]>
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> | |
7c673cae FG |
26 | #include <boost/scoped_ptr.hpp> |
27 | ||
28 | #ifdef HAVE_SYS_PARAM_H | |
29 | #include <sys/param.h> | |
30 | #endif | |
31 | ||
32 | #ifdef HAVE_SYS_MOUNT_H | |
33 | #include <sys/mount.h> | |
34 | #endif | |
35 | ||
36 | #include "osd/PG.h" | |
37 | ||
38 | #include "include/types.h" | |
39 | #include "include/compat.h" | |
11fdf7f2 | 40 | #include "include/random.h" |
7c673cae FG |
41 | |
42 | #include "OSD.h" | |
43 | #include "OSDMap.h" | |
44 | #include "Watch.h" | |
45 | #include "osdc/Objecter.h" | |
46 | ||
47 | #include "common/errno.h" | |
48 | #include "common/ceph_argparse.h" | |
224ce89b | 49 | #include "common/ceph_time.h" |
7c673cae | 50 | #include "common/version.h" |
b5b8bbf5 | 51 | #include "common/pick_address.h" |
11fdf7f2 TL |
52 | #include "common/blkdev.h" |
53 | #include "common/numa.h" | |
7c673cae FG |
54 | |
55 | #include "os/ObjectStore.h" | |
56 | #ifdef HAVE_LIBFUSE | |
57 | #include "os/FuseStore.h" | |
58 | #endif | |
59 | ||
60 | #include "PrimaryLogPG.h" | |
61 | ||
7c673cae FG |
62 | #include "msg/Messenger.h" |
63 | #include "msg/Message.h" | |
64 | ||
65 | #include "mon/MonClient.h" | |
66 | ||
67 | #include "messages/MLog.h" | |
68 | ||
69 | #include "messages/MGenericMessage.h" | |
7c673cae FG |
70 | #include "messages/MOSDPing.h" |
71 | #include "messages/MOSDFailure.h" | |
72 | #include "messages/MOSDMarkMeDown.h" | |
73 | #include "messages/MOSDFull.h" | |
74 | #include "messages/MOSDOp.h" | |
75 | #include "messages/MOSDOpReply.h" | |
76 | #include "messages/MOSDBackoff.h" | |
77 | #include "messages/MOSDBeacon.h" | |
78 | #include "messages/MOSDRepOp.h" | |
79 | #include "messages/MOSDRepOpReply.h" | |
80 | #include "messages/MOSDBoot.h" | |
81 | #include "messages/MOSDPGTemp.h" | |
11fdf7f2 | 82 | #include "messages/MOSDPGReadyToMerge.h" |
7c673cae FG |
83 | |
84 | #include "messages/MOSDMap.h" | |
85 | #include "messages/MMonGetOSDMap.h" | |
86 | #include "messages/MOSDPGNotify.h" | |
87 | #include "messages/MOSDPGQuery.h" | |
88 | #include "messages/MOSDPGLog.h" | |
89 | #include "messages/MOSDPGRemove.h" | |
90 | #include "messages/MOSDPGInfo.h" | |
91 | #include "messages/MOSDPGCreate.h" | |
11fdf7f2 | 92 | #include "messages/MOSDPGCreate2.h" |
7c673cae FG |
93 | #include "messages/MOSDPGTrim.h" |
94 | #include "messages/MOSDPGScan.h" | |
7c673cae FG |
95 | #include "messages/MBackfillReserve.h" |
96 | #include "messages/MRecoveryReserve.h" | |
c07f9fc5 | 97 | #include "messages/MOSDForceRecovery.h" |
7c673cae FG |
98 | #include "messages/MOSDECSubOpWrite.h" |
99 | #include "messages/MOSDECSubOpWriteReply.h" | |
100 | #include "messages/MOSDECSubOpRead.h" | |
101 | #include "messages/MOSDECSubOpReadReply.h" | |
102 | #include "messages/MOSDPGCreated.h" | |
103 | #include "messages/MOSDPGUpdateLogMissing.h" | |
104 | #include "messages/MOSDPGUpdateLogMissingReply.h" | |
105 | ||
11fdf7f2 TL |
106 | #include "messages/MOSDPeeringOp.h" |
107 | ||
7c673cae FG |
108 | #include "messages/MOSDAlive.h" |
109 | ||
110 | #include "messages/MOSDScrub.h" | |
11fdf7f2 | 111 | #include "messages/MOSDScrub2.h" |
7c673cae FG |
112 | #include "messages/MOSDRepScrub.h" |
113 | ||
114 | #include "messages/MMonCommand.h" | |
115 | #include "messages/MCommand.h" | |
116 | #include "messages/MCommandReply.h" | |
117 | ||
118 | #include "messages/MPGStats.h" | |
119 | #include "messages/MPGStatsAck.h" | |
120 | ||
121 | #include "messages/MWatchNotify.h" | |
122 | #include "messages/MOSDPGPush.h" | |
123 | #include "messages/MOSDPGPushReply.h" | |
124 | #include "messages/MOSDPGPull.h" | |
125 | ||
126 | #include "common/perf_counters.h" | |
127 | #include "common/Timer.h" | |
128 | #include "common/LogClient.h" | |
129 | #include "common/AsyncReserver.h" | |
130 | #include "common/HeartbeatMap.h" | |
131 | #include "common/admin_socket.h" | |
132 | #include "common/ceph_context.h" | |
133 | ||
134 | #include "global/signal_handler.h" | |
135 | #include "global/pidfile.h" | |
136 | ||
137 | #include "include/color.h" | |
138 | #include "perfglue/cpu_profiler.h" | |
139 | #include "perfglue/heap_profiler.h" | |
140 | ||
141 | #include "osd/OpRequest.h" | |
142 | ||
143 | #include "auth/AuthAuthorizeHandler.h" | |
144 | #include "auth/RotatingKeyRing.h" | |
7c673cae FG |
145 | |
146 | #include "objclass/objclass.h" | |
147 | ||
148 | #include "common/cmdparse.h" | |
149 | #include "include/str_list.h" | |
150 | #include "include/util.h" | |
151 | ||
11fdf7f2 | 152 | #include "include/ceph_assert.h" |
7c673cae FG |
153 | #include "common/config.h" |
154 | #include "common/EventTrace.h" | |
155 | ||
11fdf7f2 TL |
156 | #include "json_spirit/json_spirit_reader.h" |
157 | #include "json_spirit/json_spirit_writer.h" | |
158 | ||
7c673cae FG |
159 | #ifdef WITH_LTTNG |
160 | #define TRACEPOINT_DEFINE | |
161 | #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
162 | #include "tracing/osd.h" | |
163 | #undef TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
164 | #undef TRACEPOINT_DEFINE | |
165 | #else | |
166 | #define tracepoint(...) | |
167 | #endif | |
168 | ||
169 | #define dout_context cct | |
170 | #define dout_subsys ceph_subsys_osd | |
171 | #undef dout_prefix | |
172 | #define dout_prefix _prefix(_dout, whoami, get_osdmap_epoch()) | |
173 | ||
224ce89b | 174 | |
7c673cae FG |
175 | static ostream& _prefix(std::ostream* _dout, int whoami, epoch_t epoch) { |
176 | return *_dout << "osd." << whoami << " " << epoch << " "; | |
177 | } | |
178 | ||
7c673cae FG |
179 | //Initial features in new superblock. |
180 | //Features here are also automatically upgraded | |
181 | CompatSet OSD::get_osd_initial_compat_set() { | |
182 | CompatSet::FeatureSet ceph_osd_feature_compat; | |
183 | CompatSet::FeatureSet ceph_osd_feature_ro_compat; | |
184 | CompatSet::FeatureSet ceph_osd_feature_incompat; | |
185 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_BASE); | |
186 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_PGINFO); | |
187 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_OLOC); | |
188 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_LEC); | |
189 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_CATEGORIES); | |
190 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_HOBJECTPOOL); | |
191 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_BIGINFO); | |
192 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_LEVELDBINFO); | |
193 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_LEVELDBLOG); | |
194 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_SNAPMAPPER); | |
195 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_HINTS); | |
196 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_PGMETA); | |
197 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_MISSING); | |
198 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_FASTINFO); | |
c07f9fc5 | 199 | ceph_osd_feature_incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_RECOVERY_DELETES); |
7c673cae FG |
200 | return CompatSet(ceph_osd_feature_compat, ceph_osd_feature_ro_compat, |
201 | ceph_osd_feature_incompat); | |
202 | } | |
203 | ||
204 | //Features are added here that this OSD supports. | |
205 | CompatSet OSD::get_osd_compat_set() { | |
206 | CompatSet compat = get_osd_initial_compat_set(); | |
207 | //Any features here can be set in code, but not in initial superblock | |
208 | compat.incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_SHARDS); | |
209 | return compat; | |
210 | } | |
211 | ||
212 | OSDService::OSDService(OSD *osd) : | |
213 | osd(osd), | |
214 | cct(osd->cct), | |
7c673cae FG |
215 | whoami(osd->whoami), store(osd->store), |
216 | log_client(osd->log_client), clog(osd->clog), | |
217 | pg_recovery_stats(osd->pg_recovery_stats), | |
218 | cluster_messenger(osd->cluster_messenger), | |
219 | client_messenger(osd->client_messenger), | |
220 | logger(osd->logger), | |
221 | recoverystate_perf(osd->recoverystate_perf), | |
222 | monc(osd->monc), | |
7c673cae | 223 | class_handler(osd->class_handler), |
11fdf7f2 TL |
224 | osd_max_object_size(cct->_conf, "osd_max_object_size"), |
225 | osd_skip_data_digest(cct->_conf, "osd_skip_data_digest"), | |
226 | publish_lock{ceph::make_mutex("OSDService::publish_lock")}, | |
227 | pre_publish_lock{ceph::make_mutex("OSDService::pre_publish_lock")}, | |
7c673cae FG |
228 | max_oldest_map(0), |
229 | peer_map_epoch_lock("OSDService::peer_map_epoch_lock"), | |
230 | sched_scrub_lock("OSDService::sched_scrub_lock"), scrubs_pending(0), | |
231 | scrubs_active(0), | |
232 | agent_lock("OSDService::agent_lock"), | |
233 | agent_valid_iterator(false), | |
234 | agent_ops(0), | |
235 | flush_mode_high_count(0), | |
236 | agent_active(true), | |
237 | agent_thread(this), | |
238 | agent_stop_flag(false), | |
239 | agent_timer_lock("OSDService::agent_timer_lock"), | |
240 | agent_timer(osd->client_messenger->cct, agent_timer_lock), | |
241 | last_recalibrate(ceph_clock_now()), | |
242 | promote_max_objects(0), | |
243 | promote_max_bytes(0), | |
244 | objecter(new Objecter(osd->client_messenger->cct, osd->objecter_messenger, osd->monc, NULL, 0, 0)), | |
11fdf7f2 | 245 | m_objecter_finishers(cct->_conf->osd_objecter_finishers), |
7c673cae FG |
246 | watch_lock("OSDService::watch_lock"), |
247 | watch_timer(osd->client_messenger->cct, watch_lock), | |
248 | next_notif_id(0), | |
249 | recovery_request_lock("OSDService::recovery_request_lock"), | |
250 | recovery_request_timer(cct, recovery_request_lock, false), | |
11fdf7f2 TL |
251 | sleep_lock("OSDService::sleep_lock"), |
252 | sleep_timer(cct, sleep_lock, false), | |
7c673cae | 253 | reserver_finisher(cct), |
3efd9988 | 254 | local_reserver(cct, &reserver_finisher, cct->_conf->osd_max_backfills, |
7c673cae | 255 | cct->_conf->osd_min_recovery_priority), |
3efd9988 | 256 | remote_reserver(cct, &reserver_finisher, cct->_conf->osd_max_backfills, |
7c673cae FG |
257 | cct->_conf->osd_min_recovery_priority), |
258 | pg_temp_lock("OSDService::pg_temp_lock"), | |
3efd9988 | 259 | snap_reserver(cct, &reserver_finisher, |
7c673cae FG |
260 | cct->_conf->osd_max_trimming_pgs), |
261 | recovery_lock("OSDService::recovery_lock"), | |
262 | recovery_ops_active(0), | |
263 | recovery_ops_reserved(0), | |
264 | recovery_paused(false), | |
265 | map_cache_lock("OSDService::map_cache_lock"), | |
266 | map_cache(cct, cct->_conf->osd_map_cache_size), | |
267 | map_bl_cache(cct->_conf->osd_map_cache_size), | |
268 | map_bl_inc_cache(cct->_conf->osd_map_cache_size), | |
7c673cae FG |
269 | stat_lock("OSDService::stat_lock"), |
270 | full_status_lock("OSDService::full_status_lock"), | |
271 | cur_state(NONE), | |
11fdf7f2 | 272 | cur_ratio(0), physical_ratio(0), |
7c673cae FG |
273 | epoch_lock("OSDService::epoch_lock"), |
274 | boot_epoch(0), up_epoch(0), bind_epoch(0), | |
275 | is_stopping_lock("OSDService::is_stopping_lock") | |
276 | #ifdef PG_DEBUG_REFS | |
277 | , pgid_lock("OSDService::pgid_lock") | |
278 | #endif | |
279 | { | |
280 | objecter->init(); | |
11fdf7f2 TL |
281 | |
282 | for (int i = 0; i < m_objecter_finishers; i++) { | |
283 | ostringstream str; | |
284 | str << "objecter-finisher-" << i; | |
285 | Finisher *fin = new Finisher(osd->client_messenger->cct, str.str(), "finisher"); | |
286 | objecter_finishers.push_back(fin); | |
287 | } | |
7c673cae FG |
288 | } |
289 | ||
290 | OSDService::~OSDService() | |
291 | { | |
292 | delete objecter; | |
11fdf7f2 TL |
293 | |
294 | for (auto f : objecter_finishers) { | |
295 | delete f; | |
296 | f = NULL; | |
297 | } | |
7c673cae FG |
298 | } |
299 | ||
31f18b77 FG |
300 | |
301 | ||
302 | #ifdef PG_DEBUG_REFS | |
303 | void OSDService::add_pgid(spg_t pgid, PG *pg){ | |
11fdf7f2 | 304 | std::lock_guard l(pgid_lock); |
31f18b77 FG |
305 | if (!pgid_tracker.count(pgid)) { |
306 | live_pgs[pgid] = pg; | |
307 | } | |
308 | pgid_tracker[pgid]++; | |
309 | } | |
310 | void OSDService::remove_pgid(spg_t pgid, PG *pg) | |
311 | { | |
11fdf7f2 TL |
312 | std::lock_guard l(pgid_lock); |
313 | ceph_assert(pgid_tracker.count(pgid)); | |
314 | ceph_assert(pgid_tracker[pgid] > 0); | |
31f18b77 FG |
315 | pgid_tracker[pgid]--; |
316 | if (pgid_tracker[pgid] == 0) { | |
317 | pgid_tracker.erase(pgid); | |
318 | live_pgs.erase(pgid); | |
319 | } | |
320 | } | |
321 | void OSDService::dump_live_pgids() | |
322 | { | |
11fdf7f2 | 323 | std::lock_guard l(pgid_lock); |
31f18b77 FG |
324 | derr << "live pgids:" << dendl; |
325 | for (map<spg_t, int>::const_iterator i = pgid_tracker.cbegin(); | |
326 | i != pgid_tracker.cend(); | |
327 | ++i) { | |
328 | derr << "\t" << *i << dendl; | |
329 | live_pgs[i->first]->dump_live_ids(); | |
330 | } | |
331 | } | |
332 | #endif | |
333 | ||
334 | ||
7c673cae | 335 | |
11fdf7f2 TL |
336 | void OSDService::identify_splits_and_merges( |
337 | OSDMapRef old_map, | |
338 | OSDMapRef new_map, | |
339 | spg_t pgid, | |
340 | set<pair<spg_t,epoch_t>> *split_children, | |
341 | set<pair<spg_t,epoch_t>> *merge_pgs) | |
7c673cae | 342 | { |
11fdf7f2 | 343 | if (!old_map->have_pg_pool(pgid.pool())) { |
7c673cae | 344 | return; |
7c673cae | 345 | } |
7c673cae | 346 | int old_pgnum = old_map->get_pg_num(pgid.pool()); |
11fdf7f2 TL |
347 | auto p = osd->pg_num_history.pg_nums.find(pgid.pool()); |
348 | if (p == osd->pg_num_history.pg_nums.end()) { | |
349 | return; | |
350 | } | |
351 | dout(20) << __func__ << " " << pgid << " e" << old_map->get_epoch() | |
352 | << " to e" << new_map->get_epoch() | |
353 | << " pg_nums " << p->second << dendl; | |
354 | deque<spg_t> queue; | |
355 | queue.push_back(pgid); | |
356 | while (!queue.empty()) { | |
357 | auto cur = queue.front(); | |
358 | queue.pop_front(); | |
359 | unsigned pgnum = old_pgnum; | |
360 | for (auto q = p->second.lower_bound(old_map->get_epoch()); | |
361 | q != p->second.end() && | |
362 | q->first <= new_map->get_epoch(); | |
363 | ++q) { | |
364 | if (pgnum < q->second) { | |
365 | // split? | |
366 | if (cur.ps() < pgnum) { | |
367 | set<spg_t> children; | |
368 | if (cur.is_split(pgnum, q->second, &children)) { | |
369 | dout(20) << __func__ << " " << cur << " e" << q->first | |
370 | << " pg_num " << pgnum << " -> " << q->second | |
371 | << " children " << children << dendl; | |
372 | for (auto i : children) { | |
373 | split_children->insert(make_pair(i, q->first)); | |
374 | queue.push_back(i); | |
375 | } | |
376 | } | |
377 | } else if (cur.ps() < q->second) { | |
378 | dout(20) << __func__ << " " << cur << " e" << q->first | |
379 | << " pg_num " << pgnum << " -> " << q->second | |
380 | << " is a child" << dendl; | |
381 | // normally we'd capture this from the parent, but it's | |
382 | // possible the parent doesn't exist yet (it will be | |
383 | // fabricated to allow an intervening merge). note this PG | |
384 | // as a split child here to be sure we catch it. | |
385 | split_children->insert(make_pair(cur, q->first)); | |
386 | } else { | |
387 | dout(20) << __func__ << " " << cur << " e" << q->first | |
388 | << " pg_num " << pgnum << " -> " << q->second | |
389 | << " is post-split, skipping" << dendl; | |
390 | } | |
391 | } else if (merge_pgs) { | |
392 | // merge? | |
393 | if (cur.ps() >= q->second) { | |
394 | if (cur.ps() < pgnum) { | |
395 | spg_t parent; | |
396 | if (cur.is_merge_source(pgnum, q->second, &parent)) { | |
397 | set<spg_t> children; | |
398 | parent.is_split(q->second, pgnum, &children); | |
399 | dout(20) << __func__ << " " << cur << " e" << q->first | |
400 | << " pg_num " << pgnum << " -> " << q->second | |
401 | << " is merge source, target " << parent | |
402 | << ", source(s) " << children << dendl; | |
403 | merge_pgs->insert(make_pair(parent, q->first)); | |
404 | for (auto c : children) { | |
405 | merge_pgs->insert(make_pair(c, q->first)); | |
406 | } | |
407 | } | |
408 | } else { | |
409 | dout(20) << __func__ << " " << cur << " e" << q->first | |
410 | << " pg_num " << pgnum << " -> " << q->second | |
411 | << " is beyond old pgnum, skipping" << dendl; | |
412 | } | |
413 | } else { | |
414 | set<spg_t> children; | |
415 | if (cur.is_split(q->second, pgnum, &children)) { | |
416 | dout(20) << __func__ << " " << cur << " e" << q->first | |
417 | << " pg_num " << pgnum << " -> " << q->second | |
418 | << " is merge target, source " << children << dendl; | |
419 | for (auto c : children) { | |
420 | merge_pgs->insert(make_pair(c, q->first)); | |
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 | ||
7c673cae FG |
436 | void OSDService::start_shutdown() |
437 | { | |
438 | { | |
11fdf7f2 | 439 | std::lock_guard l(agent_timer_lock); |
7c673cae FG |
440 | agent_timer.shutdown(); |
441 | } | |
31f18b77 FG |
442 | |
443 | { | |
11fdf7f2 TL |
444 | std::lock_guard l(sleep_lock); |
445 | sleep_timer.shutdown(); | |
31f18b77 | 446 | } |
7c673cae FG |
447 | } |
448 | ||
31f18b77 | 449 | void OSDService::shutdown_reserver() |
7c673cae FG |
450 | { |
451 | reserver_finisher.wait_for_empty(); | |
452 | reserver_finisher.stop(); | |
31f18b77 FG |
453 | } |
454 | ||
455 | void OSDService::shutdown() | |
456 | { | |
7c673cae | 457 | { |
11fdf7f2 | 458 | std::lock_guard l(watch_lock); |
7c673cae FG |
459 | watch_timer.shutdown(); |
460 | } | |
461 | ||
462 | objecter->shutdown(); | |
11fdf7f2 TL |
463 | for (auto f : objecter_finishers) { |
464 | f->wait_for_empty(); | |
465 | f->stop(); | |
7c673cae FG |
466 | } |
467 | ||
468 | { | |
11fdf7f2 TL |
469 | std::lock_guard l(recovery_request_lock); |
470 | recovery_request_timer.shutdown(); | |
31f18b77 FG |
471 | } |
472 | ||
11fdf7f2 | 473 | publish_map(OSDMapRef()); |
7c673cae FG |
474 | next_osdmap = OSDMapRef(); |
475 | } | |
476 | ||
477 | void OSDService::init() | |
478 | { | |
479 | reserver_finisher.start(); | |
11fdf7f2 TL |
480 | for (auto f : objecter_finishers) { |
481 | f->start(); | |
482 | } | |
7c673cae FG |
483 | objecter->set_client_incarnation(0); |
484 | ||
485 | // deprioritize objecter in daemonperf output | |
486 | objecter->get_logger()->set_prio_adjust(-3); | |
487 | ||
488 | watch_timer.init(); | |
489 | agent_timer.init(); | |
7c673cae FG |
490 | |
491 | agent_thread.create("osd_srv_agent"); | |
492 | ||
493 | if (cct->_conf->osd_recovery_delay_start) | |
494 | defer_recovery(cct->_conf->osd_recovery_delay_start); | |
495 | } | |
496 | ||
497 | void OSDService::final_init() | |
498 | { | |
499 | objecter->start(osdmap.get()); | |
500 | } | |
501 | ||
502 | void OSDService::activate_map() | |
503 | { | |
504 | // wake/unwake the tiering agent | |
505 | agent_lock.Lock(); | |
506 | agent_active = | |
507 | !osdmap->test_flag(CEPH_OSDMAP_NOTIERAGENT) && | |
508 | osd->is_active(); | |
509 | agent_cond.Signal(); | |
510 | agent_lock.Unlock(); | |
511 | } | |
512 | ||
181888fb FG |
513 | void OSDService::request_osdmap_update(epoch_t e) |
514 | { | |
515 | osd->osdmap_subscribe(e, false); | |
516 | } | |
517 | ||
7c673cae FG |
518 | class AgentTimeoutCB : public Context { |
519 | PGRef pg; | |
520 | public: | |
521 | explicit AgentTimeoutCB(PGRef _pg) : pg(_pg) {} | |
522 | void finish(int) override { | |
523 | pg->agent_choose_mode_restart(); | |
524 | } | |
525 | }; | |
526 | ||
527 | void OSDService::agent_entry() | |
528 | { | |
529 | dout(10) << __func__ << " start" << dendl; | |
530 | agent_lock.Lock(); | |
531 | ||
532 | while (!agent_stop_flag) { | |
533 | if (agent_queue.empty()) { | |
534 | dout(20) << __func__ << " empty queue" << dendl; | |
535 | agent_cond.Wait(agent_lock); | |
536 | continue; | |
537 | } | |
538 | uint64_t level = agent_queue.rbegin()->first; | |
539 | set<PGRef>& top = agent_queue.rbegin()->second; | |
540 | dout(10) << __func__ | |
541 | << " tiers " << agent_queue.size() | |
542 | << ", top is " << level | |
543 | << " with pgs " << top.size() | |
544 | << ", ops " << agent_ops << "/" | |
545 | << cct->_conf->osd_agent_max_ops | |
546 | << (agent_active ? " active" : " NOT ACTIVE") | |
547 | << dendl; | |
548 | dout(20) << __func__ << " oids " << agent_oids << dendl; | |
549 | int max = cct->_conf->osd_agent_max_ops - agent_ops; | |
550 | int agent_flush_quota = max; | |
551 | if (!flush_mode_high_count) | |
552 | agent_flush_quota = cct->_conf->osd_agent_max_low_ops - agent_ops; | |
553 | if (agent_flush_quota <= 0 || top.empty() || !agent_active) { | |
554 | agent_cond.Wait(agent_lock); | |
555 | continue; | |
556 | } | |
557 | ||
558 | if (!agent_valid_iterator || agent_queue_pos == top.end()) { | |
559 | agent_queue_pos = top.begin(); | |
560 | agent_valid_iterator = true; | |
561 | } | |
562 | PGRef pg = *agent_queue_pos; | |
563 | dout(10) << "high_count " << flush_mode_high_count | |
564 | << " agent_ops " << agent_ops | |
565 | << " flush_quota " << agent_flush_quota << dendl; | |
566 | agent_lock.Unlock(); | |
567 | if (!pg->agent_work(max, agent_flush_quota)) { | |
11fdf7f2 | 568 | dout(10) << __func__ << " " << pg->pg_id |
7c673cae FG |
569 | << " no agent_work, delay for " << cct->_conf->osd_agent_delay_time |
570 | << " seconds" << dendl; | |
571 | ||
572 | osd->logger->inc(l_osd_tier_delay); | |
573 | // Queue a timer to call agent_choose_mode for this pg in 5 seconds | |
574 | agent_timer_lock.Lock(); | |
575 | Context *cb = new AgentTimeoutCB(pg); | |
576 | agent_timer.add_event_after(cct->_conf->osd_agent_delay_time, cb); | |
577 | agent_timer_lock.Unlock(); | |
578 | } | |
579 | agent_lock.Lock(); | |
580 | } | |
581 | agent_lock.Unlock(); | |
582 | dout(10) << __func__ << " finish" << dendl; | |
583 | } | |
584 | ||
585 | void OSDService::agent_stop() | |
586 | { | |
587 | { | |
11fdf7f2 | 588 | std::lock_guard l(agent_lock); |
7c673cae FG |
589 | |
590 | // By this time all ops should be cancelled | |
11fdf7f2 | 591 | ceph_assert(agent_ops == 0); |
7c673cae FG |
592 | // By this time all PGs are shutdown and dequeued |
593 | if (!agent_queue.empty()) { | |
594 | set<PGRef>& top = agent_queue.rbegin()->second; | |
11fdf7f2 TL |
595 | derr << "agent queue not empty, for example " << (*top.begin())->get_pgid() << dendl; |
596 | ceph_abort_msg("agent queue not empty"); | |
7c673cae FG |
597 | } |
598 | ||
599 | agent_stop_flag = true; | |
600 | agent_cond.Signal(); | |
601 | } | |
602 | agent_thread.join(); | |
603 | } | |
604 | ||
605 | // ------------------------------------- | |
606 | ||
607 | void OSDService::promote_throttle_recalibrate() | |
608 | { | |
609 | utime_t now = ceph_clock_now(); | |
610 | double dur = now - last_recalibrate; | |
611 | last_recalibrate = now; | |
612 | unsigned prob = promote_probability_millis; | |
613 | ||
614 | uint64_t target_obj_sec = cct->_conf->osd_tier_promote_max_objects_sec; | |
615 | uint64_t target_bytes_sec = cct->_conf->osd_tier_promote_max_bytes_sec; | |
616 | ||
617 | unsigned min_prob = 1; | |
618 | ||
619 | uint64_t attempts, obj, bytes; | |
620 | promote_counter.sample_and_attenuate(&attempts, &obj, &bytes); | |
621 | dout(10) << __func__ << " " << attempts << " attempts, promoted " | |
1adf2230 | 622 | << obj << " objects and " << byte_u_t(bytes) << "; target " |
7c673cae | 623 | << target_obj_sec << " obj/sec or " |
1adf2230 | 624 | << byte_u_t(target_bytes_sec) << "/sec" |
7c673cae FG |
625 | << dendl; |
626 | ||
627 | // calculate what the probability *should* be, given the targets | |
628 | unsigned new_prob; | |
629 | if (attempts && dur > 0) { | |
630 | uint64_t avg_size = 1; | |
631 | if (obj) | |
11fdf7f2 | 632 | avg_size = std::max<uint64_t>(bytes / obj, 1); |
7c673cae FG |
633 | unsigned po = (double)target_obj_sec * dur * 1000.0 / (double)attempts; |
634 | unsigned pb = (double)target_bytes_sec / (double)avg_size * dur * 1000.0 | |
635 | / (double)attempts; | |
636 | dout(20) << __func__ << " po " << po << " pb " << pb << " avg_size " | |
637 | << avg_size << dendl; | |
638 | if (target_obj_sec && target_bytes_sec) | |
11fdf7f2 | 639 | new_prob = std::min(po, pb); |
7c673cae FG |
640 | else if (target_obj_sec) |
641 | new_prob = po; | |
642 | else if (target_bytes_sec) | |
643 | new_prob = pb; | |
644 | else | |
645 | new_prob = 1000; | |
646 | } else { | |
647 | new_prob = 1000; | |
648 | } | |
649 | dout(20) << __func__ << " new_prob " << new_prob << dendl; | |
650 | ||
651 | // correct for persistent skew between target rate and actual rate, adjust | |
652 | double ratio = 1.0; | |
653 | unsigned actual = 0; | |
654 | if (attempts && obj) { | |
655 | actual = obj * 1000 / attempts; | |
656 | ratio = (double)actual / (double)prob; | |
657 | new_prob = (double)new_prob / ratio; | |
658 | } | |
11fdf7f2 TL |
659 | new_prob = std::max(new_prob, min_prob); |
660 | new_prob = std::min(new_prob, 1000u); | |
7c673cae FG |
661 | |
662 | // adjust | |
663 | prob = (prob + new_prob) / 2; | |
11fdf7f2 TL |
664 | prob = std::max(prob, min_prob); |
665 | prob = std::min(prob, 1000u); | |
7c673cae FG |
666 | dout(10) << __func__ << " actual " << actual |
667 | << ", actual/prob ratio " << ratio | |
668 | << ", adjusted new_prob " << new_prob | |
669 | << ", prob " << promote_probability_millis << " -> " << prob | |
670 | << dendl; | |
671 | promote_probability_millis = prob; | |
672 | ||
673 | // set hard limits for this interval to mitigate stampedes | |
91327a77 AA |
674 | promote_max_objects = target_obj_sec * osd->OSD_TICK_INTERVAL * 2; |
675 | promote_max_bytes = target_bytes_sec * osd->OSD_TICK_INTERVAL * 2; | |
7c673cae FG |
676 | } |
677 | ||
678 | // ------------------------------------- | |
679 | ||
680 | float OSDService::get_failsafe_full_ratio() | |
681 | { | |
682 | float full_ratio = cct->_conf->osd_failsafe_full_ratio; | |
683 | if (full_ratio > 1.0) full_ratio /= 100.0; | |
684 | return full_ratio; | |
685 | } | |
686 | ||
11fdf7f2 | 687 | OSDService::s_names OSDService::recalc_full_state(float ratio, float pratio, string &inject) |
7c673cae | 688 | { |
7c673cae FG |
689 | // The OSDMap ratios take precendence. So if the failsafe is .95 and |
690 | // the admin sets the cluster full to .96, the failsafe moves up to .96 | |
691 | // too. (Not that having failsafe == full is ideal, but it's better than | |
692 | // dropping writes before the clusters appears full.) | |
693 | OSDMapRef osdmap = get_osdmap(); | |
694 | if (!osdmap || osdmap->get_epoch() == 0) { | |
11fdf7f2 | 695 | return NONE; |
7c673cae FG |
696 | } |
697 | float nearfull_ratio = osdmap->get_nearfull_ratio(); | |
698 | float backfillfull_ratio = std::max(osdmap->get_backfillfull_ratio(), nearfull_ratio); | |
699 | float full_ratio = std::max(osdmap->get_full_ratio(), backfillfull_ratio); | |
700 | float failsafe_ratio = std::max(get_failsafe_full_ratio(), full_ratio); | |
701 | ||
31f18b77 | 702 | if (osdmap->require_osd_release < CEPH_RELEASE_LUMINOUS) { |
7c673cae FG |
703 | // use the failsafe for nearfull and full; the mon isn't using the |
704 | // flags anyway because we're mid-upgrade. | |
705 | full_ratio = failsafe_ratio; | |
706 | backfillfull_ratio = failsafe_ratio; | |
707 | nearfull_ratio = failsafe_ratio; | |
708 | } else if (full_ratio <= 0 || | |
709 | backfillfull_ratio <= 0 || | |
710 | nearfull_ratio <= 0) { | |
711 | derr << __func__ << " full_ratio, backfillfull_ratio or nearfull_ratio is <= 0" << dendl; | |
712 | // use failsafe flag. ick. the monitor did something wrong or the user | |
713 | // did something stupid. | |
714 | full_ratio = failsafe_ratio; | |
715 | backfillfull_ratio = failsafe_ratio; | |
716 | nearfull_ratio = failsafe_ratio; | |
717 | } | |
718 | ||
7c673cae | 719 | if (injectfull_state > NONE && injectfull) { |
7c673cae | 720 | inject = "(Injected)"; |
11fdf7f2 TL |
721 | return injectfull_state; |
722 | } else if (pratio > failsafe_ratio) { | |
723 | return FAILSAFE; | |
7c673cae | 724 | } else if (ratio > full_ratio) { |
11fdf7f2 | 725 | return FULL; |
7c673cae | 726 | } else if (ratio > backfillfull_ratio) { |
11fdf7f2 | 727 | return BACKFILLFULL; |
7c673cae | 728 | } else if (ratio > nearfull_ratio) { |
11fdf7f2 | 729 | return NEARFULL; |
7c673cae | 730 | } |
11fdf7f2 TL |
731 | return NONE; |
732 | } | |
733 | ||
734 | void OSDService::check_full_status(float ratio, float pratio) | |
735 | { | |
736 | std::lock_guard l(full_status_lock); | |
737 | ||
738 | cur_ratio = ratio; | |
739 | physical_ratio = pratio; | |
740 | ||
741 | string inject; | |
742 | s_names new_state; | |
743 | new_state = recalc_full_state(ratio, pratio, inject); | |
744 | ||
7c673cae | 745 | dout(20) << __func__ << " cur ratio " << ratio |
11fdf7f2 | 746 | << ", physical ratio " << pratio |
7c673cae FG |
747 | << ", new state " << get_full_state_name(new_state) |
748 | << " " << inject | |
749 | << dendl; | |
750 | ||
751 | // warn | |
752 | if (cur_state != new_state) { | |
753 | dout(10) << __func__ << " " << get_full_state_name(cur_state) | |
754 | << " -> " << get_full_state_name(new_state) << dendl; | |
755 | if (new_state == FAILSAFE) { | |
c07f9fc5 | 756 | clog->error() << "full status failsafe engaged, dropping updates, now " |
7c673cae FG |
757 | << (int)roundf(ratio * 100) << "% full"; |
758 | } else if (cur_state == FAILSAFE) { | |
c07f9fc5 FG |
759 | clog->error() << "full status failsafe disengaged, no longer dropping " |
760 | << "updates, now " << (int)roundf(ratio * 100) << "% full"; | |
7c673cae FG |
761 | } |
762 | cur_state = new_state; | |
763 | } | |
764 | } | |
765 | ||
766 | bool OSDService::need_fullness_update() | |
767 | { | |
768 | OSDMapRef osdmap = get_osdmap(); | |
769 | s_names cur = NONE; | |
770 | if (osdmap->exists(whoami)) { | |
771 | if (osdmap->get_state(whoami) & CEPH_OSD_FULL) { | |
772 | cur = FULL; | |
773 | } else if (osdmap->get_state(whoami) & CEPH_OSD_BACKFILLFULL) { | |
774 | cur = BACKFILLFULL; | |
775 | } else if (osdmap->get_state(whoami) & CEPH_OSD_NEARFULL) { | |
776 | cur = NEARFULL; | |
777 | } | |
778 | } | |
779 | s_names want = NONE; | |
780 | if (is_full()) | |
781 | want = FULL; | |
782 | else if (is_backfillfull()) | |
783 | want = BACKFILLFULL; | |
784 | else if (is_nearfull()) | |
785 | want = NEARFULL; | |
786 | return want != cur; | |
787 | } | |
788 | ||
11fdf7f2 | 789 | bool OSDService::_check_inject_full(DoutPrefixProvider *dpp, s_names type) const |
7c673cae | 790 | { |
7c673cae FG |
791 | if (injectfull && injectfull_state >= type) { |
792 | // injectfull is either a count of the number of times to return failsafe full | |
793 | // or if -1 then always return full | |
794 | if (injectfull > 0) | |
795 | --injectfull; | |
11fdf7f2 TL |
796 | ldpp_dout(dpp, 10) << __func__ << " Injected " << get_full_state_name(type) << " OSD (" |
797 | << (injectfull < 0 ? "set" : std::to_string(injectfull)) << ")" | |
798 | << dendl; | |
7c673cae FG |
799 | return true; |
800 | } | |
11fdf7f2 TL |
801 | return false; |
802 | } | |
803 | ||
804 | bool OSDService::_check_full(DoutPrefixProvider *dpp, s_names type) const | |
805 | { | |
806 | std::lock_guard l(full_status_lock); | |
807 | ||
808 | if (_check_inject_full(dpp, type)) | |
809 | return true; | |
810 | ||
811 | if (cur_state >= type) | |
812 | ldpp_dout(dpp, 10) << __func__ << " current usage is " << cur_ratio | |
813 | << " physical " << physical_ratio << dendl; | |
7c673cae | 814 | |
7c673cae FG |
815 | return cur_state >= type; |
816 | } | |
817 | ||
11fdf7f2 TL |
818 | bool OSDService::_tentative_full(DoutPrefixProvider *dpp, s_names type, uint64_t adjust_used, osd_stat_t adjusted_stat) |
819 | { | |
820 | ldpp_dout(dpp, 20) << __func__ << " type " << get_full_state_name(type) << " adjust_used " << (adjust_used >> 10) << "KiB" << dendl; | |
821 | { | |
822 | std::lock_guard l(full_status_lock); | |
823 | if (_check_inject_full(dpp, type)) { | |
824 | return true; | |
825 | } | |
826 | } | |
827 | ||
828 | float pratio; | |
829 | float ratio = compute_adjusted_ratio(adjusted_stat, &pratio, adjust_used); | |
830 | ||
831 | string notused; | |
832 | s_names tentative_state = recalc_full_state(ratio, pratio, notused); | |
833 | ||
834 | if (tentative_state >= type) | |
835 | ldpp_dout(dpp, 10) << __func__ << " tentative usage is " << ratio << dendl; | |
836 | ||
837 | return tentative_state >= type; | |
838 | } | |
839 | ||
840 | bool OSDService::check_failsafe_full(DoutPrefixProvider *dpp) const | |
841 | { | |
842 | return _check_full(dpp, FAILSAFE); | |
