | ThreadPool::TPHandle &handle) | |
9434 | { | |
9435 | uint64_t started = 0; | |
31f18b77 FG |
9436 | |
9437 | /* | |
9438 | * When the value of osd_recovery_sleep is set greater than zero, recovery | |
9439 | * ops are scheduled after osd_recovery_sleep amount of time from the previous | |
9440 | * recovery event's schedule time. This is done by adding a | |
9441 | * recovery_requeue_callback event, which re-queues the recovery op using | |
9442 | * queue_recovery_after_sleep. | |
9443 | */ | |
c07f9fc5 | 9444 | float recovery_sleep = get_osd_recovery_sleep(); |
b32b8144 | 9445 | { |
11fdf7f2 | 9446 | std::lock_guard l(service.sleep_lock); |
b32b8144 FG |
9447 | if (recovery_sleep > 0 && service.recovery_needs_sleep) { |
9448 | PGRef pgref(pg); | |
9f95a23c | 9449 | auto recovery_requeue_callback = new LambdaContext([this, pgref, queued, reserved_pushes](int r) { |
b32b8144 FG |
9450 | dout(20) << "do_recovery wake up at " |
9451 | << ceph_clock_now() | |
9452 | << ", re-queuing recovery" << dendl; | |
11fdf7f2 | 9453 | std::lock_guard l(service.sleep_lock); |
b32b8144 FG |
9454 | service.recovery_needs_sleep = false; |
9455 | service.queue_recovery_after_sleep(pgref.get(), queued, reserved_pushes); | |
9456 | }); | |
9457 | ||
9458 | // This is true for the first recovery op and when the previous recovery op | |
9459 | // has been scheduled in the past. The next recovery op is scheduled after | |
9460 | // completing the sleep from now. | |
9f95a23c TL |
9461 | |
9462 | if (auto now = ceph::real_clock::now(); | |
9463 | service.recovery_schedule_time < now) { | |
9464 | service.recovery_schedule_time = now; | |
b32b8144 | 9465 | } |
9f95a23c | 9466 | service.recovery_schedule_time += ceph::make_timespan(recovery_sleep); |
11fdf7f2 | 9467 | service.sleep_timer.add_event_at(service.recovery_schedule_time, |
9f95a23c | 9468 | recovery_requeue_callback); |
b32b8144 FG |
9469 | dout(20) << "Recovery event scheduled at " |
9470 | << service.recovery_schedule_time << dendl; | |
9471 | return; | |
9472 | } | |
7c673cae FG |
9473 | } |
9474 | ||
9475 | { | |
b32b8144 | 9476 | { |
11fdf7f2 | 9477 | std::lock_guard l(service.sleep_lock); |
b32b8144 FG |
9478 | service.recovery_needs_sleep = true; |
9479 | } | |
9480 | ||
7c673cae FG |
9481 | if (pg->pg_has_reset_since(queued)) { |
9482 | goto out; | |
9483 | } | |
9484 | ||
7c673cae FG |
9485 | dout(10) << "do_recovery starting " << reserved_pushes << " " << *pg << dendl; |
9486 | #ifdef DEBUG_RECOVERY_OIDS | |
11fdf7f2 | 9487 | dout(20) << " active was " << service.recovery_oids[pg->pg_id] << dendl; |
7c673cae FG |
9488 | #endif |
9489 | ||
11fdf7f2 | 9490 | bool do_unfound = pg->start_recovery_ops(reserved_pushes, handle, &started); |
7c673cae FG |
9491 | dout(10) << "do_recovery started " << started << "/" << reserved_pushes |
9492 | << " on " << *pg << dendl; | |
9493 | ||
11fdf7f2 | 9494 | if (do_unfound) { |
9f95a23c | 9495 | PeeringCtx rctx = create_context(); |
11fdf7f2 | 9496 | rctx.handle = &handle; |
9f95a23c | 9497 | pg->find_unfound(queued, rctx); |
11fdf7f2 | 9498 | dispatch_context(rctx, pg, pg->get_osdmap()); |
7c673cae | 9499 | } |
7c673cae FG |
9500 | } |
9501 | ||
9502 | out: | |
11fdf7f2 | 9503 | ceph_assert(started <= reserved_pushes); |
7c673cae FG |
9504 | service.release_reserved_pushes(reserved_pushes); |
9505 | } | |
9506 | ||
9507 | void OSDService::start_recovery_op(PG *pg, const hobject_t& soid) | |
9508 | { | |
11fdf7f2 | 9509 | std::lock_guard l(recovery_lock); |
7c673cae FG |
9510 | dout(10) << "start_recovery_op " << *pg << " " << soid |
9511 | << " (" << recovery_ops_active << "/" | |
9f95a23c | 9512 | << osd->get_recovery_max_active() << " rops)" |
7c673cae FG |
9513 | << dendl; |
9514 | recovery_ops_active++; | |
9515 | ||
9516 | #ifdef DEBUG_RECOVERY_OIDS | |
11fdf7f2 TL |
9517 | dout(20) << " active was " << recovery_oids[pg->pg_id] << dendl; |
9518 | ceph_assert(recovery_oids[pg->pg_id].count(soid) == 0); | |
9519 | recovery_oids[pg->pg_id].insert(soid); | |
7c673cae FG |
9520 | #endif |
9521 | } | |
9522 | ||
9523 | void OSDService::finish_recovery_op(PG *pg, const hobject_t& soid, bool dequeue) | |
9524 | { | |
11fdf7f2 | 9525 | std::lock_guard l(recovery_lock); |
7c673cae FG |
9526 | dout(10) << "finish_recovery_op " << *pg << " " << soid |
9527 | << " dequeue=" << dequeue | |
9f95a23c TL |
9528 | << " (" << recovery_ops_active << "/" |
9529 | << osd->get_recovery_max_active() << " rops)" | |
7c673cae FG |
9530 | << dendl; |
9531 | ||
9532 | // adjust count | |
11fdf7f2 | 9533 | ceph_assert(recovery_ops_active > 0); |
7c673cae FG |
9534 | recovery_ops_active--; |
9535 | ||
9536 | #ifdef DEBUG_RECOVERY_OIDS | |
11fdf7f2 TL |
9537 | dout(20) << " active oids was " << recovery_oids[pg->pg_id] << dendl; |
9538 | ceph_assert(recovery_oids[pg->pg_id].count(soid)); | |
9539 | recovery_oids[pg->pg_id].erase(soid); | |
7c673cae FG |
9540 | #endif |
9541 | ||
9542 | _maybe_queue_recovery(); | |
9543 | } | |
9544 | ||
9545 | bool OSDService::is_recovery_active() | |
9546 | { | |
eafe8130 TL |
9547 | if (cct->_conf->osd_debug_pretend_recovery_active) { |
9548 | return true; | |
9549 | } | |
b5b8bbf5 | 9550 | return local_reserver.has_reservation() || remote_reserver.has_reservation(); |
7c673cae FG |
9551 | } |
9552 | ||
11fdf7f2 TL |
9553 | void OSDService::release_reserved_pushes(uint64_t pushes) |
9554 | { | |
9555 | std::lock_guard l(recovery_lock); | |
9556 | dout(10) << __func__ << "(" << pushes << "), recovery_ops_reserved " | |
9557 | << recovery_ops_reserved << " -> " << (recovery_ops_reserved-pushes) | |
9558 | << dendl; | |
9559 | ceph_assert(recovery_ops_reserved >= pushes); | |
9560 | recovery_ops_reserved -= pushes; | |
9561 | _maybe_queue_recovery(); | |
9562 | } | |
9563 | ||
7c673cae FG |
9564 | // ========================================================= |
9565 | // OPS | |
9566 | ||
9567 | bool OSD::op_is_discardable(const MOSDOp *op) | |
9568 | { | |
9569 | // drop client request if they are not connected and can't get the | |
9570 | // reply anyway. | |
9571 | if (!op->get_connection()->is_connected()) { | |
9572 | return true; | |
9573 | } | |
9574 | return false; | |
9575 | } | |
9576 | ||
11fdf7f2 | 9577 | void OSD::enqueue_op(spg_t pg, OpRequestRef&& op, epoch_t epoch) |
7c673cae | 9578 | { |
11fdf7f2 TL |
9579 | const utime_t stamp = op->get_req()->get_recv_stamp(); |
9580 | const utime_t latency = ceph_clock_now() - stamp; | |
9581 | const unsigned priority = op->get_req()->get_priority(); | |
9582 | const int cost = op->get_req()->get_cost(); | |
9583 | const uint64_t owner = op->get_req()->get_source().num(); | |
9584 | ||
9585 | dout(15) << "enqueue_op " << op << " prio " << priority | |
9586 | << " cost " << cost | |
7c673cae FG |
9587 | << " latency " << latency |
9588 | << " epoch " << epoch | |
9589 | << " " << *(op->get_req()) << dendl; | |
9590 | op->osd_trace.event("enqueue op"); | |
11fdf7f2 TL |
9591 | op->osd_trace.keyval("priority", priority); |