843 | } | |
844 | ||
845 | bool OSDService::check_full(DoutPrefixProvider *dpp) const | |
7c673cae | 846 | { |
11fdf7f2 | 847 | return _check_full(dpp, FULL); |
7c673cae FG |
848 | } |
849 | ||
11fdf7f2 | 850 | bool OSDService::tentative_backfill_full(DoutPrefixProvider *dpp, uint64_t adjust_used, osd_stat_t stats) |
7c673cae | 851 | { |
11fdf7f2 | 852 | return _tentative_full(dpp, BACKFILLFULL, adjust_used, stats); |
7c673cae FG |
853 | } |
854 | ||
11fdf7f2 | 855 | bool OSDService::check_backfill_full(DoutPrefixProvider *dpp) const |
7c673cae | 856 | { |
11fdf7f2 | 857 | return _check_full(dpp, BACKFILLFULL); |
7c673cae FG |
858 | } |
859 | ||
11fdf7f2 | 860 | bool OSDService::check_nearfull(DoutPrefixProvider *dpp) const |
7c673cae | 861 | { |
11fdf7f2 | 862 | return _check_full(dpp, NEARFULL); |
7c673cae FG |
863 | } |
864 | ||
865 | bool OSDService::is_failsafe_full() const | |
866 | { | |
11fdf7f2 | 867 | std::lock_guard l(full_status_lock); |
7c673cae FG |
868 | return cur_state == FAILSAFE; |
869 | } | |
870 | ||
871 | bool OSDService::is_full() const | |
872 | { | |
11fdf7f2 | 873 | std::lock_guard l(full_status_lock); |
7c673cae FG |
874 | return cur_state >= FULL; |
875 | } | |
876 | ||
877 | bool OSDService::is_backfillfull() const | |
878 | { | |
11fdf7f2 | 879 | std::lock_guard l(full_status_lock); |
7c673cae FG |
880 | return cur_state >= BACKFILLFULL; |
881 | } | |
882 | ||
883 | bool OSDService::is_nearfull() const | |
884 | { | |
11fdf7f2 | 885 | std::lock_guard l(full_status_lock); |
7c673cae FG |
886 | return cur_state >= NEARFULL; |
887 | } | |
888 | ||
889 | void OSDService::set_injectfull(s_names type, int64_t count) | |
890 | { | |
11fdf7f2 | 891 | std::lock_guard l(full_status_lock); |
7c673cae FG |
892 | injectfull_state = type; |
893 | injectfull = count; | |
894 | } | |
895 | ||
11fdf7f2 TL |
896 | void OSDService::set_statfs(const struct store_statfs_t &stbuf, |
897 | osd_alert_list_t& alerts) | |
7c673cae | 898 | { |
224ce89b | 899 | uint64_t bytes = stbuf.total; |
224ce89b | 900 | uint64_t avail = stbuf.available; |
11fdf7f2 TL |
901 | uint64_t used = stbuf.get_used_raw(); |
902 | ||
903 | // For testing fake statfs values so it doesn't matter if all | |
904 | // OSDs are using the same partition. | |
905 | if (cct->_conf->fake_statfs_for_testing) { | |
906 | uint64_t total_num_bytes = 0; | |
907 | vector<PGRef> pgs; | |
908 | osd->_get_pgs(&pgs); | |
909 | for (auto p : pgs) { | |
910 | total_num_bytes += p->get_stats_num_bytes(); | |
911 | } | |
912 | bytes = cct->_conf->fake_statfs_for_testing; | |
913 | if (total_num_bytes < bytes) | |
914 | avail = bytes - total_num_bytes; | |
915 | else | |
916 | avail = 0; | |
917 | dout(0) << __func__ << " fake total " << cct->_conf->fake_statfs_for_testing | |
918 | << " adjust available " << avail | |
919 | << dendl; | |
920 | used = bytes - avail; | |
921 | } | |
7c673cae | 922 | |
224ce89b WB |
923 | osd->logger->set(l_osd_stat_bytes, bytes); |
924 | osd->logger->set(l_osd_stat_bytes_used, used); | |
925 | osd->logger->set(l_osd_stat_bytes_avail, avail); | |
7c673cae | 926 | |
11fdf7f2 TL |
927 | std::lock_guard l(stat_lock); |
928 | osd_stat.statfs = stbuf; | |
929 | osd_stat.os_alerts.clear(); | |
930 | osd_stat.os_alerts[whoami].swap(alerts); | |
931 | if (cct->_conf->fake_statfs_for_testing) { | |
932 | osd_stat.statfs.total = bytes; | |
933 | osd_stat.statfs.available = avail; | |
934 | // For testing don't want used to go negative, so clear reserved | |
935 | osd_stat.statfs.internally_reserved = 0; | |
224ce89b WB |
936 | } |
937 | } | |
7c673cae | 938 | |
11fdf7f2 TL |
939 | osd_stat_t OSDService::set_osd_stat(vector<int>& hb_peers, |
940 | int num_pgs) | |
224ce89b | 941 | { |
11fdf7f2 TL |
942 | std::lock_guard l(stat_lock); |
943 | osd_stat.hb_peers.swap(hb_peers); | |
944 | osd->op_tracker.get_age_ms_histogram(&osd_stat.op_queue_age_hist); | |
945 | osd_stat.num_pgs = num_pgs; | |
946 | return osd_stat; | |
947 | } | |
948 | ||
949 | void OSDService::inc_osd_stat_repaired() | |
950 | { | |
951 | std::lock_guard l(stat_lock); | |
952 | osd_stat.num_shards_repaired++; | |
953 | return; | |
954 | } | |
955 | ||
956 | float OSDService::compute_adjusted_ratio(osd_stat_t new_stat, float *pratio, | |
957 | uint64_t adjust_used) | |
958 | { | |
959 | *pratio = | |
960 | ((float)new_stat.statfs.get_used()) / ((float)new_stat.statfs.total); | |
961 | ||
962 | if (adjust_used) { | |
963 | dout(20) << __func__ << " Before kb_used() " << new_stat.statfs.kb_used() << dendl; | |
964 | if (new_stat.statfs.available > adjust_used) | |
965 | new_stat.statfs.available -= adjust_used; | |
966 | else | |
967 | new_stat.statfs.available = 0; | |
968 | dout(20) << __func__ << " After kb_used() " << new_stat.statfs.kb_used() << dendl; | |
7c673cae FG |
969 | } |
970 | ||
11fdf7f2 TL |
971 | // Check all pgs and adjust kb_used to include all pending backfill data |
972 | int backfill_adjusted = 0; | |
973 | vector<PGRef> pgs; | |
974 | osd->_get_pgs(&pgs); | |
975 | for (auto p : pgs) { | |
976 | backfill_adjusted += p->pg_stat_adjust(&new_stat); | |
977 | } | |
978 | if (backfill_adjusted) { | |
979 | dout(20) << __func__ << " backfill adjusted " << new_stat << dendl; | |
980 | } | |
981 | return ((float)new_stat.statfs.get_used()) / ((float)new_stat.statfs.total); | |
7c673cae FG |
982 | } |
983 | ||
984 | bool OSDService::check_osdmap_full(const set<pg_shard_t> &missing_on) | |
985 | { | |
986 | OSDMapRef osdmap = get_osdmap(); | |
987 | for (auto shard : missing_on) { | |
988 | if (osdmap->get_state(shard.osd) & CEPH_OSD_FULL) | |
989 | return true; | |
990 | } | |
991 | return false; | |
992 | } | |
993 | ||
994 | void OSDService::send_message_osd_cluster(int peer, Message *m, epoch_t from_epoch) | |
995 | { | |
996 | OSDMapRef next_map = get_nextmap_reserved(); | |
997 | // service map is always newer/newest | |
11fdf7f2 | 998 | ceph_assert(from_epoch <= next_map->get_epoch()); |
7c673cae FG |
999 | |
1000 | if (next_map->is_down(peer) || | |
1001 | next_map->get_info(peer).up_from > from_epoch) { | |
1002 | m->put(); | |
1003 | release_map(next_map); | |
1004 | return; | |
1005 | } | |
11fdf7f2 TL |
1006 | ConnectionRef peer_con = osd->cluster_messenger->connect_to_osd( |
1007 | next_map->get_cluster_addrs(peer)); | |
7c673cae FG |
1008 | share_map_peer(peer, peer_con.get(), next_map); |
1009 | peer_con->send_message(m); | |
1010 | release_map(next_map); | |
1011 | } | |
1012 | ||
1013 | ConnectionRef OSDService::get_con_osd_cluster(int peer, epoch_t from_epoch) | |
1014 | { | |
1015 | OSDMapRef next_map = get_nextmap_reserved(); | |
1016 | // service map is always newer/newest | |
11fdf7f2 | 1017 | ceph_assert(from_epoch <= next_map->get_epoch()); |
7c673cae FG |
1018 | |
1019 | if (next_map->is_down(peer) || | |
1020 | next_map->get_info(peer).up_from > from_epoch) { | |
1021 | release_map(next_map); | |
1022 | return NULL; | |
1023 | } | |
11fdf7f2 TL |
1024 | ConnectionRef con = osd->cluster_messenger->connect_to_osd( |
1025 | next_map->get_cluster_addrs(peer)); | |
7c673cae FG |
1026 | release_map(next_map); |
1027 | return con; | |
1028 | } | |
1029 | ||
1030 | pair<ConnectionRef,ConnectionRef> OSDService::get_con_osd_hb(int peer, epoch_t from_epoch) | |
1031 | { | |
1032 | OSDMapRef next_map = get_nextmap_reserved(); | |
1033 | // service map is always newer/newest | |
11fdf7f2 | 1034 | ceph_assert(from_epoch <= next_map->get_epoch()); |
7c673cae FG |
1035 | |
1036 | pair<ConnectionRef,ConnectionRef> ret; | |
1037 | if (next_map->is_down(peer) || | |
1038 | next_map->get_info(peer).up_from > from_epoch) { | |
1039 | release_map(next_map); | |
1040 | return ret; | |
1041 | } | |
11fdf7f2 TL |
1042 | ret.first = osd->hb_back_client_messenger->connect_to_osd( |
1043 | next_map->get_hb_back_addrs(peer)); | |
1044 | ret.second = osd->hb_front_client_messenger->connect_to_osd( | |
1045 | next_map->get_hb_front_addrs(peer)); | |
7c673cae FG |
1046 | release_map(next_map); |
1047 | return ret; | |
1048 | } | |
1049 | ||
11fdf7f2 TL |
1050 | entity_name_t OSDService::get_cluster_msgr_name() const |
1051 | { | |
1052 | return cluster_messenger->get_myname(); | |
1053 | } | |
7c673cae | 1054 | |
94b18763 FG |
1055 | void OSDService::queue_want_pg_temp(pg_t pgid, |
1056 | const vector<int>& want, | |
1057 | bool forced) | |
7c673cae | 1058 | { |
11fdf7f2 | 1059 | std::lock_guard l(pg_temp_lock); |
94b18763 | 1060 | auto p = pg_temp_pending.find(pgid); |
7c673cae | 1061 | if (p == pg_temp_pending.end() || |
94b18763 FG |
1062 | p->second.acting != want || |
1063 | forced) { | |
11fdf7f2 | 1064 | pg_temp_wanted[pgid] = {want, forced}; |
7c673cae FG |
1065 | } |
1066 | } | |
1067 | ||
1068 | void OSDService::remove_want_pg_temp(pg_t pgid) | |
1069 | { | |
11fdf7f2 | 1070 | std::lock_guard l(pg_temp_lock); |
7c673cae FG |
1071 | pg_temp_wanted.erase(pgid); |
1072 | pg_temp_pending.erase(pgid); | |
1073 | } | |
1074 | ||
1075 | void OSDService::_sent_pg_temp() | |
1076 | { | |
11fdf7f2 TL |
1077 | #ifdef HAVE_STDLIB_MAP_SPLICING |
1078 | pg_temp_pending.merge(pg_temp_wanted); | |
1079 | #else | |
94b18763 FG |
1080 | pg_temp_pending.insert(make_move_iterator(begin(pg_temp_wanted)), |
1081 | make_move_iterator(end(pg_temp_wanted))); | |
11fdf7f2 | 1082 | #endif |
7c673cae FG |
1083 | pg_temp_wanted.clear(); |
1084 | } | |
1085 | ||
1086 | void OSDService::requeue_pg_temp() | |
1087 | { | |
11fdf7f2 | 1088 | std::lock_guard l(pg_temp_lock); |
7c673cae FG |
1089 | // wanted overrides pending. note that remove_want_pg_temp |
1090 | // clears the item out of both. | |
1091 | unsigned old_wanted = pg_temp_wanted.size(); | |
1092 | unsigned old_pending = pg_temp_pending.size(); | |
1093 | _sent_pg_temp(); | |
1094 | pg_temp_wanted.swap(pg_temp_pending); | |
1095 | dout(10) << __func__ << " " << old_wanted << " + " << old_pending << " -> " | |
1096 | << pg_temp_wanted.size() << dendl; | |
1097 | } | |
1098 | ||
94b18763 FG |
1099 | std::ostream& operator<<(std::ostream& out, |
1100 | const OSDService::pg_temp_t& pg_temp) | |
1101 | { | |
1102 | out << pg_temp.acting; | |
1103 | if (pg_temp.forced) { | |
1104 | out << " (forced)"; | |
1105 | } | |
1106 | return out; | |
1107 | } | |
1108 | ||
7c673cae FG |
1109 | void OSDService::send_pg_temp() |
1110 | { | |
11fdf7f2 | 1111 | std::lock_guard l(pg_temp_lock); |
7c673cae FG |
1112 | if (pg_temp_wanted.empty()) |
1113 | return; | |
1114 | dout(10) << "send_pg_temp " << pg_temp_wanted << dendl; | |
94b18763 | 1115 | MOSDPGTemp *ms[2] = {nullptr, nullptr}; |
11fdf7f2 TL |
1116 | for (auto& [pgid, pg_temp] : pg_temp_wanted) { |
1117 | auto& m = ms[pg_temp.forced]; | |
94b18763 FG |
1118 | if (!m) { |
1119 | m = new MOSDPGTemp(osdmap->get_epoch()); | |
11fdf7f2 | 1120 | m->forced = pg_temp.forced; |
94b18763 | 1121 | } |
11fdf7f2 | 1122 | m->pg_temp.emplace(pgid, pg_temp.acting); |
94b18763 FG |
1123 | } |
1124 | for (auto m : ms) { | |
1125 | if (m) { | |
1126 | monc->send_mon_message(m); | |
1127 | } | |
1128 | } | |
7c673cae FG |
1129 | _sent_pg_temp(); |
1130 | } | |
1131 | ||
1132 | void OSDService::send_pg_created(pg_t pgid) | |
1133 | { | |
11fdf7f2 | 1134 | std::lock_guard l(pg_created_lock); |
7c673cae | 1135 | dout(20) << __func__ << dendl; |
11fdf7f2 TL |
1136 | auto o = get_osdmap(); |
1137 | if (o->require_osd_release >= CEPH_RELEASE_LUMINOUS) { | |
1138 | pg_created.insert(pgid); | |
c07f9fc5 FG |
1139 | monc->send_mon_message(new MOSDPGCreated(pgid)); |
1140 | } | |
7c673cae FG |
1141 | } |
1142 | ||
11fdf7f2 TL |
1143 | void OSDService::send_pg_created() |
1144 | { | |
1145 | std::lock_guard l(pg_created_lock); | |
1146 | dout(20) << __func__ << dendl; | |
1147 | auto o = get_osdmap(); | |
1148 | if (o->require_osd_release >= CEPH_RELEASE_LUMINOUS) { | |
1149 | for (auto pgid : pg_created) { | |
1150 | monc->send_mon_message(new MOSDPGCreated(pgid)); | |
1151 | } | |
1152 | } | |
1153 | } | |
1154 | ||
1155 | void OSDService::prune_pg_created() | |
1156 | { | |
1157 | std::lock_guard l(pg_created_lock); | |
1158 | dout(20) << __func__ << dendl; | |
1159 | auto o = get_osdmap(); | |
1160 | auto i = pg_created.begin(); | |
1161 | while (i != pg_created.end()) { | |
1162 | auto p = o->get_pg_pool(i->pool()); | |
1163 | if (!p || !p->has_flag(pg_pool_t::FLAG_CREATING)) { | |
1164 | dout(20) << __func__ << " pruning " << *i << dendl; | |
1165 | i = pg_created.erase(i); | |
1166 | } else { | |
1167 | dout(20) << __func__ << " keeping " << *i << dendl; | |
1168 | ++i; | |
1169 | } | |
1170 | } | |
1171 | } | |
1172 | ||
1173 | ||
7c673cae FG |
1174 | // -------------------------------------- |
1175 | // dispatch | |
1176 | ||
1177 | epoch_t OSDService::get_peer_epoch(int peer) | |
1178 | { | |
11fdf7f2 | 1179 | std::lock_guard l(peer_map_epoch_lock); |
7c673cae FG |
1180 | map<int,epoch_t>::iterator p = peer_map_epoch.find(peer); |
1181 | if (p == peer_map_epoch.end()) | |
1182 | return 0; | |
1183 | return p->second; | |
1184 | } | |
1185 | ||
1186 | epoch_t OSDService::note_peer_epoch(int peer, epoch_t e) | |
1187 | { | |
11fdf7f2 | 1188 | std::lock_guard l(peer_map_epoch_lock); |
7c673cae FG |
1189 | map<int,epoch_t>::iterator p = peer_map_epoch.find(peer); |
1190 | if (p != peer_map_epoch.end()) { | |
1191 | if (p->second < e) { | |
1192 | dout(10) << "note_peer_epoch osd." << peer << " has " << e << dendl; | |
1193 | p->second = e; | |
1194 | } else { | |
1195 | dout(30) << "note_peer_epoch osd." << peer << " has " << p->second << " >= " << e << dendl; | |
1196 | } | |
1197 | return p->second; | |
1198 | } else { | |
1199 | dout(10) << "note_peer_epoch osd." << peer << " now has " << e << dendl; | |
1200 | peer_map_epoch[peer] = e; | |
1201 | return e; | |
1202 | } | |
1203 | } | |
1204 | ||
1205 | void OSDService::forget_peer_epoch(int peer, epoch_t as_of) | |
1206 | { | |
11fdf7f2 | 1207 | std::lock_guard l(peer_map_epoch_lock); |
7c673cae FG |
1208 | map<int,epoch_t>::iterator p = peer_map_epoch.find(peer); |
1209 | if (p != peer_map_epoch.end()) { | |
1210 | if (p->second <= as_of) { | |
1211 | dout(10) << "forget_peer_epoch osd." << peer << " as_of " << as_of | |
1212 | << " had " << p->second << dendl; | |
1213 | peer_map_epoch.erase(p); | |
1214 | } else { | |
1215 | dout(10) << "forget_peer_epoch osd." << peer << " as_of " << as_of | |
1216 | << " has " << p->second << " - not forgetting" << dendl; | |
1217 | } | |
1218 | } | |
1219 | } | |
1220 | ||
1221 | bool OSDService::should_share_map(entity_name_t name, Connection *con, | |
1222 | epoch_t epoch, const OSDMapRef& osdmap, | |
1223 | const epoch_t *sent_epoch_p) | |
1224 | { | |
1225 | dout(20) << "should_share_map " | |
1226 | << name << " " << con->get_peer_addr() | |
1227 | << " " << epoch << dendl; | |
1228 | ||
1229 | // does client have old map? | |
1230 | if (name.is_client()) { | |
1231 | bool message_sendmap = epoch < osdmap->get_epoch(); | |
1232 | if (message_sendmap && sent_epoch_p) { | |
1233 | dout(20) << "client session last_sent_epoch: " | |
1234 | << *sent_epoch_p | |
1235 | << " versus osdmap epoch " << osdmap->get_epoch() << dendl; | |
1236 | if (*sent_epoch_p < osdmap->get_epoch()) { | |
1237 | return true; | |
1238 | } // else we don't need to send it out again | |
1239 | } | |
1240 | } | |
1241 | ||
1242 | if (con->get_messenger() == osd->cluster_messenger && | |
1243 | con != osd->cluster_messenger->get_loopback_connection() && | |
1244 | osdmap->is_up(name.num()) && | |
11fdf7f2 TL |
1245 | (osdmap->get_cluster_addrs(name.num()) == con->get_peer_addrs() || |
1246 | osdmap->get_hb_back_addrs(name.num()) == con->get_peer_addrs())) { | |
7c673cae | 1247 | // remember |
11fdf7f2 | 1248 | epoch_t has = std::max(get_peer_epoch(name.num()), epoch); |
7c673cae FG |
1249 | |
1250 | // share? | |
1251 | if (has < osdmap->get_epoch()) { | |
1252 | dout(10) << name << " " << con->get_peer_addr() | |
1253 | << " has old map " << epoch << " < " | |
1254 | << osdmap->get_epoch() << dendl; | |
1255 | return true; | |
1256 | } | |
1257 | } | |
1258 | ||
1259 | return false; | |
1260 | } | |
1261 | ||
1262 | void OSDService::share_map( | |
1263 | entity_name_t name, | |
1264 | Connection *con, | |
1265 | epoch_t epoch, | |
1266 | OSDMapRef& osdmap, | |
1267 | epoch_t *sent_epoch_p) | |
1268 | { | |
1269 | dout(20) << "share_map " | |
1270 | << name << " " << con->get_peer_addr() | |
1271 | << " " << epoch << dendl; | |
1272 | ||
1273 | if (!osd->is_active()) { | |
1274 | /*It is safe not to proceed as OSD is not in healthy state*/ | |
1275 | return; | |
1276 | } | |
1277 | ||
1278 | bool want_shared = should_share_map(name, con, epoch, | |
1279 | osdmap, sent_epoch_p); | |
1280 | ||
1281 | if (want_shared){ | |
1282 | if (name.is_client()) { | |
1283 | dout(10) << name << " has old map " << epoch | |
1284 | << " < " << osdmap->get_epoch() << dendl; | |
1285 | // we know the Session is valid or we wouldn't be sending | |
1286 | if (sent_epoch_p) { | |
1287 | *sent_epoch_p = osdmap->get_epoch(); | |
1288 | } | |
1289 | send_incremental_map(epoch, con, osdmap); | |
1290 | } else if (con->get_messenger() == osd->cluster_messenger && | |
1291 | osdmap->is_up(name.num()) && | |
11fdf7f2 TL |
1292 | (osdmap->get_cluster_addrs(name.num()) == con->get_peer_addrs() || |
1293 | osdmap->get_hb_back_addrs(name.num()) == con->get_peer_addrs())) { | |
1294 | dout(10) << name << " " << con->get_peer_addrs() | |
7c673cae FG |
1295 | << " has old map " << epoch << " < " |
1296 | << osdmap->get_epoch() << dendl; | |
1297 | note_peer_epoch(name.num(), osdmap->get_epoch()); | |
1298 | send_incremental_map(epoch, con, osdmap); | |
1299 | } | |
1300 | } | |
1301 | } | |
1302 | ||
1303 | void OSDService::share_map_peer(int peer, Connection *con, OSDMapRef map) | |
1304 | { | |
1305 | if (!map) | |
1306 | map = get_osdmap(); | |
1307 | ||
1308 | // send map? | |
1309 | epoch_t pe = get_peer_epoch(peer); | |
1310 | if (pe) { | |
1311 | if (pe < map->get_epoch()) { | |
1312 | send_incremental_map(pe, con, map); | |
1313 | note_peer_epoch(peer, map->get_epoch()); | |
1314 | } else | |
1315 | dout(20) << "share_map_peer " << con << " already has epoch " << pe << dendl; | |
1316 | } else { | |
1317 | dout(20) << "share_map_peer " << con << " don't know epoch, doing nothing" << dendl; | |
1318 | // no idea about peer's epoch. | |
1319 | // ??? send recent ??? | |
1320 | // do nothing. | |
1321 | } | |
1322 | } | |
1323 | ||
1324 | bool OSDService::can_inc_scrubs_pending() | |
1325 | { | |
1326 | bool can_inc = false; | |
11fdf7f2 | 1327 | std::lock_guard l(sched_scrub_lock); |
7c673cae FG |
1328 | |
1329 | if (scrubs_pending + scrubs_active < cct->_conf->osd_max_scrubs) { | |
1330 | dout(20) << __func__ << " " << scrubs_pending << " -> " << (scrubs_pending+1) | |
28e407b8 AA |
1331 | << " (max " << cct->_conf->osd_max_scrubs << ", active " << scrubs_active |
1332 | << ")" << dendl; | |
7c673cae FG |
1333 | can_inc = true; |
1334 | } else { | |
28e407b8 AA |
1335 | dout(20) << __func__ << " " << scrubs_pending << " + " << scrubs_active |
1336 | << " active >= max " << cct->_conf->osd_max_scrubs << dendl; | |
7c673cae FG |
1337 | } |
1338 | ||
1339 | return can_inc; | |
1340 | } | |
1341 | ||
1342 | bool OSDService::inc_scrubs_pending() | |
1343 | { | |
1344 | bool result = false; | |
1345 | ||
1346 | sched_scrub_lock.Lock(); | |
1347 | if (scrubs_pending + scrubs_active < cct->_conf->osd_max_scrubs) { | |
1348 | dout(20) << "inc_scrubs_pending " << scrubs_pending << " -> " << (scrubs_pending+1) | |
1349 | << " (max " << cct->_conf->osd_max_scrubs << ", active " << scrubs_active << ")" << dendl; | |
1350 | result = true; | |
1351 | ++scrubs_pending; | |
1352 | } else { | |
1353 | dout(20) << "inc_scrubs_pending " << scrubs_pending << " + " << scrubs_active << " active >= max " << cct->_conf->osd_max_scrubs << dendl; | |
1354 | } | |
1355 | sched_scrub_lock.Unlock(); | |
1356 | ||
1357 | return result; | |
1358 | } | |
1359 | ||
1360 | void OSDService::dec_scrubs_pending() | |
1361 | { | |
1362 | sched_scrub_lock.Lock(); | |
1363 | dout(20) << "dec_scrubs_pending " << scrubs_pending << " -> " << (scrubs_pending-1) | |
1364 | << " (max " << cct->_conf->osd_max_scrubs << ", active " << scrubs_active << ")" << dendl; | |
1365 | --scrubs_pending; | |
11fdf7f2 | 1366 | ceph_assert(scrubs_pending >= 0); |
7c673cae FG |
1367 | sched_scrub_lock.Unlock(); |
1368 | } | |
1369 | ||
1370 | void OSDService::inc_scrubs_active(bool reserved) | |
1371 | { | |
1372 | sched_scrub_lock.Lock(); | |
1373 | ++(scrubs_active); | |
1374 | if (reserved) { | |
1375 | --(scrubs_pending); | |
1376 | dout(20) << "inc_scrubs_active " << (scrubs_active-1) << " -> " << scrubs_active | |
1377 | << " (max " << cct->_conf->osd_max_scrubs | |
1378 | << ", pending " << (scrubs_pending+1) << " -> " << scrubs_pending << ")" << dendl; | |
11fdf7f2 | 1379 | ceph_assert(scrubs_pending >= 0); |
7c673cae FG |
1380 | } else { |
1381 | dout(20) << "inc_scrubs_active " << (scrubs_active-1) << " -> " << scrubs_active | |
1382 | << " (max " << cct->_conf->osd_max_scrubs | |
1383 | << ", pending " << scrubs_pending << ")" << dendl; | |
1384 | } | |
1385 | sched_scrub_lock.Unlock(); | |
1386 | } | |
1387 | ||
1388 | void OSDService::dec_scrubs_active() | |
1389 | { | |
1390 | sched_scrub_lock.Lock(); | |
1391 | dout(20) << "dec_scrubs_active " << scrubs_active << " -> " << (scrubs_active-1) | |
1392 | << " (max " << cct->_conf->osd_max_scrubs << ", pending " << scrubs_pending << ")" << dendl; | |
1393 | --scrubs_active; | |
11fdf7f2 | 1394 | ceph_assert(scrubs_active >= 0); |
7c673cae FG |
1395 | sched_scrub_lock.Unlock(); |
1396 | } | |
1397 | ||
1398 | void OSDService::retrieve_epochs(epoch_t *_boot_epoch, epoch_t *_up_epoch, | |
1399 | epoch_t *_bind_epoch) const | |
1400 | { | |
11fdf7f2 | 1401 | std::lock_guard l(epoch_lock); |
7c673cae FG |
1402 | if (_boot_epoch) |
1403 | *_boot_epoch = boot_epoch; | |
1404 | if (_up_epoch) | |
1405 | *_up_epoch = up_epoch; | |
1406 | if (_bind_epoch) | |
1407 | *_bind_epoch = bind_epoch; | |
1408 | } | |
1409 | ||
1410 | void OSDService::set_epochs(const epoch_t *_boot_epoch, const epoch_t *_up_epoch, | |
1411 | const epoch_t *_bind_epoch) | |
1412 | { | |
11fdf7f2 | 1413 | std::lock_guard l(epoch_lock); |
7c673cae | 1414 | if (_boot_epoch) { |
11fdf7f2 | 1415 | ceph_assert(*_boot_epoch == 0 || *_boot_epoch >= boot_epoch); |
7c673cae FG |
1416 | boot_epoch = *_boot_epoch; |
1417 | } | |
1418 | if (_up_epoch) { | |
11fdf7f2 | 1419 | ceph_assert(*_up_epoch == 0 || *_up_epoch >= up_epoch); |
7c673cae FG |
1420 | up_epoch = *_up_epoch; |
1421 | } | |
1422 | if (_bind_epoch) { | |
11fdf7f2 | 1423 | ceph_assert(*_bind_epoch == 0 || *_bind_epoch >= bind_epoch); |
7c673cae FG |
1424 | bind_epoch = *_bind_epoch; |
1425 | } | |
1426 | } | |
1427 | ||
1428 | bool OSDService::prepare_to_stop() | |
1429 | { | |
11fdf7f2 | 1430 | std::lock_guard l(is_stopping_lock); |
7c673cae FG |
1431 | if (get_state() != NOT_STOPPING) |
1432 | return false; | |
1433 | ||
1434 | OSDMapRef osdmap = get_osdmap(); | |
1435 | if (osdmap && osdmap->is_up(whoami)) { | |
1436 | dout(0) << __func__ << " telling mon we are shutting down" << dendl; | |
1437 | set_state(PREPARING_TO_STOP); | |
11fdf7f2 TL |
1438 | monc->send_mon_message( |
1439 | new MOSDMarkMeDown( | |
1440 | monc->get_fsid(), | |
1441 | whoami, | |
1442 | osdmap->get_addrs(whoami), | |
1443 | osdmap->get_epoch(), | |
1444 | true // request ack | |
1445 | )); | |
7c673cae FG |
1446 | utime_t now = ceph_clock_now(); |
1447 | utime_t timeout; | |
1448 | timeout.set_from_double(now + cct->_conf->osd_mon_shutdown_timeout); | |
1449 | while ((ceph_clock_now() < timeout) && | |
1450 | (get_state() != STOPPING)) { | |
1451 | is_stopping_cond.WaitUntil(is_stopping_lock, timeout); | |
1452 | } | |
1453 | } | |
1454 | dout(0) << __func__ << " starting shutdown" << dendl; | |
1455 | set_state(STOPPING); | |
1456 | return true; | |
1457 | } | |
1458 | ||
1459 | void OSDService::got_stop_ack() | |
1460 | { | |
11fdf7f2 | 1461 | std::lock_guard l(is_stopping_lock); |
7c673cae FG |
1462 | if (get_state() == PREPARING_TO_STOP) { |
1463 | dout(0) << __func__ << " starting shutdown" << dendl; | |
1464 | set_state(STOPPING); | |
1465 | is_stopping_cond.Signal(); | |
1466 | } else { | |
1467 | dout(10) << __func__ << " ignoring msg" << dendl; | |
1468 | } | |
1469 | } | |
1470 | ||
1471 | MOSDMap *OSDService::build_incremental_map_msg(epoch_t since, epoch_t to, | |
1472 | OSDSuperblock& sblock) | |
1473 | { | |
28e407b8 AA |
1474 | MOSDMap *m = new MOSDMap(monc->get_fsid(), |
1475 | osdmap->get_encoding_features()); | |
7c673cae FG |
1476 | m->oldest_map = max_oldest_map; |
1477 | m->newest_map = sblock.newest_map; | |
1478 | ||
11fdf7f2 TL |
1479 | int max = cct->_conf->osd_map_message_max; |
1480 | ssize_t max_bytes = cct->_conf->osd_map_message_max_bytes; | |
1481 | ||
1482 | if (since < m->oldest_map) { | |
1483 | // we don't have the next map the target wants, so start with a | |
1484 | // full map. | |
1485 | bufferlist bl; | |
1486 | dout(10) << __func__ << " oldest map " << max_oldest_map << " > since " | |
1487 | << since << ", starting with full map" << dendl; | |
1488 | since = m->oldest_map; | |
1489 | if (!get_map_bl(since, bl)) { | |
1490 | derr << __func__ << " missing full map " << since << dendl; | |
1491 | goto panic; | |
1492 | } | |
1493 | max--; | |
1494 | max_bytes -= bl.length(); | |
1495 | m->maps[since].claim(bl); | |
1496 | } | |
1497 | for (epoch_t e = since + 1; e <= to; ++e) { | |
7c673cae | 1498 | bufferlist bl; |
11fdf7f2 | 1499 | if (get_inc_map_bl(e, bl)) { |
7c673cae | 1500 | m->incremental_maps[e].claim(bl); |
11fdf7f2 TL |
1501 | } else { |
1502 | derr << __func__ << " missing incremental map " << e << dendl; | |
1503 | if (!get_map_bl(e, bl)) { | |
1504 | derr << __func__ << " also missing full map " << e << dendl; | |
1505 | goto panic; | |
1506 | } | |
7c673cae | 1507 | m->maps[e].claim(bl); |
11fdf7f2 TL |
1508 | } |
1509 | max--; | |
1510 | max_bytes -= bl.length(); | |
1511 | if (max <= 0 || max_bytes <= 0) { | |
7c673cae | 1512 | break; |
11fdf7f2 TL |
1513 | } |
1514 | } | |
1515 | return m; | |
1516 | ||
1517 | panic: | |
1518 | if (!m->maps.empty() || | |
1519 | !m->incremental_maps.empty()) { | |
1520 | // send what we have so far | |
1521 | return m; | |
1522 | } | |
1523 | // send something | |
1524 | bufferlist bl; | |
1525 | if (get_inc_map_bl(m->newest_map, bl)) { | |
1526 | m->incremental_maps[m->newest_map].claim(bl); | |
1527 | } else { | |
1528 | derr << __func__ << " unable to load latest map " << m->newest_map << dendl; | |
1529 | if (!get_map_bl(m->newest_map, bl)) { | |
1530 | derr << __func__ << " unable to load latest full map " << m->newest_map | |
7c673cae | 1531 | << dendl; |
11fdf7f2 | 1532 | ceph_abort(); |
7c673cae | 1533 | } |
11fdf7f2 | 1534 | m->maps[m->newest_map].claim(bl); |
7c673cae FG |
1535 | } |
1536 | return m; | |
1537 | } | |
1538 | ||
1539 | void OSDService::send_map(MOSDMap *m, Connection *con) | |
1540 | { | |
1541 | con->send_message(m); | |
1542 | } | |
1543 | ||
1544 | void OSDService::send_incremental_map(epoch_t since, Connection *con, | |
1545 | OSDMapRef& osdmap) | |
1546 | { | |
1547 | epoch_t to = osdmap->get_epoch(); | |
1548 | dout(10) << "send_incremental_map " << since << " -> " << to | |
1549 | << " to " << con << " " << con->get_peer_addr() << dendl; | |
1550 | ||
1551 | MOSDMap *m = NULL; | |
1552 | while (!m) { | |
1553 | OSDSuperblock sblock(get_superblock()); | |
1554 | if (since < sblock.oldest_map) { | |
1555 | // just send latest full map | |
28e407b8 AA |
1556 | MOSDMap *m = new MOSDMap(monc->get_fsid(), |
1557 | osdmap->get_encoding_features()); | |
7c673cae FG |
1558 | m->oldest_map = max_oldest_map; |
1559 | m->newest_map = sblock.newest_map; | |
1560 | get_map_bl(to, m->maps[to]); | |
1561 | send_map(m, con); | |
1562 | return; | |
1563 | } | |
1564 | ||
1565 | if (to > since && (int64_t)(to - since) > cct->_conf->osd_map_share_max_epochs) { | |
1566 | dout(10) << " " << (to - since) << " > max " << cct->_conf->osd_map_share_max_epochs | |
1567 | << ", only sending most recent" << dendl; | |
1568 | since = to - cct->_conf->osd_map_share_max_epochs; | |
1569 | } | |
1570 | ||
7c673cae FG |
1571 | m = build_incremental_map_msg(since, to, sblock); |
1572 | } | |
1573 | send_map(m, con); | |
1574 | } | |
1575 | ||
1576 | bool OSDService::_get_map_bl(epoch_t e, bufferlist& bl) | |
1577 | { | |
1578 | bool found = map_bl_cache.lookup(e, &bl); | |
31f18b77 FG |
1579 | if (found) { |
1580 | if (logger) | |
1581 | logger->inc(l_osd_map_bl_cache_hit); | |
7c673cae | 1582 | return true; |
31f18b77 FG |
1583 | } |
1584 | if (logger) | |
1585 | logger->inc(l_osd_map_bl_cache_miss); | |
11fdf7f2 | 1586 | found = store->read(meta_ch, |
31f18b77 FG |
1587 | OSD::get_osdmap_pobject_name(e), 0, 0, bl, |
1588 | CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) >= 0; | |
1589 | if (found) { | |
7c673cae | 1590 | _add_map_bl(e, bl); |
31f18b77 | 1591 | } |
7c673cae FG |
1592 | return found; |
1593 | } | |
1594 | ||
1595 | bool OSDService::get_inc_map_bl(epoch_t e, bufferlist& bl) | |
1596 | { | |
11fdf7f2 | 1597 | std::lock_guard l(map_cache_lock); |
7c673cae | 1598 | bool found = map_bl_inc_cache.lookup(e, &bl); |
31f18b77 FG |
1599 | if (found) { |
1600 | if (logger) | |
1601 | logger->inc(l_osd_map_bl_cache_hit); | |
7c673cae | 1602 | return true; |
31f18b77 FG |
1603 | } |
1604 | if (logger) | |
1605 | logger->inc(l_osd_map_bl_cache_miss); | |
11fdf7f2 | 1606 | found = store->read(meta_ch, |
31f18b77 FG |
1607 | OSD::get_inc_osdmap_pobject_name(e), 0, 0, bl, |
1608 | CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) >= 0; | |
1609 | if (found) { | |
7c673cae | 1610 | _add_map_inc_bl(e, bl); |
31f18b77 | 1611 | } |
7c673cae FG |
1612 | return found; |
1613 | } | |
1614 | ||
1615 | void OSDService::_add_map_bl(epoch_t e, bufferlist& bl) | |
1616 | { | |
1617 | dout(10) << "add_map_bl " << e << " " << bl.length() << " bytes" << dendl; | |
31f18b77 FG |
1618 | // cache a contiguous buffer |
1619 | if (bl.get_num_buffers() > 1) { | |
1620 | bl.rebuild(); | |
1621 | } | |
1622 | bl.try_assign_to_mempool(mempool::mempool_osd_mapbl); | |
7c673cae FG |
1623 | map_bl_cache.add(e, bl); |
1624 | } | |
1625 | ||
1626 | void OSDService::_add_map_inc_bl(epoch_t e, bufferlist& bl) | |
1627 | { | |
1628 | dout(10) << "add_map_inc_bl " << e << " " << bl.length() << " bytes" << dendl; | |
31f18b77 FG |
1629 | // cache a contiguous buffer |
1630 | if (bl.get_num_buffers() > 1) { | |
1631 | bl.rebuild(); | |
1632 | } | |
1633 | bl.try_assign_to_mempool(mempool::mempool_osd_mapbl); | |
7c673cae FG |
1634 | map_bl_inc_cache.add(e, bl); |
1635 | } | |
1636 | ||
11fdf7f2 | 1637 | int OSDService::get_deleted_pool_pg_num(int64_t pool) |
7c673cae | 1638 | { |
11fdf7f2 TL |
1639 | std::lock_guard l(map_cache_lock); |
1640 | auto p = deleted_pool_pg_nums.find(pool); | |
1641 | if (p != deleted_pool_pg_nums.end()) { | |
1642 | return p->second; | |
31f18b77 | 1643 | } |
11fdf7f2 TL |
1644 | dout(20) << __func__ << " " << pool << " loading" << dendl; |
1645 | ghobject_t oid = OSD::make_final_pool_info_oid(pool); | |
1646 | bufferlist bl; | |
1647 | int r = store->read(meta_ch, oid, 0, 0, bl); | |
1648 | ceph_assert(r >= 0); | |
1649 | auto blp = bl.cbegin(); | |
1650 | pg_pool_t pi; | |
1651 | ::decode(pi, blp); | |
1652 | deleted_pool_pg_nums[pool] = pi.get_pg_num(); | |
1653 | dout(20) << __func__ << " " << pool << " got " << pi.get_pg_num() << dendl; | |
1654 | return pi.get_pg_num(); | |
7c673cae FG |
1655 | } |
1656 | ||
1657 | OSDMapRef OSDService::_add_map(OSDMap *o) | |
1658 | { | |
1659 | epoch_t e = o->get_epoch(); | |
1660 | ||
1661 | if (cct->_conf->osd_map_dedup) { | |
1662 | // Dedup against an existing map at a nearby epoch | |
1663 | OSDMapRef for_dedup = map_cache.lower_bound(e); | |
1664 | if (for_dedup) { | |
1665 | OSDMap::dedup(for_dedup.get(), o); | |
1666 | } | |
1667 | } | |
1668 | bool existed; | |
1669 | OSDMapRef l = map_cache.add(e, o, &existed); | |
1670 | if (existed) { | |
1671 | delete o; | |
1672 | } | |
1673 | return l; | |
1674 | } | |