9592 | op->osd_trace.keyval("cost", cost); | |
7c673cae | 9593 | op->mark_queued_for_pg(); |
224ce89b | 9594 | logger->tinc(l_osd_op_before_queue_op_lat, latency); |
11fdf7f2 | 9595 | op_shardedwq.queue( |
9f95a23c TL |
9596 | OpSchedulerItem( |
9597 | unique_ptr<OpSchedulerItem::OpQueueable>(new PGOpItem(pg, std::move(op))), | |
11fdf7f2 | 9598 | cost, priority, stamp, owner, epoch)); |
7c673cae FG |
9599 | } |
9600 | ||
11fdf7f2 TL |
9601 | void OSD::enqueue_peering_evt(spg_t pgid, PGPeeringEventRef evt) |
9602 | { | |
9603 | dout(15) << __func__ << " " << pgid << " " << evt->get_desc() << dendl; | |
9604 | op_shardedwq.queue( | |
9f95a23c TL |
9605 | OpSchedulerItem( |
9606 | unique_ptr<OpSchedulerItem::OpQueueable>(new PGPeeringItem(pgid, evt)), | |
11fdf7f2 TL |
9607 | 10, |
9608 | cct->_conf->osd_peering_op_priority, | |
9609 | utime_t(), | |
9610 | 0, | |
9611 | evt->get_epoch_sent())); | |
9612 | } | |
7c673cae FG |
9613 | |
9614 | /* | |
9615 | * NOTE: dequeue called in worker thread, with pg lock | |
9616 | */ | |
9617 | void OSD::dequeue_op( | |
9618 | PGRef pg, OpRequestRef op, | |
9619 | ThreadPool::TPHandle &handle) | |
9620 | { | |
9f95a23c TL |
9621 | const Message *m = op->get_req(); |
9622 | ||
11fdf7f2 | 9623 | FUNCTRACE(cct); |
9f95a23c | 9624 | OID_EVENT_TRACE_WITH_MSG(m, "DEQUEUE_OP_BEGIN", false); |
7c673cae FG |
9625 | |
9626 | utime_t now = ceph_clock_now(); | |
9627 | op->set_dequeued_time(now); | |
9f95a23c TL |
9628 | |
9629 | utime_t latency = now - m->get_recv_stamp(); | |
9630 | dout(10) << "dequeue_op " << op << " prio " << m->get_priority() | |
9631 | << " cost " << m->get_cost() | |
7c673cae | 9632 | << " latency " << latency |
9f95a23c | 9633 | << " " << *m |
7c673cae FG |
9634 | << " pg " << *pg << dendl; |
9635 | ||
224ce89b WB |
9636 | logger->tinc(l_osd_op_before_dequeue_op_lat, latency); |
9637 | ||
9f95a23c TL |
9638 | service.maybe_share_map(m->get_connection().get(), |
9639 | pg->get_osdmap(), | |
9640 | op->sent_epoch); | |
7c673cae | 9641 | |
11fdf7f2 | 9642 | if (pg->is_deleting()) |
7c673cae FG |
9643 | return; |
9644 | ||
9645 | op->mark_reached_pg(); | |
9646 | op->osd_trace.event("dequeue_op"); | |
9647 | ||
9648 | pg->do_request(op, handle); | |
9649 | ||
9650 | // finish | |
9651 | dout(10) << "dequeue_op " << op << " finish" << dendl; | |
9f95a23c | 9652 | OID_EVENT_TRACE_WITH_MSG(m, "DEQUEUE_OP_END", false); |
7c673cae FG |
9653 | } |
9654 | ||
9655 | ||
11fdf7f2 TL |
9656 | void OSD::dequeue_peering_evt( |
9657 | OSDShard *sdata, | |
9658 | PG *pg, | |
9659 | PGPeeringEventRef evt, | |
9660 | ThreadPool::TPHandle& handle) | |
7c673cae | 9661 | { |
9f95a23c | 9662 | PeeringCtx rctx = create_context(); |
11fdf7f2 | 9663 | auto curmap = sdata->get_osdmap(); |
9f95a23c TL |
9664 | bool need_up_thru = false; |
9665 | epoch_t same_interval_since = 0; | |
11fdf7f2 TL |
9666 | if (!pg) { |
9667 | if (const MQuery *q = dynamic_cast<const MQuery*>(evt->evt.get())) { | |
9668 | handle_pg_query_nopg(*q); | |
7c673cae | 9669 | } else { |
11fdf7f2 TL |
9670 | derr << __func__ << " unrecognized pg-less event " << evt->get_desc() << dendl; |
9671 | ceph_abort(); | |
9672 | } | |
9f95a23c TL |
9673 | } else if (advance_pg(curmap->get_epoch(), pg, handle, rctx)) { |
9674 | pg->do_peering_event(evt, rctx); | |
11fdf7f2 | 9675 | if (pg->is_deleted()) { |
11fdf7f2 TL |
9676 | pg->unlock(); |
9677 | return; | |
7c673cae | 9678 | } |
9f95a23c | 9679 | dispatch_context(rctx, pg, curmap, &handle); |
11fdf7f2 TL |
9680 | need_up_thru = pg->get_need_up_thru(); |
9681 | same_interval_since = pg->get_same_interval_since(); | |
7c673cae FG |
9682 | pg->unlock(); |
9683 | } | |
11fdf7f2 TL |
9684 | |
9685 | if (need_up_thru) { | |
7c673cae | 9686 | queue_want_up_thru(same_interval_since); |
11fdf7f2 | 9687 | } |
7c673cae FG |
9688 | |
9689 | service.send_pg_temp(); | |
9690 | } | |
9691 | ||
11fdf7f2 TL |
9692 | void OSD::dequeue_delete( |
9693 | OSDShard *sdata, | |
9694 | PG *pg, | |
9695 | epoch_t e, | |
9696 | ThreadPool::TPHandle& handle) | |
9697 | { | |
9698 | dequeue_peering_evt( | |
9699 | sdata, | |
9700 | pg, | |
9701 | PGPeeringEventRef( | |
9702 | std::make_shared<PGPeeringEvent>( | |
9703 | e, e, | |
9f95a23c | 9704 | PeeringState::DeleteSome())), |
11fdf7f2 TL |
9705 | handle); |
9706 | } | |
9707 | ||
9708 | ||
9709 | ||
7c673cae FG |
9710 | // -------------------------------- |
9711 | ||
9712 | const char** OSD::get_tracked_conf_keys() const | |
9713 | { | |
9714 | static const char* KEYS[] = { | |
9715 | "osd_max_backfills", | |
9716 | "osd_min_recovery_priority", | |
224ce89b WB |
9717 | "osd_max_trimming_pgs", |
9718 | "osd_op_complaint_time", | |
9719 | "osd_op_log_threshold", | |
9720 | "osd_op_history_size", | |
9721 | "osd_op_history_duration", | |
9722 | "osd_op_history_slow_op_size", | |
9723 | "osd_op_history_slow_op_threshold", | |
7c673cae FG |
9724 | "osd_enable_op_tracker", |
9725 | "osd_map_cache_size", | |
11fdf7f2 | 9726 | "osd_pg_epoch_max_lag_factor", |
7c673cae | 9727 | "osd_pg_epoch_persisted_max_stale", |
7c673cae FG |
9728 | // clog & admin clog |
9729 | "clog_to_monitors", | |
9730 | "clog_to_syslog", | |
9731 | "clog_to_syslog_facility", | |
9732 | "clog_to_syslog_level", | |
9733 | "osd_objectstore_fuse", | |
9734 | "clog_to_graylog", | |
9735 | "clog_to_graylog_host", | |
9736 | "clog_to_graylog_port", | |
9737 | "host", | |
9738 | "fsid", | |
9739 | "osd_recovery_delay_start", | |
9740 | "osd_client_message_size_cap", | |
9741 | "osd_client_message_cap", | |
31f18b77 FG |
9742 | "osd_heartbeat_min_size", |
9743 | "osd_heartbeat_interval", | |
9f95a23c | 9744 | "osd_object_clean_region_max_num_intervals", |
494da23a TL |
9745 | "osd_scrub_min_interval", |
9746 | "osd_scrub_max_interval", | |
7c673cae FG |
9747 | NULL |
9748 | }; | |
9749 | return KEYS; | |
9750 | } | |
9751 | ||
11fdf7f2 | 9752 | void OSD::handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
9753 | const std::set <std::string> &changed) |
9754 | { | |
9f95a23c | 9755 | std::lock_guard l{osd_lock}; |
7c673cae FG |
9756 | if (changed.count("osd_max_backfills")) { |
9757 | service.local_reserver.set_max(cct->_conf->osd_max_backfills); | |
9758 | service.remote_reserver.set_max(cct->_conf->osd_max_backfills); | |
9759 | } | |
9760 | if (changed.count("osd_min_recovery_priority")) { | |
9761 | service.local_reserver.set_min_priority(cct->_conf->osd_min_recovery_priority); | |
9762 | service.remote_reserver.set_min_priority(cct->_conf->osd_min_recovery_priority); | |
9763 | } | |
9764 | if (changed.count("osd_max_trimming_pgs")) { | |
9765 | service.snap_reserver.set_max(cct->_conf->osd_max_trimming_pgs); | |
9766 | } | |
9767 | if (changed.count("osd_op_complaint_time") || | |
9768 | changed.count("osd_op_log_threshold")) { | |
9769 | op_tracker.set_complaint_and_threshold(cct->_conf->osd_op_complaint_time, | |
9770 | cct->_conf->osd_op_log_threshold); | |
9771 | } | |
9772 | if (changed.count("osd_op_history_size") || | |
9773 | changed.count("osd_op_history_duration")) { | |
9774 | op_tracker.set_history_size_and_duration(cct->_conf->osd_op_history_size, | |
9775 | cct->_conf->osd_op_history_duration); | |
9776 | } | |
9777 | if (changed.count("osd_op_history_slow_op_size") || | |
9778 | changed.count("osd_op_history_slow_op_threshold")) { | |
9779 | op_tracker.set_history_slow_op_size_and_threshold(cct->_conf->osd_op_history_slow_op_size, | |
9780 | cct->_conf->osd_op_history_slow_op_threshold); | |
9781 | } | |
9782 | if (changed.count("osd_enable_op_tracker")) { | |
9783 | op_tracker.set_tracking(cct->_conf->osd_enable_op_tracker); | |
9784 | } | |
7c673cae FG |
9785 | if (changed.count("osd_map_cache_size")) { |
9786 | service.map_cache.set_size(cct->_conf->osd_map_cache_size); | |
9787 | service.map_bl_cache.set_size(cct->_conf->osd_map_cache_size); | |
9788 | service.map_bl_inc_cache.set_size(cct->_conf->osd_map_cache_size); | |
9789 | } | |
9790 | if (changed.count("clog_to_monitors") || | |
9791 | changed.count("clog_to_syslog") || | |
9792 | changed.count("clog_to_syslog_level") || | |
9793 | changed.count("clog_to_syslog_facility") || | |
9794 | changed.count("clog_to_graylog") || | |
9795 | changed.count("clog_to_graylog_host") || | |
9796 | changed.count("clog_to_graylog_port") || | |
9797 | changed.count("host") || | |
9798 | changed.count("fsid")) { | |
9799 | update_log_config(); | |
9800 | } | |
11fdf7f2 TL |
9801 | if (changed.count("osd_pg_epoch_max_lag_factor")) { |
9802 | m_osd_pg_epoch_max_lag_factor = conf.get_val<double>( | |
9803 | "osd_pg_epoch_max_lag_factor"); | |
9804 | } | |
7c673cae FG |
9805 | |
9806 | #ifdef HAVE_LIBFUSE | |
9807 | if (changed.count("osd_objectstore_fuse")) { | |
9808 | if (store) { | |
9809 | enable_disable_fuse(false); | |
9810 | } | |
9811 | } | |
9812 | #endif | |
9813 | ||
9814 | if (changed.count("osd_recovery_delay_start")) { | |
9815 | service.defer_recovery(cct->_conf->osd_recovery_delay_start); | |
9816 | service.kick_recovery_queue(); | |
9817 | } | |
9818 | ||
9819 | if (changed.count("osd_client_message_cap")) { | |
9820 | uint64_t newval = cct->_conf->osd_client_message_cap; | |
9821 | Messenger::Policy pol = client_messenger->get_policy(entity_name_t::TYPE_CLIENT); | |
9822 | if (pol.throttler_messages && newval > 0) { | |
9823 | pol.throttler_messages->reset_max(newval); | |
9824 | } | |
9825 | } | |
9826 | if (changed.count("osd_client_message_size_cap")) { | |
9827 | uint64_t newval = cct->_conf->osd_client_message_size_cap; | |
9828 | Messenger::Policy pol = client_messenger->get_policy(entity_name_t::TYPE_CLIENT); | |
9829 | if (pol.throttler_bytes && newval > 0) { | |
9830 | pol.throttler_bytes->reset_max(newval); | |
9831 | } | |
9832 | } | |
9f95a23c TL |
9833 | if (changed.count("osd_object_clean_region_max_num_intervals")) { |
9834 | ObjectCleanRegions::set_max_num_intervals(cct->_conf->osd_object_clean_region_max_num_intervals); | |
9835 | } | |
7c673cae | 9836 | |
494da23a TL |
9837 | if (changed.count("osd_scrub_min_interval") || |
9838 | changed.count("osd_scrub_max_interval")) { | |
9839 | resched_all_scrubs(); | |
9840 | dout(0) << __func__ << ": scrub interval change" << dendl; | |
9841 | } | |
7c673cae FG |
9842 | check_config(); |
9843 | } | |
9844 | ||
9845 | void OSD::update_log_config() | |
9846 | { | |
9847 | map<string,string> log_to_monitors; | |
9848 | map<string,string> log_to_syslog; | |
9849 | map<string,string> log_channel; | |
9850 | map<string,string> log_prio; | |
9851 | map<string,string> log_to_graylog; | |
9852 | map<string,string> log_to_graylog_host; | |
9853 | map<string,string> log_to_graylog_port; | |
9854 | uuid_d fsid; | |
9855 | string host; | |
9856 | ||
9857 | if (parse_log_client_options(cct, log_to_monitors, log_to_syslog, | |
9858 | log_channel, log_prio, log_to_graylog, | |
9859 | log_to_graylog_host, log_to_graylog_port, | |
9860 | fsid, host) == 0) | |
9861 | clog->update_config(log_to_monitors, log_to_syslog, | |
9862 | log_channel, log_prio, log_to_graylog, | |
9863 | log_to_graylog_host, log_to_graylog_port, | |
9864 | fsid, host); | |
9865 | derr << "log_to_monitors " << log_to_monitors << dendl; | |
9866 | } | |
9867 | ||
9868 | void OSD::check_config() | |
9869 | { | |
9870 | // some sanity checks | |
7c673cae FG |
9871 | if (cct->_conf->osd_map_cache_size <= (int)cct->_conf->osd_pg_epoch_persisted_max_stale + 2) { |
9872 | clog->warn() << "osd_map_cache_size (" << cct->_conf->osd_map_cache_size << ")" | |
9873 | << " is not > osd_pg_epoch_persisted_max_stale (" | |
9874 | << cct->_conf->osd_pg_epoch_persisted_max_stale << ")"; | |
9875 | } | |
9f95a23c TL |
9876 | if (cct->_conf->osd_object_clean_region_max_num_intervals < 0) { |
9877 | clog->warn() << "osd_object_clean_region_max_num_intervals (" | |
9878 | << cct->_conf->osd_object_clean_region_max_num_intervals | |
9879 | << ") is < 0"; | |
9880 | } | |
7c673cae FG |
9881 | } |
9882 | ||
7c673cae FG |
9883 | // -------------------------------- |
9884 | ||
9885 | void OSD::get_latest_osdmap() | |
9886 | { | |
9887 | dout(10) << __func__ << " -- start" << dendl; | |
9888 | ||
9889 | C_SaferCond cond; | |
9890 | service.objecter->wait_for_latest_osdmap(&cond); | |
9891 | cond.wait(); | |
9892 | ||
9893 | dout(10) << __func__ << " -- finish" << dendl; | |
9894 | } | |
9895 | ||
9896 | // -------------------------------- | |
9897 | ||
9f95a23c TL |
9898 | void OSD::set_perf_queries(const ConfigPayload &config_payload) { |
9899 | const OSDConfigPayload &osd_config_payload = boost::get<OSDConfigPayload>(config_payload); | |
9900 | const std::map<OSDPerfMetricQuery, OSDPerfMetricLimits> &queries = osd_config_payload.config; | |
11fdf7f2 TL |
9901 | dout(10) << "setting " << queries.size() << " queries" << dendl; |
9902 | ||
9903 | std::list<OSDPerfMetricQuery> supported_queries; | |
9904 | for (auto &it : queries) { | |
9905 | auto &query = it.first; | |
9906 | if (!query.key_descriptor.empty()) { | |
9907 | supported_queries.push_back(query); | |
9908 | } | |
9909 | } | |
9910 | if (supported_queries.size() < queries.size()) { | |
9911 | dout(1) << queries.size() - supported_queries.size() | |
9912 | << " unsupported queries" << dendl; | |
9913 | } | |
11fdf7f2 | 9914 | { |
9f95a23c | 9915 | std::lock_guard locker{m_perf_queries_lock}; |
11fdf7f2 TL |
9916 | m_perf_queries = supported_queries; |
9917 | m_perf_limits = queries; | |
9918 | } | |
11fdf7f2 TL |
9919 | std::vector<PGRef> pgs; |
9920 | _get_pgs(&pgs); | |
9921 | for (auto& pg : pgs) { | |
9f95a23c | 9922 | std::scoped_lock l{*pg}; |
eafe8130 | 9923 | pg->set_dynamic_perf_stats_queries(supported_queries); |
7c673cae | 9924 | } |
7c673cae FG |
9925 | } |
9926 | ||
9f95a23c TL |
9927 | MetricPayload OSD::get_perf_reports() { |
9928 | OSDMetricPayload payload; | |
9929 | std::map<OSDPerfMetricQuery, OSDPerfMetricReport> &reports = payload.report; | |
9930 | ||
11fdf7f2 TL |
9931 | std::vector<PGRef> pgs; |