1675 | ||
1676 | OSDMapRef OSDService::try_get_map(epoch_t epoch) | |
1677 | { | |
11fdf7f2 | 1678 | std::lock_guard l(map_cache_lock); |
7c673cae FG |
1679 | OSDMapRef retval = map_cache.lookup(epoch); |
1680 | if (retval) { | |
1681 | dout(30) << "get_map " << epoch << " -cached" << dendl; | |
1682 | if (logger) { | |
1683 | logger->inc(l_osd_map_cache_hit); | |
1684 | } | |
1685 | return retval; | |
1686 | } | |
1687 | if (logger) { | |
1688 | logger->inc(l_osd_map_cache_miss); | |
1689 | epoch_t lb = map_cache.cached_key_lower_bound(); | |
1690 | if (epoch < lb) { | |
1691 | dout(30) << "get_map " << epoch << " - miss, below lower bound" << dendl; | |
1692 | logger->inc(l_osd_map_cache_miss_low); | |
1693 | logger->inc(l_osd_map_cache_miss_low_avg, lb - epoch); | |
1694 | } | |
1695 | } | |
1696 | ||
1697 | OSDMap *map = new OSDMap; | |
1698 | if (epoch > 0) { | |
1699 | dout(20) << "get_map " << epoch << " - loading and decoding " << map << dendl; | |
1700 | bufferlist bl; | |
1701 | if (!_get_map_bl(epoch, bl) || bl.length() == 0) { | |
1702 | derr << "failed to load OSD map for epoch " << epoch << ", got " << bl.length() << " bytes" << dendl; | |
1703 | delete map; | |
1704 | return OSDMapRef(); | |
1705 | } | |
1706 | map->decode(bl); | |
1707 | } else { | |
1708 | dout(20) << "get_map " << epoch << " - return initial " << map << dendl; | |
1709 | } | |
1710 | return _add_map(map); | |
1711 | } | |
1712 | ||
1713 | // ops | |
1714 | ||
1715 | ||
1716 | void OSDService::reply_op_error(OpRequestRef op, int err) | |
1717 | { | |
1718 | reply_op_error(op, err, eversion_t(), 0); | |
1719 | } | |
1720 | ||
1721 | void OSDService::reply_op_error(OpRequestRef op, int err, eversion_t v, | |
1722 | version_t uv) | |
1723 | { | |
1724 | const MOSDOp *m = static_cast<const MOSDOp*>(op->get_req()); | |
11fdf7f2 | 1725 | ceph_assert(m->get_type() == CEPH_MSG_OSD_OP); |
7c673cae FG |
1726 | int flags; |
1727 | flags = m->get_flags() & (CEPH_OSD_FLAG_ACK|CEPH_OSD_FLAG_ONDISK); | |
1728 | ||
11fdf7f2 | 1729 | MOSDOpReply *reply = new MOSDOpReply(m, err, osdmap->get_epoch(), flags, true); |
7c673cae FG |
1730 | reply->set_reply_versions(v, uv); |
1731 | m->get_connection()->send_message(reply); | |
1732 | } | |
1733 | ||
1734 | void OSDService::handle_misdirected_op(PG *pg, OpRequestRef op) | |
1735 | { | |
31f18b77 FG |
1736 | if (!cct->_conf->osd_debug_misdirected_ops) { |
1737 | return; | |
1738 | } | |
1739 | ||
7c673cae | 1740 | const MOSDOp *m = static_cast<const MOSDOp*>(op->get_req()); |
11fdf7f2 | 1741 | ceph_assert(m->get_type() == CEPH_MSG_OSD_OP); |
7c673cae | 1742 | |
11fdf7f2 | 1743 | ceph_assert(m->get_map_epoch() >= pg->get_history().same_primary_since); |
7c673cae FG |
1744 | |
1745 | if (pg->is_ec_pg()) { | |
1746 | /** | |
1747 | * OSD recomputes op target based on current OSDMap. With an EC pg, we | |
1748 | * can get this result: | |
1749 | * 1) client at map 512 sends an op to osd 3, pg_t 3.9 based on mapping | |
1750 | * [CRUSH_ITEM_NONE, 2, 3]/3 | |
1751 | * 2) OSD 3 at map 513 remaps op to osd 3, spg_t 3.9s0 based on mapping | |
1752 | * [3, 2, 3]/3 | |
1753 | * 3) PG 3.9s0 dequeues the op at epoch 512 and notices that it isn't primary | |
1754 | * -- misdirected op | |
1755 | * 4) client resends and this time PG 3.9s0 having caught up to 513 gets | |
1756 | * it and fulfils it | |
1757 | * | |
1758 | * We can't compute the op target based on the sending map epoch due to | |
1759 | * splitting. The simplest thing is to detect such cases here and drop | |
1760 | * them without an error (the client will resend anyway). | |
1761 | */ | |
11fdf7f2 | 1762 | ceph_assert(m->get_map_epoch() <= superblock.newest_map); |
7c673cae FG |
1763 | OSDMapRef opmap = try_get_map(m->get_map_epoch()); |
1764 | if (!opmap) { | |
1765 | dout(7) << __func__ << ": " << *pg << " no longer have map for " | |
1766 | << m->get_map_epoch() << ", dropping" << dendl; | |
1767 | return; | |
1768 | } | |
1769 | pg_t _pgid = m->get_raw_pg(); | |
1770 | spg_t pgid; | |
1771 | if ((m->get_flags() & CEPH_OSD_FLAG_PGOP) == 0) | |
1772 | _pgid = opmap->raw_pg_to_pg(_pgid); | |
1773 | if (opmap->get_primary_shard(_pgid, &pgid) && | |
11fdf7f2 | 1774 | pgid.shard != pg->pg_id.shard) { |
7c673cae FG |
1775 | dout(7) << __func__ << ": " << *pg << " primary changed since " |
1776 | << m->get_map_epoch() << ", dropping" << dendl; | |
1777 | return; | |
1778 | } | |
1779 | } | |
1780 | ||
1781 | dout(7) << *pg << " misdirected op in " << m->get_map_epoch() << dendl; | |
1782 | clog->warn() << m->get_source_inst() << " misdirected " << m->get_reqid() | |
1783 | << " pg " << m->get_raw_pg() | |
1784 | << " to osd." << whoami | |
11fdf7f2 | 1785 | << " not " << pg->get_acting() |
7c673cae | 1786 | << " in e" << m->get_map_epoch() << "/" << osdmap->get_epoch(); |
7c673cae FG |
1787 | } |
1788 | ||
11fdf7f2 | 1789 | void OSDService::enqueue_back(OpQueueItem&& qi) |
7c673cae | 1790 | { |
11fdf7f2 | 1791 | osd->op_shardedwq.queue(std::move(qi)); |
7c673cae FG |
1792 | } |
1793 | ||
11fdf7f2 | 1794 | void OSDService::enqueue_front(OpQueueItem&& qi) |
7c673cae | 1795 | { |
11fdf7f2 | 1796 | osd->op_shardedwq.queue_front(std::move(qi)); |
7c673cae FG |
1797 | } |
1798 | ||
11fdf7f2 TL |
1799 | void OSDService::queue_recovery_context( |
1800 | PG *pg, | |
1801 | GenContext<ThreadPool::TPHandle&> *c) | |
7c673cae | 1802 | { |
11fdf7f2 TL |
1803 | epoch_t e = get_osdmap_epoch(); |
1804 | enqueue_back( | |
1805 | OpQueueItem( | |
1806 | unique_ptr<OpQueueItem::OpQueueable>( | |
1807 | new PGRecoveryContext(pg->get_pgid(), c, e)), | |
1808 | cct->_conf->osd_recovery_cost, | |
1809 | cct->_conf->osd_recovery_priority, | |
1810 | ceph_clock_now(), | |
1811 | 0, | |
1812 | e)); | |
7c673cae FG |
1813 | } |
1814 | ||
1815 | void OSDService::queue_for_snap_trim(PG *pg) | |
1816 | { | |
1817 | dout(10) << "queueing " << *pg << " for snaptrim" << dendl; | |
11fdf7f2 TL |
1818 | enqueue_back( |
1819 | OpQueueItem( | |
1820 | unique_ptr<OpQueueItem::OpQueueable>( | |
1821 | new PGSnapTrim(pg->get_pgid(), pg->get_osdmap_epoch())), | |
1822 | cct->_conf->osd_snap_trim_cost, | |
1823 | cct->_conf->osd_snap_trim_priority, | |
1824 | ceph_clock_now(), | |
1825 | 0, | |
1826 | pg->get_osdmap_epoch())); | |
1827 | } | |
1828 | ||
1829 | void OSDService::queue_for_scrub(PG *pg, bool with_high_priority) | |
1830 | { | |
1831 | unsigned scrub_queue_priority = pg->scrubber.priority; | |
1832 | if (with_high_priority && scrub_queue_priority < cct->_conf->osd_client_op_priority) { | |
1833 | scrub_queue_priority = cct->_conf->osd_client_op_priority; | |
1834 | } | |
1835 | const auto epoch = pg->get_osdmap_epoch(); | |
1836 | enqueue_back( | |
1837 | OpQueueItem( | |
1838 | unique_ptr<OpQueueItem::OpQueueable>(new PGScrub(pg->get_pgid(), epoch)), | |
1839 | cct->_conf->osd_scrub_cost, | |
1840 | scrub_queue_priority, | |
1841 | ceph_clock_now(), | |
1842 | 0, | |
1843 | epoch)); | |
1844 | } | |
1845 | ||
1846 | void OSDService::queue_for_pg_delete(spg_t pgid, epoch_t e) | |
1847 | { | |
1848 | dout(10) << __func__ << " on " << pgid << " e " << e << dendl; | |
1849 | enqueue_back( | |
1850 | OpQueueItem( | |
1851 | unique_ptr<OpQueueItem::OpQueueable>( | |
1852 | new PGDelete(pgid, e)), | |
1853 | cct->_conf->osd_pg_delete_cost, | |
1854 | cct->_conf->osd_pg_delete_priority, | |
1855 | ceph_clock_now(), | |
1856 | 0, | |
1857 | e)); | |
1858 | } | |
1859 | ||
1860 | bool OSDService::try_finish_pg_delete(PG *pg, unsigned old_pg_num) | |
1861 | { | |
1862 | return osd->try_finish_pg_delete(pg, old_pg_num); | |
1863 | } | |
1864 | ||
1865 | // --- | |
1866 | ||
1867 | void OSDService::set_ready_to_merge_source(PG *pg, eversion_t version) | |
1868 | { | |
1869 | std::lock_guard l(merge_lock); | |
1870 | dout(10) << __func__ << " " << pg->pg_id << dendl; | |
1871 | ready_to_merge_source[pg->pg_id.pgid] = version; | |
1872 | assert(not_ready_to_merge_source.count(pg->pg_id.pgid) == 0); | |
1873 | _send_ready_to_merge(); | |
1874 | } | |
1875 | ||
1876 | void OSDService::set_ready_to_merge_target(PG *pg, | |
1877 | eversion_t version, | |
1878 | epoch_t last_epoch_started, | |
1879 | epoch_t last_epoch_clean) | |
1880 | { | |
1881 | std::lock_guard l(merge_lock); | |
1882 | dout(10) << __func__ << " " << pg->pg_id << dendl; | |
1883 | ready_to_merge_target.insert(make_pair(pg->pg_id.pgid, | |
1884 | make_tuple(version, | |
1885 | last_epoch_started, | |
1886 | last_epoch_clean))); | |
1887 | assert(not_ready_to_merge_target.count(pg->pg_id.pgid) == 0); | |
1888 | _send_ready_to_merge(); | |
1889 | } | |
1890 | ||
1891 | void OSDService::set_not_ready_to_merge_source(pg_t source) | |
1892 | { | |
1893 | std::lock_guard l(merge_lock); | |
1894 | dout(10) << __func__ << " " << source << dendl; | |
1895 | not_ready_to_merge_source.insert(source); | |
1896 | assert(ready_to_merge_source.count(source) == 0); | |
1897 | _send_ready_to_merge(); | |
1898 | } | |
1899 | ||
1900 | void OSDService::set_not_ready_to_merge_target(pg_t target, pg_t source) | |
1901 | { | |
1902 | std::lock_guard l(merge_lock); | |
1903 | dout(10) << __func__ << " " << target << " source " << source << dendl; | |
1904 | not_ready_to_merge_target[target] = source; | |
1905 | assert(ready_to_merge_target.count(target) == 0); | |
1906 | _send_ready_to_merge(); | |
1907 | } | |
1908 | ||
1909 | void OSDService::send_ready_to_merge() | |
1910 | { | |
1911 | std::lock_guard l(merge_lock); | |
1912 | _send_ready_to_merge(); | |
1913 | } | |
1914 | ||
1915 | void OSDService::_send_ready_to_merge() | |
1916 | { | |
1917 | dout(20) << __func__ | |
1918 | << " ready_to_merge_source " << ready_to_merge_source | |
1919 | << " not_ready_to_merge_source " << not_ready_to_merge_source | |
1920 | << " ready_to_merge_target " << ready_to_merge_target | |
1921 | << " not_ready_to_merge_target " << not_ready_to_merge_target | |
1922 | << " sent_ready_to_merge_source " << sent_ready_to_merge_source | |
1923 | << dendl; | |
1924 | for (auto src : not_ready_to_merge_source) { | |
1925 | if (sent_ready_to_merge_source.count(src) == 0) { | |
1926 | monc->send_mon_message(new MOSDPGReadyToMerge( | |
1927 | src, | |
1928 | {}, {}, 0, 0, | |
1929 | false, | |
1930 | osdmap->get_epoch())); | |
1931 | sent_ready_to_merge_source.insert(src); | |
1932 | } | |
1933 | } | |
1934 | for (auto p : not_ready_to_merge_target) { | |
1935 | if (sent_ready_to_merge_source.count(p.second) == 0) { | |
1936 | monc->send_mon_message(new MOSDPGReadyToMerge( | |
1937 | p.second, | |
1938 | {}, {}, 0, 0, | |
1939 | false, | |
1940 | osdmap->get_epoch())); | |
1941 | sent_ready_to_merge_source.insert(p.second); | |
1942 | } | |
1943 | } | |
1944 | for (auto src : ready_to_merge_source) { | |
1945 | if (not_ready_to_merge_source.count(src.first) || | |
1946 | not_ready_to_merge_target.count(src.first.get_parent())) { | |
1947 | continue; | |
1948 | } | |
1949 | auto p = ready_to_merge_target.find(src.first.get_parent()); | |
1950 | if (p != ready_to_merge_target.end() && | |
1951 | sent_ready_to_merge_source.count(src.first) == 0) { | |
1952 | monc->send_mon_message(new MOSDPGReadyToMerge( | |
1953 | src.first, // source pgid | |
1954 | src.second, // src version | |
1955 | std::get<0>(p->second), // target version | |
1956 | std::get<1>(p->second), // PG's last_epoch_started | |
1957 | std::get<2>(p->second), // PG's last_epoch_clean | |
1958 | true, | |
1959 | osdmap->get_epoch())); | |
1960 | sent_ready_to_merge_source.insert(src.first); | |
1961 | } | |
1962 | } | |
1963 | } | |
1964 | ||
1965 | void OSDService::clear_ready_to_merge(PG *pg) | |
1966 | { | |
1967 | std::lock_guard l(merge_lock); | |
1968 | dout(10) << __func__ << " " << pg->pg_id << dendl; | |
1969 | ready_to_merge_source.erase(pg->pg_id.pgid); | |
1970 | ready_to_merge_target.erase(pg->pg_id.pgid); | |
1971 | not_ready_to_merge_source.erase(pg->pg_id.pgid); | |
1972 | not_ready_to_merge_target.erase(pg->pg_id.pgid); | |
1973 | sent_ready_to_merge_source.erase(pg->pg_id.pgid); | |
1974 | } | |
1975 | ||
1976 | void OSDService::clear_sent_ready_to_merge() | |
1977 | { | |
1978 | std::lock_guard l(merge_lock); | |
1979 | sent_ready_to_merge_source.clear(); | |
1980 | } | |
1981 | ||
1982 | void OSDService::prune_sent_ready_to_merge(OSDMapRef& osdmap) | |
1983 | { | |
1984 | std::lock_guard l(merge_lock); | |
1985 | auto i = sent_ready_to_merge_source.begin(); | |
1986 | while (i != sent_ready_to_merge_source.end()) { | |
1987 | if (!osdmap->pg_exists(*i)) { | |
1988 | dout(10) << __func__ << " " << *i << dendl; | |
1989 | i = sent_ready_to_merge_source.erase(i); | |
1990 | } else { | |
1991 | ++i; | |
1992 | } | |
1993 | } | |
7c673cae FG |
1994 | } |
1995 | ||
11fdf7f2 TL |
1996 | // --- |
1997 | ||
1998 | void OSDService::_queue_for_recovery( | |
1999 | std::pair<epoch_t, PGRef> p, | |
2000 | uint64_t reserved_pushes) | |
2001 | { | |
2002 | ceph_assert(recovery_lock.is_locked_by_me()); | |
2003 | enqueue_back( | |
2004 | OpQueueItem( | |
2005 | unique_ptr<OpQueueItem::OpQueueable>( | |
2006 | new PGRecovery( | |
2007 | p.second->get_pgid(), p.first, reserved_pushes)), | |
2008 | cct->_conf->osd_recovery_cost, | |
2009 | cct->_conf->osd_recovery_priority, | |
2010 | ceph_clock_now(), | |
2011 | 0, | |
2012 | p.first)); | |
2013 | } | |
7c673cae FG |
2014 | |
2015 | // ==================================================================== | |
2016 | // OSD | |
2017 | ||
2018 | #undef dout_prefix | |
2019 | #define dout_prefix *_dout | |
2020 | ||
2021 | // Commands shared between OSD's console and admin console: | |
2022 | namespace ceph { | |
2023 | namespace osd_cmds { | |
2024 | ||
11fdf7f2 | 2025 | int heap(CephContext& cct, const cmdmap_t& cmdmap, Formatter& f, std::ostream& os); |
7c673cae FG |
2026 | |
2027 | }} // namespace ceph::osd_cmds | |
2028 | ||
11fdf7f2 | 2029 | int OSD::mkfs(CephContext *cct, ObjectStore *store, uuid_d fsid, int whoami) |
7c673cae FG |
2030 | { |
2031 | int ret; | |
2032 | ||
7c673cae FG |
2033 | OSDSuperblock sb; |
2034 | bufferlist sbbl; | |
11fdf7f2 | 2035 | ObjectStore::CollectionHandle ch; |
7c673cae FG |
2036 | |
2037 | // if we are fed a uuid for this osd, use it. | |
2038 | store->set_fsid(cct->_conf->osd_uuid); | |
2039 | ||
2040 | ret = store->mkfs(); | |
2041 | if (ret) { | |
224ce89b WB |
2042 | derr << "OSD::mkfs: ObjectStore::mkfs failed with error " |
2043 | << cpp_strerror(ret) << dendl; | |
7c673cae FG |
2044 | goto free_store; |
2045 | } | |
2046 | ||
31f18b77 | 2047 | store->set_cache_shards(1); // doesn't matter for mkfs! |
7c673cae FG |
2048 | |
2049 | ret = store->mount(); | |
2050 | if (ret) { | |
224ce89b WB |
2051 | derr << "OSD::mkfs: couldn't mount ObjectStore: error " |
2052 | << cpp_strerror(ret) << dendl; | |
7c673cae FG |
2053 | goto free_store; |
2054 | } | |
2055 | ||
11fdf7f2 TL |
2056 | ch = store->open_collection(coll_t::meta()); |
2057 | if (ch) { | |
2058 | ret = store->read(ch, OSD_SUPERBLOCK_GOBJECT, 0, 0, sbbl); | |
2059 | if (ret < 0) { | |
2060 | derr << "OSD::mkfs: have meta collection but no superblock" << dendl; | |
2061 | goto free_store; | |
2062 | } | |
7c673cae FG |
2063 | /* if we already have superblock, check content of superblock */ |
2064 | dout(0) << " have superblock" << dendl; | |
11fdf7f2 TL |
2065 | auto p = sbbl.cbegin(); |
2066 | decode(sb, p); | |
7c673cae FG |
2067 | if (whoami != sb.whoami) { |
2068 | derr << "provided osd id " << whoami << " != superblock's " << sb.whoami | |
2069 | << dendl; | |
2070 | ret = -EINVAL; | |
2071 | goto umount_store; | |
2072 | } | |
2073 | if (fsid != sb.cluster_fsid) { | |
2074 | derr << "provided cluster fsid " << fsid | |
2075 | << " != superblock's " << sb.cluster_fsid << dendl; | |
2076 | ret = -EINVAL; | |
2077 | goto umount_store; | |
2078 | } | |
2079 | } else { | |
2080 | // create superblock | |
2081 | sb.cluster_fsid = fsid; | |
2082 | sb.osd_fsid = store->get_fsid(); | |
2083 | sb.whoami = whoami; | |
2084 | sb.compat_features = get_osd_initial_compat_set(); | |
2085 | ||
2086 | bufferlist bl; | |
11fdf7f2 | 2087 | encode(sb, bl); |
7c673cae | 2088 | |
11fdf7f2 TL |
2089 | ObjectStore::CollectionHandle ch = store->create_new_collection( |
2090 | coll_t::meta()); | |
7c673cae FG |
2091 | ObjectStore::Transaction t; |
2092 | t.create_collection(coll_t::meta(), 0); | |
2093 | t.write(coll_t::meta(), OSD_SUPERBLOCK_GOBJECT, 0, bl.length(), bl); | |
11fdf7f2 | 2094 | ret = store->queue_transaction(ch, std::move(t)); |
7c673cae FG |
2095 | if (ret) { |
2096 | derr << "OSD::mkfs: error while writing OSD_SUPERBLOCK_GOBJECT: " | |
11fdf7f2 | 2097 | << "queue_transaction returned " << cpp_strerror(ret) << dendl; |
7c673cae FG |
2098 | goto umount_store; |
2099 | } | |
2100 | } | |
2101 | ||
3efd9988 | 2102 | ret = write_meta(cct, store, sb.cluster_fsid, sb.osd_fsid, whoami); |
7c673cae | 2103 | if (ret) { |
224ce89b WB |
2104 | derr << "OSD::mkfs: failed to write fsid file: error " |
2105 | << cpp_strerror(ret) << dendl; | |
7c673cae FG |
2106 | goto umount_store; |
2107 | } | |
2108 | ||
2109 | umount_store: | |
11fdf7f2 TL |
2110 | if (ch) { |
2111 | ch.reset(); | |
2112 | } | |
7c673cae FG |
2113 | store->umount(); |
2114 | free_store: | |
2115 | delete store; | |
2116 | return ret; | |
2117 | } | |
2118 | ||
3efd9988 | 2119 | int OSD::write_meta(CephContext *cct, ObjectStore *store, uuid_d& cluster_fsid, uuid_d& osd_fsid, int whoami) |
7c673cae FG |
2120 | { |
2121 | char val[80]; | |
2122 | int r; | |
2123 | ||
2124 | snprintf(val, sizeof(val), "%s", CEPH_OSD_ONDISK_MAGIC); | |
2125 | r = store->write_meta("magic", val); | |
2126 | if (r < 0) | |
2127 | return r; | |
2128 | ||
2129 | snprintf(val, sizeof(val), "%d", whoami); | |
2130 | r = store->write_meta("whoami", val); | |
2131 | if (r < 0) | |
2132 | return r; | |
2133 | ||
2134 | cluster_fsid.print(val); | |
2135 | r = store->write_meta("ceph_fsid", val); | |
2136 | if (r < 0) | |
2137 | return r; | |
2138 | ||
11fdf7f2 | 2139 | string key = cct->_conf.get_val<string>("key"); |
3efd9988 FG |
2140 | if (key.size()) { |
2141 | r = store->write_meta("osd_key", key); | |
2142 | if (r < 0) | |
2143 | return r; | |
b32b8144 | 2144 | } else { |
11fdf7f2 | 2145 | string keyfile = cct->_conf.get_val<string>("keyfile"); |
b32b8144 FG |
2146 | if (!keyfile.empty()) { |
2147 | bufferlist keybl; | |
2148 | string err; | |
11fdf7f2 | 2149 | r = keybl.read_file(keyfile.c_str(), &err); |
b32b8144 FG |
2150 | if (r < 0) { |
2151 | derr << __func__ << " failed to read keyfile " << keyfile << ": " | |
2152 | << err << ": " << cpp_strerror(r) << dendl; | |
2153 | return r; | |
2154 | } | |
2155 | r = store->write_meta("osd_key", keybl.to_str()); | |
2156 | if (r < 0) | |
2157 | return r; | |
2158 | } | |
3efd9988 FG |
2159 | } |
2160 | ||
7c673cae FG |
2161 | r = store->write_meta("ready", "ready"); |
2162 | if (r < 0) | |
2163 | return r; | |
2164 | ||
2165 | return 0; | |
2166 | } | |
2167 | ||
11fdf7f2 TL |
2168 | int OSD::peek_meta(ObjectStore *store, |
2169 | std::string *magic, | |
2170 | uuid_d *cluster_fsid, | |
2171 | uuid_d *osd_fsid, | |
2172 | int *whoami, | |
2173 | int *require_osd_release) | |
7c673cae FG |
2174 | { |
2175 | string val; | |
2176 | ||
2177 | int r = store->read_meta("magic", &val); | |
2178 | if (r < 0) | |
2179 | return r; | |
11fdf7f2 | 2180 | *magic = val; |
7c673cae FG |
2181 | |
2182 | r = store->read_meta("whoami", &val); | |
2183 | if (r < 0) | |
2184 | return r; | |
11fdf7f2 | 2185 | *whoami = atoi(val.c_str()); |
7c673cae FG |
2186 | |
2187 | r = store->read_meta("ceph_fsid", &val); | |
2188 | if (r < 0) | |
2189 | return r; | |
11fdf7f2 | 2190 | r = cluster_fsid->parse(val.c_str()); |
7c673cae FG |
2191 | if (!r) |
2192 | return -EINVAL; | |
2193 | ||
2194 | r = store->read_meta("fsid", &val); | |
2195 | if (r < 0) { | |
11fdf7f2 | 2196 | *osd_fsid = uuid_d(); |
7c673cae | 2197 | } else { |
11fdf7f2 | 2198 | r = osd_fsid->parse(val.c_str()); |
7c673cae FG |
2199 | if (!r) |
2200 | return -EINVAL; | |
2201 | } | |
2202 | ||
11fdf7f2 TL |
2203 | r = store->read_meta("require_osd_release", &val); |
2204 | if (r >= 0) { | |
2205 | *require_osd_release = atoi(val.c_str()); | |
2206 | } | |
2207 | ||
7c673cae FG |
2208 | return 0; |
2209 | } | |
2210 | ||
2211 | ||
2212 | #undef dout_prefix | |
2213 | #define dout_prefix _prefix(_dout, whoami, get_osdmap_epoch()) | |
2214 | ||
2215 | // cons/des | |
2216 | ||
2217 | OSD::OSD(CephContext *cct_, ObjectStore *store_, | |
2218 | int id, | |
2219 | Messenger *internal_messenger, | |
2220 | Messenger *external_messenger, | |
2221 | Messenger *hb_client_front, | |
2222 | Messenger *hb_client_back, | |
2223 | Messenger *hb_front_serverm, | |
2224 | Messenger *hb_back_serverm, | |
2225 | Messenger *osdc_messenger, | |
2226 | MonClient *mc, | |
2227 | const std::string &dev, const std::string &jdev) : | |
2228 | Dispatcher(cct_), | |
2229 | osd_lock("OSD::osd_lock"), | |
2230 | tick_timer(cct, osd_lock), | |
2231 | tick_timer_lock("OSD::tick_timer_lock"), | |
2232 | tick_timer_without_osd_lock(cct, tick_timer_lock), | |
11fdf7f2 | 2233 | gss_ktfile_client(cct->_conf.get_val<std::string>("gss_ktab_client_file")), |
7c673cae FG |
2234 | cluster_messenger(internal_messenger), |
2235 | client_messenger(external_messenger), | |
2236 | objecter_messenger(osdc_messenger), | |
2237 | monc(mc), | |
2238 | mgrc(cct_, client_messenger), | |
2239 | logger(NULL), | |
2240 | recoverystate_perf(NULL), | |
2241 | store(store_), | |
2242 | log_client(cct, client_messenger, &mc->monmap, LogClient::NO_FLAGS), | |
2243 | clog(log_client.create_channel()), | |
2244 | whoami(id), | |
2245 | dev_path(dev), journal_path(jdev), | |
31f18b77 | 2246 | store_is_rotational(store->is_rotational()), |
7c673cae FG |
2247 | trace_endpoint("0.0.0.0", 0, "osd"), |
2248 | asok_hook(NULL), | |
11fdf7f2 TL |
2249 | m_osd_pg_epoch_max_lag_factor(cct->_conf.get_val<double>( |
2250 | "osd_pg_epoch_max_lag_factor")), | |
7c673cae | 2251 | osd_compat(get_osd_compat_set()), |
7c673cae | 2252 | osd_op_tp(cct, "OSD::osd_op_tp", "tp_osd_tp", |
31f18b77 | 2253 | get_num_op_threads()), |
7c673cae FG |
2254 | command_tp(cct, "OSD::command_tp", "tp_osd_cmd", 1), |
2255 | session_waiting_lock("OSD::session_waiting_lock"), | |
181888fb | 2256 | osdmap_subscribe_lock("OSD::osdmap_subscribe_lock"), |
7c673cae FG |
2257 | heartbeat_lock("OSD::heartbeat_lock"), |
2258 | heartbeat_stop(false), | |
2259 | heartbeat_need_update(true), | |
2260 | hb_front_client_messenger(hb_client_front), | |
2261 | hb_back_client_messenger(hb_client_back), | |
2262 | hb_front_server_messenger(hb_front_serverm), | |
2263 | hb_back_server_messenger(hb_back_serverm), | |
2264 | daily_loadavg(0.0), | |
2265 | heartbeat_thread(this), | |
2266 | heartbeat_dispatcher(this), | |
2267 | op_tracker(cct, cct->_conf->osd_enable_op_tracker, | |
2268 | cct->_conf->osd_num_op_tracker_shard), | |
2269 | test_ops_hook(NULL), | |
2270 | op_queue(get_io_queue()), | |
2271 | op_prio_cutoff(get_io_prio_cut()), | |
2272 | op_shardedwq( | |
7c673cae FG |
2273 | this, |
2274 | cct->_conf->osd_op_thread_timeout, | |
2275 | cct->_conf->osd_op_thread_suicide_timeout, | |
2276 | &osd_op_tp), | |
7c673cae | 2277 | map_lock("OSD::map_lock"), |
7c673cae FG |
2278 | last_pg_create_epoch(0), |
2279 | mon_report_lock("OSD::mon_report_lock"), | |
11fdf7f2 | 2280 | boot_finisher(cct), |
7c673cae FG |
2281 | up_thru_wanted(0), |
2282 | requested_full_first(0), | |
2283 | requested_full_last(0), | |
7c673cae FG |
2284 | command_wq( |
2285 | this, | |
2286 | cct->_conf->osd_command_thread_timeout, | |
2287 | cct->_conf->osd_command_thread_suicide_timeout, | |
2288 | &command_tp), | |
7c673cae FG |
2289 | service(this) |
2290 | { | |
11fdf7f2 TL |
2291 | |
2292 | if (!gss_ktfile_client.empty()) { | |
2293 | // Assert we can export environment variable | |
2294 | /* | |
2295 | The default client keytab is used, if it is present and readable, | |
2296 | to automatically obtain initial credentials for GSSAPI client | |
2297 | applications. The principal name of the first entry in the client | |
2298 | keytab is used by default when obtaining initial credentials. | |
2299 | 1. The KRB5_CLIENT_KTNAME environment variable. | |
2300 | 2. The default_client_keytab_name profile variable in [libdefaults]. | |
2301 | 3. The hardcoded default, DEFCKTNAME. | |
2302 | */ | |
2303 | const int32_t set_result(setenv("KRB5_CLIENT_KTNAME", | |
2304 | gss_ktfile_client.c_str(), 1)); | |
2305 | ceph_assert(set_result == 0); | |
2306 | } | |
2307 | ||
7c673cae FG |
2308 | monc->set_messenger(client_messenger); |
2309 | op_tracker.set_complaint_and_threshold(cct->_conf->osd_op_complaint_time, | |
2310 | cct->_conf->osd_op_log_threshold); | |
2311 | op_tracker.set_history_size_and_duration(cct->_conf->osd_op_history_size, | |
2312 | cct->_conf->osd_op_history_duration); | |
2313 | op_tracker.set_history_slow_op_size_and_threshold(cct->_conf->osd_op_history_slow_op_size, | |
2314 | cct->_conf->osd_op_history_slow_op_threshold); | |
2315 | #ifdef WITH_BLKIN | |
2316 | std::stringstream ss; | |
2317 | ss << "osd." << whoami; | |
2318 | trace_endpoint.copy_name(ss.str()); | |
2319 | #endif | |
11fdf7f2 TL |
2320 | |
2321 | // initialize shards | |
2322 | num_shards = get_num_op_shards(); | |
2323 | for (uint32_t i = 0; i < num_shards; i++) { | |
2324 | OSDShard *one_shard = new OSDShard( | |
2325 | i, | |
2326 | cct, | |
2327 | this, | |
2328 | cct->_conf->osd_op_pq_max_tokens_per_priority, | |
2329 | cct->_conf->osd_op_pq_min_cost, | |
2330 | op_queue); | |
2331 | shards.push_back(one_shard); | |
2332 | } | |
7c673cae FG |
2333 | } |
2334 | ||
2335 | OSD::~OSD() | |
2336 | { | |
11fdf7f2 TL |
2337 | while (!shards.empty()) { |
2338 | delete shards.back(); | |
2339 | shards.pop_back(); | |
2340 | } | |
7c673cae FG |
2341 | delete class_handler; |
2342 | cct->get_perfcounters_collection()->remove(recoverystate_perf); | |
2343 | cct->get_perfcounters_collection()->remove(logger); | |
2344 | delete recoverystate_perf; | |
2345 | delete logger; | |
2346 | delete store; | |
2347 | } | |
2348 | ||
91327a77 AA |
2349 | double OSD::get_tick_interval() const |
2350 | { | |
2351 | // vary +/- 5% to avoid scrub scheduling livelocks | |
2352 | constexpr auto delta = 0.05; | |
91327a77 | 2353 | return (OSD_TICK_INTERVAL * |
11fdf7f2 | 2354 | ceph::util::generate_random_number(1.0 - delta, 1.0 + delta)); |
91327a77 AA |
2355 | } |
2356 | ||
7c673cae FG |
2357 | void cls_initialize(ClassHandler *ch); |
2358 | ||
2359 | void OSD::handle_signal(int signum) | |
2360 | { | |
11fdf7f2 | 2361 | ceph_assert(signum == SIGINT || signum == SIGTERM); |
7c673cae FG |
2362 | derr << "*** Got signal " << sig_str(signum) << " ***" << dendl; |
2363 | shutdown(); | |
2364 | } | |
2365 | ||
2366 | int OSD::pre_init() | |
2367 | { | |
11fdf7f2 | 2368 | std::lock_guard lock(osd_lock); |
7c673cae FG |
2369 | if (is_stopping()) |
2370 | return 0; | |
2371 | ||
2372 | if (store->test_mount_in_use()) { | |
2373 | derr << "OSD::pre_init: object store '" << dev_path << "' is " | |
2374 | << "currently in use. (Is ceph-osd already running?)" << dendl; | |
2375 | return -EBUSY; | |
2376 | } | |
2377 | ||
11fdf7f2 TL |
2378 | cct->_conf.add_observer(this); |
2379 | return 0; | |
2380 | } | |
2381 | ||
2382 | int OSD::set_numa_affinity() | |
2383 | { | |
2384 | // storage numa node | |
2385 | int store_node = -1; | |
2386 | store->get_numa_node(&store_node, nullptr, nullptr); | |
2387 | if (store_node >= 0) { | |
2388 | dout(1) << __func__ << " storage numa node " << store_node << dendl; | |
2389 | } | |
2390 | ||
2391 | // check network numa node(s) | |
2392 | int front_node = -1, back_node = -1; | |
2393 | string front_iface = pick_iface( | |
2394 | cct, | |
2395 | client_messenger->get_myaddrs().front().get_sockaddr_storage()); | |
2396 | string back_iface = pick_iface( | |
2397 | cct, | |
2398 | cluster_messenger->get_myaddrs().front().get_sockaddr_storage()); | |
2399 | int r = get_iface_numa_node(front_iface, &front_node); | |
2400 | if (r >= 0) { | |
2401 | dout(1) << __func__ << " public network " << front_iface << " numa node " | |
2402 | << front_node << dendl; | |
2403 | r = get_iface_numa_node(back_iface, &back_node); | |
2404 | if (r >= 0) { | |
2405 | dout(1) << __func__ << " cluster network " << back_iface << " numa node " | |
2406 | << back_node << dendl; | |
2407 | if (front_node == back_node && | |
2408 | front_node == store_node) { | |
2409 | dout(1) << " objectstore and network numa nodes all match" << dendl; | |
2410 | if (g_conf().get_val<bool>("osd_numa_auto_affinity")) { | |
2411 | numa_node = front_node; | |
2412 | } | |
2413 | } else { | |
2414 | dout(1) << __func__ << " objectstore and network numa nodes do not match" | |
2415 | << dendl; | |
2416 | } | |
2417 | } | |
2418 | } else { | |
2419 | derr << __func__ << " unable to identify public interface '" << front_iface | |
2420 | << "' numa node: " << cpp_strerror(r) << dendl; | |
2421 | } | |
2422 | if (int node = g_conf().get_val<int64_t>("osd_numa_node"); node >= 0) { | |
2423 | // this takes precedence over the automagic logic above | |
2424 | numa_node = node; | |
2425 | } | |
2426 | if (numa_node >= 0) { | |
2427 | int r = get_numa_node_cpu_set(numa_node, &numa_cpu_set_size, &numa_cpu_set); | |
2428 | if (r < 0) { | |
2429 | dout(1) << __func__ << " unable to determine numa node " << numa_node | |
2430 | << " CPUs" << dendl; | |
2431 | numa_node = -1; | |
2432 | } else { | |
2433 | dout(1) << __func__ << " setting numa affinity to node " << numa_node | |
2434 | << " cpus " | |
2435 | << cpu_set_to_str_list(numa_cpu_set_size, &numa_cpu_set) | |
2436 | << dendl; | |
2437 | r = sched_setaffinity(getpid(), numa_cpu_set_size, &numa_cpu_set); | |
2438 | if (r < 0) { | |
2439 | r = -errno; | |
2440 | derr << __func__ << " failed to set numa affinity: " << cpp_strerror(r) | |
2441 | << dendl; | |
2442 | numa_node = -1; | |
2443 | } | |
2444 | } | |
2445 | } else { | |
2446 | dout(1) << __func__ << " not setting numa affinity" << dendl; | |
2447 | } | |
7c673cae FG |
2448 | return 0; |
2449 | } | |
2450 | ||
2451 | // asok | |
2452 | ||
2453 | class OSDSocketHook : public AdminSocketHook { | |
2454 | OSD *osd; | |
2455 | public: | |
2456 | explicit OSDSocketHook(OSD *o) : osd(o) {} | |
11fdf7f2 TL |
2457 | bool call(std::string_view admin_command, const cmdmap_t& cmdmap, |
2458 | std::string_view format, bufferlist& out) override { | |
7c673cae | 2459 | stringstream ss; |
11fdf7f2 TL |
2460 | bool r = true; |
2461 | try { | |
2462 | r = osd->asok_command(admin_command, cmdmap, format, ss); | |
2463 | } catch (const bad_cmd_get& e) { | |
2464 | ss << e.what(); | |
2465 | r = true; | |
2466 | } | |
7c673cae FG |
2467 | out.append(ss); |
2468 | return r; | |
2469 | } | |
2470 | }; | |
2471 | ||
11fdf7f2 TL |
2472 | std::set<int64_t> OSD::get_mapped_pools() |
2473 | { | |
2474 | std::set<int64_t> pools; | |
2475 | std::vector<spg_t> pgids; | |
2476 | _get_pgids(&pgids); | |
2477 | for (const auto &pgid : pgids) { | |
2478 | pools.insert(pgid.pool()); | |
2479 | } | |
2480 | return pools; | |
2481 | } | |
2482 | ||
2483 | bool OSD::asok_command(std::string_view admin_command, const cmdmap_t& cmdmap, | |
2484 | std::string_view format, ostream& ss) | |
7c673cae FG |
2485 | { |
2486 | Formatter *f = Formatter::create(format, "json-pretty", "json-pretty"); | |
2487 | if (admin_command == "status") { | |
2488 | f->open_object_section("status"); | |
2489 | f->dump_stream("cluster_fsid") << superblock.cluster_fsid; | |
2490 | f->dump_stream("osd_fsid") << superblock.osd_fsid; | |
2491 | f->dump_unsigned("whoami", superblock.whoami); | |
2492 | f->dump_string("state", get_state_name(get_state())); | |
2493 | f->dump_unsigned("oldest_map", superblock.oldest_map); | |
2494 | f->dump_unsigned("newest_map", superblock.newest_map); | |
11fdf7f2 | 2495 | f->dump_unsigned("num_pgs", num_pgs); |
7c673cae FG |
2496 | f->close_section(); |
2497 | } else if (admin_command == "flush_journal") { | |
2498 | store->flush_journal(); | |
2499 | } else if (admin_command == "dump_ops_in_flight" || | |
c07f9fc5 FG |
2500 | admin_command == "ops" || |
2501 | admin_command == "dump_blocked_ops" || | |