9932 | _get_pgs(&pgs); | |
9933 | DynamicPerfStats dps; | |
9934 | for (auto& pg : pgs) { | |
eafe8130 TL |
9935 | // m_perf_queries can be modified only in set_perf_queries by mgr client |
9936 | // request, and it is protected by by mgr client's lock, which is held | |
9937 | // when set_perf_queries/get_perf_reports are called, so we may not hold | |
9938 | // m_perf_queries_lock here. | |
9939 | DynamicPerfStats pg_dps(m_perf_queries); | |
9940 | pg->lock(); | |
9941 | pg->get_dynamic_perf_stats(&pg_dps); | |
9942 | pg->unlock(); | |
9943 | dps.merge(pg_dps); | |
11fdf7f2 | 9944 | } |
9f95a23c TL |
9945 | dps.add_to_reports(m_perf_limits, &reports); |
9946 | dout(20) << "reports for " << reports.size() << " queries" << dendl; | |
9947 | ||
9948 | return payload; | |
11fdf7f2 | 9949 | } |
224ce89b | 9950 | |
7c673cae FG |
9951 | // ============================================================= |
9952 | ||
9953 | #undef dout_context | |
11fdf7f2 | 9954 | #define dout_context cct |
7c673cae | 9955 | #undef dout_prefix |
11fdf7f2 | 9956 | #define dout_prefix *_dout << "osd." << osd->get_nodeid() << ":" << shard_id << "." << __func__ << " " |
7c673cae | 9957 | |
11fdf7f2 | 9958 | void OSDShard::_attach_pg(OSDShardPGSlot *slot, PG *pg) |
7c673cae | 9959 | { |
11fdf7f2 TL |
9960 | dout(10) << pg->pg_id << " " << pg << dendl; |
9961 | slot->pg = pg; | |
9962 | pg->osd_shard = this; | |
9963 | pg->pg_slot = slot; | |
9964 | osd->inc_num_pgs(); | |
9965 | ||
9966 | slot->epoch = pg->get_osdmap_epoch(); | |
9967 | pg_slots_by_epoch.insert(*slot); | |
9968 | } | |
9969 | ||
9970 | void OSDShard::_detach_pg(OSDShardPGSlot *slot) | |
9971 | { | |
9972 | dout(10) << slot->pg->pg_id << " " << slot->pg << dendl; | |
9973 | slot->pg->osd_shard = nullptr; | |
9974 | slot->pg->pg_slot = nullptr; | |
9975 | slot->pg = nullptr; | |
9976 | osd->dec_num_pgs(); | |
9977 | ||
9978 | pg_slots_by_epoch.erase(pg_slots_by_epoch.iterator_to(*slot)); | |
9979 | slot->epoch = 0; | |
9980 | if (waiting_for_min_pg_epoch) { | |
9981 | min_pg_epoch_cond.notify_all(); | |
9982 | } | |
9983 | } | |
9984 | ||
9985 | void OSDShard::update_pg_epoch(OSDShardPGSlot *slot, epoch_t e) | |
9986 | { | |
9987 | std::lock_guard l(shard_lock); | |
9988 | dout(30) << "min was " << pg_slots_by_epoch.begin()->epoch | |
9989 | << " on " << pg_slots_by_epoch.begin()->pg->pg_id << dendl; | |
9990 | pg_slots_by_epoch.erase(pg_slots_by_epoch.iterator_to(*slot)); | |
9991 | dout(20) << slot->pg->pg_id << " " << slot->epoch << " -> " << e << dendl; | |
9992 | slot->epoch = e; | |
9993 | pg_slots_by_epoch.insert(*slot); | |
9994 | dout(30) << "min is now " << pg_slots_by_epoch.begin()->epoch | |
9995 | << " on " << pg_slots_by_epoch.begin()->pg->pg_id << dendl; | |
9996 | if (waiting_for_min_pg_epoch) { | |
9997 | min_pg_epoch_cond.notify_all(); | |
9998 | } | |
9999 | } | |
10000 | ||
10001 | epoch_t OSDShard::get_min_pg_epoch() | |
10002 | { | |
10003 | std::lock_guard l(shard_lock); | |
10004 | auto p = pg_slots_by_epoch.begin(); | |
10005 | if (p == pg_slots_by_epoch.end()) { | |
10006 | return 0; | |
10007 | } | |
10008 | return p->epoch; | |
10009 | } | |
10010 | ||
10011 | void OSDShard::wait_min_pg_epoch(epoch_t need) | |
10012 | { | |
10013 | std::unique_lock l{shard_lock}; | |
10014 | ++waiting_for_min_pg_epoch; | |
10015 | min_pg_epoch_cond.wait(l, [need, this] { | |
10016 | if (pg_slots_by_epoch.empty()) { | |
10017 | return true; | |
10018 | } else if (pg_slots_by_epoch.begin()->epoch >= need) { | |
10019 | return true; | |
10020 | } else { | |
10021 | dout(10) << need << " waiting on " | |
10022 | << pg_slots_by_epoch.begin()->epoch << dendl; | |
10023 | return false; | |
10024 | } | |
10025 | }); | |
10026 | --waiting_for_min_pg_epoch; | |
10027 | } | |
10028 | ||
10029 | epoch_t OSDShard::get_max_waiting_epoch() | |
10030 | { | |
10031 | std::lock_guard l(shard_lock); | |
10032 | epoch_t r = 0; | |
10033 | for (auto& i : pg_slots) { | |
10034 | if (!i.second->waiting_peering.empty()) { | |
10035 | r = std::max(r, i.second->waiting_peering.rbegin()->first); | |
10036 | } | |
10037 | } | |
10038 | return r; | |
10039 | } | |
10040 | ||
10041 | void OSDShard::consume_map( | |
9f95a23c | 10042 | const OSDMapRef& new_osdmap, |
11fdf7f2 TL |
10043 | unsigned *pushes_to_free) |
10044 | { | |
10045 | std::lock_guard l(shard_lock); | |
10046 | OSDMapRef old_osdmap; | |
7c673cae | 10047 | { |
11fdf7f2 TL |
10048 | std::lock_guard l(osdmap_lock); |
10049 | old_osdmap = std::move(shard_osdmap); | |
10050 | shard_osdmap = new_osdmap; | |
10051 | } | |
10052 | dout(10) << new_osdmap->get_epoch() | |
10053 | << " (was " << (old_osdmap ? old_osdmap->get_epoch() : 0) << ")" | |
10054 | << dendl; | |
10055 | bool queued = false; | |
10056 | ||
10057 | // check slots | |
10058 | auto p = pg_slots.begin(); | |
10059 | while (p != pg_slots.end()) { | |
10060 | OSDShardPGSlot *slot = p->second.get(); | |
10061 | const spg_t& pgid = p->first; | |
10062 | dout(20) << __func__ << " " << pgid << dendl; | |
10063 | if (!slot->waiting_for_split.empty()) { | |
10064 | dout(20) << __func__ << " " << pgid | |
10065 | << " waiting for split " << slot->waiting_for_split << dendl; | |
10066 | ++p; | |
10067 | continue; | |
10068 | } | |
10069 | if (slot->waiting_for_merge_epoch > new_osdmap->get_epoch()) { | |
10070 | dout(20) << __func__ << " " << pgid | |
10071 | << " waiting for merge by epoch " << slot->waiting_for_merge_epoch | |
10072 | << dendl; | |
10073 | ++p; | |
10074 | continue; | |
10075 | } | |
10076 | if (!slot->waiting_peering.empty()) { | |
10077 | epoch_t first = slot->waiting_peering.begin()->first; | |
10078 | if (first <= new_osdmap->get_epoch()) { | |
10079 | dout(20) << __func__ << " " << pgid | |
10080 | << " pending_peering first epoch " << first | |
10081 | << " <= " << new_osdmap->get_epoch() << ", requeueing" << dendl; | |
10082 | _wake_pg_slot(pgid, slot); | |
10083 | queued = true; | |
10084 | } | |
10085 | ++p; | |
10086 | continue; | |
10087 | } | |
10088 | if (!slot->waiting.empty()) { | |
10089 | if (new_osdmap->is_up_acting_osd_shard(pgid, osd->get_nodeid())) { | |
10090 | dout(20) << __func__ << " " << pgid << " maps to us, keeping" | |
10091 | << dendl; | |
10092 | ++p; | |
10093 | continue; | |
7c673cae | 10094 | } |
11fdf7f2 TL |
10095 | while (!slot->waiting.empty() && |
10096 | slot->waiting.front().get_map_epoch() <= new_osdmap->get_epoch()) { | |
10097 | auto& qi = slot->waiting.front(); | |
10098 | dout(20) << __func__ << " " << pgid | |
10099 | << " waiting item " << qi | |
10100 | << " epoch " << qi.get_map_epoch() | |
10101 | << " <= " << new_osdmap->get_epoch() | |
10102 | << ", " | |
10103 | << (qi.get_map_epoch() < new_osdmap->get_epoch() ? "stale" : | |
10104 | "misdirected") | |
10105 | << ", dropping" << dendl; | |
10106 | *pushes_to_free += qi.get_reserved_pushes(); | |
10107 | slot->waiting.pop_front(); | |
10108 | } | |
10109 | } | |
10110 | if (slot->waiting.empty() && | |