2502 | admin_command == "dump_historic_ops" || | |
2503 | admin_command == "dump_historic_ops_by_duration" || | |
2504 | admin_command == "dump_historic_slow_ops") { | |
2505 | ||
2506 | const string error_str = "op_tracker tracking is not enabled now, so no ops are tracked currently, \ | |
2507 | even those get stuck. Please enable \"osd_enable_op_tracker\", and the tracker \ | |
2508 | will start to track new ops received afterwards."; | |
2509 | ||
2510 | set<string> filters; | |
2511 | vector<string> filter_str; | |
2512 | if (cmd_getval(cct, cmdmap, "filterstr", filter_str)) { | |
2513 | copy(filter_str.begin(), filter_str.end(), | |
2514 | inserter(filters, filters.end())); | |
2515 | } | |
2516 | ||
2517 | if (admin_command == "dump_ops_in_flight" || | |
2518 | admin_command == "ops") { | |
2519 | if (!op_tracker.dump_ops_in_flight(f, false, filters)) { | |
2520 | ss << error_str; | |
2521 | } | |
2522 | } | |
2523 | if (admin_command == "dump_blocked_ops") { | |
2524 | if (!op_tracker.dump_ops_in_flight(f, true, filters)) { | |
2525 | ss << error_str; | |
2526 | } | |
2527 | } | |
2528 | if (admin_command == "dump_historic_ops") { | |
2529 | if (!op_tracker.dump_historic_ops(f, false, filters)) { | |
2530 | ss << error_str; | |
2531 | } | |
2532 | } | |
2533 | if (admin_command == "dump_historic_ops_by_duration") { | |
2534 | if (!op_tracker.dump_historic_ops(f, true, filters)) { | |
2535 | ss << error_str; | |
2536 | } | |
2537 | } | |
2538 | if (admin_command == "dump_historic_slow_ops") { | |
2539 | if (!op_tracker.dump_historic_slow_ops(f, filters)) { | |
2540 | ss << error_str; | |
2541 | } | |
7c673cae FG |
2542 | } |
2543 | } else if (admin_command == "dump_op_pq_state") { | |
2544 | f->open_object_section("pq"); | |
2545 | op_shardedwq.dump(f); | |
2546 | f->close_section(); | |
2547 | } else if (admin_command == "dump_blacklist") { | |
2548 | list<pair<entity_addr_t,utime_t> > bl; | |
2549 | OSDMapRef curmap = service.get_osdmap(); | |
2550 | ||
2551 | f->open_array_section("blacklist"); | |
2552 | curmap->get_blacklist(&bl); | |
2553 | for (list<pair<entity_addr_t,utime_t> >::iterator it = bl.begin(); | |
2554 | it != bl.end(); ++it) { | |
224ce89b | 2555 | f->open_object_section("entry"); |
7c673cae FG |
2556 | f->open_object_section("entity_addr_t"); |
2557 | it->first.dump(f); | |
2558 | f->close_section(); //entity_addr_t | |
2559 | it->second.localtime(f->dump_stream("expire_time")); | |
2560 | f->close_section(); //entry | |
2561 | } | |
2562 | f->close_section(); //blacklist | |
2563 | } else if (admin_command == "dump_watchers") { | |
2564 | list<obj_watch_item_t> watchers; | |
2565 | // scan pg's | |
11fdf7f2 TL |
2566 | vector<PGRef> pgs; |
2567 | _get_pgs(&pgs); | |
2568 | for (auto& pg : pgs) { | |
2569 | list<obj_watch_item_t> pg_watchers; | |
2570 | pg->get_watchers(&pg_watchers); | |
2571 | watchers.splice(watchers.end(), pg_watchers); | |
7c673cae FG |
2572 | } |
2573 | ||
2574 | f->open_array_section("watchers"); | |
2575 | for (list<obj_watch_item_t>::iterator it = watchers.begin(); | |
2576 | it != watchers.end(); ++it) { | |
2577 | ||
224ce89b | 2578 | f->open_object_section("watch"); |
7c673cae FG |
2579 | |
2580 | f->dump_string("namespace", it->obj.nspace); | |
2581 | f->dump_string("object", it->obj.oid.name); | |
2582 | ||
2583 | f->open_object_section("entity_name"); | |
2584 | it->wi.name.dump(f); | |
2585 | f->close_section(); //entity_name_t | |
2586 | ||
224ce89b WB |
2587 | f->dump_unsigned("cookie", it->wi.cookie); |
2588 | f->dump_unsigned("timeout", it->wi.timeout_seconds); | |
7c673cae FG |
2589 | |
2590 | f->open_object_section("entity_addr_t"); | |
2591 | it->wi.addr.dump(f); | |
2592 | f->close_section(); //entity_addr_t | |
2593 | ||
2594 | f->close_section(); //watch | |
2595 | } | |
2596 | ||
2597 | f->close_section(); //watchers | |
2598 | } else if (admin_command == "dump_reservations") { | |
2599 | f->open_object_section("reservations"); | |
2600 | f->open_object_section("local_reservations"); | |
2601 | service.local_reserver.dump(f); | |
2602 | f->close_section(); | |
2603 | f->open_object_section("remote_reservations"); | |
2604 | service.remote_reserver.dump(f); | |
2605 | f->close_section(); | |
2606 | f->close_section(); | |
2607 | } else if (admin_command == "get_latest_osdmap") { | |
2608 | get_latest_osdmap(); | |
2609 | } else if (admin_command == "heap") { | |
2610 | auto result = ceph::osd_cmds::heap(*cct, cmdmap, *f, ss); | |
2611 | ||
2612 | // Note: Failed heap profile commands won't necessarily trigger an error: | |
2613 | f->open_object_section("result"); | |
2614 | f->dump_string("error", cpp_strerror(result)); | |
2615 | f->dump_bool("success", result >= 0); | |
2616 | f->close_section(); | |
2617 | } else if (admin_command == "set_heap_property") { | |
2618 | string property; | |
2619 | int64_t value = 0; | |
2620 | string error; | |
2621 | bool success = false; | |
2622 | if (!cmd_getval(cct, cmdmap, "property", property)) { | |
2623 | error = "unable to get property"; | |
2624 | success = false; | |
2625 | } else if (!cmd_getval(cct, cmdmap, "value", value)) { | |
2626 | error = "unable to get value"; | |
2627 | success = false; | |
2628 | } else if (value < 0) { | |
2629 | error = "negative value not allowed"; | |
2630 | success = false; | |
2631 | } else if (!ceph_heap_set_numeric_property(property.c_str(), (size_t)value)) { | |
2632 | error = "invalid property"; | |
2633 | success = false; | |
2634 | } else { | |
2635 | success = true; | |
2636 | } | |
2637 | f->open_object_section("result"); | |
2638 | f->dump_string("error", error); | |
2639 | f->dump_bool("success", success); | |
2640 | f->close_section(); | |
2641 | } else if (admin_command == "get_heap_property") { | |
2642 | string property; | |
2643 | size_t value = 0; | |
2644 | string error; | |
2645 | bool success = false; | |
2646 | if (!cmd_getval(cct, cmdmap, "property", property)) { | |
2647 | error = "unable to get property"; | |
2648 | success = false; | |
2649 | } else if (!ceph_heap_get_numeric_property(property.c_str(), &value)) { | |
2650 | error = "invalid property"; | |
2651 | success = false; | |
2652 | } else { | |
2653 | success = true; | |
2654 | } | |
2655 | f->open_object_section("result"); | |
2656 | f->dump_string("error", error); | |
2657 | f->dump_bool("success", success); | |
2658 | f->dump_int("value", value); | |
2659 | f->close_section(); | |
2660 | } else if (admin_command == "dump_objectstore_kv_stats") { | |
2661 | store->get_db_statistics(f); | |
2662 | } else if (admin_command == "dump_scrubs") { | |
2663 | service.dumps_scrub(f); | |
2664 | } else if (admin_command == "calc_objectstore_db_histogram") { | |
2665 | store->generate_db_histogram(f); | |
2666 | } else if (admin_command == "flush_store_cache") { | |
11fdf7f2 | 2667 | store->flush_cache(&ss); |
7c673cae FG |
2668 | } else if (admin_command == "dump_pgstate_history") { |
2669 | f->open_object_section("pgstate_history"); | |
11fdf7f2 TL |
2670 | vector<PGRef> pgs; |
2671 | _get_pgs(&pgs); | |
2672 | for (auto& pg : pgs) { | |
2673 | f->dump_stream("pg") << pg->pg_id; | |
2674 | pg->dump_pgstate_history(f); | |
7c673cae FG |
2675 | } |
2676 | f->close_section(); | |
224ce89b WB |
2677 | } else if (admin_command == "compact") { |
2678 | dout(1) << "triggering manual compaction" << dendl; | |
2679 | auto start = ceph::coarse_mono_clock::now(); | |
2680 | store->compact(); | |
2681 | auto end = ceph::coarse_mono_clock::now(); | |
11fdf7f2 | 2682 | double duration = std::chrono::duration<double>(end-start).count(); |
224ce89b | 2683 | dout(1) << "finished manual compaction in " |
11fdf7f2 | 2684 | << duration |
224ce89b WB |
2685 | << " seconds" << dendl; |
2686 | f->open_object_section("compact_result"); | |
11fdf7f2 TL |
2687 | f->dump_float("elapsed_time", duration); |
2688 | f->close_section(); | |
2689 | } else if (admin_command == "get_mapped_pools") { | |
2690 | f->open_array_section("mapped_pools"); | |
2691 | set<int64_t> poollist = get_mapped_pools(); | |
2692 | for (auto pool : poollist) { | |
2693 | f->dump_int("pool_id", pool); | |
2694 | } | |
2695 | f->close_section(); | |
2696 | } else if (admin_command == "smart") { | |
2697 | string devid; | |
2698 | cmd_getval(cct, cmdmap, "devid", devid); | |
2699 | probe_smart(devid, ss); | |
2700 | } else if (admin_command == "list_devices") { | |
2701 | set<string> devnames; | |
2702 | store->get_devices(&devnames); | |
2703 | f->open_object_section("list_devices"); | |
2704 | for (auto dev : devnames) { | |
2705 | if (dev.find("dm-") == 0) { | |
2706 | continue; | |
2707 | } | |
2708 | f->dump_string("device", "/dev/" + dev); | |
2709 | } | |
224ce89b | 2710 | f->close_section(); |
11fdf7f2 TL |
2711 | } else if (admin_command == "send_beacon") { |
2712 | if (is_active()) { | |
2713 | send_beacon(ceph::coarse_mono_clock::now()); | |
2714 | } | |
7c673cae | 2715 | } else { |
11fdf7f2 | 2716 | ceph_abort_msg("broken asok registration"); |
7c673cae FG |
2717 | } |
2718 | f->flush(ss); | |
2719 | delete f; | |
2720 | return true; | |
2721 | } | |
2722 | ||
2723 | class TestOpsSocketHook : public AdminSocketHook { | |
2724 | OSDService *service; | |
2725 | ObjectStore *store; | |
2726 | public: | |
2727 | TestOpsSocketHook(OSDService *s, ObjectStore *st) : service(s), store(st) {} | |
11fdf7f2 TL |
2728 | bool call(std::string_view command, const cmdmap_t& cmdmap, |
2729 | std::string_view format, bufferlist& out) override { | |
7c673cae | 2730 | stringstream ss; |
11fdf7f2 TL |
2731 | try { |
2732 | test_ops(service, store, command, cmdmap, ss); | |
2733 | } catch (const bad_cmd_get& e) { | |
2734 | ss << e.what(); | |
2735 | } | |
7c673cae FG |
2736 | out.append(ss); |
2737 | return true; | |
2738 | } | |
2739 | void test_ops(OSDService *service, ObjectStore *store, | |
11fdf7f2 | 2740 | std::string_view command, const cmdmap_t& cmdmap, ostream &ss); |
7c673cae FG |
2741 | |
2742 | }; | |
2743 | ||
2744 | class OSD::C_Tick : public Context { | |
2745 | OSD *osd; | |
2746 | public: | |
2747 | explicit C_Tick(OSD *o) : osd(o) {} | |
2748 | void finish(int r) override { | |
2749 | osd->tick(); | |
2750 | } | |
2751 | }; | |
2752 | ||
2753 | class OSD::C_Tick_WithoutOSDLock : public Context { | |
2754 | OSD *osd; | |
2755 | public: | |
2756 | explicit C_Tick_WithoutOSDLock(OSD *o) : osd(o) {} | |
2757 | void finish(int r) override { | |
2758 | osd->tick_without_osd_lock(); | |
2759 | } | |
2760 | }; | |
2761 | ||
2762 | int OSD::enable_disable_fuse(bool stop) | |
2763 | { | |
2764 | #ifdef HAVE_LIBFUSE | |
2765 | int r; | |
2766 | string mntpath = cct->_conf->osd_data + "/fuse"; | |
2767 | if (fuse_store && (stop || !cct->_conf->osd_objectstore_fuse)) { | |
2768 | dout(1) << __func__ << " disabling" << dendl; | |
2769 | fuse_store->stop(); | |
2770 | delete fuse_store; | |
2771 | fuse_store = NULL; | |
2772 | r = ::rmdir(mntpath.c_str()); | |
7c673cae | 2773 | if (r < 0) { |
c07f9fc5 FG |
2774 | r = -errno; |
2775 | derr << __func__ << " failed to rmdir " << mntpath << ": " | |
2776 | << cpp_strerror(r) << dendl; | |
7c673cae FG |
2777 | return r; |
2778 | } | |
2779 | return 0; | |
2780 | } | |
2781 | if (!fuse_store && cct->_conf->osd_objectstore_fuse) { | |
2782 | dout(1) << __func__ << " enabling" << dendl; | |
2783 | r = ::mkdir(mntpath.c_str(), 0700); | |
2784 | if (r < 0) | |
2785 | r = -errno; | |
2786 | if (r < 0 && r != -EEXIST) { | |
2787 | derr << __func__ << " unable to create " << mntpath << ": " | |
2788 | << cpp_strerror(r) << dendl; | |
2789 | return r; | |
2790 | } | |
2791 | fuse_store = new FuseStore(store, mntpath); | |
2792 | r = fuse_store->start(); | |
2793 | if (r < 0) { | |
2794 | derr << __func__ << " unable to start fuse: " << cpp_strerror(r) << dendl; | |
2795 | delete fuse_store; | |
2796 | fuse_store = NULL; | |
2797 | return r; | |
2798 | } | |
2799 | } | |
2800 | #endif // HAVE_LIBFUSE | |
2801 | return 0; | |
2802 | } | |
2803 | ||
31f18b77 FG |
2804 | int OSD::get_num_op_shards() |
2805 | { | |
2806 | if (cct->_conf->osd_op_num_shards) | |
2807 | return cct->_conf->osd_op_num_shards; | |
2808 | if (store_is_rotational) | |
2809 | return cct->_conf->osd_op_num_shards_hdd; | |
2810 | else | |
2811 | return cct->_conf->osd_op_num_shards_ssd; | |
2812 | } | |
2813 | ||
2814 | int OSD::get_num_op_threads() | |
2815 | { | |
2816 | if (cct->_conf->osd_op_num_threads_per_shard) | |
2817 | return get_num_op_shards() * cct->_conf->osd_op_num_threads_per_shard; | |
2818 | if (store_is_rotational) | |
2819 | return get_num_op_shards() * cct->_conf->osd_op_num_threads_per_shard_hdd; | |
2820 | else | |
2821 | return get_num_op_shards() * cct->_conf->osd_op_num_threads_per_shard_ssd; | |
2822 | } | |
2823 | ||
c07f9fc5 FG |
2824 | float OSD::get_osd_recovery_sleep() |
2825 | { | |
2826 | if (cct->_conf->osd_recovery_sleep) | |
2827 | return cct->_conf->osd_recovery_sleep; | |
d2e6a577 | 2828 | if (!store_is_rotational && !journal_is_rotational) |
c07f9fc5 | 2829 | return cct->_conf->osd_recovery_sleep_ssd; |
d2e6a577 | 2830 | else if (store_is_rotational && !journal_is_rotational) |
11fdf7f2 | 2831 | return cct->_conf.get_val<double>("osd_recovery_sleep_hybrid"); |
d2e6a577 FG |
2832 | else |
2833 | return cct->_conf->osd_recovery_sleep_hdd; | |
c07f9fc5 FG |
2834 | } |
2835 | ||
11fdf7f2 TL |
2836 | float OSD::get_osd_delete_sleep() |
2837 | { | |
2838 | float osd_delete_sleep = cct->_conf.get_val<double>("osd_delete_sleep"); | |
2839 | if (osd_delete_sleep > 0) | |
2840 | return osd_delete_sleep; | |
2841 | if (!store_is_rotational && !journal_is_rotational) | |
2842 | return cct->_conf.get_val<double>("osd_delete_sleep_ssd"); | |
2843 | if (store_is_rotational && !journal_is_rotational) | |
2844 | return cct->_conf.get_val<double>("osd_delete_sleep_hybrid"); | |
2845 | return cct->_conf.get_val<double>("osd_delete_sleep_hdd"); | |
2846 | } | |
2847 | ||
7c673cae FG |
2848 | int OSD::init() |
2849 | { | |
2850 | CompatSet initial, diff; | |
11fdf7f2 | 2851 | std::lock_guard lock(osd_lock); |
7c673cae FG |
2852 | if (is_stopping()) |
2853 | return 0; | |
2854 | ||
2855 | tick_timer.init(); | |
2856 | tick_timer_without_osd_lock.init(); | |
2857 | service.recovery_request_timer.init(); | |
11fdf7f2 TL |
2858 | service.sleep_timer.init(); |
2859 | ||
2860 | boot_finisher.start(); | |
2861 | ||
2862 | { | |
2863 | string val; | |
2864 | store->read_meta("require_osd_release", &val); | |
2865 | last_require_osd_release = atoi(val.c_str()); | |
2866 | } | |
7c673cae FG |
2867 | |
2868 | // mount. | |
31f18b77 FG |
2869 | dout(2) << "init " << dev_path |
2870 | << " (looks like " << (store_is_rotational ? "hdd" : "ssd") << ")" | |
2871 | << dendl; | |
d2e6a577 | 2872 | dout(2) << "journal " << journal_path << dendl; |
11fdf7f2 | 2873 | ceph_assert(store); // call pre_init() first! |
7c673cae | 2874 | |
31f18b77 | 2875 | store->set_cache_shards(get_num_op_shards()); |
7c673cae FG |
2876 | |
2877 | int r = store->mount(); | |
2878 | if (r < 0) { | |
2879 | derr << "OSD:init: unable to mount object store" << dendl; | |
2880 | return r; | |
2881 | } | |
d2e6a577 FG |
2882 | journal_is_rotational = store->is_journal_rotational(); |
2883 | dout(2) << "journal looks like " << (journal_is_rotational ? "hdd" : "ssd") | |
2884 | << dendl; | |
7c673cae FG |
2885 | |
2886 | enable_disable_fuse(false); | |
2887 | ||
2888 | dout(2) << "boot" << dendl; | |
2889 | ||
11fdf7f2 TL |
2890 | service.meta_ch = store->open_collection(coll_t::meta()); |
2891 | ||
7c673cae FG |
2892 | // initialize the daily loadavg with current 15min loadavg |
2893 | double loadavgs[3]; | |
2894 | if (getloadavg(loadavgs, 3) == 3) { | |
2895 | daily_loadavg = loadavgs[2]; | |
2896 | } else { | |
2897 | derr << "OSD::init() : couldn't read loadavgs\n" << dendl; | |
2898 | daily_loadavg = 1.0; | |
2899 | } | |
2900 | ||
2901 | int rotating_auth_attempts = 0; | |
11fdf7f2 TL |
2902 | auto rotating_auth_timeout = |
2903 | g_conf().get_val<int64_t>("rotating_keys_bootstrap_timeout"); | |
7c673cae FG |
2904 | |
2905 | // sanity check long object name handling | |
2906 | { | |
2907 | hobject_t l; | |
2908 | l.oid.name = string(cct->_conf->osd_max_object_name_len, 'n'); | |
2909 | l.set_key(string(cct->_conf->osd_max_object_name_len, 'k')); | |
2910 | l.nspace = string(cct->_conf->osd_max_object_namespace_len, 's'); | |
2911 | r = store->validate_hobject_key(l); | |
2912 | if (r < 0) { | |
2913 | derr << "backend (" << store->get_type() << ") is unable to support max " | |
2914 | << "object name[space] len" << dendl; | |
2915 | derr << " osd max object name len = " | |
2916 | << cct->_conf->osd_max_object_name_len << dendl; | |
2917 | derr << " osd max object namespace len = " | |
2918 | << cct->_conf->osd_max_object_namespace_len << dendl; | |
2919 | derr << cpp_strerror(r) << dendl; | |
2920 | if (cct->_conf->osd_check_max_object_name_len_on_startup) { | |
2921 | goto out; | |
2922 | } | |
2923 | derr << "osd_check_max_object_name_len_on_startup = false, starting anyway" | |
2924 | << dendl; | |
2925 | } else { | |
2926 | dout(20) << "configured osd_max_object_name[space]_len looks ok" << dendl; | |
2927 | } | |
2928 | } | |
2929 | ||
2930 | // read superblock | |
2931 | r = read_superblock(); | |
2932 | if (r < 0) { | |
2933 | derr << "OSD::init() : unable to read osd superblock" << dendl; | |
2934 | r = -EINVAL; | |
2935 | goto out; | |
2936 | } | |
2937 | ||
2938 | if (osd_compat.compare(superblock.compat_features) < 0) { | |
2939 | derr << "The disk uses features unsupported by the executable." << dendl; | |
2940 | derr << " ondisk features " << superblock.compat_features << dendl; | |
2941 | derr << " daemon features " << osd_compat << dendl; | |
2942 | ||
2943 | if (osd_compat.writeable(superblock.compat_features)) { | |
2944 | CompatSet diff = osd_compat.unsupported(superblock.compat_features); | |
2945 | derr << "it is still writeable, though. Missing features: " << diff << dendl; | |
2946 | r = -EOPNOTSUPP; | |
2947 | goto out; | |
2948 | } | |
2949 | else { | |
2950 | CompatSet diff = osd_compat.unsupported(superblock.compat_features); | |
2951 | derr << "Cannot write to disk! Missing features: " << diff << dendl; | |
2952 | r = -EOPNOTSUPP; | |
2953 | goto out; | |
2954 | } | |
2955 | } | |
2956 | ||
2957 | assert_warn(whoami == superblock.whoami); | |
2958 | if (whoami != superblock.whoami) { | |
2959 | derr << "OSD::init: superblock says osd" | |
2960 | << superblock.whoami << " but I am osd." << whoami << dendl; | |
2961 | r = -EINVAL; | |
2962 | goto out; | |
2963 | } | |
2964 | ||
11fdf7f2 TL |
2965 | // load up "current" osdmap |
2966 | assert_warn(!osdmap); | |
2967 | if (osdmap) { | |
2968 | derr << "OSD::init: unable to read current osdmap" << dendl; | |
2969 | r = -EINVAL; | |
2970 | goto out; | |
2971 | } | |
2972 | osdmap = get_map(superblock.current_epoch); | |
2973 | ||
2974 | // make sure we don't have legacy pgs deleting | |
2975 | { | |
2976 | vector<coll_t> ls; | |
2977 | int r = store->list_collections(ls); | |
2978 | ceph_assert(r >= 0); | |
2979 | for (auto c : ls) { | |
2980 | spg_t pgid; | |
2981 | if (c.is_pg(&pgid) && | |
2982 | !osdmap->have_pg_pool(pgid.pool())) { | |
2983 | ghobject_t oid = make_final_pool_info_oid(pgid.pool()); | |
2984 | if (!store->exists(service.meta_ch, oid)) { | |
2985 | derr << __func__ << " missing pg_pool_t for deleted pool " | |
2986 | << pgid.pool() << " for pg " << pgid | |
2987 | << "; please downgrade to luminous and allow " | |
2988 | << "pg deletion to complete before upgrading" << dendl; | |
2989 | ceph_abort(); | |
2990 | } | |
2991 | } | |
2992 | } | |
2993 | } | |
2994 | ||
7c673cae FG |
2995 | initial = get_osd_initial_compat_set(); |
2996 | diff = superblock.compat_features.unsupported(initial); | |
2997 | if (superblock.compat_features.merge(initial)) { | |
2998 | // We need to persist the new compat_set before we | |
2999 | // do anything else | |
3000 | dout(5) << "Upgrading superblock adding: " << diff << dendl; | |
3001 | ObjectStore::Transaction t; | |
3002 | write_superblock(t); | |
11fdf7f2 | 3003 | r = store->queue_transaction(service.meta_ch, std::move(t)); |
7c673cae FG |
3004 | if (r < 0) |
3005 | goto out; | |
3006 | } | |
3007 | ||
3008 | // make sure snap mapper object exists | |
11fdf7f2 | 3009 | if (!store->exists(service.meta_ch, OSD::make_snapmapper_oid())) { |
7c673cae FG |
3010 | dout(10) << "init creating/touching snapmapper object" << dendl; |
3011 | ObjectStore::Transaction t; | |
3012 | t.touch(coll_t::meta(), OSD::make_snapmapper_oid()); | |
11fdf7f2 | 3013 | r = store->queue_transaction(service.meta_ch, std::move(t)); |
7c673cae FG |
3014 | if (r < 0) |
3015 | goto out; | |
3016 | } | |
3017 | ||
3018 | class_handler = new ClassHandler(cct); | |
3019 | cls_initialize(class_handler); | |
3020 | ||
3021 | if (cct->_conf->osd_open_classes_on_start) { | |
3022 | int r = class_handler->open_all_classes(); | |
3023 | if (r) | |
3024 | dout(1) << "warning: got an error loading one or more classes: " << cpp_strerror(r) << dendl; | |
3025 | } | |
3026 | ||
11fdf7f2 | 3027 | check_osdmap_features(); |
7c673cae FG |
3028 | |
3029 | create_recoverystate_perf(); | |
3030 | ||
3031 | { | |
3032 | epoch_t bind_epoch = osdmap->get_epoch(); | |
3033 | service.set_epochs(NULL, NULL, &bind_epoch); | |
3034 | } | |
3035 | ||
3036 | clear_temp_objects(); | |
3037 | ||
d2e6a577 | 3038 | // initialize osdmap references in sharded wq |
11fdf7f2 TL |
3039 | for (auto& shard : shards) { |
3040 | std::lock_guard l(shard->osdmap_lock); | |
3041 | shard->shard_osdmap = osdmap; | |
3042 | } | |
d2e6a577 | 3043 | |
7c673cae FG |
3044 | // load up pgs (as they previously existed) |
3045 | load_pgs(); | |
3046 | ||
3047 | dout(2) << "superblock: I am osd." << superblock.whoami << dendl; | |
3048 | dout(0) << "using " << op_queue << " op queue with priority op cut off at " << | |
3049 | op_prio_cutoff << "." << dendl; | |
3050 | ||
3051 | create_logger(); | |
3052 | ||
11fdf7f2 TL |
3053 | // prime osd stats |
3054 | { | |
3055 | struct store_statfs_t stbuf; | |
3056 | osd_alert_list_t alerts; | |
3057 | int r = store->statfs(&stbuf, &alerts); | |
3058 | ceph_assert(r == 0); | |
3059 | service.set_statfs(stbuf, alerts); | |
3060 | } | |
3061 | ||
3062 | // client_messenger auth_client is already set up by monc. | |
3063 | for (auto m : { cluster_messenger, | |
3064 | objecter_messenger, | |
3065 | hb_front_client_messenger, | |
3066 | hb_back_client_messenger, | |
3067 | hb_front_server_messenger, | |
3068 | hb_back_server_messenger } ) { | |
3069 | m->set_auth_client(monc); | |
3070 | } | |
3071 | for (auto m : { client_messenger, | |
3072 | cluster_messenger, | |
3073 | hb_front_server_messenger, | |
3074 | hb_back_server_messenger }) { | |
3075 | m->set_auth_server(monc); | |
3076 | } | |
3077 | monc->set_handle_authentication_dispatcher(this); | |
7c673cae FG |
3078 | |
3079 | monc->set_want_keys(CEPH_ENTITY_TYPE_MON | CEPH_ENTITY_TYPE_OSD | |
3080 | | CEPH_ENTITY_TYPE_MGR); | |
3081 | r = monc->init(); | |
3082 | if (r < 0) | |
3083 | goto out; | |
3084 | ||
11fdf7f2 TL |
3085 | mgrc.set_pgstats_cb([this](){ return collect_pg_stats(); }); |
3086 | mgrc.set_perf_metric_query_cb( | |
3087 | [this](const std::map<OSDPerfMetricQuery, OSDPerfMetricLimits> &queries) { | |
3088 | set_perf_queries(queries); | |
3089 | }, | |
3090 | [this](std::map<OSDPerfMetricQuery, OSDPerfMetricReport> *reports) { | |
3091 | get_perf_reports(reports); | |
3092 | }); | |
7c673cae | 3093 | mgrc.init(); |
7c673cae FG |
3094 | |
3095 | // tell monc about log_client so it will know about mon session resets | |
3096 | monc->set_log_client(&log_client); | |
3097 | update_log_config(); | |
3098 | ||
11fdf7f2 TL |
3099 | // i'm ready! |
3100 | client_messenger->add_dispatcher_tail(&mgrc); | |
3101 | client_messenger->add_dispatcher_tail(this); | |
3102 | cluster_messenger->add_dispatcher_head(this); | |
3103 | ||
3104 | hb_front_client_messenger->add_dispatcher_head(&heartbeat_dispatcher); | |
3105 | hb_back_client_messenger->add_dispatcher_head(&heartbeat_dispatcher); | |
3106 | hb_front_server_messenger->add_dispatcher_head(&heartbeat_dispatcher); | |
3107 | hb_back_server_messenger->add_dispatcher_head(&heartbeat_dispatcher); | |
3108 | ||
3109 | objecter_messenger->add_dispatcher_head(service.objecter); | |
3110 | ||
28e407b8 AA |
3111 | service.init(); |
3112 | service.publish_map(osdmap); | |
3113 | service.publish_superblock(superblock); | |
3114 | service.max_oldest_map = superblock.oldest_map; | |
3115 | ||
11fdf7f2 TL |
3116 | for (auto& shard : shards) { |
3117 | // put PGs in a temporary set because we may modify pg_slots | |
3118 | // unordered_map below. | |
3119 | set<PGRef> pgs; | |
3120 | for (auto& i : shard->pg_slots) { | |
3121 | PGRef pg = i.second->pg; | |
3122 | if (!pg) { | |
3123 | continue; | |
3124 | } | |
3125 | pgs.insert(pg); | |
3126 | } | |
3127 | for (auto pg : pgs) { | |
3128 | pg->lock(); | |
3129 | set<pair<spg_t,epoch_t>> new_children; | |
3130 | set<pair<spg_t,epoch_t>> merge_pgs; | |
3131 | service.identify_splits_and_merges(pg->get_osdmap(), osdmap, pg->pg_id, | |
3132 | &new_children, &merge_pgs); | |
3133 | if (!new_children.empty()) { | |
3134 | for (auto shard : shards) { | |
3135 | shard->prime_splits(osdmap, &new_children); | |
3136 | } | |
3137 | assert(new_children.empty()); | |
3138 | } | |
3139 | if (!merge_pgs.empty()) { | |
3140 | for (auto shard : shards) { | |
3141 | shard->prime_merges(osdmap, &merge_pgs); | |
3142 | } | |
3143 | assert(merge_pgs.empty()); | |
3144 | } | |
3145 | pg->unlock(); | |
3146 | } | |
3147 | } | |
3148 | ||
7c673cae | 3149 | osd_op_tp.start(); |
7c673cae FG |
3150 | command_tp.start(); |
3151 | ||
7c673cae FG |
3152 | // start the heartbeat |
3153 | heartbeat_thread.create("osd_srv_heartbt"); | |
3154 | ||
3155 | // tick | |
91327a77 AA |
3156 | tick_timer.add_event_after(get_tick_interval(), |
3157 | new C_Tick(this)); | |
7c673cae | 3158 | { |
11fdf7f2 | 3159 | std::lock_guard l(tick_timer_lock); |
91327a77 AA |
3160 | tick_timer_without_osd_lock.add_event_after(get_tick_interval(), |
3161 | new C_Tick_WithoutOSDLock(this)); | |
7c673cae FG |
3162 | } |
3163 | ||
7c673cae FG |
3164 | osd_lock.Unlock(); |
3165 | ||
3166 | r = monc->authenticate(); | |
3167 | if (r < 0) { | |
c07f9fc5 FG |
3168 | derr << __func__ << " authentication failed: " << cpp_strerror(r) |
3169 | << dendl; | |
11fdf7f2 | 3170 | exit(1); |
7c673cae FG |
3171 | } |
3172 | ||
11fdf7f2 | 3173 | while (monc->wait_auth_rotating(rotating_auth_timeout) < 0) { |
7c673cae FG |
3174 | derr << "unable to obtain rotating service keys; retrying" << dendl; |
3175 | ++rotating_auth_attempts; | |
11fdf7f2 | 3176 | if (rotating_auth_attempts > g_conf()->max_rotating_auth_attempts) { |
c07f9fc5 | 3177 | derr << __func__ << " wait_auth_rotating timed out" << dendl; |
11fdf7f2 | 3178 | exit(1); |
7c673cae FG |
3179 | } |
3180 | } | |
3181 | ||
3182 | r = update_crush_device_class(); | |
3183 | if (r < 0) { | |
d2e6a577 FG |
3184 | derr << __func__ << " unable to update_crush_device_class: " |
3185 | << cpp_strerror(r) << dendl; | |
11fdf7f2 | 3186 | exit(1); |
7c673cae FG |
3187 | } |
3188 | ||
3189 | r = update_crush_location(); | |
3190 | if (r < 0) { | |
d2e6a577 | 3191 | derr << __func__ << " unable to update_crush_location: " |
c07f9fc5 | 3192 | << cpp_strerror(r) << dendl; |
11fdf7f2 | 3193 | exit(1); |
7c673cae FG |
3194 | } |
3195 | ||
3196 | osd_lock.Lock(); | |
3197 | if (is_stopping()) | |
3198 | return 0; | |
3199 | ||
3200 | // start objecter *after* we have authenticated, so that we don't ignore | |
3201 | // the OSDMaps it requests. | |
3202 | service.final_init(); | |
3203 | ||
3204 | check_config(); | |
3205 | ||
3206 | dout(10) << "ensuring pgs have consumed prior maps" << dendl; | |
3207 | consume_map(); | |
7c673cae FG |
3208 | |
3209 | dout(0) << "done with init, starting boot process" << dendl; | |
3210 | ||
3211 | // subscribe to any pg creations | |
3212 | monc->sub_want("osd_pg_creates", last_pg_create_epoch, 0); | |
3213 | ||
3214 | // MgrClient needs this (it doesn't have MonClient reference itself) | |
3215 | monc->sub_want("mgrmap", 0, 0); | |
3216 | ||
3217 | // we don't need to ask for an osdmap here; objecter will | |
3218 | //monc->sub_want("osdmap", osdmap->get_epoch(), CEPH_SUBSCRIBE_ONETIME); | |
3219 | ||
3220 | monc->renew_subs(); | |
3221 | ||
3222 | start_boot(); | |
3223 | ||
3224 | return 0; | |
7c673cae FG |
3225 | |
3226 | out: | |
3227 | enable_disable_fuse(true); | |
3228 | store->umount(); | |
3229 | delete store; | |
3230 | store = NULL; | |
3231 | return r; | |
3232 | } | |
3233 | ||
3234 | void OSD::final_init() | |
3235 | { | |
3236 | AdminSocket *admin_socket = cct->get_admin_socket(); | |
3237 | asok_hook = new OSDSocketHook(this); | |
3238 | int r = admin_socket->register_command("status", "status", asok_hook, | |
3239 | "high-level status of OSD"); | |
11fdf7f2 | 3240 | ceph_assert(r == 0); |
7c673cae FG |
3241 | r = admin_socket->register_command("flush_journal", "flush_journal", |
3242 | asok_hook, | |
3243 | "flush the journal to permanent store"); | |
11fdf7f2 | 3244 | ceph_assert(r == 0); |
7c673cae | 3245 | r = admin_socket->register_command("dump_ops_in_flight", |
c07f9fc5 FG |
3246 | "dump_ops_in_flight " \ |
3247 | "name=filterstr,type=CephString,n=N,req=false", | |
3248 | asok_hook, | |
7c673cae | 3249 | "show the ops currently in flight"); |
11fdf7f2 | 3250 | ceph_assert(r == 0); |
7c673cae | 3251 | r = admin_socket->register_command("ops", |
c07f9fc5 FG |
3252 | "ops " \ |
3253 | "name=filterstr,type=CephString,n=N,req=false", | |
3254 | asok_hook, | |
7c673cae | 3255 | "show the ops currently in flight"); |
11fdf7f2 | 3256 | ceph_assert(r == 0); |
7c673cae | 3257 | r = admin_socket->register_command("dump_blocked_ops", |
c07f9fc5 FG |
3258 | "dump_blocked_ops " \ |
3259 | "name=filterstr,type=CephString,n=N,req=false", | |
3260 | asok_hook, | |
7c673cae | 3261 | "show the blocked ops currently in flight"); |
11fdf7f2 | 3262 | ceph_assert(r == 0); |
c07f9fc5 FG |
3263 | r = admin_socket->register_command("dump_historic_ops", |
3264 | "dump_historic_ops " \ | |
3265 | "name=filterstr,type=CephString,n=N,req=false", | |
7c673cae FG |
3266 | asok_hook, |
3267 | "show recent ops"); | |
11fdf7f2 | 3268 | ceph_assert(r == 0); |
c07f9fc5 FG |
3269 | r = admin_socket->register_command("dump_historic_slow_ops", |
3270 | "dump_historic_slow_ops " \ | |
3271 | "name=filterstr,type=CephString,n=N,req=false", | |
7c673cae FG |
3272 | asok_hook, |
3273 | "show slowest recent ops"); | |
11fdf7f2 | 3274 | ceph_assert(r == 0); |
c07f9fc5 FG |
3275 | r = admin_socket->register_command("dump_historic_ops_by_duration", |
3276 | "dump_historic_ops_by_duration " \ | |
3277 | "name=filterstr,type=CephString,n=N,req=false", | |
7c673cae FG |
3278 | asok_hook, |
3279 | "show slowest recent ops, sorted by duration"); | |
11fdf7f2 | 3280 | ceph_assert(r == 0); |
7c673cae FG |
3281 | r = admin_socket->register_command("dump_op_pq_state", "dump_op_pq_state", |
3282 | asok_hook, | |
3283 | "dump op priority queue state"); | |
11fdf7f2 | 3284 | ceph_assert(r == 0); |
7c673cae FG |
3285 | r = admin_socket->register_command("dump_blacklist", "dump_blacklist", |
3286 | asok_hook, | |
3287 | "dump blacklisted clients and times"); | |
11fdf7f2 | 3288 | ceph_assert(r == 0); |
7c673cae FG |
3289 | r = admin_socket->register_command("dump_watchers", "dump_watchers", |
3290 | asok_hook, | |
3291 | "show clients which have active watches," | |
3292 | " and on which objects"); | |
11fdf7f2 | 3293 | ceph_assert(r == 0); |
7c673cae FG |
3294 | r = admin_socket->register_command("dump_reservations", "dump_reservations", |
3295 | asok_hook, | |
3296 | "show recovery reservations"); | |
11fdf7f2 | 3297 | ceph_assert(r == 0); |
7c673cae FG |
3298 | r = admin_socket->register_command("get_latest_osdmap", "get_latest_osdmap", |
3299 | asok_hook, | |
3300 | "force osd to update the latest map from " | |
3301 | "the mon"); | |
11fdf7f2 | 3302 | ceph_assert(r == 0); |
7c673cae FG |
3303 | |
3304 | r = admin_socket->register_command( "heap", | |