10111 | slot->num_running == 0 && | |
10112 | slot->waiting_for_split.empty() && | |
10113 | !slot->pg) { | |
10114 | dout(20) << __func__ << " " << pgid << " empty, pruning" << dendl; | |
10115 | p = pg_slots.erase(p); | |
10116 | continue; | |
7c673cae | 10117 | } |
11fdf7f2 TL |
10118 | |
10119 | ++p; | |
7c673cae | 10120 | } |
7c673cae | 10121 | if (queued) { |
11fdf7f2 TL |
10122 | std::lock_guard l{sdata_wait_lock}; |
10123 | sdata_cond.notify_one(); | |
7c673cae FG |
10124 | } |
10125 | } | |
10126 | ||
11fdf7f2 TL |
10127 | void OSDShard::_wake_pg_slot( |
10128 | spg_t pgid, | |
10129 | OSDShardPGSlot *slot) | |
10130 | { | |
10131 | dout(20) << __func__ << " " << pgid | |
10132 | << " to_process " << slot->to_process | |
10133 | << " waiting " << slot->waiting | |
10134 | << " waiting_peering " << slot->waiting_peering << dendl; | |
10135 | for (auto i = slot->to_process.rbegin(); | |
10136 | i != slot->to_process.rend(); | |
10137 | ++i) { | |
9f95a23c | 10138 | scheduler->enqueue_front(std::move(*i)); |
11fdf7f2 TL |
10139 | } |
10140 | slot->to_process.clear(); | |
10141 | for (auto i = slot->waiting.rbegin(); | |
10142 | i != slot->waiting.rend(); | |
10143 | ++i) { | |
9f95a23c | 10144 | scheduler->enqueue_front(std::move(*i)); |
11fdf7f2 TL |
10145 | } |
10146 | slot->waiting.clear(); | |
10147 | for (auto i = slot->waiting_peering.rbegin(); | |
10148 | i != slot->waiting_peering.rend(); | |
10149 | ++i) { | |
10150 | // this is overkill; we requeue everything, even if some of these | |
10151 | // items are waiting for maps we don't have yet. FIXME, maybe, | |
10152 | // someday, if we decide this inefficiency matters | |
10153 | for (auto j = i->second.rbegin(); j != i->second.rend(); ++j) { | |
9f95a23c | 10154 | scheduler->enqueue_front(std::move(*j)); |
11fdf7f2 TL |
10155 | } |
10156 | } | |
10157 | slot->waiting_peering.clear(); | |
10158 | ++slot->requeue_seq; | |
10159 | } | |
10160 | ||
10161 | void OSDShard::identify_splits_and_merges( | |
10162 | const OSDMapRef& as_of_osdmap, | |
10163 | set<pair<spg_t,epoch_t>> *split_pgs, | |
10164 | set<pair<spg_t,epoch_t>> *merge_pgs) | |
10165 | { | |
10166 | std::lock_guard l(shard_lock); | |
10167 | if (shard_osdmap) { | |
10168 | for (auto& i : pg_slots) { | |
10169 | const spg_t& pgid = i.first; | |
10170 | auto *slot = i.second.get(); | |
10171 | if (slot->pg) { | |
10172 | osd->service.identify_splits_and_merges( | |
10173 | shard_osdmap, as_of_osdmap, pgid, | |
10174 | split_pgs, merge_pgs); | |
10175 | } else if (!slot->waiting_for_split.empty()) { | |
10176 | osd->service.identify_splits_and_merges( | |
10177 | shard_osdmap, as_of_osdmap, pgid, | |
10178 | split_pgs, nullptr); | |
10179 | } else { | |
10180 | dout(20) << __func__ << " slot " << pgid | |
9f95a23c | 10181 | << " has no pg and waiting_for_split " << dendl; |
7c673cae | 10182 | } |
11fdf7f2 TL |
10183 | } |
10184 | } | |
10185 | } | |
10186 | ||
10187 | void OSDShard::prime_splits(const OSDMapRef& as_of_osdmap, | |
10188 | set<pair<spg_t,epoch_t>> *pgids) | |
10189 | { | |
10190 | std::lock_guard l(shard_lock); | |
10191 | _prime_splits(pgids); | |
10192 | if (shard_osdmap->get_epoch() > as_of_osdmap->get_epoch()) { | |
10193 | set<pair<spg_t,epoch_t>> newer_children; | |
10194 | for (auto i : *pgids) { | |
10195 | osd->service.identify_splits_and_merges( | |
10196 | as_of_osdmap, shard_osdmap, i.first, | |
10197 | &newer_children, nullptr); | |
10198 | } | |
10199 | newer_children.insert(pgids->begin(), pgids->end()); | |
10200 | dout(10) << "as_of_osdmap " << as_of_osdmap->get_epoch() << " < shard " | |
10201 | << shard_osdmap->get_epoch() << ", new children " << newer_children | |
10202 | << dendl; | |
10203 | _prime_splits(&newer_children); | |
10204 | // note: we don't care what is left over here for other shards. | |
10205 | // if this shard is ahead of us and one isn't, e.g., one thread is | |
10206 | // calling into prime_splits via _process (due to a newly created | |
10207 | // pg) and this shard has a newer map due to a racing consume_map, | |
10208 | // then any grandchildren left here will be identified (or were | |
10209 | // identified) when the slower shard's osdmap is advanced. | |
10210 | // _prime_splits() will tolerate the case where the pgid is | |
10211 | // already primed. | |
10212 | } | |
10213 | } | |
10214 | ||
10215 | void OSDShard::_prime_splits(set<pair<spg_t,epoch_t>> *pgids) | |
10216 | { | |
10217 | dout(10) << *pgids << dendl; | |
10218 | auto p = pgids->begin(); | |
10219 | while (p != pgids->end()) { | |
10220 | unsigned shard_index = p->first.hash_to_shard(osd->num_shards); | |
10221 | if (shard_index == shard_id) { | |
10222 | auto r = pg_slots.emplace(p->first, nullptr); | |
10223 | if (r.second) { | |
10224 | dout(10) << "priming slot " << p->first << " e" << p->second << dendl; | |
10225 | r.first->second = make_unique<OSDShardPGSlot>(); | |
10226 | r.first->second->waiting_for_split.insert(p->second); | |
7c673cae | 10227 | } else { |
11fdf7f2 TL |
10228 | auto q = r.first; |
10229 | ceph_assert(q != pg_slots.end()); | |
10230 | dout(10) << "priming (existing) slot " << p->first << " e" << p->second | |
10231 | << dendl; | |
10232 | q->second->waiting_for_split.insert(p->second); | |
7c673cae | 10233 | } |
11fdf7f2 TL |
10234 | p = pgids->erase(p); |
10235 | } else { | |
10236 | ++p; | |
7c673cae FG |
10237 | } |
10238 | } | |
11fdf7f2 TL |
10239 | } |
10240 | ||
10241 | void OSDShard::prime_merges(const OSDMapRef& as_of_osdmap, | |
10242 | set<pair<spg_t,epoch_t>> *merge_pgs) | |
10243 | { | |
10244 | std::lock_guard l(shard_lock); | |
10245 | dout(20) << __func__ << " checking shard " << shard_id | |
10246 | << " for remaining merge pgs " << merge_pgs << dendl; | |
10247 | auto p = merge_pgs->begin(); | |
10248 | while (p != merge_pgs->end()) { | |
10249 | spg_t pgid = p->first; | |
10250 | epoch_t epoch = p->second; | |
10251 | unsigned shard_index = pgid.hash_to_shard(osd->num_shards); | |
10252 | if (shard_index != shard_id) { | |
10253 | ++p; | |
10254 | continue; | |
10255 | } | |
10256 | OSDShardPGSlot *slot; | |
10257 | auto r = pg_slots.emplace(pgid, nullptr); | |
10258 | if (r.second) { | |
10259 | r.first->second = make_unique<OSDShardPGSlot>(); | |
10260 | } | |
10261 | slot = r.first->second.get(); | |
10262 | if (slot->pg) { | |
10263 | // already have pg | |
10264 | dout(20) << __func__ << " have merge participant pg " << pgid | |
10265 | << " " << slot->pg << dendl; | |
10266 | } else if (!slot->waiting_for_split.empty() && | |
10267 | *slot->waiting_for_split.begin() < epoch) { | |
10268 | dout(20) << __func__ << " pending split on merge participant pg " << pgid | |
10269 | << " " << slot->waiting_for_split << dendl; | |
10270 | } else { | |
10271 | dout(20) << __func__ << " creating empty merge participant " << pgid | |
10272 | << " for merge in " << epoch << dendl; | |
10273 | // leave history zeroed; PG::merge_from() will fill it in. | |
10274 | pg_history_t history; | |
10275 | PGCreateInfo cinfo(pgid, epoch - 1, | |
10276 | history, PastIntervals(), false); | |
10277 | PGRef pg = osd->handle_pg_create_info(shard_osdmap, &cinfo); | |
10278 | _attach_pg(r.first->second.get(), pg.get()); | |
10279 | _wake_pg_slot(pgid, slot); | |
10280 | pg->unlock(); | |
10281 | } | |
10282 | // mark slot for merge | |
10283 | dout(20) << __func__ << " marking merge participant " << pgid << dendl; | |
10284 | slot->waiting_for_merge_epoch = epoch; | |
10285 | p = merge_pgs->erase(p); | |
7c673cae FG |
10286 | } |
10287 | } | |
10288 | ||
11fdf7f2 | 10289 | void OSDShard::register_and_wake_split_child(PG *pg) |
7c673cae | 10290 | { |
11fdf7f2 TL |
10291 | epoch_t epoch; |
10292 | { | |
10293 | std::lock_guard l(shard_lock); | |
10294 | dout(10) << pg->pg_id << " " << pg << dendl; | |
10295 | auto p = pg_slots.find(pg->pg_id); | |
10296 | ceph_assert(p != pg_slots.end()); | |
10297 | auto *slot = p->second.get(); | |
10298 | dout(20) << pg->pg_id << " waiting_for_split " << slot->waiting_for_split | |
10299 | << dendl; | |
10300 | ceph_assert(!slot->pg); | |
10301 | ceph_assert(!slot->waiting_for_split.empty()); | |
10302 | _attach_pg(slot, pg); | |
10303 | ||
10304 | epoch = pg->get_osdmap_epoch(); | |
10305 | ceph_assert(slot->waiting_for_split.count(epoch)); | |
10306 | slot->waiting_for_split.erase(epoch); | |
10307 | if (slot->waiting_for_split.empty()) { | |
10308 | _wake_pg_slot(pg->pg_id, slot); | |
10309 | } else { | |
10310 | dout(10) << __func__ << " still waiting for split on " | |
10311 | << slot->waiting_for_split << dendl; | |
10312 | } | |
7c673cae | 10313 | } |
11fdf7f2 TL |
10314 | |
10315 | // kick child to ensure it pulls up to the latest osdmap | |
10316 | osd->enqueue_peering_evt( | |
10317 | pg->pg_id, | |
10318 | PGPeeringEventRef( | |
10319 | std::make_shared<PGPeeringEvent>( | |
10320 | epoch, | |
10321 | epoch, | |
10322 | NullEvt()))); | |
10323 | ||
10324 | std::lock_guard l{sdata_wait_lock}; | |
10325 | sdata_cond.notify_one(); | |
7c673cae FG |
10326 | } |
10327 | ||
11fdf7f2 | 10328 | void OSDShard::unprime_split_children(spg_t parent, unsigned old_pg_num) |
7c673cae | 10329 | { |
11fdf7f2 TL |
10330 | std::lock_guard l(shard_lock); |
10331 | vector<spg_t> to_delete; | |
10332 | for (auto& i : pg_slots) { | |
10333 | if (i.first != parent && | |
10334 | i.first.get_ancestor(old_pg_num) == parent) { | |
10335 | dout(10) << __func__ << " parent " << parent << " clearing " << i.first | |
10336 | << dendl; | |
10337 | _wake_pg_slot(i.first, i.second.get()); | |
10338 | to_delete.push_back(i.first); | |
10339 | } | |
10340 | } | |
10341 | for (auto pgid : to_delete) { | |
10342 | pg_slots.erase(pgid); | |
10343 | } | |
10344 | } | |
10345 | ||
9f95a23c TL |
10346 | OSDShard::OSDShard( |
10347 | int id, | |
10348 | CephContext *cct, | |
10349 | OSD *osd) | |
10350 | : shard_id(id), | |
10351 | cct(cct), | |
10352 | osd(osd), | |
10353 | shard_name(string("OSDShard.") + stringify(id)), | |
10354 | sdata_wait_lock_name(shard_name + "::sdata_wait_lock"), | |
10355 | sdata_wait_lock{make_mutex(sdata_wait_lock_name)}, | |
10356 | osdmap_lock{make_mutex(shard_name + "::osdmap_lock")}, | |
10357 | shard_lock_name(shard_name + "::shard_lock"), | |
10358 | shard_lock{make_mutex(shard_lock_name)}, | |
10359 | scheduler(ceph::osd::scheduler::make_scheduler(cct)), | |
10360 | context_queue(sdata_wait_lock, sdata_cond) | |
10361 | { | |
10362 | dout(0) << "using op scheduler " << *scheduler << dendl; | |
10363 | } | |
10364 | ||
11fdf7f2 TL |
10365 | |
10366 | // ============================================================= | |
10367 | ||
10368 | #undef dout_context | |
10369 | #define dout_context osd->cct | |
10370 | #undef dout_prefix | |
10371 | #define dout_prefix *_dout << "osd." << osd->whoami << " op_wq " | |
10372 | ||
10373 | void OSD::ShardedOpWQ::_add_slot_waiter( | |
10374 | spg_t pgid, | |
10375 | OSDShardPGSlot *slot, | |
9f95a23c | 10376 | OpSchedulerItem&& qi) |
11fdf7f2 TL |
10377 | { |
10378 | if (qi.is_peering()) { | |
10379 | dout(20) << __func__ << " " << pgid | |
10380 | << " peering, item epoch is " | |
10381 | << qi.get_map_epoch() | |
10382 | << ", will wait on " << qi << dendl; | |
10383 | slot->waiting_peering[qi.get_map_epoch()].push_back(std::move(qi)); | |
10384 | } else { | |
10385 | dout(20) << __func__ << " " << pgid | |
10386 | << " item epoch is " | |
10387 | << qi.get_map_epoch() | |
10388 | << ", will wait on " << qi << dendl; | |
10389 | slot->waiting.push_back(std::move(qi)); | |
7c673cae FG |
10390 | } |
10391 | } | |
10392 | ||
10393 | #undef dout_prefix | |
10394 | #define dout_prefix *_dout << "osd." << osd->whoami << " op_wq(" << shard_index << ") " | |
10395 | ||
10396 | void OSD::ShardedOpWQ::_process(uint32_t thread_index, heartbeat_handle_d *hb) | |
10397 | { | |
11fdf7f2 TL |
10398 | uint32_t shard_index = thread_index % osd->num_shards; |
10399 | auto& sdata = osd->shards[shard_index]; | |
10400 | ceph_assert(sdata); | |
10401 | ||
10402 | // If all threads of shards do oncommits, there is a out-of-order | |
10403 | // problem. So we choose the thread which has the smallest | |
10404 | // thread_index(thread_index < num_shards) of shard to do oncommit | |
10405 | // callback. | |
10406 | bool is_smallest_thread_index = thread_index < osd->num_shards; | |
7c673cae FG |
10407 | |
10408 | // peek at spg_t | |
11fdf7f2 | 10409 | sdata->shard_lock.lock(); |
9f95a23c | 10410 | if (sdata->scheduler->empty() && |
11fdf7f2 TL |
10411 | (!is_smallest_thread_index || sdata->context_queue.empty())) { |
10412 | std::unique_lock wait_lock{sdata->sdata_wait_lock}; | |
10413 | if (is_smallest_thread_index && !sdata->context_queue.empty()) { | |
10414 | // we raced with a context_queue addition, don't wait | |
10415 | wait_lock.unlock(); | |
10416 | } else if (!sdata->stop_waiting) { | |
10417 | dout(20) << __func__ << " empty q, waiting" << dendl; | |
10418 | osd->cct->get_heartbeat_map()->clear_timeout(hb); | |
10419 | sdata->shard_lock.unlock(); | |
10420 | sdata->sdata_cond.wait(wait_lock); | |
10421 | wait_lock.unlock(); | |
10422 | sdata->shard_lock.lock(); | |
9f95a23c | 10423 | if (sdata->scheduler->empty() && |
11fdf7f2 TL |
10424 | !(is_smallest_thread_index && !sdata->context_queue.empty())) { |
10425 | sdata->shard_lock.unlock(); | |
10426 | return; | |
10427 | } | |
10428 | osd->cct->get_heartbeat_map()->reset_timeout(hb, | |
10429 | osd->cct->_conf->threadpool_default_timeout, 0); | |
10430 | } else { | |
10431 | dout(20) << __func__ << " need return immediately" << dendl; | |
10432 | wait_lock.unlock(); | |
10433 | sdata->shard_lock.unlock(); | |
7c673cae FG |
10434 | return; |
10435 | } | |
10436 | } | |
11fdf7f2 TL |
10437 | |
10438 | list<Context *> oncommits; | |
9f95a23c TL |
10439 | if (is_smallest_thread_index) { |
10440 | sdata->context_queue.move_to(oncommits); | |
7c673cae | 10441 | } |
11fdf7f2 | 10442 | |
9f95a23c | 10443 | if (sdata->scheduler->empty()) { |