3305 | "heap " \ | |
11fdf7f2 TL |
3306 | "name=heapcmd,type=CephString " \ |
3307 | "name=value,type=CephString,req=false", | |
7c673cae FG |
3308 | asok_hook, |
3309 | "show heap usage info (available only if " | |
3310 | "compiled with tcmalloc)"); | |
11fdf7f2 | 3311 | ceph_assert(r == 0); |
7c673cae FG |
3312 | |
3313 | r = admin_socket->register_command("set_heap_property", | |
3314 | "set_heap_property " \ | |
3315 | "name=property,type=CephString " \ | |
3316 | "name=value,type=CephInt", | |
3317 | asok_hook, | |
3318 | "update malloc extension heap property"); | |
11fdf7f2 | 3319 | ceph_assert(r == 0); |
7c673cae FG |
3320 | |
3321 | r = admin_socket->register_command("get_heap_property", | |
3322 | "get_heap_property " \ | |
3323 | "name=property,type=CephString", | |
3324 | asok_hook, | |
3325 | "get malloc extension heap property"); | |
11fdf7f2 | 3326 | ceph_assert(r == 0); |
7c673cae FG |
3327 | |
3328 | r = admin_socket->register_command("dump_objectstore_kv_stats", | |
3329 | "dump_objectstore_kv_stats", | |
3330 | asok_hook, | |
3331 | "print statistics of kvdb which used by bluestore"); | |
11fdf7f2 | 3332 | ceph_assert(r == 0); |
7c673cae FG |
3333 | |
3334 | r = admin_socket->register_command("dump_scrubs", | |
3335 | "dump_scrubs", | |
3336 | asok_hook, | |
3337 | "print scheduled scrubs"); | |
11fdf7f2 | 3338 | ceph_assert(r == 0); |
7c673cae FG |
3339 | |
3340 | r = admin_socket->register_command("calc_objectstore_db_histogram", | |
3341 | "calc_objectstore_db_histogram", | |
3342 | asok_hook, | |
3343 | "Generate key value histogram of kvdb(rocksdb) which used by bluestore"); | |
11fdf7f2 | 3344 | ceph_assert(r == 0); |
7c673cae FG |
3345 | |
3346 | r = admin_socket->register_command("flush_store_cache", | |
3347 | "flush_store_cache", | |
3348 | asok_hook, | |
3349 | "Flush bluestore internal cache"); | |
11fdf7f2 | 3350 | ceph_assert(r == 0); |
7c673cae FG |
3351 | r = admin_socket->register_command("dump_pgstate_history", "dump_pgstate_history", |
3352 | asok_hook, | |
3353 | "show recent state history"); | |
11fdf7f2 | 3354 | ceph_assert(r == 0); |
7c673cae | 3355 | |
224ce89b WB |
3356 | r = admin_socket->register_command("compact", "compact", |
3357 | asok_hook, | |
3358 | "Commpact object store's omap." | |
3359 | " WARNING: Compaction probably slows your requests"); | |
11fdf7f2 TL |
3360 | ceph_assert(r == 0); |
3361 | ||
3362 | r = admin_socket->register_command("get_mapped_pools", "get_mapped_pools", | |
3363 | asok_hook, | |
3364 | "dump pools whose PG(s) are mapped to this OSD."); | |
3365 | ||
3366 | ceph_assert(r == 0); | |
3367 | ||
3368 | r = admin_socket->register_command("smart", "smart name=devid,type=CephString,req=False", | |
3369 | asok_hook, | |
3370 | "probe OSD devices for SMART data."); | |
3371 | ||
3372 | ceph_assert(r == 0); | |
3373 | ||
3374 | r = admin_socket->register_command("list_devices", "list_devices", | |
3375 | asok_hook, | |
3376 | "list OSD devices."); | |
3377 | r = admin_socket->register_command("send_beacon", "send_beacon", | |
3378 | asok_hook, | |
3379 | "send OSD beacon to mon immediately"); | |
224ce89b | 3380 | |
7c673cae FG |
3381 | test_ops_hook = new TestOpsSocketHook(&(this->service), this->store); |
3382 | // Note: pools are CephString instead of CephPoolname because | |
3383 | // these commands traditionally support both pool names and numbers | |
3384 | r = admin_socket->register_command( | |
3385 | "setomapval", | |
3386 | "setomapval " \ | |
3387 | "name=pool,type=CephString " \ | |
3388 | "name=objname,type=CephObjectname " \ | |
3389 | "name=key,type=CephString "\ | |
3390 | "name=val,type=CephString", | |
3391 | test_ops_hook, | |
3392 | "set omap key"); | |
11fdf7f2 | 3393 | ceph_assert(r == 0); |
7c673cae FG |
3394 | r = admin_socket->register_command( |
3395 | "rmomapkey", | |
3396 | "rmomapkey " \ | |
3397 | "name=pool,type=CephString " \ | |
3398 | "name=objname,type=CephObjectname " \ | |
3399 | "name=key,type=CephString", | |
3400 | test_ops_hook, | |
3401 | "remove omap key"); | |
11fdf7f2 | 3402 | ceph_assert(r == 0); |
7c673cae FG |
3403 | r = admin_socket->register_command( |
3404 | "setomapheader", | |
3405 | "setomapheader " \ | |
3406 | "name=pool,type=CephString " \ | |
3407 | "name=objname,type=CephObjectname " \ | |
3408 | "name=header,type=CephString", | |
3409 | test_ops_hook, | |
3410 | "set omap header"); | |
11fdf7f2 | 3411 | ceph_assert(r == 0); |
7c673cae FG |
3412 | |
3413 | r = admin_socket->register_command( | |
3414 | "getomap", | |
3415 | "getomap " \ | |
3416 | "name=pool,type=CephString " \ | |
3417 | "name=objname,type=CephObjectname", | |
3418 | test_ops_hook, | |
3419 | "output entire object map"); | |
11fdf7f2 | 3420 | ceph_assert(r == 0); |
7c673cae FG |
3421 | |
3422 | r = admin_socket->register_command( | |
3423 | "truncobj", | |
3424 | "truncobj " \ | |
3425 | "name=pool,type=CephString " \ | |
3426 | "name=objname,type=CephObjectname " \ | |
3427 | "name=len,type=CephInt", | |
3428 | test_ops_hook, | |
3429 | "truncate object to length"); | |
11fdf7f2 | 3430 | ceph_assert(r == 0); |
7c673cae FG |
3431 | |
3432 | r = admin_socket->register_command( | |
3433 | "injectdataerr", | |
3434 | "injectdataerr " \ | |
3435 | "name=pool,type=CephString " \ | |
3436 | "name=objname,type=CephObjectname " \ | |
3437 | "name=shardid,type=CephInt,req=false,range=0|255", | |
3438 | test_ops_hook, | |
3439 | "inject data error to an object"); | |
11fdf7f2 | 3440 | ceph_assert(r == 0); |
7c673cae FG |
3441 | |
3442 | r = admin_socket->register_command( | |
3443 | "injectmdataerr", | |
3444 | "injectmdataerr " \ | |
3445 | "name=pool,type=CephString " \ | |
3446 | "name=objname,type=CephObjectname " \ | |
3447 | "name=shardid,type=CephInt,req=false,range=0|255", | |
3448 | test_ops_hook, | |
3449 | "inject metadata error to an object"); | |
11fdf7f2 | 3450 | ceph_assert(r == 0); |
7c673cae FG |
3451 | r = admin_socket->register_command( |
3452 | "set_recovery_delay", | |
3453 | "set_recovery_delay " \ | |
3454 | "name=utime,type=CephInt,req=false", | |
3455 | test_ops_hook, | |
3456 | "Delay osd recovery by specified seconds"); | |
11fdf7f2 | 3457 | ceph_assert(r == 0); |
7c673cae FG |
3458 | r = admin_socket->register_command( |
3459 | "trigger_scrub", | |
3460 | "trigger_scrub " \ | |
a8e16298 TL |
3461 | "name=pgid,type=CephString " \ |
3462 | "name=time,type=CephInt,req=false", | |
7c673cae FG |
3463 | test_ops_hook, |
3464 | "Trigger a scheduled scrub "); | |
11fdf7f2 | 3465 | ceph_assert(r == 0); |
a8e16298 TL |
3466 | r = admin_socket->register_command( |
3467 | "trigger_deep_scrub", | |
3468 | "trigger_deep_scrub " \ | |
3469 | "name=pgid,type=CephString " \ | |
3470 | "name=time,type=CephInt,req=false", | |
3471 | test_ops_hook, | |
3472 | "Trigger a scheduled deep scrub "); | |
3473 | ceph_assert(r == 0); | |
7c673cae FG |
3474 | r = admin_socket->register_command( |
3475 | "injectfull", | |
3476 | "injectfull " \ | |
3477 | "name=type,type=CephString,req=false " \ | |
3478 | "name=count,type=CephInt,req=false ", | |
3479 | test_ops_hook, | |
3480 | "Inject a full disk (optional count times)"); | |
11fdf7f2 | 3481 | ceph_assert(r == 0); |
7c673cae FG |
3482 | } |
3483 | ||
3484 | void OSD::create_logger() | |
3485 | { | |
3486 | dout(10) << "create_logger" << dendl; | |
3487 | ||
3488 | PerfCountersBuilder osd_plb(cct, "osd", l_osd_first, l_osd_last); | |
3489 | ||
3490 | // Latency axis configuration for op histograms, values are in nanoseconds | |
3491 | PerfHistogramCommon::axis_config_d op_hist_x_axis_config{ | |
3492 | "Latency (usec)", | |
3493 | PerfHistogramCommon::SCALE_LOG2, ///< Latency in logarithmic scale | |
3494 | 0, ///< Start at 0 | |
3495 | 100000, ///< Quantization unit is 100usec | |
3496 | 32, ///< Enough to cover much longer than slow requests | |
3497 | }; | |
3498 | ||
3499 | // Op size axis configuration for op histograms, values are in bytes | |
3500 | PerfHistogramCommon::axis_config_d op_hist_y_axis_config{ | |
3501 | "Request size (bytes)", | |
3502 | PerfHistogramCommon::SCALE_LOG2, ///< Request size in logarithmic scale | |
3503 | 0, ///< Start at 0 | |
3504 | 512, ///< Quantization unit is 512 bytes | |
3505 | 32, ///< Enough to cover requests larger than GB | |
3506 | }; | |
3507 | ||
3508 | ||
3efd9988 FG |
3509 | // All the basic OSD operation stats are to be considered useful |
3510 | osd_plb.set_prio_default(PerfCountersBuilder::PRIO_USEFUL); | |
3511 | ||
7c673cae FG |
3512 | osd_plb.add_u64( |
3513 | l_osd_op_wip, "op_wip", | |
3514 | "Replication operations currently being processed (primary)"); | |
3515 | osd_plb.add_u64_counter( | |
3516 | l_osd_op, "op", | |
3517 | "Client operations", | |
3518 | "ops", PerfCountersBuilder::PRIO_CRITICAL); | |
3519 | osd_plb.add_u64_counter( | |
3520 | l_osd_op_inb, "op_in_bytes", | |
3521 | "Client operations total write size", | |
11fdf7f2 | 3522 | "wr", PerfCountersBuilder::PRIO_INTERESTING, unit_t(UNIT_BYTES)); |
7c673cae FG |
3523 | osd_plb.add_u64_counter( |
3524 | l_osd_op_outb, "op_out_bytes", | |
3525 | "Client operations total read size", | |
11fdf7f2 | 3526 | "rd", PerfCountersBuilder::PRIO_INTERESTING, unit_t(UNIT_BYTES)); |
7c673cae FG |
3527 | osd_plb.add_time_avg( |
3528 | l_osd_op_lat, "op_latency", | |
3529 | "Latency of client operations (including queue time)", | |
3530 | "l", 9); | |
3531 | osd_plb.add_time_avg( | |
3532 | l_osd_op_process_lat, "op_process_latency", | |
3533 | "Latency of client operations (excluding queue time)"); | |
3534 | osd_plb.add_time_avg( | |
3535 | l_osd_op_prepare_lat, "op_prepare_latency", | |
3536 | "Latency of client operations (excluding queue time and wait for finished)"); | |
3537 | ||
3538 | osd_plb.add_u64_counter( | |
3539 | l_osd_op_r, "op_r", "Client read operations"); | |
3540 | osd_plb.add_u64_counter( | |
11fdf7f2 | 3541 | l_osd_op_r_outb, "op_r_out_bytes", "Client data read", NULL, PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); |
7c673cae FG |
3542 | osd_plb.add_time_avg( |
3543 | l_osd_op_r_lat, "op_r_latency", | |
3544 | "Latency of read operation (including queue time)"); | |
31f18b77 | 3545 | osd_plb.add_u64_counter_histogram( |
7c673cae FG |
3546 | l_osd_op_r_lat_outb_hist, "op_r_latency_out_bytes_histogram", |
3547 | op_hist_x_axis_config, op_hist_y_axis_config, | |
3548 | "Histogram of operation latency (including queue time) + data read"); | |
3549 | osd_plb.add_time_avg( | |
3550 | l_osd_op_r_process_lat, "op_r_process_latency", | |
3551 | "Latency of read operation (excluding queue time)"); | |
3552 | osd_plb.add_time_avg( | |
3553 | l_osd_op_r_prepare_lat, "op_r_prepare_latency", | |
3554 | "Latency of read operations (excluding queue time and wait for finished)"); | |
3555 | osd_plb.add_u64_counter( | |
3556 | l_osd_op_w, "op_w", "Client write operations"); | |
3557 | osd_plb.add_u64_counter( | |
3558 | l_osd_op_w_inb, "op_w_in_bytes", "Client data written"); | |
3559 | osd_plb.add_time_avg( | |
3560 | l_osd_op_w_lat, "op_w_latency", | |
3561 | "Latency of write operation (including queue time)"); | |
31f18b77 | 3562 | osd_plb.add_u64_counter_histogram( |
7c673cae FG |
3563 | l_osd_op_w_lat_inb_hist, "op_w_latency_in_bytes_histogram", |
3564 | op_hist_x_axis_config, op_hist_y_axis_config, | |
3565 | "Histogram of operation latency (including queue time) + data written"); | |
3566 | osd_plb.add_time_avg( | |
3567 | l_osd_op_w_process_lat, "op_w_process_latency", | |
3568 | "Latency of write operation (excluding queue time)"); | |
3569 | osd_plb.add_time_avg( | |
3570 | l_osd_op_w_prepare_lat, "op_w_prepare_latency", | |
3571 | "Latency of write operations (excluding queue time and wait for finished)"); | |
3572 | osd_plb.add_u64_counter( | |
3573 | l_osd_op_rw, "op_rw", | |
3574 | "Client read-modify-write operations"); | |
3575 | osd_plb.add_u64_counter( | |
3576 | l_osd_op_rw_inb, "op_rw_in_bytes", | |
11fdf7f2 | 3577 | "Client read-modify-write operations write in", NULL, PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); |
7c673cae FG |
3578 | osd_plb.add_u64_counter( |
3579 | l_osd_op_rw_outb,"op_rw_out_bytes", | |
11fdf7f2 | 3580 | "Client read-modify-write operations read out ", NULL, PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); |
7c673cae FG |
3581 | osd_plb.add_time_avg( |
3582 | l_osd_op_rw_lat, "op_rw_latency", | |
3583 | "Latency of read-modify-write operation (including queue time)"); | |
31f18b77 | 3584 | osd_plb.add_u64_counter_histogram( |
7c673cae FG |
3585 | l_osd_op_rw_lat_inb_hist, "op_rw_latency_in_bytes_histogram", |
3586 | op_hist_x_axis_config, op_hist_y_axis_config, | |
3587 | "Histogram of rw operation latency (including queue time) + data written"); | |
31f18b77 | 3588 | osd_plb.add_u64_counter_histogram( |
7c673cae FG |
3589 | l_osd_op_rw_lat_outb_hist, "op_rw_latency_out_bytes_histogram", |
3590 | op_hist_x_axis_config, op_hist_y_axis_config, | |
3591 | "Histogram of rw operation latency (including queue time) + data read"); | |
3592 | osd_plb.add_time_avg( | |
3593 | l_osd_op_rw_process_lat, "op_rw_process_latency", | |
3594 | "Latency of read-modify-write operation (excluding queue time)"); | |
3595 | osd_plb.add_time_avg( | |
3596 | l_osd_op_rw_prepare_lat, "op_rw_prepare_latency", | |
3597 | "Latency of read-modify-write operations (excluding queue time and wait for finished)"); | |
3598 | ||
3efd9988 FG |
3599 | // Now we move on to some more obscure stats, revert to assuming things |
3600 | // are low priority unless otherwise specified. | |
3601 | osd_plb.set_prio_default(PerfCountersBuilder::PRIO_DEBUGONLY); | |
3602 | ||
224ce89b WB |
3603 | osd_plb.add_time_avg(l_osd_op_before_queue_op_lat, "op_before_queue_op_lat", |
3604 | "Latency of IO before calling queue(before really queue into ShardedOpWq)"); // client io before queue op_wq latency | |
3605 | osd_plb.add_time_avg(l_osd_op_before_dequeue_op_lat, "op_before_dequeue_op_lat", | |
3606 | "Latency of IO before calling dequeue_op(already dequeued and get PG lock)"); // client io before dequeue_op latency | |
3607 | ||
7c673cae FG |
3608 | osd_plb.add_u64_counter( |
3609 | l_osd_sop, "subop", "Suboperations"); | |
3610 | osd_plb.add_u64_counter( | |
11fdf7f2 | 3611 | l_osd_sop_inb, "subop_in_bytes", "Suboperations total size", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
3612 | osd_plb.add_time_avg(l_osd_sop_lat, "subop_latency", "Suboperations latency"); |
3613 | ||
3614 | osd_plb.add_u64_counter(l_osd_sop_w, "subop_w", "Replicated writes"); | |
3615 | osd_plb.add_u64_counter( | |
11fdf7f2 | 3616 | l_osd_sop_w_inb, "subop_w_in_bytes", "Replicated written data size", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
3617 | osd_plb.add_time_avg( |
3618 | l_osd_sop_w_lat, "subop_w_latency", "Replicated writes latency"); | |
3619 | osd_plb.add_u64_counter( | |
3620 | l_osd_sop_pull, "subop_pull", "Suboperations pull requests"); | |
3621 | osd_plb.add_time_avg( | |
3622 | l_osd_sop_pull_lat, "subop_pull_latency", "Suboperations pull latency"); | |
3623 | osd_plb.add_u64_counter( | |
3624 | l_osd_sop_push, "subop_push", "Suboperations push messages"); | |
3625 | osd_plb.add_u64_counter( | |
11fdf7f2 | 3626 | l_osd_sop_push_inb, "subop_push_in_bytes", "Suboperations pushed size", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
3627 | osd_plb.add_time_avg( |
3628 | l_osd_sop_push_lat, "subop_push_latency", "Suboperations push latency"); | |
3629 | ||
3630 | osd_plb.add_u64_counter(l_osd_pull, "pull", "Pull requests sent"); | |
3631 | osd_plb.add_u64_counter(l_osd_push, "push", "Push messages sent"); | |
11fdf7f2 | 3632 | osd_plb.add_u64_counter(l_osd_push_outb, "push_out_bytes", "Pushed size", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
3633 | |
3634 | osd_plb.add_u64_counter( | |
3635 | l_osd_rop, "recovery_ops", | |
3636 | "Started recovery operations", | |
3637 | "rop", PerfCountersBuilder::PRIO_INTERESTING); | |
3638 | ||
11fdf7f2 TL |
3639 | osd_plb.add_u64_counter( |
3640 | l_osd_rbytes, "recovery_bytes", | |
3641 | "recovery bytes", | |
3642 | "rbt", PerfCountersBuilder::PRIO_INTERESTING); | |
3643 | ||
7c673cae | 3644 | osd_plb.add_u64(l_osd_loadavg, "loadavg", "CPU load"); |
7c673cae FG |
3645 | osd_plb.add_u64( |
3646 | l_osd_cached_crc, "cached_crc", "Total number getting crc from crc_cache"); | |
3647 | osd_plb.add_u64( | |
3648 | l_osd_cached_crc_adjusted, "cached_crc_adjusted", | |
3649 | "Total number getting crc from crc_cache with adjusting"); | |
3650 | osd_plb.add_u64(l_osd_missed_crc, "missed_crc", | |
3651 | "Total number of crc cache misses"); | |
3652 | ||
3653 | osd_plb.add_u64(l_osd_pg, "numpg", "Placement groups", | |
3654 | "pgs", PerfCountersBuilder::PRIO_USEFUL); | |
3655 | osd_plb.add_u64( | |
3656 | l_osd_pg_primary, "numpg_primary", | |
3657 | "Placement groups for which this osd is primary"); | |
3658 | osd_plb.add_u64( | |
3659 | l_osd_pg_replica, "numpg_replica", | |
3660 | "Placement groups for which this osd is replica"); | |
3661 | osd_plb.add_u64( | |
3662 | l_osd_pg_stray, "numpg_stray", | |
3663 | "Placement groups ready to be deleted from this osd"); | |
94b18763 FG |
3664 | osd_plb.add_u64( |
3665 | l_osd_pg_removing, "numpg_removing", | |
3666 | "Placement groups queued for local deletion", "pgsr", | |
3667 | PerfCountersBuilder::PRIO_USEFUL); | |
7c673cae FG |
3668 | osd_plb.add_u64( |
3669 | l_osd_hb_to, "heartbeat_to_peers", "Heartbeat (ping) peers we send to"); | |
3670 | osd_plb.add_u64_counter(l_osd_map, "map_messages", "OSD map messages"); | |
3671 | osd_plb.add_u64_counter(l_osd_mape, "map_message_epochs", "OSD map epochs"); | |
3672 | osd_plb.add_u64_counter( | |
3673 | l_osd_mape_dup, "map_message_epoch_dups", "OSD map duplicates"); | |
3674 | osd_plb.add_u64_counter( | |
3675 | l_osd_waiting_for_map, "messages_delayed_for_map", | |
3676 | "Operations waiting for OSD map"); | |
31f18b77 | 3677 | |
7c673cae FG |
3678 | osd_plb.add_u64_counter( |
3679 | l_osd_map_cache_hit, "osd_map_cache_hit", "osdmap cache hit"); | |
3680 | osd_plb.add_u64_counter( | |
3681 | l_osd_map_cache_miss, "osd_map_cache_miss", "osdmap cache miss"); | |
3682 | osd_plb.add_u64_counter( | |
3683 | l_osd_map_cache_miss_low, "osd_map_cache_miss_low", | |
3684 | "osdmap cache miss below cache lower bound"); | |
3685 | osd_plb.add_u64_avg( | |
3686 | l_osd_map_cache_miss_low_avg, "osd_map_cache_miss_low_avg", | |
3687 | "osdmap cache miss, avg distance below cache lower bound"); | |
31f18b77 FG |
3688 | osd_plb.add_u64_counter( |
3689 | l_osd_map_bl_cache_hit, "osd_map_bl_cache_hit", | |
3690 | "OSDMap buffer cache hits"); | |
3691 | osd_plb.add_u64_counter( | |
3692 | l_osd_map_bl_cache_miss, "osd_map_bl_cache_miss", | |
3693 | "OSDMap buffer cache misses"); | |
7c673cae | 3694 | |
3efd9988 FG |
3695 | osd_plb.add_u64( |
3696 | l_osd_stat_bytes, "stat_bytes", "OSD size", "size", | |
11fdf7f2 | 3697 | PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); |
3efd9988 FG |
3698 | osd_plb.add_u64( |
3699 | l_osd_stat_bytes_used, "stat_bytes_used", "Used space", "used", | |
11fdf7f2 TL |
3700 | PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); |
3701 | osd_plb.add_u64(l_osd_stat_bytes_avail, "stat_bytes_avail", "Available space", NULL, 0, unit_t(UNIT_BYTES)); | |
7c673cae FG |
3702 | |
3703 | osd_plb.add_u64_counter( | |
3704 | l_osd_copyfrom, "copyfrom", "Rados \"copy-from\" operations"); | |
3705 | ||
3706 | osd_plb.add_u64_counter(l_osd_tier_promote, "tier_promote", "Tier promotions"); | |
3707 | osd_plb.add_u64_counter(l_osd_tier_flush, "tier_flush", "Tier flushes"); | |
3708 | osd_plb.add_u64_counter( | |
3709 | l_osd_tier_flush_fail, "tier_flush_fail", "Failed tier flushes"); | |
3710 | osd_plb.add_u64_counter( | |
3711 | l_osd_tier_try_flush, "tier_try_flush", "Tier flush attempts"); | |
3712 | osd_plb.add_u64_counter( | |
3713 | l_osd_tier_try_flush_fail, "tier_try_flush_fail", | |
3714 | "Failed tier flush attempts"); | |
3715 | osd_plb.add_u64_counter( | |
3716 | l_osd_tier_evict, "tier_evict", "Tier evictions"); | |
3717 | osd_plb.add_u64_counter( | |
3718 | l_osd_tier_whiteout, "tier_whiteout", "Tier whiteouts"); | |
3719 | osd_plb.add_u64_counter( | |
3720 | l_osd_tier_dirty, "tier_dirty", "Dirty tier flag set"); | |
3721 | osd_plb.add_u64_counter( | |
3722 | l_osd_tier_clean, "tier_clean", "Dirty tier flag cleaned"); | |
3723 | osd_plb.add_u64_counter( | |
3724 | l_osd_tier_delay, "tier_delay", "Tier delays (agent waiting)"); | |
3725 | osd_plb.add_u64_counter( | |
3726 | l_osd_tier_proxy_read, "tier_proxy_read", "Tier proxy reads"); | |
3727 | osd_plb.add_u64_counter( | |
3728 | l_osd_tier_proxy_write, "tier_proxy_write", "Tier proxy writes"); | |
3729 | ||
3730 | osd_plb.add_u64_counter( | |
3731 | l_osd_agent_wake, "agent_wake", "Tiering agent wake up"); | |
3732 | osd_plb.add_u64_counter( | |
3733 | l_osd_agent_skip, "agent_skip", "Objects skipped by agent"); | |
3734 | osd_plb.add_u64_counter( | |
3735 | l_osd_agent_flush, "agent_flush", "Tiering agent flushes"); | |
3736 | osd_plb.add_u64_counter( | |
3737 | l_osd_agent_evict, "agent_evict", "Tiering agent evictions"); | |
3738 | ||
3739 | osd_plb.add_u64_counter( | |
3740 | l_osd_object_ctx_cache_hit, "object_ctx_cache_hit", "Object context cache hits"); | |
3741 | osd_plb.add_u64_counter( | |
3742 | l_osd_object_ctx_cache_total, "object_ctx_cache_total", "Object context cache lookups"); | |
3743 | ||
3744 | osd_plb.add_u64_counter(l_osd_op_cache_hit, "op_cache_hit"); | |
3745 | osd_plb.add_time_avg( | |
3746 | l_osd_tier_flush_lat, "osd_tier_flush_lat", "Object flush latency"); | |
3747 | osd_plb.add_time_avg( | |
3748 | l_osd_tier_promote_lat, "osd_tier_promote_lat", "Object promote latency"); | |
3749 | osd_plb.add_time_avg( | |
3750 | l_osd_tier_r_lat, "osd_tier_r_lat", "Object proxy read latency"); | |
3751 | ||
3752 | osd_plb.add_u64_counter( | |
3753 | l_osd_pg_info, "osd_pg_info", "PG updated its info (using any method)"); | |
3754 | osd_plb.add_u64_counter( | |
3755 | l_osd_pg_fastinfo, "osd_pg_fastinfo", | |
3756 | "PG updated its info using fastinfo attr"); | |
3757 | osd_plb.add_u64_counter( | |
3758 | l_osd_pg_biginfo, "osd_pg_biginfo", "PG updated its biginfo attr"); | |
3759 | ||
3760 | logger = osd_plb.create_perf_counters(); | |
3761 | cct->get_perfcounters_collection()->add(logger); | |
3762 | } | |
3763 | ||
3764 | void OSD::create_recoverystate_perf() | |
3765 | { | |
3766 | dout(10) << "create_recoverystate_perf" << dendl; | |
3767 | ||
3768 | PerfCountersBuilder rs_perf(cct, "recoverystate_perf", rs_first, rs_last); | |
3769 | ||
3770 | rs_perf.add_time_avg(rs_initial_latency, "initial_latency", "Initial recovery state latency"); | |
3771 | rs_perf.add_time_avg(rs_started_latency, "started_latency", "Started recovery state latency"); | |
3772 | rs_perf.add_time_avg(rs_reset_latency, "reset_latency", "Reset recovery state latency"); | |
3773 | rs_perf.add_time_avg(rs_start_latency, "start_latency", "Start recovery state latency"); | |
3774 | rs_perf.add_time_avg(rs_primary_latency, "primary_latency", "Primary recovery state latency"); | |
3775 | rs_perf.add_time_avg(rs_peering_latency, "peering_latency", "Peering recovery state latency"); | |
3776 | rs_perf.add_time_avg(rs_backfilling_latency, "backfilling_latency", "Backfilling recovery state latency"); | |
3777 | rs_perf.add_time_avg(rs_waitremotebackfillreserved_latency, "waitremotebackfillreserved_latency", "Wait remote backfill reserved recovery state latency"); | |
3778 | rs_perf.add_time_avg(rs_waitlocalbackfillreserved_latency, "waitlocalbackfillreserved_latency", "Wait local backfill reserved recovery state latency"); | |
3779 | rs_perf.add_time_avg(rs_notbackfilling_latency, "notbackfilling_latency", "Notbackfilling recovery state latency"); | |
3780 | rs_perf.add_time_avg(rs_repnotrecovering_latency, "repnotrecovering_latency", "Repnotrecovering recovery state latency"); | |
3781 | rs_perf.add_time_avg(rs_repwaitrecoveryreserved_latency, "repwaitrecoveryreserved_latency", "Rep wait recovery reserved recovery state latency"); | |
3782 | rs_perf.add_time_avg(rs_repwaitbackfillreserved_latency, "repwaitbackfillreserved_latency", "Rep wait backfill reserved recovery state latency"); | |
3783 | rs_perf.add_time_avg(rs_reprecovering_latency, "reprecovering_latency", "RepRecovering recovery state latency"); | |
3784 | rs_perf.add_time_avg(rs_activating_latency, "activating_latency", "Activating recovery state latency"); | |
3785 | rs_perf.add_time_avg(rs_waitlocalrecoveryreserved_latency, "waitlocalrecoveryreserved_latency", "Wait local recovery reserved recovery state latency"); | |
3786 | rs_perf.add_time_avg(rs_waitremoterecoveryreserved_latency, "waitremoterecoveryreserved_latency", "Wait remote recovery reserved recovery state latency"); | |
3787 | rs_perf.add_time_avg(rs_recovering_latency, "recovering_latency", "Recovering recovery state latency"); | |
3788 | rs_perf.add_time_avg(rs_recovered_latency, "recovered_latency", "Recovered recovery state latency"); | |
3789 | rs_perf.add_time_avg(rs_clean_latency, "clean_latency", "Clean recovery state latency"); | |
3790 | rs_perf.add_time_avg(rs_active_latency, "active_latency", "Active recovery state latency"); | |
3791 | rs_perf.add_time_avg(rs_replicaactive_latency, "replicaactive_latency", "Replicaactive recovery state latency"); | |
3792 | rs_perf.add_time_avg(rs_stray_latency, "stray_latency", "Stray recovery state latency"); | |
3793 | rs_perf.add_time_avg(rs_getinfo_latency, "getinfo_latency", "Getinfo recovery state latency"); | |
3794 | rs_perf.add_time_avg(rs_getlog_latency, "getlog_latency", "Getlog recovery state latency"); | |
3795 | rs_perf.add_time_avg(rs_waitactingchange_latency, "waitactingchange_latency", "Waitactingchange recovery state latency"); | |
3796 | rs_perf.add_time_avg(rs_incomplete_latency, "incomplete_latency", "Incomplete recovery state latency"); | |
3797 | rs_perf.add_time_avg(rs_down_latency, "down_latency", "Down recovery state latency"); | |
3798 | rs_perf.add_time_avg(rs_getmissing_latency, "getmissing_latency", "Getmissing recovery state latency"); | |
3799 | rs_perf.add_time_avg(rs_waitupthru_latency, "waitupthru_latency", "Waitupthru recovery state latency"); | |
3800 | rs_perf.add_time_avg(rs_notrecovering_latency, "notrecovering_latency", "Notrecovering recovery state latency"); | |
3801 | ||
3802 | recoverystate_perf = rs_perf.create_perf_counters(); | |
3803 | cct->get_perfcounters_collection()->add(recoverystate_perf); | |
3804 | } | |
3805 | ||
3806 | int OSD::shutdown() | |
3807 | { | |
3808 | if (!service.prepare_to_stop()) | |
3809 | return 0; // already shutting down | |
3810 | osd_lock.Lock(); | |
3811 | if (is_stopping()) { | |
3812 | osd_lock.Unlock(); | |
3813 | return 0; | |
3814 | } | |
11fdf7f2 | 3815 | dout(0) << "shutdown" << dendl; |
7c673cae FG |
3816 | |
3817 | set_state(STATE_STOPPING); | |
3818 | ||
3819 | // Debugging | |
11fdf7f2 TL |
3820 | if (cct->_conf.get_val<bool>("osd_debug_shutdown")) { |
3821 | cct->_conf.set_val("debug_osd", "100"); | |
3822 | cct->_conf.set_val("debug_journal", "100"); | |
3823 | cct->_conf.set_val("debug_filestore", "100"); | |
3824 | cct->_conf.set_val("debug_bluestore", "100"); | |
3825 | cct->_conf.set_val("debug_ms", "100"); | |
3826 | cct->_conf.apply_changes(nullptr); | |
3efd9988 | 3827 | } |
7c673cae FG |
3828 | |
3829 | // stop MgrClient earlier as it's more like an internal consumer of OSD | |
3830 | mgrc.shutdown(); | |
3831 | ||
3832 | service.start_shutdown(); | |
3833 | ||
3834 | // stop sending work to pgs. this just prevents any new work in _process | |
3835 | // from racing with on_shutdown and potentially entering the pg after. | |
3836 | op_shardedwq.drain(); | |
3837 | ||
3838 | // Shutdown PGs | |
3839 | { | |
11fdf7f2 TL |
3840 | vector<PGRef> pgs; |
3841 | _get_pgs(&pgs); | |
3842 | for (auto pg : pgs) { | |
3843 | pg->shutdown(); | |
7c673cae FG |
3844 | } |
3845 | } | |
7c673cae FG |
3846 | |
3847 | // drain op queue again (in case PGs requeued something) | |
3848 | op_shardedwq.drain(); | |
3849 | { | |
3850 | finished.clear(); // zap waiters (bleh, this is messy) | |
11fdf7f2 | 3851 | waiting_for_osdmap.clear(); |
7c673cae FG |
3852 | } |
3853 | ||
7c673cae | 3854 | // unregister commands |
11fdf7f2 | 3855 | cct->get_admin_socket()->unregister_commands(asok_hook); |
7c673cae FG |
3856 | delete asok_hook; |
3857 | asok_hook = NULL; | |
3858 | ||
11fdf7f2 | 3859 | cct->get_admin_socket()->unregister_commands(test_ops_hook); |
7c673cae FG |
3860 | delete test_ops_hook; |
3861 | test_ops_hook = NULL; | |
3862 | ||
3863 | osd_lock.Unlock(); | |
3864 | ||
3865 | heartbeat_lock.Lock(); | |
3866 | heartbeat_stop = true; | |
3867 | heartbeat_cond.Signal(); | |
3868 | heartbeat_lock.Unlock(); | |
3869 | heartbeat_thread.join(); | |
3870 | ||
7c673cae FG |
3871 | osd_op_tp.drain(); |
3872 | osd_op_tp.stop(); | |
3873 | dout(10) << "op sharded tp stopped" << dendl; | |
3874 | ||
3875 | command_tp.drain(); | |
3876 | command_tp.stop(); | |
3877 | dout(10) << "command tp stopped" << dendl; | |
3878 | ||
7c673cae FG |
3879 | dout(10) << "stopping agent" << dendl; |
3880 | service.agent_stop(); | |
3881 | ||
11fdf7f2 TL |
3882 | boot_finisher.wait_for_empty(); |
3883 | ||
7c673cae FG |
3884 | osd_lock.Lock(); |
3885 | ||
11fdf7f2 | 3886 | boot_finisher.stop(); |
7c673cae FG |
3887 | reset_heartbeat_peers(); |
3888 | ||
3889 | tick_timer.shutdown(); | |
3890 | ||
3891 | { | |
11fdf7f2 | 3892 | std::lock_guard l(tick_timer_lock); |
7c673cae FG |
3893 | tick_timer_without_osd_lock.shutdown(); |
3894 | } | |
3895 | ||
3896 | // note unmount epoch | |
3897 | dout(10) << "noting clean unmount in epoch " << osdmap->get_epoch() << dendl; | |
3898 | superblock.mounted = service.get_boot_epoch(); | |
3899 | superblock.clean_thru = osdmap->get_epoch(); | |
3900 | ObjectStore::Transaction t; | |
3901 | write_superblock(t); | |
11fdf7f2 | 3902 | int r = store->queue_transaction(service.meta_ch, std::move(t)); |
7c673cae FG |
3903 | if (r) { |
3904 | derr << "OSD::shutdown: error writing superblock: " | |
3905 | << cpp_strerror(r) << dendl; | |
3906 | } | |
3907 | ||
3908 | ||
31f18b77 FG |
3909 | service.shutdown_reserver(); |
3910 | ||
7c673cae FG |
3911 | // Remove PGs |
3912 | #ifdef PG_DEBUG_REFS | |
3913 | service.dump_live_pgids(); | |
3914 | #endif | |
11fdf7f2 TL |
3915 | while (true) { |
3916 | vector<PGRef> pgs; | |
3917 | _get_pgs(&pgs, true); | |
3918 | if (pgs.empty()) { | |
3919 | break; | |
3920 | } | |
3921 | for (auto& pg : pgs) { | |
3922 | if (pg->is_deleted()) { | |
3923 | continue; | |
3924 | } | |
3925 | dout(20) << " kicking pg " << pg << dendl; | |
3926 | pg->lock(); | |
3927 | if (pg->get_num_ref() != 1) { | |
3928 | derr << "pgid " << pg->get_pgid() << " has ref count of " | |
3929 | << pg->get_num_ref() << dendl; | |
7c673cae | 3930 | #ifdef PG_DEBUG_REFS |
11fdf7f2 | 3931 | pg->dump_live_ids(); |
7c673cae | 3932 | #endif |
31f18b77 FG |
3933 | if (cct->_conf->osd_shutdown_pgref_assert) { |
3934 | ceph_abort(); | |
3935 | } | |
7c673cae | 3936 | } |
11fdf7f2 TL |
3937 | pg->ch.reset(); |
3938 | pg->unlock(); | |
7c673cae | 3939 | } |
7c673cae FG |
3940 | } |
3941 | #ifdef PG_DEBUG_REFS | |
3942 | service.dump_live_pgids(); | |
3943 | #endif | |
f64942e4 AA |
3944 | |
3945 | osd_lock.Unlock(); | |
11fdf7f2 | 3946 | cct->_conf.remove_observer(this); |
f64942e4 | 3947 | osd_lock.Lock(); |
7c673cae | 3948 | |
11fdf7f2 TL |
3949 | service.meta_ch.reset(); |
3950 | ||
7c673cae FG |
3951 | dout(10) << "syncing store" << dendl; |
3952 | enable_disable_fuse(true); | |
3953 | ||
3954 | if (cct->_conf->osd_journal_flush_on_shutdown) { | |
3955 | dout(10) << "flushing journal" << dendl; | |
3956 | store->flush_journal(); | |
3957 | } | |
3958 | ||
7c673cae FG |
3959 | monc->shutdown(); |
3960 | osd_lock.Unlock(); | |
3961 | ||
11fdf7f2 | 3962 | map_lock.get_write(); |
7c673cae | 3963 | osdmap = OSDMapRef(); |
11fdf7f2 TL |
3964 | map_lock.put_write(); |
3965 | ||
3966 | for (auto s : shards) { | |
3967 | std::lock_guard l(s->osdmap_lock); | |
3968 | s->shard_osdmap = OSDMapRef(); | |
3969 | } | |
7c673cae | 3970 | service.shutdown(); |
11fdf7f2 TL |
3971 | |
3972 | std::lock_guard lock(osd_lock); | |
3973 | store->umount(); | |
3974 | delete store; | |
3975 | store = nullptr; | |
3976 | dout(10) << "Store synced" << dendl; | |
3977 | ||
7c673cae FG |
3978 | op_tracker.on_shutdown(); |
3979 | ||
3980 | class_handler->shutdown(); | |
3981 | client_messenger->shutdown(); | |
3982 | cluster_messenger->shutdown(); | |
3983 | hb_front_client_messenger->shutdown(); | |
3984 | hb_back_client_messenger->shutdown(); | |
3985 | objecter_messenger->shutdown(); | |
3986 | hb_front_server_messenger->shutdown(); | |
3987 | hb_back_server_messenger->shutdown(); | |
3988 | ||
7c673cae FG |
3989 | return r; |
3990 | } | |
3991 | ||
3992 | int OSD::mon_cmd_maybe_osd_create(string &cmd) | |
3993 | { | |
3994 | bool created = false; | |
3995 | while (true) { | |
3996 | dout(10) << __func__ << " cmd: " << cmd << dendl; | |
3997 | vector<string> vcmd{cmd}; | |
3998 | bufferlist inbl; | |
3999 | C_SaferCond w; | |
4000 | string outs; | |
4001 | monc->start_mon_command(vcmd, inbl, NULL, &outs, &w); | |
4002 | int r = w.wait(); | |
4003 | if (r < 0) { | |
4004 | if (r == -ENOENT && !created) { | |
4005 | string newcmd = "{\"prefix\": \"osd create\", \"id\": " + stringify(whoami) | |
4006 | + ", \"uuid\": \"" + stringify(superblock.osd_fsid) + "\"}"; | |
4007 | vector<string> vnewcmd{newcmd}; | |
4008 | bufferlist inbl; | |
4009 | C_SaferCond w; | |
4010 | string outs; | |
4011 | monc->start_mon_command(vnewcmd, inbl, NULL, &outs, &w); | |
4012 | int r = w.wait(); | |
4013 | if (r < 0) { | |
4014 | derr << __func__ << " fail: osd does not exist and created failed: " | |
4015 | << cpp_strerror(r) << dendl; | |
4016 | return r; | |
4017 | } | |
4018 | created = true; | |
4019 | continue; | |
4020 | } | |
4021 | derr << __func__ << " fail: '" << outs << "': " << cpp_strerror(r) << dendl; | |
4022 | return r; | |
4023 | } | |
4024 | break; | |
4025 | } | |
4026 | ||
4027 | return 0; | |
4028 | } | |
4029 | ||
4030 | int OSD::update_crush_location() | |
4031 | { | |
4032 | if (!cct->_conf->osd_crush_update_on_start) { | |
4033 | dout(10) << __func__ << " osd_crush_update_on_start = false" << dendl; | |
4034 | return 0; | |
4035 | } | |
4036 | ||
4037 | char weight[32]; | |
4038 | if (cct->_conf->osd_crush_initial_weight >= 0) { | |
4039 | snprintf(weight, sizeof(weight), "%.4lf", cct->_conf->osd_crush_initial_weight); | |
4040 | } else { | |
4041 | struct store_statfs_t st; | |
11fdf7f2 TL |
4042 | osd_alert_list_t alerts; |
4043 | int r = store->statfs(&st, &alerts); | |
7c673cae FG |
4044 | if (r < 0) { |
4045 | derr << "statfs: " << cpp_strerror(r) << dendl; | |
4046 | return r; | |
4047 | } | |
4048 | snprintf(weight, sizeof(weight), "%.4lf", | |
11fdf7f2 TL |
4049 | std::max(.00001, |
4050 | double(st.total) / | |
4051 | double(1ull << 40 /* TB */))); | |
7c673cae FG |
4052 | } |
4053 | ||
4054 | std::multimap<string,string> loc = cct->crush_location.get_location(); | |
4055 | dout(10) << __func__ << " crush location is " << loc << dendl; | |
4056 | ||
4057 | string cmd = | |
4058 | string("{\"prefix\": \"osd crush create-or-move\", ") + | |
4059 | string("\"id\": ") + stringify(whoami) + string(", ") + | |
4060 | string("\"weight\":") + weight + string(", ") + | |
4061 | string("\"args\": ["); | |
4062 | for (multimap<string,string>::iterator p = loc.begin(); p != loc.end(); ++p) { | |
4063 | if (p != loc.begin()) | |
4064 | cmd += ", "; | |
4065 | cmd += "\"" + p->first + "=" + p->second + "\""; | |
4066 | } | |
4067 | cmd += "]}"; | |
4068 | ||
4069 | return mon_cmd_maybe_osd_create(cmd); | |
4070 | } | |
4071 | ||
4072 | int OSD::update_crush_device_class() | |
4073 | { | |
224ce89b WB |
4074 | if (!cct->_conf->osd_class_update_on_start) { |
4075 | dout(10) << __func__ << " osd_class_update_on_start = false" << dendl; | |
4076 | return 0; | |
4077 | } | |
4078 | ||
7c673cae FG |
4079 | string device_class; |
4080 | int r = store->read_meta("crush_device_class", &device_class); | |
224ce89b WB |
4081 | if (r < 0 || device_class.empty()) { |
4082 | device_class = store->get_default_device_class(); | |
4083 | } | |
4084 | ||
4085 | if (device_class.empty()) { | |
d2e6a577 | 4086 | dout(20) << __func__ << " no device class stored locally" << dendl; |
7c673cae | 4087 | return 0; |
224ce89b | 4088 | } |
7c673cae FG |
4089 | |
4090 | string cmd = | |
4091 | string("{\"prefix\": \"osd crush set-device-class\", ") + | |
224ce89b WB |
4092 | string("\"class\": \"") + device_class + string("\", ") + |
4093 | string("\"ids\": [\"") + stringify(whoami) + string("\"]}"); | |
7c673cae | 4094 | |
224ce89b | 4095 | r = mon_cmd_maybe_osd_create(cmd); |
11fdf7f2 TL |
4096 | if (r == -EBUSY) { |
4097 | // good, already bound to a device-class | |
4098 | return 0; | |
4099 | } else { | |
4100 | return r; | |
4101 | } | |
7c673cae FG |
4102 | } |
4103 | ||
4104 | void OSD::write_superblock(ObjectStore::Transaction& t) | |
4105 | { | |
4106 | dout(10) << "write_superblock " << superblock << dendl; | |
4107 | ||
4108 | //hack: at minimum it's using the baseline feature set | |
4109 | if (!superblock.compat_features.incompat.contains(CEPH_OSD_FEATURE_INCOMPAT_BASE)) | |
4110 | superblock.compat_features.incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_BASE); | |
4111 | ||
4112 | bufferlist bl; | |
11fdf7f2 | 4113 | encode(superblock, bl); |
7c673cae FG |
4114 | t.write(coll_t::meta(), OSD_SUPERBLOCK_GOBJECT, 0, bl.length(), bl); |
4115 | } | |
4116 | ||
4117 | int OSD::read_superblock() | |
4118 | { | |
4119 | bufferlist bl; | |
11fdf7f2 | 4120 | int r = store->read(service.meta_ch, OSD_SUPERBLOCK_GOBJECT, 0, 0, bl); |
7c673cae FG |
4121 | if (r < 0) |
4122 | return r; | |
4123 | ||
11fdf7f2 TL |
4124 | auto p = bl.cbegin(); |
4125 | decode(superblock, p); | |
7c673cae FG |
4126 | |
4127 | dout(10) << "read_superblock " << superblock << dendl; | |
4128 | ||
4129 | return 0; | |
4130 | } | |
4131 | ||
4132 | void OSD::clear_temp_objects() | |
4133 | { | |
4134 | dout(10) << __func__ << dendl; | |
4135 | vector<coll_t> ls; | |
4136 | store->list_collections(ls); | |
4137 | for (vector<coll_t>::iterator p = ls.begin(); p != ls.end(); ++p) { | |
4138 | spg_t pgid; | |
4139 | if (!p->is_pg(&pgid)) | |
4140 | continue; | |
4141 | ||
4142 | // list temp objects | |
4143 | dout(20) << " clearing temps in " << *p << " pgid " << pgid << dendl; | |
4144 | ||
4145 | vector<ghobject_t> temps; | |
4146 | ghobject_t next; | |
4147 | while (1) { | |
4148 | vector<ghobject_t> objects; | |
11fdf7f2 TL |
4149 | auto ch = store->open_collection(*p); |
4150 | ceph_assert(ch); | |
4151 | store->collection_list(ch, next, ghobject_t::get_max(), | |
7c673cae FG |
4152 | store->get_ideal_list_max(), |
4153 | &objects, &next); | |
4154 | if (objects.empty()) | |
4155 | break; | |
4156 | vector<ghobject_t>::iterator q; | |
4157 | for (q = objects.begin(); q != objects.end(); ++q) { | |
4158 | // Hammer set pool for temps to -1, so check for clean-up | |
4159 | if (q->hobj.is_temp() || (q->hobj.pool == -1)) { | |
4160 | temps.push_back(*q); | |
4161 | } else { | |
4162 | break; | |
4163 | } | |
4164 | } | |
4165 | // If we saw a non-temp object and hit the break above we can | |
4166 | // break out of the while loop too. | |
4167 | if (q != objects.end()) | |
4168 | break; | |
4169 | } | |
4170 | if (!temps.empty()) { | |
4171 | ObjectStore::Transaction t; | |
4172 | int removed = 0; | |
4173 | for (vector<ghobject_t>::iterator q = temps.begin(); q != temps.end(); ++q) { | |
4174 | dout(20) << " removing " << *p << " object " << *q << dendl; | |
4175 | t.remove(*p, *q); | |
4176 | if (++removed > cct->_conf->osd_target_transaction_size) { | |
11fdf7f2 | 4177 | store->queue_transaction(service.meta_ch, std::move(t)); |
7c673cae FG |
4178 | t = ObjectStore::Transaction(); |
4179 | removed = 0; | |
4180 | } | |
4181 | } | |
4182 | if (removed) { | |
11fdf7f2 | 4183 | store->queue_transaction(service.meta_ch, std::move(t)); |
7c673cae FG |
4184 | } |
4185 | } | |
4186 | } | |
4187 | } | |
4188 | ||
4189 | void OSD::recursive_remove_collection(CephContext* cct, | |
4190 | ObjectStore *store, spg_t pgid, | |
4191 | coll_t tmp) | |
4192 | { | |
4193 | OSDriver driver( | |
4194 | store, | |
4195 | coll_t(), | |
4196 | make_snapmapper_oid()); | |
4197 | ||
11fdf7f2 | 4198 | ObjectStore::CollectionHandle ch = store->open_collection(tmp); |
7c673cae FG |
4199 | ObjectStore::Transaction t; |
4200 | SnapMapper mapper(cct, &driver, 0, 0, 0, pgid.shard); | |
4201 | ||
11fdf7f2 TL |
4202 | ghobject_t next; |
4203 | int max = cct->_conf->osd_target_transaction_size; | |
7c673cae | 4204 | vector<ghobject_t> objects; |
11fdf7f2 TL |
4205 | objects.reserve(max); |
4206 | while (true) { | |
4207 | objects.clear(); | |
4208 | store->collection_list(ch, next, ghobject_t::get_max(), | |
4209 | max, &objects, &next); | |
4210 | generic_dout(10) << __func__ << " " << objects << dendl; | |
4211 | if (objects.empty()) | |
4212 | break; | |
4213 | for (auto& p: objects) { | |
4214 | OSDriver::OSTransaction _t(driver.get_transaction(&t)); | |
4215 | int r = mapper.remove_oid(p.hobj, &_t); | |
4216 | if (r != 0 && r != -ENOENT) | |
4217 | ceph_abort(); | |
4218 | t.remove(tmp, p); | |
7c673cae | 4219 | } |
11fdf7f2 TL |
4220 | int r = store->queue_transaction(ch, std::move(t)); |
4221 | ceph_assert(r == 0); | |
4222 | t = ObjectStore::Transaction(); | |
7c673cae FG |
4223 | } |
4224 | t.remove_collection(tmp); | |
11fdf7f2 TL |
4225 | int r = store->queue_transaction(ch, std::move(t)); |
4226 | ceph_assert(r == 0); | |
7c673cae FG |
4227 | |
4228 | C_SaferCond waiter; | |
11fdf7f2 | 4229 | if (!ch->flush_commit(&waiter)) { |
7c673cae FG |
4230 | waiter.wait(); |
4231 | } | |
4232 | } | |
4233 | ||
4234 | ||
4235 | // ====================================================== | |
4236 | // PG's | |
4237 | ||
7c673cae FG |
4238 | PG* OSD::_make_pg( |
4239 | OSDMapRef createmap, | |
4240 | spg_t pgid) | |
4241 | { | |
11fdf7f2 TL |
4242 | dout(10) << __func__ << " " << pgid << dendl; |
4243 | pg_pool_t pi; | |
4244 | map<string,string> ec_profile; | |
4245 | string name; | |
4246 | if (createmap->have_pg_pool(pgid.pool())) { | |
4247 | pi = *createmap->get_pg_pool(pgid.pool()); | |
4248 | name = createmap->get_pool_name(pgid.pool()); | |
4249 | if (pi.is_erasure()) { | |
4250 | ec_profile = createmap->get_erasure_code_profile(pi.erasure_code_profile); | |
4251 | } | |
4252 | } else { | |
4253 | // pool was deleted; grab final pg_pool_t off disk. | |
4254 | ghobject_t oid = make_final_pool_info_oid(pgid.pool()); | |
4255 | bufferlist bl; | |
4256 | int r = store->read(service.meta_ch, oid, 0, 0, bl); | |
4257 | if (r < 0) { | |
4258 | derr << __func__ << " missing pool " << pgid.pool() << " tombstone" | |
4259 | << dendl; | |
4260 | return nullptr; | |
4261 | } | |
4262 | ceph_assert(r >= 0); | |
4263 | auto p = bl.cbegin(); | |
4264 | decode(pi, p); | |
4265 | decode(name, p); | |
4266 | if (p.end()) { // dev release v13.0.2 did not include ec_profile | |
4267 | derr << __func__ << " missing ec_profile from pool " << pgid.pool() | |
4268 | << " tombstone" << dendl; | |
4269 | return nullptr; | |
4270 | } | |
4271 | decode(ec_profile, p); | |
4272 | } | |
4273 | PGPool pool(cct, createmap, pgid.pool(), pi, name); | |
7c673cae | 4274 | PG *pg; |
11fdf7f2 TL |
4275 | if (pi.type == pg_pool_t::TYPE_REPLICATED || |
4276 | pi.type == pg_pool_t::TYPE_ERASURE) | |
4277 | pg = new PrimaryLogPG(&service, createmap, pool, ec_profile, pgid); | |
7c673cae FG |
4278 | else |
4279 | ceph_abort(); | |
7c673cae FG |
4280 | return pg; |
4281 | } | |
4282 | ||
11fdf7f2 | 4283 | void OSD::_get_pgs(vector<PGRef> *v, bool clear_too) |
7c673cae | 4284 | { |
11fdf7f2 TL |
4285 | v->clear(); |
4286 | v->reserve(get_num_pgs()); | |
4287 | for (auto& s : shards) { | |
4288 | std::lock_guard l(s->shard_lock); | |
4289 | for (auto& j : s->pg_slots) { | |
4290 | if (j.second->pg && | |
4291 | !j.second->pg->is_deleted()) { | |
4292 | v->push_back(j.second->pg); | |
4293 | if (clear_too) { | |
4294 | s->_detach_pg(j.second.get()); | |
4295 | } | |
4296 | } | |
7c673cae | 4297 | } |
7c673cae | 4298 | } |
7c673cae FG |
4299 | } |
4300 | ||
11fdf7f2 | 4301 | void OSD::_get_pgids(vector<spg_t> *v) |
7c673cae | 4302 | { |
11fdf7f2 TL |
4303 | v->clear(); |
4304 | v->reserve(get_num_pgs()); | |
4305 | for (auto& s : shards) { | |
4306 | std::lock_guard l(s->shard_lock); | |
4307 | for (auto& j : s->pg_slots) { | |
4308 | if (j.second->pg && | |
4309 | !j.second->pg->is_deleted()) { | |
4310 | v->push_back(j.first); | |
4311 | } | |
7c673cae FG |
4312 | } |
4313 | } | |
7c673cae FG |
4314 | } |
4315 | ||
11fdf7f2 | 4316 | void OSD::register_pg(PGRef pg) |
7c673cae | 4317 | { |
11fdf7f2 TL |
4318 | spg_t pgid = pg->get_pgid(); |
4319 | uint32_t shard_index = pgid.hash_to_shard(num_shards); | |
4320 | auto sdata = shards[shard_index]; | |
4321 | std::lock_guard l(sdata->shard_lock); | |
4322 | auto r = sdata->pg_slots.emplace(pgid, make_unique<OSDShardPGSlot>()); | |
4323 | ceph_assert(r.second); | |
4324 | auto *slot = r.first->second.get(); | |
4325 | dout(20) << __func__ << " " << pgid << " " << pg << dendl; | |
4326 | sdata->_attach_pg(slot, pg.get()); | |
4327 | } | |
7c673cae | 4328 | |
11fdf7f2 TL |
4329 | bool OSD::try_finish_pg_delete(PG *pg, unsigned old_pg_num) |
4330 | { | |
4331 | auto sdata = pg->osd_shard; | |
4332 | ceph_assert(sdata); | |
4333 | { | |
4334 | std::lock_guard l(sdata->shard_lock); | |
4335 | auto p = sdata->pg_slots.find(pg->pg_id); | |
4336 | if (p == sdata->pg_slots.end() || | |
4337 | !p->second->pg) { | |
4338 | dout(20) << __func__ << " " << pg->pg_id << " not found" << dendl; | |
4339 | return false; | |
4340 | } | |
4341 | if (p->second->waiting_for_merge_epoch) { | |
4342 | dout(20) << __func__ << " " << pg->pg_id << " waiting for merge" << dendl; | |
4343 | return false; | |
4344 | } | |
4345 | dout(20) << __func__ << " " << pg->pg_id << " " << pg << dendl; | |
4346 | sdata->_detach_pg(p->second.get()); | |
4347 | } | |
7c673cae | 4348 | |
11fdf7f2 TL |
4349 | for (auto shard : shards) { |
4350 | shard->unprime_split_children(pg->pg_id, old_pg_num); | |
4351 | } | |
7c673cae | 4352 | |
11fdf7f2 TL |
4353 | // update pg count now since we might not get an osdmap any time soon. |
4354 | if (pg->is_primary()) | |
4355 | service.logger->dec(l_osd_pg_primary); | |
4356 | else if (pg->is_replica()) | |
4357 | service.logger->dec(l_osd_pg_replica); | |
4358 | else | |
4359 | service.logger->dec(l_osd_pg_stray); | |
7c673cae | 4360 | |
11fdf7f2 | 4361 | return true; |
7c673cae FG |
4362 | } |
4363 | ||
11fdf7f2 | 4364 | PGRef OSD::_lookup_pg(spg_t pgid) |
7c673cae | 4365 | { |
11fdf7f2 TL |
4366 | uint32_t shard_index = pgid.hash_to_shard(num_shards); |
4367 | auto sdata = shards[shard_index]; | |
4368 | std::lock_guard l(sdata->shard_lock); | |
4369 | auto p = sdata->pg_slots.find(pgid); | |
4370 | if (p == sdata->pg_slots.end()) { | |
7c673cae | 4371 | return nullptr; |
11fdf7f2 TL |
4372 | } |
4373 | return p->second->pg; | |
7c673cae FG |
4374 | } |
4375 | ||
11fdf7f2 | 4376 | PGRef OSD::_lookup_lock_pg(spg_t pgid) |
31f18b77 | 4377 | { |
11fdf7f2 TL |
4378 | PGRef pg = _lookup_pg(pgid); |
4379 | if (!pg) { | |
4380 | return nullptr; | |
4381 | } | |
4382 | pg->lock(); | |
4383 | if (!pg->is_deleted()) { | |
4384 | return pg; | |
4385 | } | |
4386 | pg->unlock(); | |
4387 | return nullptr; | |
31f18b77 FG |
4388 | } |
4389 | ||
11fdf7f2 | 4390 | PGRef OSD::lookup_lock_pg(spg_t pgid) |
7c673cae | 4391 | { |
11fdf7f2 | 4392 | return _lookup_lock_pg(pgid); |
7c673cae FG |
4393 | } |
4394 | ||
4395 | void OSD::load_pgs() | |
4396 | { | |
11fdf7f2 | 4397 | ceph_assert(osd_lock.is_locked()); |
7c673cae | 4398 | dout(0) << "load_pgs" << dendl; |
11fdf7f2 | 4399 | |
7c673cae | 4400 | { |
11fdf7f2 TL |
4401 | auto pghist = make_pg_num_history_oid(); |
4402 | bufferlist bl; | |
4403 | int r = store->read(service.meta_ch, pghist, 0, 0, bl, 0); | |
4404 | if (r >= 0 && bl.length() > 0) { | |
4405 | auto p = bl.cbegin(); | |
4406 | decode(pg_num_history, p); | |
4407 | } | |
4408 | dout(20) << __func__ << " pg_num_history " << pg_num_history << dendl; | |
7c673cae FG |
4409 | } |
4410 | ||
4411 | vector<coll_t> ls; | |
4412 | int r = store->list_collections(ls); | |
4413 | if (r < 0) { | |
4414 | derr << "failed to list pgs: " << cpp_strerror(-r) << dendl; | |
4415 | } | |
4416 | ||
11fdf7f2 | 4417 | int num = 0; |
7c673cae FG |
4418 | for (vector<coll_t>::iterator it = ls.begin(); |
4419 | it != ls.end(); | |
4420 | ++it) { | |
4421 | spg_t pgid; | |
4422 | if (it->is_temp(&pgid) || | |
4423 | (it->is_pg(&pgid) && PG::_has_removal_flag(store, pgid))) { | |
11fdf7f2 TL |
4424 | dout(10) << "load_pgs " << *it |
4425 | << " removing, legacy or flagged for removal pg" << dendl; | |
7c673cae FG |
4426 | recursive_remove_collection(cct, store, pgid, *it); |
4427 | continue; | |
4428 | } | |
4429 | ||
4430 | if (!it->is_pg(&pgid)) { | |
4431 | dout(10) << "load_pgs ignoring unrecognized " << *it << dendl; | |
4432 | continue; | |
4433 | } | |
4434 | ||
7c673cae | 4435 | dout(10) << "pgid " << pgid << " coll " << coll_t(pgid) << dendl; |
7c673cae | 4436 | epoch_t map_epoch = 0; |
11fdf7f2 | 4437 | int r = PG::peek_map_epoch(store, pgid, &map_epoch); |
7c673cae FG |
4438 | if (r < 0) { |
4439 | derr << __func__ << " unable to peek at " << pgid << " metadata, skipping" | |
4440 | << dendl; | |
4441 | continue; | |
4442 | } | |
4443 | ||
11fdf7f2 | 4444 | PGRef pg; |
7c673cae FG |
4445 | if (map_epoch > 0) { |
4446 | OSDMapRef pgosdmap = service.try_get_map(map_epoch); | |
4447 | if (!pgosdmap) { | |
4448 | if (!osdmap->have_pg_pool(pgid.pool())) { | |
4449 | derr << __func__ << ": could not find map for epoch " << map_epoch | |
4450 | << " on pg " << pgid << ", but the pool is not present in the " | |
4451 | << "current map, so this is probably a result of bug 10617. " | |
4452 | << "Skipping the pg for now, you can use ceph-objectstore-tool " | |
4453 | << "to clean it up later." << dendl; | |
4454 | continue; | |
4455 | } else { | |
4456 | derr << __func__ << ": have pgid " << pgid << " at epoch " | |
4457 | << map_epoch << ", but missing map. Crashing." | |
4458 | << dendl; | |
11fdf7f2 | 4459 | ceph_abort_msg("Missing map in load_pgs"); |
7c673cae FG |
4460 | } |
4461 | } | |
11fdf7f2 | 4462 | pg = _make_pg(pgosdmap, pgid); |
7c673cae | 4463 | } else { |
11fdf7f2 | 4464 | pg = _make_pg(osdmap, pgid); |
7c673cae | 4465 | } |
11fdf7f2 TL |
4466 | if (!pg) { |
4467 | recursive_remove_collection(cct, store, pgid, *it); | |
4468 | continue; | |
4469 | } | |
4470 | ||
4471 | // there can be no waiters here, so we don't call _wake_pg_slot | |
7c673cae | 4472 | |
11fdf7f2 | 4473 | pg->lock(); |
7c673cae FG |
4474 | pg->ch = store->open_collection(pg->coll); |
4475 | ||
4476 | // read pg state, log | |
11fdf7f2 | 4477 | pg->read_state(store); |
7c673cae | 4478 | |
94b18763 FG |
4479 | if (pg->dne()) { |
4480 | dout(10) << "load_pgs " << *it << " deleting dne" << dendl; | |
4481 | pg->ch = nullptr; | |
94b18763 | 4482 | pg->unlock(); |
94b18763 FG |
4483 | recursive_remove_collection(cct, store, pgid, *it); |
4484 | continue; | |
4485 | } | |
11fdf7f2 TL |
4486 | { |
4487 | uint32_t shard_index = pgid.hash_to_shard(shards.size()); | |
4488 | assert(NULL != shards[shard_index]); | |
4489 | store->set_collection_commit_queue(pg->coll, &(shards[shard_index]->context_queue)); | |
4490 | } | |
7c673cae FG |
4491 | |
4492 | pg->reg_next_scrub(); | |
4493 | ||
11fdf7f2 | 4494 | dout(10) << __func__ << " loaded " << *pg << dendl; |
7c673cae | 4495 | pg->unlock(); |
7c673cae | 4496 | |
11fdf7f2 TL |
4497 | register_pg(pg); |
4498 | ++num; | |
7c673cae | 4499 | } |
11fdf7f2 | 4500 | dout(0) << __func__ << " opened " << num << " pgs" << dendl; |
7c673cae FG |
4501 | } |
4502 | ||
4503 | ||
11fdf7f2 TL |
4504 | PGRef OSD::handle_pg_create_info(const OSDMapRef& osdmap, |
4505 | const PGCreateInfo *info) | |
4506 | { | |
4507 | spg_t pgid = info->pgid; | |
7c673cae | 4508 | |
11fdf7f2 TL |
4509 | if (maybe_wait_for_max_pg(osdmap, pgid, info->by_mon)) { |
4510 | dout(10) << __func__ << " hit max pg, dropping" << dendl; | |
4511 | return nullptr; | |
4512 | } | |
3efd9988 | 4513 | |
11fdf7f2 | 4514 | PG::RecoveryCtx rctx = create_context(); |
7c673cae | 4515 | |
11fdf7f2 | 4516 | OSDMapRef startmap = get_map(info->epoch); |
7c673cae | 4517 | |
11fdf7f2 TL |
4518 | if (info->by_mon) { |
4519 | int64_t pool_id = pgid.pgid.pool(); | |
4520 | const pg_pool_t *pool = osdmap->get_pg_pool(pool_id); | |
4521 | if (!pool) { | |
4522 | dout(10) << __func__ << " ignoring " << pgid << ", pool dne" << dendl; | |
4523 | return nullptr; | |
4524 | } | |
4525 | if (osdmap->require_osd_release >= CEPH_RELEASE_NAUTILUS && | |
4526 | !pool->has_flag(pg_pool_t::FLAG_CREATING)) { | |
4527 | // this ensures we do not process old creating messages after the | |
4528 | // pool's initial pgs have been created (and pg are subsequently | |
4529 | // allowed to split or merge). | |
4530 | dout(20) << __func__ << " dropping " << pgid | |
4531 | << "create, pool does not have CREATING flag set" << dendl; | |
4532 | return nullptr; | |
7c673cae FG |
4533 | } |
4534 | } | |
7c673cae | 4535 | |
11fdf7f2 TL |
4536 | int up_primary, acting_primary; |
4537 | vector<int> up, acting; | |
4538 | startmap->pg_to_up_acting_osds( | |
4539 | pgid.pgid, &up, &up_primary, &acting, &acting_primary); | |
7c673cae | 4540 | |
11fdf7f2 TL |
4541 | const pg_pool_t* pp = startmap->get_pg_pool(pgid.pool()); |
4542 | if (pp->has_flag(pg_pool_t::FLAG_EC_OVERWRITES) && | |
4543 | store->get_type() != "bluestore") { | |
4544 | clog->warn() << "pg " << pgid | |
4545 | << " is at risk of silent data corruption: " | |
4546 | << "the pool allows ec overwrites but is not stored in " | |
4547 | << "bluestore, so deep scrubbing will not detect bitrot"; | |
7c673cae | 4548 | } |
11fdf7f2 TL |
4549 | PG::_create(*rctx.transaction, pgid, pgid.get_split_bits(pp->get_pg_num())); |
4550 | PG::_init(*rctx.transaction, pgid, pp); | |
7c673cae | 4551 | |
11fdf7f2 TL |
4552 | int role = startmap->calc_pg_role(whoami, acting, acting.size()); |
4553 | if (!pp->is_replicated() && role != pgid.shard) { | |
4554 | role = -1; | |
7c673cae FG |
4555 | } |
4556 | ||
11fdf7f2 TL |
4557 | PGRef pg = _make_pg(startmap, pgid); |
4558 | pg->ch = store->create_new_collection(pg->coll); | |
7c673cae | 4559 | |
11fdf7f2 TL |
4560 | { |
4561 | uint32_t shard_index = pgid.hash_to_shard(shards.size()); | |
4562 | assert(NULL != shards[shard_index]); | |
4563 | store->set_collection_commit_queue(pg->coll, &(shards[shard_index]->context_queue)); | |
7c673cae | 4564 | } |
7c673cae | 4565 | |
11fdf7f2 | 4566 | pg->lock(true); |
7c673cae | 4567 | |
11fdf7f2 TL |
4568 | // we are holding the shard lock |
4569 | ceph_assert(!pg->is_deleted()); | |
4570 | ||
4571 | pg->init( | |
4572 | role, | |
4573 | up, | |
4574 | up_primary, | |
4575 | acting, | |
4576 | acting_primary, | |
4577 | info->history, | |
4578 | info->past_intervals, | |
4579 | false, | |
4580 | rctx.transaction); | |
7c673cae | 4581 | |
11fdf7f2 TL |
4582 | if (pg->is_primary()) { |
4583 | Mutex::Locker locker(m_perf_queries_lock); | |
4584 | pg->set_dynamic_perf_stats_queries(m_perf_queries); | |
4585 | } | |
7c673cae | 4586 | |
11fdf7f2 TL |
4587 | pg->handle_initialize(&rctx); |
4588 | pg->handle_activate_map(&rctx); | |
7c673cae | 4589 | |
11fdf7f2 | 4590 | dispatch_context(rctx, pg.get(), osdmap, nullptr); |
7c673cae | 4591 | |
11fdf7f2 TL |
4592 | dout(10) << __func__ << " new pg " << *pg << dendl; |
4593 | return pg; | |
7c673cae FG |
4594 | } |
4595 | ||
11fdf7f2 TL |
4596 | bool OSD::maybe_wait_for_max_pg(const OSDMapRef& osdmap, |
4597 | spg_t pgid, | |
4598 | bool is_mon_create) | |
3efd9988 FG |
4599 | { |
4600 | const auto max_pgs_per_osd = | |
11fdf7f2 TL |
4601 | (cct->_conf.get_val<uint64_t>("mon_max_pg_per_osd") * |
4602 | cct->_conf.get_val<double>("osd_max_pg_per_osd_hard_ratio")); | |
3efd9988 | 4603 | |
11fdf7f2 | 4604 | if (num_pgs < max_pgs_per_osd) { |
3efd9988 FG |
4605 | return false; |
4606 | } | |
11fdf7f2 TL |
4607 | |
4608 | std::lock_guard l(pending_creates_lock); | |
3efd9988 FG |
4609 | if (is_mon_create) { |
4610 | pending_creates_from_mon++; | |
4611 | } else { | |
b32b8144 FG |
4612 | bool is_primary = osdmap->get_pg_acting_rank(pgid.pgid, whoami) == 0; |
4613 | pending_creates_from_osd.emplace(pgid.pgid, is_primary); | |
3efd9988 | 4614 | } |
1adf2230 | 4615 | dout(1) << __func__ << " withhold creation of pg " << pgid |
11fdf7f2 | 4616 | << ": " << num_pgs << " >= "<< max_pgs_per_osd << dendl; |
3efd9988 FG |
4617 | return true; |
4618 | } | |
4619 | ||
4620 | // to re-trigger a peering, we have to twiddle the pg mapping a little bit, | |
4621 | // see PG::should_restart_peering(). OSDMap::pg_to_up_acting_osds() will turn | |
4622 | // to up set if pg_temp is empty. so an empty pg_temp won't work. | |
4623 | static vector<int32_t> twiddle(const vector<int>& acting) { | |
4624 | if (acting.size() > 1) { | |
4625 | return {acting[0]}; | |
4626 | } else { | |
4627 | vector<int32_t> twiddled(acting.begin(), acting.end()); | |
4628 | twiddled.push_back(-1); | |
4629 | return twiddled; | |
4630 | } | |
4631 | } | |
4632 | ||
4633 | void OSD::resume_creating_pg() | |
4634 | { | |
4635 | bool do_sub_pg_creates = false; | |
b32b8144 | 4636 | bool have_pending_creates = false; |
3efd9988 FG |
4637 | { |
4638 | const auto max_pgs_per_osd = | |
11fdf7f2 TL |
4639 | (cct->_conf.get_val<uint64_t>("mon_max_pg_per_osd") * |
4640 | cct->_conf.get_val<double>("osd_max_pg_per_osd_hard_ratio")); | |
4641 | if (max_pgs_per_osd <= num_pgs) { | |
3efd9988 FG |
4642 | // this could happen if admin decreases this setting before a PG is removed |
4643 | return; | |
4644 | } | |
11fdf7f2 TL |
4645 | unsigned spare_pgs = max_pgs_per_osd - num_pgs; |
4646 | std::lock_guard l(pending_creates_lock); | |
3efd9988 | 4647 | if (pending_creates_from_mon > 0) { |
11fdf7f2 TL |
4648 | dout(20) << __func__ << " pending_creates_from_mon " |
4649 | << pending_creates_from_mon << dendl; | |
3efd9988 FG |
4650 | do_sub_pg_creates = true; |
4651 | if (pending_creates_from_mon >= spare_pgs) { | |
4652 | spare_pgs = pending_creates_from_mon = 0; | |
4653 | } else { | |
4654 | spare_pgs -= pending_creates_from_mon; | |
4655 | pending_creates_from_mon = 0; | |
4656 | } | |
4657 | } | |
4658 | auto pg = pending_creates_from_osd.cbegin(); | |
4659 | while (spare_pgs > 0 && pg != pending_creates_from_osd.cend()) { | |
94b18763 | 4660 | dout(20) << __func__ << " pg " << pg->first << dendl; |
3efd9988 | 4661 | vector<int> acting; |
b32b8144 | 4662 | osdmap->pg_to_up_acting_osds(pg->first, nullptr, nullptr, &acting, nullptr); |
94b18763 | 4663 | service.queue_want_pg_temp(pg->first, twiddle(acting), true); |
3efd9988 | 4664 | pg = pending_creates_from_osd.erase(pg); |
94b18763 | 4665 | do_sub_pg_creates = true; |
3efd9988 FG |
4666 | spare_pgs--; |
4667 | } | |
b32b8144 FG |
4668 | have_pending_creates = (pending_creates_from_mon > 0 || |
4669 | !pending_creates_from_osd.empty()); | |
3efd9988 | 4670 | } |
b32b8144 FG |
4671 | |
4672 | bool do_renew_subs = false; | |
3efd9988 FG |
4673 | if (do_sub_pg_creates) { |
4674 | if (monc->sub_want("osd_pg_creates", last_pg_create_epoch, 0)) { | |
4675 | dout(4) << __func__ << ": resolicit pg creates from mon since " | |
4676 | << last_pg_create_epoch << dendl; | |
b32b8144 | 4677 | do_renew_subs = true; |
3efd9988 FG |
4678 | } |
4679 | } | |
b32b8144 FG |
4680 | version_t start = osdmap->get_epoch() + 1; |
4681 | if (have_pending_creates) { | |
4682 | // don't miss any new osdmap deleting PGs | |
4683 | if (monc->sub_want("osdmap", start, 0)) { | |
4684 | dout(4) << __func__ << ": resolicit osdmap from mon since " | |
4685 | << start << dendl; | |
4686 | do_renew_subs = true; | |
4687 | } | |
94b18763 | 4688 | } else if (do_sub_pg_creates) { |
b32b8144 FG |
4689 | // no need to subscribe the osdmap continuously anymore |
4690 | // once the pgtemp and/or mon_subscribe(pg_creates) is sent | |
4691 | if (monc->sub_want_increment("osdmap", start, CEPH_SUBSCRIBE_ONETIME)) { | |
11fdf7f2 | 4692 | dout(4) << __func__ << ": re-subscribe osdmap(onetime) since " |
b32b8144 FG |
4693 | << start << dendl; |
4694 | do_renew_subs = true; | |
4695 | } | |
4696 | } | |
4697 | ||
4698 | if (do_renew_subs) { | |
4699 | monc->renew_subs(); | |
4700 | } | |
4701 | ||
94b18763 | 4702 | service.send_pg_temp(); |
3efd9988 | 4703 | } |
7c673cae FG |
4704 | |
4705 | void OSD::build_initial_pg_history( | |
4706 | spg_t pgid, | |
4707 | epoch_t created, | |
4708 | utime_t created_stamp, | |
4709 | pg_history_t *h, | |
4710 | PastIntervals *pi) | |
4711 | { | |
4712 | dout(10) << __func__ << " " << pgid << " created " << created << dendl; | |
4713 | h->epoch_created = created; | |
31f18b77 | 4714 | h->epoch_pool_created = created; |
7c673cae FG |
4715 | h->same_interval_since = created; |
4716 | h->same_up_since = created; | |
4717 | h->same_primary_since = created; | |
4718 | h->last_scrub_stamp = created_stamp; | |
4719 | h->last_deep_scrub_stamp = created_stamp; | |
4720 | h->last_clean_scrub_stamp = created_stamp; | |
4721 | ||
4722 | OSDMapRef lastmap = service.get_map(created); | |
4723 | int up_primary, acting_primary; | |
4724 | vector<int> up, acting; | |
4725 | lastmap->pg_to_up_acting_osds( | |
4726 | pgid.pgid, &up, &up_primary, &acting, &acting_primary); | |
4727 | ||
4728 | ostringstream debug; | |
4729 | for (epoch_t e = created + 1; e <= osdmap->get_epoch(); ++e) { | |
4730 | OSDMapRef osdmap = service.get_map(e); | |
4731 | int new_up_primary, new_acting_primary; | |
4732 | vector<int> new_up, new_acting; | |
4733 | osdmap->pg_to_up_acting_osds( | |
4734 | pgid.pgid, &new_up, &new_up_primary, &new_acting, &new_acting_primary); | |
4735 | ||
4736 | // this is a bit imprecise, but sufficient? | |
4737 | struct min_size_predicate_t : public IsPGRecoverablePredicate { | |
4738 | const pg_pool_t *pi; | |
4739 | bool operator()(const set<pg_shard_t> &have) const { | |
4740 | return have.size() >= pi->min_size; | |
4741 | } | |
11fdf7f2 | 4742 | explicit min_size_predicate_t(const pg_pool_t *i) : pi(i) {} |
7c673cae FG |
4743 | } min_size_predicate(osdmap->get_pg_pool(pgid.pgid.pool())); |
4744 | ||
4745 | bool new_interval = PastIntervals::check_new_interval( | |
4746 | acting_primary, | |
4747 | new_acting_primary, | |
4748 | acting, new_acting, | |
4749 | up_primary, | |
4750 | new_up_primary, | |
4751 | up, new_up, | |
4752 | h->same_interval_since, | |
4753 | h->last_epoch_clean, | |
4754 | osdmap, | |
4755 | lastmap, | |
4756 | pgid.pgid, | |
4757 | &min_size_predicate, | |
4758 | pi, | |
4759 | &debug); | |
4760 | if (new_interval) { | |
4761 | h->same_interval_since = e; | |
181888fb FG |
4762 | if (up != new_up) { |
4763 | h->same_up_since = e; | |
4764 | } | |
4765 | if (acting_primary != new_acting_primary) { | |
4766 | h->same_primary_since = e; | |
4767 | } | |
4768 | if (pgid.pgid.is_split(lastmap->get_pg_num(pgid.pgid.pool()), | |
4769 | osdmap->get_pg_num(pgid.pgid.pool()), | |
4770 | nullptr)) { | |
4771 | h->last_epoch_split = e; | |
4772 | } | |
4773 | up = new_up; | |
4774 | acting = new_acting; | |
4775 | up_primary = new_up_primary; | |
4776 | acting_primary = new_acting_primary; | |
c07f9fc5 | 4777 | } |
7c673cae FG |
4778 | lastmap = osdmap; |
4779 | } | |
4780 | dout(20) << __func__ << " " << debug.str() << dendl; | |
4781 | dout(10) << __func__ << " " << *h << " " << *pi | |
4782 | << " [" << (pi->empty() ? pair<epoch_t,epoch_t>(0,0) : | |
4783 | pi->get_bounds()) << ")" | |
4784 | << dendl; | |
4785 | } | |
4786 | ||
7c673cae FG |
4787 | void OSD::_add_heartbeat_peer(int p) |
4788 | { | |
4789 | if (p == whoami) | |
4790 | return; | |
4791 | HeartbeatInfo *hi; | |
4792 | ||
4793 | map<int,HeartbeatInfo>::iterator i = heartbeat_peers.find(p); | |
4794 | if (i == heartbeat_peers.end()) { | |
4795 | pair<ConnectionRef,ConnectionRef> cons = service.get_con_osd_hb(p, osdmap->get_epoch()); | |
4796 | if (!cons.first) | |
4797 | return; | |
4798 | hi = &heartbeat_peers[p]; | |
4799 | hi->peer = p; | |