11fdf7f2 TL |
10444 | if (osd->is_stopping()) { |
10445 | sdata->shard_lock.unlock(); | |
10446 | for (auto c : oncommits) { | |
10447 | dout(10) << __func__ << " discarding in-flight oncommit " << c << dendl; | |
10448 | delete c; | |
10449 | } | |
10450 | return; // OSD shutdown, discard. | |
7c673cae | 10451 | } |
11fdf7f2 TL |
10452 | sdata->shard_lock.unlock(); |
10453 | handle_oncommits(oncommits); | |
10454 | return; | |
7c673cae | 10455 | } |
7c673cae | 10456 | |
9f95a23c | 10457 | OpSchedulerItem item = sdata->scheduler->dequeue(); |
11fdf7f2 TL |
10458 | if (osd->is_stopping()) { |
10459 | sdata->shard_lock.unlock(); | |
10460 | for (auto c : oncommits) { | |
10461 | dout(10) << __func__ << " discarding in-flight oncommit " << c << dendl; | |
10462 | delete c; | |
10463 | } | |
10464 | return; // OSD shutdown, discard. | |
10465 | } | |
7c673cae | 10466 | |
11fdf7f2 TL |
10467 | const auto token = item.get_ordering_token(); |
10468 | auto r = sdata->pg_slots.emplace(token, nullptr); | |
10469 | if (r.second) { | |
10470 | r.first->second = make_unique<OSDShardPGSlot>(); | |
7c673cae | 10471 | } |
11fdf7f2 TL |
10472 | OSDShardPGSlot *slot = r.first->second.get(); |
10473 | dout(20) << __func__ << " " << token | |
10474 | << (r.second ? " (new)" : "") | |
10475 | << " to_process " << slot->to_process | |
10476 | << " waiting " << slot->waiting | |
10477 | << " waiting_peering " << slot->waiting_peering | |
10478 | << dendl; | |
10479 | slot->to_process.push_back(std::move(item)); | |
10480 | dout(20) << __func__ << " " << slot->to_process.back() | |
10481 | << " queued" << dendl; | |
7c673cae | 10482 | |
11fdf7f2 TL |
10483 | retry_pg: |
10484 | PGRef pg = slot->pg; | |
7c673cae | 10485 | |
11fdf7f2 TL |
10486 | // lock pg (if we have it) |
10487 | if (pg) { | |
10488 | // note the requeue seq now... | |
10489 | uint64_t requeue_seq = slot->requeue_seq; | |
10490 | ++slot->num_running; | |
7c673cae | 10491 | |
11fdf7f2 TL |
10492 | sdata->shard_lock.unlock(); |
10493 | osd->service.maybe_inject_dispatch_delay(); | |
10494 | pg->lock(); | |
10495 | osd->service.maybe_inject_dispatch_delay(); | |
10496 | sdata->shard_lock.lock(); | |
7c673cae | 10497 | |
11fdf7f2 TL |
10498 | auto q = sdata->pg_slots.find(token); |
10499 | if (q == sdata->pg_slots.end()) { | |
10500 | // this can happen if we race with pg removal. | |
10501 | dout(20) << __func__ << " slot " << token << " no longer there" << dendl; | |
10502 | pg->unlock(); | |
10503 | sdata->shard_lock.unlock(); | |
10504 | handle_oncommits(oncommits); | |
10505 | return; | |
10506 | } | |
10507 | slot = q->second.get(); | |
10508 | --slot->num_running; | |
7c673cae | 10509 | |
11fdf7f2 TL |
10510 | if (slot->to_process.empty()) { |
10511 | // raced with _wake_pg_slot or consume_map | |
10512 | dout(20) << __func__ << " " << token | |
10513 | << " nothing queued" << dendl; | |
7c673cae | 10514 | pg->unlock(); |
11fdf7f2 TL |
10515 | sdata->shard_lock.unlock(); |
10516 | handle_oncommits(oncommits); | |
10517 | return; | |
7c673cae | 10518 | } |
11fdf7f2 TL |
10519 | if (requeue_seq != slot->requeue_seq) { |
10520 | dout(20) << __func__ << " " << token | |
10521 | << " requeue_seq " << slot->requeue_seq << " > our " | |
10522 | << requeue_seq << ", we raced with _wake_pg_slot" | |
10523 | << dendl; | |
7c673cae | 10524 | pg->unlock(); |
11fdf7f2 TL |
10525 | sdata->shard_lock.unlock(); |
10526 | handle_oncommits(oncommits); | |
10527 | return; | |
7c673cae | 10528 | } |
11fdf7f2 TL |
10529 | if (slot->pg != pg) { |
10530 | // this can happen if we race with pg removal. | |
10531 | dout(20) << __func__ << " slot " << token << " no longer attached to " | |
10532 | << pg << dendl; | |
7c673cae | 10533 | pg->unlock(); |
11fdf7f2 | 10534 | goto retry_pg; |
7c673cae | 10535 | } |
7c673cae FG |
10536 | } |
10537 | ||
11fdf7f2 TL |
10538 | dout(20) << __func__ << " " << token |
10539 | << " to_process " << slot->to_process | |
10540 | << " waiting " << slot->waiting | |
10541 | << " waiting_peering " << slot->waiting_peering << dendl; | |
10542 | ||
10543 | ThreadPool::TPHandle tp_handle(osd->cct, hb, timeout_interval, | |
10544 | suicide_interval); | |
10545 | ||
7c673cae | 10546 | // take next item |
11fdf7f2 TL |
10547 | auto qi = std::move(slot->to_process.front()); |
10548 | slot->to_process.pop_front(); | |
10549 | dout(20) << __func__ << " " << qi << " pg " << pg << dendl; | |
10550 | set<pair<spg_t,epoch_t>> new_children; | |
10551 | OSDMapRef osdmap; | |
7c673cae | 10552 | |
11fdf7f2 | 10553 | while (!pg) { |
7c673cae | 10554 | // should this pg shard exist on this osd in this (or a later) epoch? |
11fdf7f2 TL |
10555 | osdmap = sdata->shard_osdmap; |
10556 | const PGCreateInfo *create_info = qi.creates_pg(); | |
10557 | if (!slot->waiting_for_split.empty()) { | |
10558 | dout(20) << __func__ << " " << token | |
10559 | << " splitting " << slot->waiting_for_split << dendl; | |
10560 | _add_slot_waiter(token, slot, std::move(qi)); | |
10561 | } else if (qi.get_map_epoch() > osdmap->get_epoch()) { | |
10562 | dout(20) << __func__ << " " << token | |
10563 | << " map " << qi.get_map_epoch() << " > " | |
10564 | << osdmap->get_epoch() << dendl; | |
10565 | _add_slot_waiter(token, slot, std::move(qi)); | |
10566 | } else if (qi.is_peering()) { | |
10567 | if (!qi.peering_requires_pg()) { | |
10568 | // for pg-less events, we run them under the ordering lock, since | |
10569 | // we don't have the pg lock to keep them ordered. | |
10570 | qi.run(osd, sdata, pg, tp_handle); | |
10571 | } else if (osdmap->is_up_acting_osd_shard(token, osd->whoami)) { | |
10572 | if (create_info) { | |
10573 | if (create_info->by_mon && | |
10574 | osdmap->get_pg_acting_primary(token.pgid) != osd->whoami) { | |
10575 | dout(20) << __func__ << " " << token | |
10576 | << " no pg, no longer primary, ignoring mon create on " | |
10577 | << qi << dendl; | |
10578 | } else { | |
10579 | dout(20) << __func__ << " " << token | |
10580 | << " no pg, should create on " << qi << dendl; | |
10581 | pg = osd->handle_pg_create_info(osdmap, create_info); | |
10582 | if (pg) { | |
10583 | // we created the pg! drop out and continue "normally"! | |
10584 | sdata->_attach_pg(slot, pg.get()); | |
10585 | sdata->_wake_pg_slot(token, slot); | |
10586 | ||
10587 | // identify split children between create epoch and shard epoch. | |
10588 | osd->service.identify_splits_and_merges( | |
10589 | pg->get_osdmap(), osdmap, pg->pg_id, &new_children, nullptr); | |
10590 | sdata->_prime_splits(&new_children); | |
10591 | // distribute remaining split children to other shards below! | |
10592 | break; | |
10593 | } | |
10594 | dout(20) << __func__ << " ignored create on " << qi << dendl; | |
10595 | } | |
10596 | } else { | |
10597 | dout(20) << __func__ << " " << token | |
10598 | << " no pg, peering, !create, discarding " << qi << dendl; | |
10599 | } | |
10600 | } else { | |
10601 | dout(20) << __func__ << " " << token | |