11fdf7f2 | 4800 | RefCountedPtr s{new HeartbeatSession{p}, false}; |
7c673cae | 4801 | hi->con_back = cons.first.get(); |
11fdf7f2 | 4802 | hi->con_back->set_priv(s); |
7c673cae FG |
4803 | if (cons.second) { |
4804 | hi->con_front = cons.second.get(); | |
11fdf7f2 | 4805 | hi->con_front->set_priv(s); |
7c673cae FG |
4806 | dout(10) << "_add_heartbeat_peer: new peer osd." << p |
4807 | << " " << hi->con_back->get_peer_addr() | |
4808 | << " " << hi->con_front->get_peer_addr() | |
4809 | << dendl; | |
4810 | } else { | |
4811 | hi->con_front.reset(NULL); | |
4812 | dout(10) << "_add_heartbeat_peer: new peer osd." << p | |
4813 | << " " << hi->con_back->get_peer_addr() | |
4814 | << dendl; | |
4815 | } | |
7c673cae FG |
4816 | } else { |
4817 | hi = &i->second; | |
4818 | } | |
4819 | hi->epoch = osdmap->get_epoch(); | |
4820 | } | |
4821 | ||
4822 | void OSD::_remove_heartbeat_peer(int n) | |
4823 | { | |
4824 | map<int,HeartbeatInfo>::iterator q = heartbeat_peers.find(n); | |
11fdf7f2 | 4825 | ceph_assert(q != heartbeat_peers.end()); |
7c673cae FG |
4826 | dout(20) << " removing heartbeat peer osd." << n |
4827 | << " " << q->second.con_back->get_peer_addr() | |
4828 | << " " << (q->second.con_front ? q->second.con_front->get_peer_addr() : entity_addr_t()) | |
4829 | << dendl; | |
4830 | q->second.con_back->mark_down(); | |
4831 | if (q->second.con_front) { | |
4832 | q->second.con_front->mark_down(); | |
4833 | } | |
4834 | heartbeat_peers.erase(q); | |
4835 | } | |
4836 | ||
4837 | void OSD::need_heartbeat_peer_update() | |
4838 | { | |
4839 | if (is_stopping()) | |
4840 | return; | |
4841 | dout(20) << "need_heartbeat_peer_update" << dendl; | |
4842 | heartbeat_set_peers_need_update(); | |
4843 | } | |
4844 | ||
4845 | void OSD::maybe_update_heartbeat_peers() | |
4846 | { | |
11fdf7f2 | 4847 | ceph_assert(osd_lock.is_locked()); |
7c673cae | 4848 | |
11fdf7f2 | 4849 | if (is_waiting_for_healthy() || is_active()) { |
7c673cae FG |
4850 | utime_t now = ceph_clock_now(); |
4851 | if (last_heartbeat_resample == utime_t()) { | |
4852 | last_heartbeat_resample = now; | |
4853 | heartbeat_set_peers_need_update(); | |
4854 | } else if (!heartbeat_peers_need_update()) { | |
4855 | utime_t dur = now - last_heartbeat_resample; | |
4856 | if (dur > cct->_conf->osd_heartbeat_grace) { | |
4857 | dout(10) << "maybe_update_heartbeat_peers forcing update after " << dur << " seconds" << dendl; | |
4858 | heartbeat_set_peers_need_update(); | |
4859 | last_heartbeat_resample = now; | |
11fdf7f2 TL |
4860 | if (is_waiting_for_healthy()) { |
4861 | reset_heartbeat_peers(); // we want *new* peers! | |
4862 | } | |
7c673cae FG |
4863 | } |
4864 | } | |
4865 | } | |
4866 | ||
4867 | if (!heartbeat_peers_need_update()) | |
4868 | return; | |
4869 | heartbeat_clear_peers_need_update(); | |
4870 | ||
11fdf7f2 | 4871 | std::lock_guard l(heartbeat_lock); |
7c673cae FG |
4872 | |
4873 | dout(10) << "maybe_update_heartbeat_peers updating" << dendl; | |
4874 | ||
4875 | ||
4876 | // build heartbeat from set | |
4877 | if (is_active()) { | |
11fdf7f2 TL |
4878 | vector<PGRef> pgs; |
4879 | _get_pgs(&pgs); | |
4880 | for (auto& pg : pgs) { | |
4881 | pg->with_heartbeat_peers([&](int peer) { | |
4882 | if (osdmap->is_up(peer)) { | |
4883 | _add_heartbeat_peer(peer); | |
4884 | } | |
4885 | }); | |
7c673cae FG |
4886 | } |
4887 | } | |
4888 | ||
4889 | // include next and previous up osds to ensure we have a fully-connected set | |
4890 | set<int> want, extras; | |
11fdf7f2 | 4891 | const int next = osdmap->get_next_up_osd_after(whoami); |
7c673cae FG |
4892 | if (next >= 0) |
4893 | want.insert(next); | |
4894 | int prev = osdmap->get_previous_up_osd_before(whoami); | |
4895 | if (prev >= 0 && prev != next) | |
4896 | want.insert(prev); | |
4897 | ||
11fdf7f2 TL |
4898 | // make sure we have at least **min_down** osds coming from different |
4899 | // subtree level (e.g., hosts) for fast failure detection. | |
4900 | auto min_down = cct->_conf.get_val<uint64_t>("mon_osd_min_down_reporters"); | |
4901 | auto subtree = cct->_conf.get_val<string>("mon_osd_reporter_subtree_level"); | |
4902 | osdmap->get_random_up_osds_by_subtree( | |
4903 | whoami, subtree, min_down, want, &want); | |
4904 | ||
7c673cae FG |
4905 | for (set<int>::iterator p = want.begin(); p != want.end(); ++p) { |
4906 | dout(10) << " adding neighbor peer osd." << *p << dendl; | |
4907 | extras.insert(*p); | |
4908 | _add_heartbeat_peer(*p); | |
4909 | } | |
4910 | ||
4911 | // remove down peers; enumerate extras | |
4912 | map<int,HeartbeatInfo>::iterator p = heartbeat_peers.begin(); | |
4913 | while (p != heartbeat_peers.end()) { | |
4914 | if (!osdmap->is_up(p->first)) { | |
4915 | int o = p->first; | |
4916 | ++p; | |
4917 | _remove_heartbeat_peer(o); | |
4918 | continue; | |
4919 | } | |
4920 | if (p->second.epoch < osdmap->get_epoch()) { | |
4921 | extras.insert(p->first); | |
4922 | } | |
4923 | ++p; | |
4924 | } | |
4925 | ||
4926 | // too few? | |
11fdf7f2 | 4927 | for (int n = next; n >= 0; ) { |
7c673cae FG |
4928 | if ((int)heartbeat_peers.size() >= cct->_conf->osd_heartbeat_min_peers) |
4929 | break; | |
4930 | if (!extras.count(n) && !want.count(n) && n != whoami) { | |
4931 | dout(10) << " adding random peer osd." << n << dendl; | |
4932 | extras.insert(n); | |
4933 | _add_heartbeat_peer(n); | |
4934 | } | |
4935 | n = osdmap->get_next_up_osd_after(n); | |
11fdf7f2 | 4936 | if (n == next) |
7c673cae FG |
4937 | break; // came full circle; stop |
4938 | } | |
4939 | ||
4940 | // too many? | |
4941 | for (set<int>::iterator p = extras.begin(); | |
4942 | (int)heartbeat_peers.size() > cct->_conf->osd_heartbeat_min_peers && p != extras.end(); | |
4943 | ++p) { | |
4944 | if (want.count(*p)) | |
4945 | continue; | |
4946 | _remove_heartbeat_peer(*p); | |
4947 | } | |
4948 | ||
4949 | dout(10) << "maybe_update_heartbeat_peers " << heartbeat_peers.size() << " peers, extras " << extras << dendl; | |
4950 | } | |
4951 | ||
4952 | void OSD::reset_heartbeat_peers() | |
4953 | { | |
11fdf7f2 | 4954 | ceph_assert(osd_lock.is_locked()); |
7c673cae | 4955 | dout(10) << "reset_heartbeat_peers" << dendl; |
11fdf7f2 | 4956 | std::lock_guard l(heartbeat_lock); |
7c673cae FG |
4957 | while (!heartbeat_peers.empty()) { |
4958 | HeartbeatInfo& hi = heartbeat_peers.begin()->second; | |
4959 | hi.con_back->mark_down(); | |
4960 | if (hi.con_front) { | |
4961 | hi.con_front->mark_down(); | |
4962 | } | |
4963 | heartbeat_peers.erase(heartbeat_peers.begin()); | |
4964 | } | |
4965 | failure_queue.clear(); | |
4966 | } | |
4967 | ||
4968 | void OSD::handle_osd_ping(MOSDPing *m) | |
4969 | { | |
4970 | if (superblock.cluster_fsid != m->fsid) { | |
4971 | dout(20) << "handle_osd_ping from " << m->get_source_inst() | |
4972 | << " bad fsid " << m->fsid << " != " << superblock.cluster_fsid << dendl; | |
4973 | m->put(); | |
4974 | return; | |
4975 | } | |
4976 | ||
4977 | int from = m->get_source().num(); | |
4978 | ||
4979 | heartbeat_lock.Lock(); | |
4980 | if (is_stopping()) { | |
4981 | heartbeat_lock.Unlock(); | |
4982 | m->put(); | |
4983 | return; | |
4984 | } | |
4985 | ||
4986 | OSDMapRef curmap = service.get_osdmap(); | |
c07f9fc5 FG |
4987 | if (!curmap) { |
4988 | heartbeat_lock.Unlock(); | |
4989 | m->put(); | |
4990 | return; | |
4991 | } | |
7c673cae FG |
4992 | |
4993 | switch (m->op) { | |
4994 | ||
4995 | case MOSDPing::PING: | |
4996 | { | |
4997 | if (cct->_conf->osd_debug_drop_ping_probability > 0) { | |
4998 | auto heartbeat_drop = debug_heartbeat_drops_remaining.find(from); | |
4999 | if (heartbeat_drop != debug_heartbeat_drops_remaining.end()) { | |
5000 | if (heartbeat_drop->second == 0) { | |
5001 | debug_heartbeat_drops_remaining.erase(heartbeat_drop); | |
5002 | } else { | |
5003 | --heartbeat_drop->second; | |
5004 | dout(5) << "Dropping heartbeat from " << from | |
5005 | << ", " << heartbeat_drop->second | |
5006 | << " remaining to drop" << dendl; | |
5007 | break; | |
5008 | } | |
5009 | } else if (cct->_conf->osd_debug_drop_ping_probability > | |
5010 | ((((double)(rand()%100))/100.0))) { | |
5011 | heartbeat_drop = | |
5012 | debug_heartbeat_drops_remaining.insert(std::make_pair(from, | |
5013 | cct->_conf->osd_debug_drop_ping_duration)).first; | |
5014 | dout(5) << "Dropping heartbeat from " << from | |
5015 | << ", " << heartbeat_drop->second | |
5016 | << " remaining to drop" << dendl; | |
5017 | break; | |
5018 | } | |
5019 | } | |
5020 | ||
5021 | if (!cct->get_heartbeat_map()->is_healthy()) { | |
5022 | dout(10) << "internal heartbeat not healthy, dropping ping request" << dendl; | |
5023 | break; | |
5024 | } | |
5025 | ||
5026 | Message *r = new MOSDPing(monc->get_fsid(), | |
5027 | curmap->get_epoch(), | |
31f18b77 FG |
5028 | MOSDPing::PING_REPLY, m->stamp, |
5029 | cct->_conf->osd_heartbeat_min_size); | |
7c673cae FG |
5030 | m->get_connection()->send_message(r); |
5031 | ||
5032 | if (curmap->is_up(from)) { | |
5033 | service.note_peer_epoch(from, m->map_epoch); | |
5034 | if (is_active()) { | |
5035 | ConnectionRef con = service.get_con_osd_cluster(from, curmap->get_epoch()); | |
5036 | if (con) { | |
5037 | service.share_map_peer(from, con.get()); | |
5038 | } | |
5039 | } | |
5040 | } else if (!curmap->exists(from) || | |
5041 | curmap->get_down_at(from) > m->map_epoch) { | |
5042 | // tell them they have died | |
5043 | Message *r = new MOSDPing(monc->get_fsid(), | |
5044 | curmap->get_epoch(), | |
5045 | MOSDPing::YOU_DIED, | |
31f18b77 FG |
5046 | m->stamp, |
5047 | cct->_conf->osd_heartbeat_min_size); | |
7c673cae FG |
5048 | m->get_connection()->send_message(r); |
5049 | } | |
5050 | } | |
5051 | break; | |
5052 | ||
5053 | case MOSDPing::PING_REPLY: | |
5054 | { | |
5055 | map<int,HeartbeatInfo>::iterator i = heartbeat_peers.find(from); | |
5056 | if (i != heartbeat_peers.end()) { | |
11fdf7f2 TL |
5057 | auto acked = i->second.ping_history.find(m->stamp); |
5058 | if (acked != i->second.ping_history.end()) { | |
5059 | utime_t now = ceph_clock_now(); | |
5060 | int &unacknowledged = acked->second.second; | |
5061 | if (m->get_connection() == i->second.con_back) { | |
5062 | dout(25) << "handle_osd_ping got reply from osd." << from | |
5063 | << " first_tx " << i->second.first_tx | |
5064 | << " last_tx " << i->second.last_tx | |
5065 | << " last_rx_back " << i->second.last_rx_back << " -> " << now | |
5066 | << " last_rx_front " << i->second.last_rx_front | |
5067 | << dendl; | |
5068 | i->second.last_rx_back = now; | |
5069 | ceph_assert(unacknowledged > 0); | |
5070 | --unacknowledged; | |
5071 | // if there is no front con, set both stamps. | |
5072 | if (i->second.con_front == NULL) { | |
5073 | i->second.last_rx_front = now; | |
5074 | ceph_assert(unacknowledged > 0); | |
5075 | --unacknowledged; | |
5076 | } | |
5077 | } else if (m->get_connection() == i->second.con_front) { | |
5078 | dout(25) << "handle_osd_ping got reply from osd." << from | |
5079 | << " first_tx " << i->second.first_tx | |
5080 | << " last_tx " << i->second.last_tx | |
5081 | << " last_rx_back " << i->second.last_rx_back | |
5082 | << " last_rx_front " << i->second.last_rx_front << " -> " << now | |
5083 | << dendl; | |
5084 | i->second.last_rx_front = now; | |
5085 | ceph_assert(unacknowledged > 0); | |
5086 | --unacknowledged; | |
5087 | } | |
7c673cae | 5088 | |
11fdf7f2 TL |
5089 | if (unacknowledged == 0) { |
5090 | // succeeded in getting all replies | |
5091 | dout(25) << "handle_osd_ping got all replies from osd." << from | |
5092 | << " , erase pending ping(sent at " << m->stamp << ")" | |
5093 | << " and older pending ping(s)" | |
5094 | << dendl; | |
5095 | i->second.ping_history.erase(i->second.ping_history.begin(), ++acked); | |
7c673cae FG |
5096 | } |
5097 | ||
11fdf7f2 TL |
5098 | if (i->second.is_healthy(now)) { |
5099 | // Cancel false reports | |
5100 | auto failure_queue_entry = failure_queue.find(from); | |
5101 | if (failure_queue_entry != failure_queue.end()) { | |
5102 | dout(10) << "handle_osd_ping canceling queued " | |
5103 | << "failure report for osd." << from << dendl; | |
5104 | failure_queue.erase(failure_queue_entry); | |
5105 | } | |
5106 | ||
5107 | auto failure_pending_entry = failure_pending.find(from); | |
5108 | if (failure_pending_entry != failure_pending.end()) { | |
5109 | dout(10) << "handle_osd_ping canceling in-flight " | |
5110 | << "failure report for osd." << from << dendl; | |
5111 | send_still_alive(curmap->get_epoch(), | |
5112 | from, | |
5113 | failure_pending_entry->second.second); | |
5114 | failure_pending.erase(failure_pending_entry); | |
5115 | } | |
7c673cae | 5116 | } |
11fdf7f2 TL |
5117 | } else { |
5118 | // old replies, deprecated by newly sent pings. | |
5119 | dout(10) << "handle_osd_ping no pending ping(sent at " << m->stamp | |
5120 | << ") is found, treat as covered by newly sent pings " | |
5121 | << "and ignore" | |
5122 | << dendl; | |
7c673cae FG |
5123 | } |
5124 | } | |
5125 | ||
5126 | if (m->map_epoch && | |
5127 | curmap->is_up(from)) { | |
5128 | service.note_peer_epoch(from, m->map_epoch); | |
5129 | if (is_active()) { | |
5130 | ConnectionRef con = service.get_con_osd_cluster(from, curmap->get_epoch()); | |
5131 | if (con) { | |
5132 | service.share_map_peer(from, con.get()); | |
5133 | } | |
5134 | } | |
5135 | } | |
5136 | } | |
5137 | break; | |
5138 | ||
5139 | case MOSDPing::YOU_DIED: | |
5140 | dout(10) << "handle_osd_ping " << m->get_source_inst() | |
5141 | << " says i am down in " << m->map_epoch << dendl; | |
5142 | osdmap_subscribe(curmap->get_epoch()+1, false); | |
5143 | break; | |
5144 | } | |
5145 | ||
5146 | heartbeat_lock.Unlock(); | |
5147 | m->put(); | |
5148 | } | |
5149 | ||
5150 | void OSD::heartbeat_entry() | |
5151 | { | |
11fdf7f2 | 5152 | std::lock_guard l(heartbeat_lock); |
7c673cae FG |
5153 | if (is_stopping()) |
5154 | return; | |
5155 | while (!heartbeat_stop) { | |
5156 | heartbeat(); | |
5157 | ||
5158 | double wait = .5 + ((float)(rand() % 10)/10.0) * (float)cct->_conf->osd_heartbeat_interval; | |
5159 | utime_t w; | |
5160 | w.set_from_double(wait); | |
5161 | dout(30) << "heartbeat_entry sleeping for " << wait << dendl; | |
5162 | heartbeat_cond.WaitInterval(heartbeat_lock, w); | |
5163 | if (is_stopping()) | |
5164 | return; | |
5165 | dout(30) << "heartbeat_entry woke up" << dendl; | |
5166 | } | |
5167 | } | |
5168 | ||
5169 | void OSD::heartbeat_check() | |
5170 | { | |
11fdf7f2 | 5171 | ceph_assert(heartbeat_lock.is_locked()); |
7c673cae FG |
5172 | utime_t now = ceph_clock_now(); |
5173 | ||
11fdf7f2 | 5174 | // check for incoming heartbeats (move me elsewhere?) |
7c673cae FG |
5175 | for (map<int,HeartbeatInfo>::iterator p = heartbeat_peers.begin(); |
5176 | p != heartbeat_peers.end(); | |
5177 | ++p) { | |
5178 | ||
5179 | if (p->second.first_tx == utime_t()) { | |
5180 | dout(25) << "heartbeat_check we haven't sent ping to osd." << p->first | |
11fdf7f2 | 5181 | << " yet, skipping" << dendl; |
7c673cae FG |
5182 | continue; |
5183 | } | |
5184 | ||
5185 | dout(25) << "heartbeat_check osd." << p->first | |
5186 | << " first_tx " << p->second.first_tx | |
5187 | << " last_tx " << p->second.last_tx | |
5188 | << " last_rx_back " << p->second.last_rx_back | |
5189 | << " last_rx_front " << p->second.last_rx_front | |
5190 | << dendl; | |
11fdf7f2 TL |
5191 | if (p->second.is_unhealthy(now)) { |
5192 | utime_t oldest_deadline = p->second.ping_history.begin()->second.first; | |
7c673cae FG |
5193 | if (p->second.last_rx_back == utime_t() || |
5194 | p->second.last_rx_front == utime_t()) { | |
11fdf7f2 TL |
5195 | derr << "heartbeat_check: no reply from " |
5196 | << p->second.con_front->get_peer_addr().get_sockaddr() | |
5197 | << " osd." << p->first | |
5198 | << " ever on either front or back, first ping sent " | |
5199 | << p->second.first_tx | |
5200 | << " (oldest deadline " << oldest_deadline << ")" | |
5201 | << dendl; | |
7c673cae | 5202 | // fail |
11fdf7f2 | 5203 | failure_queue[p->first] = p->second.first_tx; |
7c673cae | 5204 | } else { |
11fdf7f2 TL |
5205 | derr << "heartbeat_check: no reply from " |
5206 | << p->second.con_front->get_peer_addr().get_sockaddr() | |
7c673cae FG |
5207 | << " osd." << p->first << " since back " << p->second.last_rx_back |
5208 | << " front " << p->second.last_rx_front | |
11fdf7f2 TL |
5209 | << " (oldest deadline " << oldest_deadline << ")" |
5210 | << dendl; | |
7c673cae | 5211 | // fail |
11fdf7f2 | 5212 | failure_queue[p->first] = std::min(p->second.last_rx_back, p->second.last_rx_front); |
7c673cae FG |
5213 | } |
5214 | } | |
5215 | } | |
5216 | } | |
5217 | ||
5218 | void OSD::heartbeat() | |
5219 | { | |
5220 | dout(30) << "heartbeat" << dendl; | |
5221 | ||
5222 | // get CPU load avg | |
5223 | double loadavgs[1]; | |
11fdf7f2 TL |
5224 | int hb_interval = cct->_conf->osd_heartbeat_interval; |
5225 | int n_samples = 86400; | |
5226 | if (hb_interval > 1) { | |
5227 | n_samples /= hb_interval; | |
5228 | if (n_samples < 1) | |
5229 | n_samples = 1; | |
5230 | } | |
5231 | ||
7c673cae FG |
5232 | if (getloadavg(loadavgs, 1) == 1) { |
5233 | logger->set(l_osd_loadavg, 100 * loadavgs[0]); | |
5234 | daily_loadavg = (daily_loadavg * (n_samples - 1) + loadavgs[0]) / n_samples; | |
5235 | dout(30) << "heartbeat: daily_loadavg " << daily_loadavg << dendl; | |
5236 | } | |
5237 | ||
5238 | dout(30) << "heartbeat checking stats" << dendl; | |
5239 | ||
11fdf7f2 | 5240 | // refresh peer list and osd stats |
7c673cae FG |
5241 | vector<int> hb_peers; |
5242 | for (map<int,HeartbeatInfo>::iterator p = heartbeat_peers.begin(); | |
5243 | p != heartbeat_peers.end(); | |
5244 | ++p) | |
5245 | hb_peers.push_back(p->first); | |
7c673cae | 5246 | |
11fdf7f2 TL |
5247 | auto new_stat = service.set_osd_stat(hb_peers, get_num_pgs()); |
5248 | dout(5) << __func__ << " " << new_stat << dendl; | |
5249 | ceph_assert(new_stat.statfs.total); | |
5250 | ||
5251 | float pratio; | |
5252 | float ratio = service.compute_adjusted_ratio(new_stat, &pratio); | |
5253 | ||
5254 | service.check_full_status(ratio, pratio); | |
7c673cae FG |
5255 | |
5256 | utime_t now = ceph_clock_now(); | |
11fdf7f2 TL |
5257 | utime_t deadline = now; |
5258 | deadline += cct->_conf->osd_heartbeat_grace; | |
7c673cae FG |
5259 | |
5260 | // send heartbeats | |
5261 | for (map<int,HeartbeatInfo>::iterator i = heartbeat_peers.begin(); | |
5262 | i != heartbeat_peers.end(); | |
5263 | ++i) { | |
5264 | int peer = i->first; | |
5265 | i->second.last_tx = now; | |
5266 | if (i->second.first_tx == utime_t()) | |
5267 | i->second.first_tx = now; | |
11fdf7f2 TL |
5268 | i->second.ping_history[now] = make_pair(deadline, |
5269 | HeartbeatInfo::HEARTBEAT_MAX_CONN); | |
7c673cae FG |
5270 | dout(30) << "heartbeat sending ping to osd." << peer << dendl; |
5271 | i->second.con_back->send_message(new MOSDPing(monc->get_fsid(), | |
11fdf7f2 | 5272 | service.get_osdmap_epoch(), |
31f18b77 FG |
5273 | MOSDPing::PING, now, |
5274 | cct->_conf->osd_heartbeat_min_size)); | |
7c673cae FG |
5275 | |
5276 | if (i->second.con_front) | |
5277 | i->second.con_front->send_message(new MOSDPing(monc->get_fsid(), | |
11fdf7f2 | 5278 | service.get_osdmap_epoch(), |
31f18b77 FG |
5279 | MOSDPing::PING, now, |
5280 | cct->_conf->osd_heartbeat_min_size)); | |
7c673cae FG |
5281 | } |
5282 | ||
5283 | logger->set(l_osd_hb_to, heartbeat_peers.size()); | |
5284 | ||
5285 | // hmm.. am i all alone? | |
5286 | dout(30) << "heartbeat lonely?" << dendl; | |
5287 | if (heartbeat_peers.empty()) { | |
5288 | if (now - last_mon_heartbeat > cct->_conf->osd_mon_heartbeat_interval && is_active()) { | |
5289 | last_mon_heartbeat = now; | |
5290 | dout(10) << "i have no heartbeat peers; checking mon for new map" << dendl; | |
5291 | osdmap_subscribe(osdmap->get_epoch() + 1, false); | |
5292 | } | |
5293 | } | |
5294 | ||
5295 | dout(30) << "heartbeat done" << dendl; | |
5296 | } | |
5297 | ||
5298 | bool OSD::heartbeat_reset(Connection *con) | |
5299 | { | |
11fdf7f2 TL |
5300 | std::lock_guard l(heartbeat_lock); |
5301 | auto s = con->get_priv(); | |
5302 | con->set_priv(nullptr); | |
7c673cae | 5303 | if (s) { |
7c673cae | 5304 | if (is_stopping()) { |
7c673cae FG |
5305 | return true; |
5306 | } | |
11fdf7f2 TL |
5307 | auto heartbeat_session = static_cast<HeartbeatSession*>(s.get()); |
5308 | auto p = heartbeat_peers.find(heartbeat_session->peer); | |
7c673cae FG |
5309 | if (p != heartbeat_peers.end() && |
5310 | (p->second.con_back == con || | |
5311 | p->second.con_front == con)) { | |
5312 | dout(10) << "heartbeat_reset failed hb con " << con << " for osd." << p->second.peer | |
5313 | << ", reopening" << dendl; | |
5314 | if (con != p->second.con_back) { | |
5315 | p->second.con_back->mark_down(); | |
5316 | } | |
5317 | p->second.con_back.reset(NULL); | |
5318 | if (p->second.con_front && con != p->second.con_front) { | |
5319 | p->second.con_front->mark_down(); | |
5320 | } | |
5321 | p->second.con_front.reset(NULL); | |
5322 | pair<ConnectionRef,ConnectionRef> newcon = service.get_con_osd_hb(p->second.peer, p->second.epoch); | |
5323 | if (newcon.first) { | |
5324 | p->second.con_back = newcon.first.get(); | |
11fdf7f2 | 5325 | p->second.con_back->set_priv(s); |
7c673cae FG |
5326 | if (newcon.second) { |
5327 | p->second.con_front = newcon.second.get(); | |
11fdf7f2 | 5328 | p->second.con_front->set_priv(s); |
7c673cae | 5329 | } |
11fdf7f2 | 5330 | p->second.ping_history.clear(); |
7c673cae FG |
5331 | } else { |
5332 | dout(10) << "heartbeat_reset failed hb con " << con << " for osd." << p->second.peer | |
5333 | << ", raced with osdmap update, closing out peer" << dendl; | |
5334 | heartbeat_peers.erase(p); | |
5335 | } | |
5336 | } else { | |
5337 | dout(10) << "heartbeat_reset closing (old) failed hb con " << con << dendl; | |
5338 | } | |
7c673cae FG |
5339 | } |
5340 | return true; | |
5341 | } | |
5342 | ||
5343 | ||
5344 | ||
5345 | // ========================================= | |
5346 | ||
5347 | void OSD::tick() | |
5348 | { | |
11fdf7f2 | 5349 | ceph_assert(osd_lock.is_locked()); |
7c673cae FG |
5350 | dout(10) << "tick" << dendl; |
5351 | ||
5352 | if (is_active() || is_waiting_for_healthy()) { | |
5353 | maybe_update_heartbeat_peers(); | |
5354 | } | |
5355 | ||
5356 | if (is_waiting_for_healthy()) { | |
5357 | start_boot(); | |
11fdf7f2 TL |
5358 | if (is_waiting_for_healthy()) { |
5359 | // failed to boot | |
5360 | std::lock_guard l(heartbeat_lock); | |
5361 | utime_t now = ceph_clock_now(); | |
5362 | if (now - last_mon_heartbeat > cct->_conf->osd_mon_heartbeat_interval) { | |
5363 | last_mon_heartbeat = now; | |
5364 | dout(1) << __func__ << " checking mon for new map" << dendl; | |
5365 | osdmap_subscribe(osdmap->get_epoch() + 1, false); | |
5366 | } | |
5367 | } | |
7c673cae FG |
5368 | } |
5369 | ||
5370 | do_waiters(); | |
5371 | ||
91327a77 | 5372 | tick_timer.add_event_after(get_tick_interval(), new C_Tick(this)); |
7c673cae FG |
5373 | } |
5374 | ||
5375 | void OSD::tick_without_osd_lock() | |
5376 | { | |
11fdf7f2 | 5377 | ceph_assert(tick_timer_lock.is_locked()); |
7c673cae FG |
5378 | dout(10) << "tick_without_osd_lock" << dendl; |
5379 | ||
7c673cae FG |
5380 | logger->set(l_osd_cached_crc, buffer::get_cached_crc()); |
5381 | logger->set(l_osd_cached_crc_adjusted, buffer::get_cached_crc_adjusted()); | |
5382 | logger->set(l_osd_missed_crc, buffer::get_missed_crc()); | |
11fdf7f2 TL |
5383 | |
5384 | // refresh osd stats | |
5385 | struct store_statfs_t stbuf; | |
5386 | osd_alert_list_t alerts; | |
5387 | int r = store->statfs(&stbuf, &alerts); | |
5388 | ceph_assert(r == 0); | |
5389 | service.set_statfs(stbuf, alerts); | |
7c673cae FG |
5390 | |
5391 | // osd_lock is not being held, which means the OSD state | |
5392 | // might change when doing the monitor report | |
5393 | if (is_active() || is_waiting_for_healthy()) { | |
5394 | heartbeat_lock.Lock(); | |
5395 | heartbeat_check(); | |
5396 | heartbeat_lock.Unlock(); | |
5397 | ||
5398 | map_lock.get_read(); | |
11fdf7f2 | 5399 | std::lock_guard l(mon_report_lock); |
7c673cae FG |
5400 | |
5401 | // mon report? | |
7c673cae | 5402 | utime_t now = ceph_clock_now(); |
11fdf7f2 TL |
5403 | if (service.need_fullness_update() || |
5404 | now - last_mon_report > cct->_conf->osd_mon_report_interval) { | |
7c673cae | 5405 | last_mon_report = now; |
7c673cae FG |
5406 | send_full_update(); |
5407 | send_failures(); | |
7c673cae FG |
5408 | } |
5409 | map_lock.put_read(); | |
11fdf7f2 TL |
5410 | |
5411 | epoch_t max_waiting_epoch = 0; | |
5412 | for (auto s : shards) { | |
5413 | max_waiting_epoch = std::max(max_waiting_epoch, | |
5414 | s->get_max_waiting_epoch()); | |
5415 | } | |
5416 | if (max_waiting_epoch > get_osdmap()->get_epoch()) { | |
5417 | dout(20) << __func__ << " max_waiting_epoch " << max_waiting_epoch | |
5418 | << ", requesting new map" << dendl; | |
5419 | osdmap_subscribe(superblock.newest_map + 1, false); | |
5420 | } | |
7c673cae FG |
5421 | } |
5422 | ||
5423 | if (is_active()) { | |
5424 | if (!scrub_random_backoff()) { | |
5425 | sched_scrub(); | |
5426 | } | |
5427 | service.promote_throttle_recalibrate(); | |
3efd9988 | 5428 | resume_creating_pg(); |
224ce89b WB |
5429 | bool need_send_beacon = false; |
5430 | const auto now = ceph::coarse_mono_clock::now(); | |
5431 | { | |
5432 | // borrow lec lock to pretect last_sent_beacon from changing | |
11fdf7f2 | 5433 | std::lock_guard l{min_last_epoch_clean_lock}; |
224ce89b WB |
5434 | const auto elapsed = now - last_sent_beacon; |
5435 | if (chrono::duration_cast<chrono::seconds>(elapsed).count() > | |
5436 | cct->_conf->osd_beacon_report_interval) { | |
5437 | need_send_beacon = true; | |
5438 | } | |
5439 | } | |
5440 | if (need_send_beacon) { | |
5441 | send_beacon(now); | |
5442 | } | |
7c673cae FG |
5443 | } |
5444 | ||
11fdf7f2 | 5445 | mgrc.update_daemon_health(get_health_metrics()); |
7c673cae | 5446 | service.kick_recovery_queue(); |
91327a77 AA |
5447 | tick_timer_without_osd_lock.add_event_after(get_tick_interval(), |
5448 | new C_Tick_WithoutOSDLock(this)); | |
7c673cae FG |
5449 | } |
5450 | ||
7c673cae FG |
5451 | // Usage: |
5452 | // setomapval <pool-id> [namespace/]<obj-name> <key> <val> | |
5453 | // rmomapkey <pool-id> [namespace/]<obj-name> <key> | |
5454 | // setomapheader <pool-id> [namespace/]<obj-name> <header> | |
5455 | // getomap <pool> [namespace/]<obj-name> | |
5456 | // truncobj <pool-id> [namespace/]<obj-name> <newlen> | |
5457 | // injectmdataerr [namespace/]<obj-name> [shardid] | |
5458 | // injectdataerr [namespace/]<obj-name> [shardid] | |
5459 | // | |
5460 | // set_recovery_delay [utime] | |
5461 | void TestOpsSocketHook::test_ops(OSDService *service, ObjectStore *store, | |
11fdf7f2 TL |
5462 | std::string_view command, |
5463 | const cmdmap_t& cmdmap, ostream &ss) | |
7c673cae FG |
5464 | { |
5465 | //Test support | |
5466 | //Support changing the omap on a single osd by using the Admin Socket to | |
5467 | //directly request the osd make a change. | |
5468 | if (command == "setomapval" || command == "rmomapkey" || | |
5469 | command == "setomapheader" || command == "getomap" || | |
5470 | command == "truncobj" || command == "injectmdataerr" || | |
5471 | command == "injectdataerr" | |
5472 | ) { | |
5473 | pg_t rawpg; | |
5474 | int64_t pool; | |
5475 | OSDMapRef curmap = service->get_osdmap(); | |
5476 | int r = -1; | |
5477 | ||
5478 | string poolstr; | |
5479 | ||
5480 | cmd_getval(service->cct, cmdmap, "pool", poolstr); | |
5481 | pool = curmap->lookup_pg_pool_name(poolstr); | |
5482 | //If we can't find it by name then maybe id specified | |
5483 | if (pool < 0 && isdigit(poolstr[0])) | |
5484 | pool = atoll(poolstr.c_str()); | |
5485 | if (pool < 0) { | |
b5b8bbf5 | 5486 | ss << "Invalid pool '" << poolstr << "''"; |
7c673cae FG |
5487 | return; |
5488 | } | |
5489 | ||
5490 | string objname, nspace; | |
5491 | cmd_getval(service->cct, cmdmap, "objname", objname); | |
5492 | std::size_t found = objname.find_first_of('/'); | |
5493 | if (found != string::npos) { | |
5494 | nspace = objname.substr(0, found); | |
5495 | objname = objname.substr(found+1); | |
5496 | } | |
5497 | object_locator_t oloc(pool, nspace); | |
5498 | r = curmap->object_locator_to_pg(object_t(objname), oloc, rawpg); | |
5499 | ||
5500 | if (r < 0) { | |
5501 | ss << "Invalid namespace/objname"; | |
5502 | return; | |
5503 | } | |
5504 | ||
5505 | int64_t shardid; | |
5506 | cmd_getval(service->cct, cmdmap, "shardid", shardid, int64_t(shard_id_t::NO_SHARD)); | |
5507 | hobject_t obj(object_t(objname), string(""), CEPH_NOSNAP, rawpg.ps(), pool, nspace); | |
5508 | ghobject_t gobj(obj, ghobject_t::NO_GEN, shard_id_t(uint8_t(shardid))); | |
5509 | spg_t pgid(curmap->raw_pg_to_pg(rawpg), shard_id_t(shardid)); | |
5510 | if (curmap->pg_is_ec(rawpg)) { | |
5511 | if ((command != "injectdataerr") && (command != "injectmdataerr")) { | |
5512 | ss << "Must not call on ec pool, except injectdataerr or injectmdataerr"; | |
5513 | return; | |
5514 | } | |
5515 | } | |
5516 | ||
5517 | ObjectStore::Transaction t; | |
5518 | ||
5519 | if (command == "setomapval") { | |
5520 | map<string, bufferlist> newattrs; | |
5521 | bufferlist val; | |
5522 | string key, valstr; | |
5523 | cmd_getval(service->cct, cmdmap, "key", key); | |
5524 | cmd_getval(service->cct, cmdmap, "val", valstr); | |
5525 | ||
5526 | val.append(valstr); | |
5527 | newattrs[key] = val; | |
5528 | t.omap_setkeys(coll_t(pgid), ghobject_t(obj), newattrs); | |
11fdf7f2 | 5529 | r = store->queue_transaction(service->meta_ch, std::move(t)); |
7c673cae FG |
5530 | if (r < 0) |
5531 | ss << "error=" << r; | |
5532 | else | |
5533 | ss << "ok"; | |
5534 | } else if (command == "rmomapkey") { | |
5535 | string key; | |
5536 | set<string> keys; | |
5537 | cmd_getval(service->cct, cmdmap, "key", key); | |
5538 | ||
5539 | keys.insert(key); | |
5540 | t.omap_rmkeys(coll_t(pgid), ghobject_t(obj), keys); | |
11fdf7f2 | 5541 | r = store->queue_transaction(service->meta_ch, std::move(t)); |
7c673cae FG |
5542 | if (r < 0) |
5543 | ss << "error=" << r; | |
5544 | else | |
5545 | ss << "ok"; | |
5546 | } else if (command == "setomapheader") { | |
5547 | bufferlist newheader; | |
5548 | string headerstr; | |
5549 | ||
5550 | cmd_getval(service->cct, cmdmap, "header", headerstr); | |
5551 | newheader.append(headerstr); | |
5552 | t.omap_setheader(coll_t(pgid), ghobject_t(obj), newheader); | |
11fdf7f2 | 5553 | r = store->queue_transaction(service->meta_ch, std::move(t)); |
7c673cae FG |
5554 | if (r < 0) |
5555 | ss << "error=" << r; | |
5556 | else | |
5557 | ss << "ok"; | |
5558 | } else if (command == "getomap") { | |
5559 | //Debug: Output entire omap | |
5560 | bufferlist hdrbl; | |
5561 | map<string, bufferlist> keyvals; | |
11fdf7f2 TL |
5562 | auto ch = store->open_collection(coll_t(pgid)); |
5563 | if (!ch) { | |
5564 | ss << "unable to open collection for " << pgid; | |
5565 | r = -ENOENT; | |
5566 | } else { | |
5567 | r = store->omap_get(ch, ghobject_t(obj), &hdrbl, &keyvals); | |
5568 | if (r >= 0) { | |
7c673cae FG |
5569 | ss << "header=" << string(hdrbl.c_str(), hdrbl.length()); |
5570 | for (map<string, bufferlist>::iterator it = keyvals.begin(); | |
11fdf7f2 | 5571 | it != keyvals.end(); ++it) |
7c673cae FG |
5572 | ss << " key=" << (*it).first << " val=" |
5573 | << string((*it).second.c_str(), (*it).second.length()); | |
11fdf7f2 | 5574 | } else { |
7c673cae | 5575 | ss << "error=" << r; |
11fdf7f2 | 5576 | } |
7c673cae FG |
5577 | } |
5578 | } else if (command == "truncobj") { | |
5579 | int64_t trunclen; | |
5580 | cmd_getval(service->cct, cmdmap, "len", trunclen); | |
5581 | t.truncate(coll_t(pgid), ghobject_t(obj), trunclen); | |
11fdf7f2 | 5582 | r = store->queue_transaction(service->meta_ch, std::move(t)); |
7c673cae FG |
5583 | if (r < 0) |
5584 | ss << "error=" << r; | |
5585 | else | |
5586 | ss << "ok"; | |
5587 | } else if (command == "injectdataerr") { | |
5588 | store->inject_data_error(gobj); | |
5589 | ss << "ok"; | |
5590 | } else if (command == "injectmdataerr") { | |
5591 | store->inject_mdata_error(gobj); | |
5592 | ss << "ok"; | |
5593 | } | |
5594 | return; | |
5595 | } | |
5596 | if (command == "set_recovery_delay") { | |
5597 | int64_t delay; | |