10602 | << " no pg, peering, doesn't map here e" << osdmap->get_epoch() | |
10603 | << ", discarding " << qi | |
10604 | << dendl; | |
10605 | } | |
10606 | } else if (osdmap->is_up_acting_osd_shard(token, osd->whoami)) { | |
10607 | dout(20) << __func__ << " " << token | |
10608 | << " no pg, should exist e" << osdmap->get_epoch() | |
10609 | << ", will wait on " << qi << dendl; | |
10610 | _add_slot_waiter(token, slot, std::move(qi)); | |
7c673cae | 10611 | } else { |
11fdf7f2 TL |
10612 | dout(20) << __func__ << " " << token |
10613 | << " no pg, shouldn't exist e" << osdmap->get_epoch() | |
10614 | << ", dropping " << qi << dendl; | |
7c673cae | 10615 | // share map with client? |
9f95a23c TL |
10616 | if (std::optional<OpRequestRef> _op = qi.maybe_get_op()) { |
10617 | osd->service.maybe_share_map((*_op)->get_req()->get_connection().get(), | |
10618 | sdata->shard_osdmap, | |
10619 | (*_op)->sent_epoch); | |
7c673cae | 10620 | } |
11fdf7f2 | 10621 | unsigned pushes_to_free = qi.get_reserved_pushes(); |
7c673cae | 10622 | if (pushes_to_free > 0) { |
11fdf7f2 | 10623 | sdata->shard_lock.unlock(); |
7c673cae | 10624 | osd->service.release_reserved_pushes(pushes_to_free); |
11fdf7f2 | 10625 | handle_oncommits(oncommits); |
7c673cae FG |
10626 | return; |
10627 | } | |
10628 | } | |
11fdf7f2 TL |
10629 | sdata->shard_lock.unlock(); |
10630 | handle_oncommits(oncommits); | |
7c673cae FG |
10631 | return; |
10632 | } | |
11fdf7f2 TL |
10633 | if (qi.is_peering()) { |
10634 | OSDMapRef osdmap = sdata->shard_osdmap; | |
10635 | if (qi.get_map_epoch() > osdmap->get_epoch()) { | |
10636 | _add_slot_waiter(token, slot, std::move(qi)); | |
10637 | sdata->shard_lock.unlock(); | |
10638 | pg->unlock(); | |
10639 | handle_oncommits(oncommits); | |
10640 | return; | |
10641 | } | |
10642 | } | |
10643 | sdata->shard_lock.unlock(); | |
7c673cae | 10644 | |
11fdf7f2 TL |
10645 | if (!new_children.empty()) { |
10646 | for (auto shard : osd->shards) { | |
10647 | shard->prime_splits(osdmap, &new_children); | |
10648 | } | |
10649 | ceph_assert(new_children.empty()); | |
10650 | } | |
7c673cae FG |
10651 | |
10652 | // osd_opwq_process marks the point at which an operation has been dequeued | |
10653 | // and will begin to be handled by a worker thread. | |
10654 | { | |
10655 | #ifdef WITH_LTTNG | |
10656 | osd_reqid_t reqid; | |
9f95a23c | 10657 | if (std::optional<OpRequestRef> _op = qi.maybe_get_op()) { |
7c673cae FG |
10658 | reqid = (*_op)->get_reqid(); |
10659 | } | |
10660 | #endif | |
10661 | tracepoint(osd, opwq_process_start, reqid.name._type, | |
10662 | reqid.name._num, reqid.tid, reqid.inc); | |
10663 | } | |
10664 | ||
10665 | lgeneric_subdout(osd->cct, osd, 30) << "dequeue status: "; | |
10666 | Formatter *f = Formatter::create("json"); | |
10667 | f->open_object_section("q"); | |
10668 | dump(f); | |
10669 | f->close_section(); | |
10670 | f->flush(*_dout); | |
10671 | delete f; | |
10672 | *_dout << dendl; | |
10673 | ||
11fdf7f2 | 10674 | qi.run(osd, sdata, pg, tp_handle); |
7c673cae FG |
10675 | |
10676 | { | |
10677 | #ifdef WITH_LTTNG | |
10678 | osd_reqid_t reqid; | |
9f95a23c | 10679 | if (std::optional<OpRequestRef> _op = qi.maybe_get_op()) { |
7c673cae FG |
10680 | reqid = (*_op)->get_reqid(); |
10681 | } | |
10682 | #endif | |
10683 | tracepoint(osd, opwq_process_finish, reqid.name._type, | |
10684 | reqid.name._num, reqid.tid, reqid.inc); | |
10685 | } | |
10686 | ||
11fdf7f2 | 10687 | handle_oncommits(oncommits); |
7c673cae FG |
10688 | } |
10689 | ||
9f95a23c | 10690 | void OSD::ShardedOpWQ::_enqueue(OpSchedulerItem&& item) { |
7c673cae | 10691 | uint32_t shard_index = |
11fdf7f2 | 10692 | item.get_ordering_token().hash_to_shard(osd->shards.size()); |
7c673cae | 10693 | |
9f95a23c TL |
10694 | dout(20) << __func__ << " " << item << dendl; |
10695 | ||
11fdf7f2 | 10696 | OSDShard* sdata = osd->shards[shard_index]; |
7c673cae | 10697 | assert (NULL != sdata); |
7c673cae | 10698 | |
9f95a23c TL |
10699 | bool empty = true; |
10700 | { | |
10701 | std::lock_guard l{sdata->shard_lock}; | |
10702 | empty = sdata->scheduler->empty(); | |
10703 | sdata->scheduler->enqueue(std::move(item)); | |
10704 | } | |
7c673cae | 10705 | |
9f95a23c TL |
10706 | if (empty) { |
10707 | std::lock_guard l{sdata->sdata_wait_lock}; | |
10708 | sdata->sdata_cond.notify_one(); | |
10709 | } | |
7c673cae FG |
10710 | } |
10711 | ||
9f95a23c | 10712 | void OSD::ShardedOpWQ::_enqueue_front(OpSchedulerItem&& item) |
7c673cae | 10713 | { |
11fdf7f2 TL |
10714 | auto shard_index = item.get_ordering_token().hash_to_shard(osd->shards.size()); |
10715 | auto& sdata = osd->shards[shard_index]; | |
10716 | ceph_assert(sdata); | |
10717 | sdata->shard_lock.lock(); | |
10718 | auto p = sdata->pg_slots.find(item.get_ordering_token()); | |
10719 | if (p != sdata->pg_slots.end() && | |
10720 | !p->second->to_process.empty()) { | |
7c673cae | 10721 | // we may be racing with _process, which has dequeued a new item |
9f95a23c | 10722 | // from scheduler, put it on to_process, and is now busy taking the |
7c673cae FG |
10723 | // pg lock. ensure this old requeued item is ordered before any |
10724 | // such newer item in to_process. | |
11fdf7f2 TL |
10725 | p->second->to_process.push_front(std::move(item)); |
10726 | item = std::move(p->second->to_process.back()); | |
10727 | p->second->to_process.pop_back(); | |
10728 | dout(20) << __func__ | |
10729 | << " " << p->second->to_process.front() | |
10730 | << " shuffled w/ " << item << dendl; | |
7c673cae | 10731 | } else { |
11fdf7f2 | 10732 | dout(20) << __func__ << " " << item << dendl; |
7c673cae | 10733 | } |
9f95a23c | 10734 | sdata->scheduler->enqueue_front(std::move(item)); |
11fdf7f2 TL |
10735 | sdata->shard_lock.unlock(); |
10736 | std::lock_guard l{sdata->sdata_wait_lock}; | |
10737 | sdata->sdata_cond.notify_one(); | |
7c673cae FG |
10738 | } |
10739 | ||
10740 | namespace ceph { | |
10741 | namespace osd_cmds { | |
10742 | ||
11fdf7f2 TL |
10743 | int heap(CephContext& cct, const cmdmap_t& cmdmap, Formatter& f, |
10744 | std::ostream& os) | |
7c673cae FG |
10745 | { |
10746 | if (!ceph_using_tcmalloc()) { | |
10747 | os << "could not issue heap profiler command -- not using tcmalloc!"; | |
10748 | return -EOPNOTSUPP; | |
10749 | } | |
10750 | ||
10751 | string cmd; | |
9f95a23c | 10752 | if (!cmd_getval(cmdmap, "heapcmd", cmd)) { |
7c673cae FG |
10753 | os << "unable to get value for command \"" << cmd << "\""; |
10754 | return -EINVAL; | |
11fdf7f2 | 10755 | } |
7c673cae FG |
10756 | |
10757 | std::vector<std::string> cmd_vec; | |
10758 | get_str_vec(cmd, cmd_vec); | |
11fdf7f2 TL |
10759 | |
10760 | string val; | |
9f95a23c | 10761 | if (cmd_getval(cmdmap, "value", val)) { |
11fdf7f2 TL |
10762 | cmd_vec.push_back(val); |
10763 | } | |
7c673cae FG |
10764 | |
10765 | ceph_heap_profiler_handle_command(cmd_vec, os); | |
10766 | ||
10767 | return 0; | |
10768 | } | |
10769 | ||
10770 | }} // namespace ceph::osd_cmds |