5598 | cmd_getval(service->cct, cmdmap, "utime", delay, (int64_t)0); | |
5599 | ostringstream oss; | |
5600 | oss << delay; | |
11fdf7f2 | 5601 | int r = service->cct->_conf.set_val("osd_recovery_delay_start", |
7c673cae FG |
5602 | oss.str().c_str()); |
5603 | if (r != 0) { | |
5604 | ss << "set_recovery_delay: error setting " | |
5605 | << "osd_recovery_delay_start to '" << delay << "': error " | |
5606 | << r; | |
5607 | return; | |
5608 | } | |
11fdf7f2 | 5609 | service->cct->_conf.apply_changes(nullptr); |
7c673cae FG |
5610 | ss << "set_recovery_delay: set osd_recovery_delay_start " |
5611 | << "to " << service->cct->_conf->osd_recovery_delay_start; | |
5612 | return; | |
5613 | } | |
a8e16298 | 5614 | if (command == "trigger_scrub" || command == "trigger_deep_scrub") { |
7c673cae | 5615 | spg_t pgid; |
a8e16298 | 5616 | bool deep = (command == "trigger_deep_scrub"); |
7c673cae FG |
5617 | OSDMapRef curmap = service->get_osdmap(); |
5618 | ||
5619 | string pgidstr; | |
5620 | ||
5621 | cmd_getval(service->cct, cmdmap, "pgid", pgidstr); | |
5622 | if (!pgid.parse(pgidstr.c_str())) { | |
5623 | ss << "Invalid pgid specified"; | |
5624 | return; | |
5625 | } | |
5626 | ||
a8e16298 TL |
5627 | int64_t time; |
5628 | cmd_getval(service->cct, cmdmap, "time", time, (int64_t)0); | |
5629 | ||
11fdf7f2 | 5630 | PGRef pg = service->osd->_lookup_lock_pg(pgid); |
7c673cae FG |
5631 | if (pg == nullptr) { |
5632 | ss << "Can't find pg " << pgid; | |
5633 | return; | |
5634 | } | |
5635 | ||
5636 | if (pg->is_primary()) { | |
5637 | pg->unreg_next_scrub(); | |
5638 | const pg_pool_t *p = curmap->get_pg_pool(pgid.pool()); | |
5639 | double pool_scrub_max_interval = 0; | |
a8e16298 TL |
5640 | double scrub_max_interval; |
5641 | if (deep) { | |
5642 | p->opts.get(pool_opts_t::DEEP_SCRUB_INTERVAL, &pool_scrub_max_interval); | |
5643 | scrub_max_interval = pool_scrub_max_interval > 0 ? | |
11fdf7f2 | 5644 | pool_scrub_max_interval : g_conf()->osd_deep_scrub_interval; |
a8e16298 TL |
5645 | } else { |
5646 | p->opts.get(pool_opts_t::SCRUB_MAX_INTERVAL, &pool_scrub_max_interval); | |
5647 | scrub_max_interval = pool_scrub_max_interval > 0 ? | |
11fdf7f2 | 5648 | pool_scrub_max_interval : g_conf()->osd_scrub_max_interval; |
a8e16298 | 5649 | } |
7c673cae FG |
5650 | // Instead of marking must_scrub force a schedule scrub |
5651 | utime_t stamp = ceph_clock_now(); | |
a8e16298 TL |
5652 | if (time == 0) |
5653 | stamp -= scrub_max_interval; | |
5654 | else | |
5655 | stamp -= (float)time; | |
5656 | stamp -= 100.0; // push back last scrub more for good measure | |
5657 | if (deep) { | |
5658 | pg->set_last_deep_scrub_stamp(stamp); | |
5659 | } else { | |
5660 | pg->set_last_scrub_stamp(stamp); | |
5661 | } | |
7c673cae | 5662 | pg->reg_next_scrub(); |
a8e16298 TL |
5663 | pg->publish_stats_to_osd(); |
5664 | ss << "ok - set" << (deep ? " deep" : "" ) << " stamp " << stamp; | |
7c673cae FG |
5665 | } else { |
5666 | ss << "Not primary"; | |
5667 | } | |
5668 | pg->unlock(); | |
5669 | return; | |
5670 | } | |
5671 | if (command == "injectfull") { | |
5672 | int64_t count; | |
5673 | string type; | |
5674 | OSDService::s_names state; | |
5675 | cmd_getval(service->cct, cmdmap, "type", type, string("full")); | |
5676 | cmd_getval(service->cct, cmdmap, "count", count, (int64_t)-1); | |
5677 | if (type == "none" || count == 0) { | |
5678 | type = "none"; | |
5679 | count = 0; | |
5680 | } | |
5681 | state = service->get_full_state(type); | |
5682 | if (state == OSDService::s_names::INVALID) { | |
5683 | ss << "Invalid type use (none, nearfull, backfillfull, full, failsafe)"; | |
5684 | return; | |
5685 | } | |
5686 | service->set_injectfull(state, count); | |
5687 | return; | |
5688 | } | |
5689 | ss << "Internal error - command=" << command; | |
5690 | } | |
5691 | ||
7c673cae FG |
5692 | // ========================================= |
5693 | ||
5694 | void OSD::ms_handle_connect(Connection *con) | |
5695 | { | |
5696 | dout(10) << __func__ << " con " << con << dendl; | |
5697 | if (con->get_peer_type() == CEPH_ENTITY_TYPE_MON) { | |
11fdf7f2 | 5698 | std::lock_guard l(osd_lock); |
7c673cae FG |
5699 | if (is_stopping()) |
5700 | return; | |
5701 | dout(10) << __func__ << " on mon" << dendl; | |
5702 | ||
5703 | if (is_preboot()) { | |
5704 | start_boot(); | |
5705 | } else if (is_booting()) { | |
5706 | _send_boot(); // resend boot message | |
5707 | } else { | |
5708 | map_lock.get_read(); | |
11fdf7f2 | 5709 | std::lock_guard l2(mon_report_lock); |
7c673cae FG |
5710 | |
5711 | utime_t now = ceph_clock_now(); | |
5712 | last_mon_report = now; | |
5713 | ||
5714 | // resend everything, it's a new session | |
5715 | send_full_update(); | |
5716 | send_alive(); | |
5717 | service.requeue_pg_temp(); | |
11fdf7f2 | 5718 | service.clear_sent_ready_to_merge(); |
7c673cae | 5719 | service.send_pg_temp(); |
11fdf7f2 TL |
5720 | service.send_ready_to_merge(); |
5721 | service.send_pg_created(); | |
7c673cae FG |
5722 | requeue_failures(); |
5723 | send_failures(); | |
7c673cae FG |
5724 | |
5725 | map_lock.put_read(); | |
5726 | if (is_active()) { | |
5727 | send_beacon(ceph::coarse_mono_clock::now()); | |
5728 | } | |
5729 | } | |
5730 | ||
5731 | // full map requests may happen while active or pre-boot | |
5732 | if (requested_full_first) { | |
5733 | rerequest_full_maps(); | |
5734 | } | |
5735 | } | |
5736 | } | |
5737 | ||
5738 | void OSD::ms_handle_fast_connect(Connection *con) | |
5739 | { | |
5740 | if (con->get_peer_type() != CEPH_ENTITY_TYPE_MON && | |
5741 | con->get_peer_type() != CEPH_ENTITY_TYPE_MGR) { | |
11fdf7f2 TL |
5742 | auto priv = con->get_priv(); |
5743 | auto s = static_cast<Session*>(priv.get()); | |
7c673cae | 5744 | if (!s) { |
11fdf7f2 TL |
5745 | s = new Session{cct, con}; |
5746 | con->set_priv(RefCountedPtr{s, false}); | |
7c673cae FG |
5747 | dout(10) << " new session (outgoing) " << s << " con=" << s->con |
5748 | << " addr=" << s->con->get_peer_addr() << dendl; | |
5749 | // we don't connect to clients | |
11fdf7f2 | 5750 | ceph_assert(con->get_peer_type() == CEPH_ENTITY_TYPE_OSD); |
7c673cae FG |
5751 | s->entity_name.set_type(CEPH_ENTITY_TYPE_OSD); |
5752 | } | |
7c673cae FG |
5753 | } |
5754 | } | |
5755 | ||
5756 | void OSD::ms_handle_fast_accept(Connection *con) | |
5757 | { | |
5758 | if (con->get_peer_type() != CEPH_ENTITY_TYPE_MON && | |
5759 | con->get_peer_type() != CEPH_ENTITY_TYPE_MGR) { | |
11fdf7f2 TL |
5760 | auto priv = con->get_priv(); |
5761 | auto s = static_cast<Session*>(priv.get()); | |
7c673cae | 5762 | if (!s) { |
11fdf7f2 TL |
5763 | s = new Session{cct, con}; |
5764 | con->set_priv(RefCountedPtr{s, false}); | |
7c673cae FG |
5765 | dout(10) << "new session (incoming)" << s << " con=" << con |
5766 | << " addr=" << con->get_peer_addr() | |
5767 | << " must have raced with connect" << dendl; | |
11fdf7f2 | 5768 | ceph_assert(con->get_peer_type() == CEPH_ENTITY_TYPE_OSD); |
7c673cae FG |
5769 | s->entity_name.set_type(CEPH_ENTITY_TYPE_OSD); |
5770 | } | |
7c673cae FG |
5771 | } |
5772 | } | |
5773 | ||
5774 | bool OSD::ms_handle_reset(Connection *con) | |
5775 | { | |
11fdf7f2 TL |
5776 | auto s = con->get_priv(); |
5777 | auto session = static_cast<Session*>(s.get()); | |
7c673cae FG |
5778 | dout(2) << "ms_handle_reset con " << con << " session " << session << dendl; |
5779 | if (!session) | |
5780 | return false; | |
5781 | session->wstate.reset(con); | |
11fdf7f2 TL |
5782 | session->con->set_priv(nullptr); |
5783 | session->con.reset(); // break con <-> session ref cycle | |
7c673cae FG |
5784 | // note that we break session->con *before* the session_handle_reset |
5785 | // cleanup below. this avoids a race between us and | |
5786 | // PG::add_backoff, Session::check_backoff, etc. | |
11fdf7f2 | 5787 | session_handle_reset(SessionRef{session}); |
7c673cae FG |
5788 | return true; |
5789 | } | |
5790 | ||
5791 | bool OSD::ms_handle_refused(Connection *con) | |
5792 | { | |
5793 | if (!cct->_conf->osd_fast_fail_on_connection_refused) | |
5794 | return false; | |
5795 | ||
11fdf7f2 TL |
5796 | auto priv = con->get_priv(); |
5797 | auto session = static_cast<Session*>(priv.get()); | |
7c673cae FG |
5798 | dout(2) << "ms_handle_refused con " << con << " session " << session << dendl; |
5799 | if (!session) | |
5800 | return false; | |
5801 | int type = con->get_peer_type(); | |
5802 | // handle only OSD failures here | |
5803 | if (monc && (type == CEPH_ENTITY_TYPE_OSD)) { | |
5804 | OSDMapRef osdmap = get_osdmap(); | |
5805 | if (osdmap) { | |
5806 | int id = osdmap->identify_osd_on_all_channels(con->get_peer_addr()); | |
5807 | if (id >= 0 && osdmap->is_up(id)) { | |
5808 | // I'm cheating mon heartbeat grace logic, because we know it's not going | |
5809 | // to respawn alone. +1 so we won't hit any boundary case. | |
11fdf7f2 TL |
5810 | monc->send_mon_message( |
5811 | new MOSDFailure( | |
5812 | monc->get_fsid(), | |
5813 | id, | |
5814 | osdmap->get_addrs(id), | |
5815 | cct->_conf->osd_heartbeat_grace + 1, | |
5816 | osdmap->get_epoch(), | |
5817 | MOSDFailure::FLAG_IMMEDIATE | MOSDFailure::FLAG_FAILED | |
5818 | )); | |
7c673cae FG |
5819 | } |
5820 | } | |
5821 | } | |
7c673cae FG |
5822 | return true; |
5823 | } | |
5824 | ||
5825 | struct C_OSD_GetVersion : public Context { | |
5826 | OSD *osd; | |
5827 | uint64_t oldest, newest; | |
5828 | explicit C_OSD_GetVersion(OSD *o) : osd(o), oldest(0), newest(0) {} | |
5829 | void finish(int r) override { | |
5830 | if (r >= 0) | |
5831 | osd->_got_mon_epochs(oldest, newest); | |
5832 | } | |
5833 | }; | |
5834 | ||
5835 | void OSD::start_boot() | |
5836 | { | |
5837 | if (!_is_healthy()) { | |
5838 | // if we are not healthy, do not mark ourselves up (yet) | |
5839 | dout(1) << "not healthy; waiting to boot" << dendl; | |
5840 | if (!is_waiting_for_healthy()) | |
5841 | start_waiting_for_healthy(); | |
5842 | // send pings sooner rather than later | |
5843 | heartbeat_kick(); | |
5844 | return; | |
5845 | } | |
5846 | dout(1) << __func__ << dendl; | |
5847 | set_state(STATE_PREBOOT); | |
5848 | dout(10) << "start_boot - have maps " << superblock.oldest_map | |
5849 | << ".." << superblock.newest_map << dendl; | |
5850 | C_OSD_GetVersion *c = new C_OSD_GetVersion(this); | |
5851 | monc->get_version("osdmap", &c->newest, &c->oldest, c); | |
5852 | } | |
5853 | ||
5854 | void OSD::_got_mon_epochs(epoch_t oldest, epoch_t newest) | |
5855 | { | |
11fdf7f2 | 5856 | std::lock_guard l(osd_lock); |
7c673cae FG |
5857 | if (is_preboot()) { |
5858 | _preboot(oldest, newest); | |
5859 | } | |
5860 | } | |
5861 | ||
5862 | void OSD::_preboot(epoch_t oldest, epoch_t newest) | |
5863 | { | |
11fdf7f2 | 5864 | ceph_assert(is_preboot()); |
7c673cae FG |
5865 | dout(10) << __func__ << " _preboot mon has osdmaps " |
5866 | << oldest << ".." << newest << dendl; | |
5867 | ||
5868 | // ensure our local fullness awareness is accurate | |
5869 | heartbeat(); | |
5870 | ||
5871 | // if our map within recent history, try to add ourselves to the osdmap. | |
31f18b77 FG |
5872 | if (osdmap->get_epoch() == 0) { |
5873 | derr << "waiting for initial osdmap" << dendl; | |
c07f9fc5 | 5874 | } else if (osdmap->is_destroyed(whoami)) { |
b32b8144 FG |
5875 | derr << "osdmap says I am destroyed" << dendl; |
5876 | // provide a small margin so we don't livelock seeing if we | |
5877 | // un-destroyed ourselves. | |
5878 | if (osdmap->get_epoch() > newest - 1) { | |
5879 | exit(0); | |
5880 | } | |
31f18b77 | 5881 | } else if (osdmap->test_flag(CEPH_OSDMAP_NOUP) || osdmap->is_noup(whoami)) { |
7c673cae FG |
5882 | derr << "osdmap NOUP flag is set, waiting for it to clear" << dendl; |
5883 | } else if (!osdmap->test_flag(CEPH_OSDMAP_SORTBITWISE)) { | |
5884 | derr << "osdmap SORTBITWISE OSDMap flag is NOT set; please set it" | |
5885 | << dendl; | |
11fdf7f2 TL |
5886 | } else if (osdmap->require_osd_release < CEPH_RELEASE_LUMINOUS) { |
5887 | derr << "osdmap require_osd_release < luminous; please upgrade to luminous" | |
7c673cae | 5888 | << dendl; |
7c673cae FG |
5889 | } else if (service.need_fullness_update()) { |
5890 | derr << "osdmap fullness state needs update" << dendl; | |
5891 | send_full_update(); | |
5892 | } else if (osdmap->get_epoch() >= oldest - 1 && | |
5893 | osdmap->get_epoch() + cct->_conf->osd_map_message_max > newest) { | |
11fdf7f2 TL |
5894 | |
5895 | // wait for pgs to fully catch up in a different thread, since | |
5896 | // this thread might be required for splitting and merging PGs to | |
5897 | // make progress. | |
5898 | boot_finisher.queue( | |
5899 | new FunctionContext( | |
5900 | [this](int r) { | |
5901 | std::lock_guard l(osd_lock); | |
5902 | if (is_preboot()) { | |
5903 | dout(10) << __func__ << " waiting for peering work to drain" | |
5904 | << dendl; | |
5905 | osd_lock.Unlock(); | |
5906 | for (auto shard : shards) { | |
5907 | shard->wait_min_pg_epoch(osdmap->get_epoch()); | |
5908 | } | |
5909 | osd_lock.Lock(); | |
5910 | } | |
5911 | if (is_preboot()) { | |
5912 | _send_boot(); | |
5913 | } | |
5914 | })); | |
5915 | return; | |
7c673cae FG |
5916 | } |
5917 | ||
5918 | // get all the latest maps | |
5919 | if (osdmap->get_epoch() + 1 >= oldest) | |
5920 | osdmap_subscribe(osdmap->get_epoch() + 1, false); | |
5921 | else | |
5922 | osdmap_subscribe(oldest - 1, true); | |
5923 | } | |
5924 | ||
5925 | void OSD::send_full_update() | |
5926 | { | |
5927 | if (!service.need_fullness_update()) | |
5928 | return; | |
5929 | unsigned state = 0; | |
5930 | if (service.is_full()) { | |
5931 | state = CEPH_OSD_FULL; | |
5932 | } else if (service.is_backfillfull()) { | |
5933 | state = CEPH_OSD_BACKFILLFULL; | |
5934 | } else if (service.is_nearfull()) { | |
5935 | state = CEPH_OSD_NEARFULL; | |
5936 | } | |
5937 | set<string> s; | |
5938 | OSDMap::calc_state_set(state, s); | |
5939 | dout(10) << __func__ << " want state " << s << dendl; | |
5940 | monc->send_mon_message(new MOSDFull(osdmap->get_epoch(), state)); | |
5941 | } | |
5942 | ||
5943 | void OSD::start_waiting_for_healthy() | |
5944 | { | |
5945 | dout(1) << "start_waiting_for_healthy" << dendl; | |
5946 | set_state(STATE_WAITING_FOR_HEALTHY); | |
5947 | last_heartbeat_resample = utime_t(); | |
181888fb FG |
5948 | |
5949 | // subscribe to osdmap updates, in case our peers really are known to be dead | |
5950 | osdmap_subscribe(osdmap->get_epoch() + 1, false); | |
7c673cae FG |
5951 | } |
5952 | ||
5953 | bool OSD::_is_healthy() | |
5954 | { | |
5955 | if (!cct->get_heartbeat_map()->is_healthy()) { | |
5956 | dout(1) << "is_healthy false -- internal heartbeat failed" << dendl; | |
5957 | return false; | |
5958 | } | |
5959 | ||
5960 | if (is_waiting_for_healthy()) { | |
11fdf7f2 TL |
5961 | utime_t now = ceph_clock_now(); |
5962 | utime_t grace = utime_t(cct->_conf->osd_max_markdown_period, 0); | |
5963 | while (!osd_markdown_log.empty() && | |
5964 | osd_markdown_log.front() + grace < now) | |
5965 | osd_markdown_log.pop_front(); | |
5966 | if (osd_markdown_log.size() <= 1) { | |
5967 | dout(5) << __func__ << " first time marked as down," | |
5968 | << " try reboot unconditionally" << dendl; | |
5969 | return true; | |
5970 | } | |
5971 | std::lock_guard l(heartbeat_lock); | |
7c673cae FG |
5972 | int num = 0, up = 0; |
5973 | for (map<int,HeartbeatInfo>::iterator p = heartbeat_peers.begin(); | |
5974 | p != heartbeat_peers.end(); | |
5975 | ++p) { | |
11fdf7f2 | 5976 | if (p->second.is_healthy(now)) |
7c673cae FG |
5977 | ++up; |
5978 | ++num; | |
5979 | } | |
5980 | if ((float)up < (float)num * cct->_conf->osd_heartbeat_min_healthy_ratio) { | |
5981 | dout(1) << "is_healthy false -- only " << up << "/" << num << " up peers (less than " | |
5982 | << int(cct->_conf->osd_heartbeat_min_healthy_ratio * 100.0) << "%)" << dendl; | |
5983 | return false; | |
5984 | } | |
5985 | } | |
5986 | ||
5987 | return true; | |
5988 | } | |
5989 | ||
5990 | void OSD::_send_boot() | |
5991 | { | |
5992 | dout(10) << "_send_boot" << dendl; | |
11fdf7f2 TL |
5993 | Connection *local_connection = |
5994 | cluster_messenger->get_loopback_connection().get(); | |
5995 | entity_addrvec_t client_addrs = client_messenger->get_myaddrs(); | |
5996 | entity_addrvec_t cluster_addrs = cluster_messenger->get_myaddrs(); | |
5997 | entity_addrvec_t hb_back_addrs = hb_back_server_messenger->get_myaddrs(); | |
5998 | entity_addrvec_t hb_front_addrs = hb_front_server_messenger->get_myaddrs(); | |
5999 | ||
6000 | dout(20) << " initial client_addrs " << client_addrs | |
6001 | << ", cluster_addrs " << cluster_addrs | |
6002 | << ", hb_back_addrs " << hb_back_addrs | |
6003 | << ", hb_front_addrs " << hb_front_addrs | |
6004 | << dendl; | |
6005 | if (cluster_messenger->set_addr_unknowns(client_addrs)) { | |
6006 | dout(10) << " assuming cluster_addrs match client_addrs " | |
6007 | << client_addrs << dendl; | |
6008 | cluster_addrs = cluster_messenger->get_myaddrs(); | |
6009 | } | |
6010 | if (auto session = local_connection->get_priv(); !session) { | |
6011 | cluster_messenger->ms_deliver_handle_fast_connect(local_connection); | |
7c673cae FG |
6012 | } |
6013 | ||
7c673cae | 6014 | local_connection = hb_back_server_messenger->get_loopback_connection().get(); |
11fdf7f2 TL |
6015 | if (hb_back_server_messenger->set_addr_unknowns(cluster_addrs)) { |
6016 | dout(10) << " assuming hb_back_addrs match cluster_addrs " | |
6017 | << cluster_addrs << dendl; | |
6018 | hb_back_addrs = hb_back_server_messenger->get_myaddrs(); | |
7c673cae | 6019 | } |
11fdf7f2 TL |
6020 | if (auto session = local_connection->get_priv(); !session) { |
6021 | hb_back_server_messenger->ms_deliver_handle_fast_connect(local_connection); | |
7c673cae FG |
6022 | } |
6023 | ||
11fdf7f2 TL |
6024 | local_connection = hb_front_server_messenger->get_loopback_connection().get(); |
6025 | if (hb_front_server_messenger->set_addr_unknowns(client_addrs)) { | |
6026 | dout(10) << " assuming hb_front_addrs match client_addrs " | |
6027 | << client_addrs << dendl; | |
6028 | hb_front_addrs = hb_front_server_messenger->get_myaddrs(); | |
6029 | } | |
6030 | if (auto session = local_connection->get_priv(); !session) { | |
6031 | hb_front_server_messenger->ms_deliver_handle_fast_connect(local_connection); | |
6032 | } | |
6033 | ||
6034 | // we now know what our front and back addrs will be, and we are | |
6035 | // about to tell the mon what our metadata (including numa bindings) | |
6036 | // are, so now is a good time! | |
6037 | set_numa_affinity(); | |
6038 | ||
6039 | MOSDBoot *mboot = new MOSDBoot( | |
6040 | superblock, get_osdmap_epoch(), service.get_boot_epoch(), | |
6041 | hb_back_addrs, hb_front_addrs, cluster_addrs, | |
6042 | CEPH_FEATURES_ALL); | |
6043 | dout(10) << " final client_addrs " << client_addrs | |
6044 | << ", cluster_addrs " << cluster_addrs | |
6045 | << ", hb_back_addrs " << hb_back_addrs | |
6046 | << ", hb_front_addrs " << hb_front_addrs | |
7c673cae FG |
6047 | << dendl; |
6048 | _collect_metadata(&mboot->metadata); | |
6049 | monc->send_mon_message(mboot); | |
6050 | set_state(STATE_BOOTING); | |
6051 | } | |
6052 | ||
6053 | void OSD::_collect_metadata(map<string,string> *pm) | |
6054 | { | |
6055 | // config info | |
6056 | (*pm)["osd_data"] = dev_path; | |
c07f9fc5 FG |
6057 | if (store->get_type() == "filestore") { |
6058 | // not applicable for bluestore | |
6059 | (*pm)["osd_journal"] = journal_path; | |
6060 | } | |
11fdf7f2 TL |
6061 | (*pm)["front_addr"] = stringify(client_messenger->get_myaddrs()); |
6062 | (*pm)["back_addr"] = stringify(cluster_messenger->get_myaddrs()); | |
6063 | (*pm)["hb_front_addr"] = stringify(hb_front_server_messenger->get_myaddrs()); | |
6064 | (*pm)["hb_back_addr"] = stringify(hb_back_server_messenger->get_myaddrs()); | |
7c673cae FG |
6065 | |
6066 | // backend | |
6067 | (*pm)["osd_objectstore"] = store->get_type(); | |
31f18b77 | 6068 | (*pm)["rotational"] = store_is_rotational ? "1" : "0"; |
d2e6a577 | 6069 | (*pm)["journal_rotational"] = journal_is_rotational ? "1" : "0"; |
c07f9fc5 | 6070 | (*pm)["default_device_class"] = store->get_default_device_class(); |
7c673cae FG |
6071 | store->collect_metadata(pm); |
6072 | ||
6073 | collect_sys_info(pm, cct); | |
6074 | ||
11fdf7f2 TL |
6075 | (*pm)["front_iface"] = pick_iface( |
6076 | cct, | |
6077 | client_messenger->get_myaddrs().front().get_sockaddr_storage()); | |
6078 | (*pm)["back_iface"] = pick_iface( | |
6079 | cct, | |
6080 | cluster_messenger->get_myaddrs().front().get_sockaddr_storage()); | |
6081 | ||
6082 | // network numa | |
6083 | { | |
6084 | int node = -1; | |
6085 | set<int> nodes; | |
6086 | set<string> unknown; | |
6087 | for (auto nm : { "front_iface", "back_iface" }) { | |
6088 | if (!(*pm)[nm].size()) { | |
6089 | unknown.insert(nm); | |
6090 | continue; | |
6091 | } | |
6092 | int n = -1; | |
6093 | int r = get_iface_numa_node((*pm)[nm], &n); | |
6094 | if (r < 0) { | |
6095 | unknown.insert((*pm)[nm]); | |
6096 | continue; | |
6097 | } | |
6098 | nodes.insert(n); | |
6099 | if (node < 0) { | |
6100 | node = n; | |
6101 | } | |
6102 | } | |
6103 | if (unknown.size()) { | |
6104 | (*pm)["network_numa_unknown_ifaces"] = stringify(unknown); | |
6105 | } | |
6106 | if (!nodes.empty()) { | |
6107 | (*pm)["network_numa_nodes"] = stringify(nodes); | |
6108 | } | |
6109 | if (node >= 0 && nodes.size() == 1 && unknown.empty()) { | |
6110 | (*pm)["network_numa_node"] = stringify(node); | |
6111 | } | |
6112 | } | |
6113 | ||
6114 | if (numa_node >= 0) { | |
6115 | (*pm)["numa_node"] = stringify(numa_node); | |
6116 | (*pm)["numa_node_cpus"] = cpu_set_to_str_list(numa_cpu_set_size, | |
6117 | &numa_cpu_set); | |
6118 | } | |
6119 | ||
6120 | set<string> devnames; | |
6121 | store->get_devices(&devnames); | |
6122 | (*pm)["devices"] = stringify(devnames); | |
6123 | string devids; | |
6124 | for (auto& dev : devnames) { | |
6125 | string err; | |
6126 | string id = get_device_id(dev, &err); | |
6127 | if (id.size()) { | |
6128 | if (!devids.empty()) { | |
6129 | devids += ","; | |
6130 | } | |
6131 | devids += dev + "=" + id; | |
6132 | } else { | |
6133 | dout(10) << __func__ << " no unique device id for " << dev << ": " | |
6134 | << err << dendl; | |
6135 | } | |
6136 | } | |
6137 | (*pm)["device_ids"] = devids; | |
b5b8bbf5 | 6138 | |
7c673cae FG |
6139 | dout(10) << __func__ << " " << *pm << dendl; |
6140 | } | |
6141 | ||
6142 | void OSD::queue_want_up_thru(epoch_t want) | |
6143 | { | |
6144 | map_lock.get_read(); | |
6145 | epoch_t cur = osdmap->get_up_thru(whoami); | |
11fdf7f2 | 6146 | std::lock_guard l(mon_report_lock); |
7c673cae FG |
6147 | if (want > up_thru_wanted) { |
6148 | dout(10) << "queue_want_up_thru now " << want << " (was " << up_thru_wanted << ")" | |
6149 | << ", currently " << cur | |
6150 | << dendl; | |
6151 | up_thru_wanted = want; | |
6152 | send_alive(); | |
6153 | } else { | |
6154 | dout(10) << "queue_want_up_thru want " << want << " <= queued " << up_thru_wanted | |
6155 | << ", currently " << cur | |
6156 | << dendl; | |
6157 | } | |
6158 | map_lock.put_read(); | |
6159 | } | |
6160 | ||
6161 | void OSD::send_alive() | |
6162 | { | |
11fdf7f2 | 6163 | ceph_assert(mon_report_lock.is_locked()); |
7c673cae FG |
6164 | if (!osdmap->exists(whoami)) |
6165 | return; | |
6166 | epoch_t up_thru = osdmap->get_up_thru(whoami); | |
6167 | dout(10) << "send_alive up_thru currently " << up_thru << " want " << up_thru_wanted << dendl; | |
6168 | if (up_thru_wanted > up_thru) { | |
6169 | dout(10) << "send_alive want " << up_thru_wanted << dendl; | |
6170 | monc->send_mon_message(new MOSDAlive(osdmap->get_epoch(), up_thru_wanted)); | |
6171 | } | |
6172 | } | |
6173 | ||
6174 | void OSD::request_full_map(epoch_t first, epoch_t last) | |
6175 | { | |
6176 | dout(10) << __func__ << " " << first << ".." << last | |
6177 | << ", previously requested " | |
6178 | << requested_full_first << ".." << requested_full_last << dendl; | |
11fdf7f2 TL |
6179 | ceph_assert(osd_lock.is_locked()); |
6180 | ceph_assert(first > 0 && last > 0); | |
6181 | ceph_assert(first <= last); | |
6182 | ceph_assert(first >= requested_full_first); // we shouldn't ever ask for older maps | |
7c673cae FG |
6183 | if (requested_full_first == 0) { |
6184 | // first request | |
6185 | requested_full_first = first; | |
6186 | requested_full_last = last; | |
6187 | } else if (last <= requested_full_last) { | |
6188 | // dup | |
6189 | return; | |
6190 | } else { | |
6191 | // additional request | |
6192 | first = requested_full_last + 1; | |
6193 | requested_full_last = last; | |
6194 | } | |
6195 | MMonGetOSDMap *req = new MMonGetOSDMap; | |
6196 | req->request_full(first, last); | |
6197 | monc->send_mon_message(req); | |
6198 | } | |
6199 | ||
6200 | void OSD::got_full_map(epoch_t e) | |
6201 | { | |
11fdf7f2 TL |
6202 | ceph_assert(requested_full_first <= requested_full_last); |
6203 | ceph_assert(osd_lock.is_locked()); | |
7c673cae FG |
6204 | if (requested_full_first == 0) { |
6205 | dout(20) << __func__ << " " << e << ", nothing requested" << dendl; | |
6206 | return; | |
6207 | } | |
6208 | if (e < requested_full_first) { | |
6209 | dout(10) << __func__ << " " << e << ", requested " << requested_full_first | |
6210 | << ".." << requested_full_last | |
6211 | << ", ignoring" << dendl; | |
6212 | return; | |
6213 | } | |
6214 | if (e >= requested_full_last) { | |
6215 | dout(10) << __func__ << " " << e << ", requested " << requested_full_first | |
6216 | << ".." << requested_full_last << ", resetting" << dendl; | |
6217 | requested_full_first = requested_full_last = 0; | |
6218 | return; | |
6219 | } | |
6220 | ||
6221 | requested_full_first = e + 1; | |
6222 | ||
6223 | dout(10) << __func__ << " " << e << ", requested " << requested_full_first | |
6224 | << ".." << requested_full_last | |
6225 | << ", still need more" << dendl; | |
6226 | } | |
6227 | ||
6228 | void OSD::requeue_failures() | |
6229 | { | |
11fdf7f2 | 6230 | std::lock_guard l(heartbeat_lock); |
7c673cae FG |
6231 | unsigned old_queue = failure_queue.size(); |
6232 | unsigned old_pending = failure_pending.size(); | |
11fdf7f2 | 6233 | for (auto p = failure_pending.begin(); p != failure_pending.end(); ) { |
7c673cae FG |
6234 | failure_queue[p->first] = p->second.first; |
6235 | failure_pending.erase(p++); | |
6236 | } | |
6237 | dout(10) << __func__ << " " << old_queue << " + " << old_pending << " -> " | |
6238 | << failure_queue.size() << dendl; | |
6239 | } | |
6240 | ||
6241 | void OSD::send_failures() | |
6242 | { | |
11fdf7f2 TL |
6243 | ceph_assert(map_lock.is_locked()); |
6244 | ceph_assert(mon_report_lock.is_locked()); | |
6245 | std::lock_guard l(heartbeat_lock); | |
7c673cae FG |
6246 | utime_t now = ceph_clock_now(); |
6247 | while (!failure_queue.empty()) { | |
6248 | int osd = failure_queue.begin()->first; | |
7c673cae FG |
6249 | if (!failure_pending.count(osd)) { |
6250 | int failed_for = (int)(double)(now - failure_queue.begin()->second); | |
11fdf7f2 TL |
6251 | monc->send_mon_message( |
6252 | new MOSDFailure( | |
6253 | monc->get_fsid(), | |
6254 | osd, | |
6255 | osdmap->get_addrs(osd), | |
6256 | failed_for, | |
6257 | osdmap->get_epoch())); | |
6258 | failure_pending[osd] = make_pair(failure_queue.begin()->second, | |
6259 | osdmap->get_addrs(osd)); | |
7c673cae FG |
6260 | } |
6261 | failure_queue.erase(osd); | |
6262 | } | |
6263 | } | |
6264 | ||
11fdf7f2 | 6265 | void OSD::send_still_alive(epoch_t epoch, int osd, const entity_addrvec_t &addrs) |
7c673cae | 6266 | { |
11fdf7f2 TL |
6267 | MOSDFailure *m = new MOSDFailure(monc->get_fsid(), osd, addrs, 0, epoch, |
6268 | MOSDFailure::FLAG_ALIVE); | |
7c673cae FG |
6269 | monc->send_mon_message(m); |
6270 | } | |
6271 | ||
11fdf7f2 | 6272 | void OSD::cancel_pending_failures() |
7c673cae | 6273 | { |
11fdf7f2 TL |
6274 | std::lock_guard l(heartbeat_lock); |
6275 | auto it = failure_pending.begin(); | |
6276 | while (it != failure_pending.end()) { | |
6277 | dout(10) << __func__ << " canceling in-flight failure report for osd." | |
6278 | << it->first << dendl; | |
6279 | send_still_alive(osdmap->get_epoch(), it->first, it->second.second); | |
6280 | failure_pending.erase(it++); | |
7c673cae | 6281 | } |
7c673cae FG |
6282 | } |
6283 | ||
6284 | void OSD::send_beacon(const ceph::coarse_mono_clock::time_point& now) | |
6285 | { | |
6286 | const auto& monmap = monc->monmap; | |
6287 | // send beacon to mon even if we are just connected, and the monmap is not | |
6288 | // initialized yet by then. | |
6289 | if (monmap.epoch > 0 && | |
6290 | monmap.get_required_features().contains_all( | |
6291 | ceph::features::mon::FEATURE_LUMINOUS)) { | |
6292 | dout(20) << __func__ << " sending" << dendl; | |
7c673cae FG |
6293 | MOSDBeacon* beacon = nullptr; |
6294 | { | |
11fdf7f2 | 6295 | std::lock_guard l{min_last_epoch_clean_lock}; |
7c673cae FG |
6296 | beacon = new MOSDBeacon(osdmap->get_epoch(), min_last_epoch_clean); |
6297 | std::swap(beacon->pgs, min_last_epoch_clean_pgs); | |
224ce89b | 6298 | last_sent_beacon = now; |
7c673cae FG |
6299 | } |
6300 | monc->send_mon_message(beacon); | |
6301 | } else { | |
6302 | dout(20) << __func__ << " not sending" << dendl; | |
6303 | } | |
6304 | } | |
6305 | ||
6306 | void OSD::handle_command(MMonCommand *m) | |
6307 | { | |
6308 | if (!require_mon_peer(m)) { | |
6309 | m->put(); | |
6310 | return; | |
6311 | } | |
6312 | ||
6313 | Command *c = new Command(m->cmd, m->get_tid(), m->get_data(), NULL); | |
6314 | command_wq.queue(c); | |
6315 | m->put(); | |
6316 | } | |
6317 | ||
6318 | void OSD::handle_command(MCommand *m) | |
6319 | { | |
6320 | ConnectionRef con = m->get_connection(); | |
11fdf7f2 TL |
6321 | auto priv = con->get_priv(); |
6322 | auto session = static_cast<Session *>(priv.get()); | |
7c673cae FG |
6323 | if (!session) { |
6324 | con->send_message(new MCommandReply(m, -EPERM)); | |
6325 | m->put(); | |
6326 | return; | |
6327 | } | |
6328 | ||
6329 | OSDCap& caps = session->caps; | |
11fdf7f2 | 6330 | priv.reset(); |
7c673cae FG |
6331 | |
6332 | if (!caps.allow_all() || m->get_source().is_mon()) { | |
6333 | con->send_message(new MCommandReply(m, -EPERM)); | |
6334 | m->put(); | |
6335 | return; | |
6336 | } | |
6337 | ||
6338 | Command *c = new Command(m->cmd, m->get_tid(), m->get_data(), con.get()); | |
6339 | command_wq.queue(c); | |
6340 | ||
6341 | m->put(); | |
6342 | } | |
6343 | ||
6344 | struct OSDCommand { | |
6345 | string cmdstring; | |
6346 | string helpstring; | |
6347 | string module; | |
6348 | string perm; | |
7c673cae FG |
6349 | } osd_commands[] = { |
6350 | ||
11fdf7f2 TL |
6351 | #define COMMAND(parsesig, helptext, module, perm) \ |
6352 | {parsesig, helptext, module, perm}, | |
7c673cae FG |
6353 | |
6354 | // yes, these are really pg commands, but there's a limit to how | |
6355 | // much work it's worth. The OSD returns all of them. Make this | |
6356 | // form (pg <pgid> <cmd>) valid only for the cli. | |
6357 | // Rest uses "tell <pgid> <cmd>" | |
6358 | ||
6359 | COMMAND("pg " \ | |
6360 | "name=pgid,type=CephPgid " \ | |
6361 | "name=cmd,type=CephChoices,strings=query", \ | |
11fdf7f2 | 6362 | "show details of a specific pg", "osd", "r") |
7c673cae FG |
6363 | COMMAND("pg " \ |
6364 | "name=pgid,type=CephPgid " \ | |
6365 | "name=cmd,type=CephChoices,strings=mark_unfound_lost " \ | |
6366 | "name=mulcmd,type=CephChoices,strings=revert|delete", \ | |
6367 | "mark all unfound objects in this pg as lost, either removing or reverting to a prior version if one is available", | |
11fdf7f2 | 6368 | "osd", "rw") |
7c673cae FG |
6369 | COMMAND("pg " \ |
6370 | "name=pgid,type=CephPgid " \ | |
11fdf7f2 | 6371 | "name=cmd,type=CephChoices,strings=list_unfound " \ |
7c673cae | 6372 | "name=offset,type=CephString,req=false", |
11fdf7f2 TL |
6373 | "list unfound objects on this pg, perhaps starting at an offset given in JSON", |
6374 | "osd", "r") | |
7c673cae FG |
6375 | |
6376 | // new form: tell <pgid> <cmd> for both cli and rest | |
6377 | ||
6378 | COMMAND("query", | |
11fdf7f2 | 6379 | "show details of a specific pg", "osd", "r") |
7c673cae FG |
6380 | COMMAND("mark_unfound_lost " \ |
6381 | "name=mulcmd,type=CephChoices,strings=revert|delete", \ | |
6382 | "mark all unfound objects in this pg as lost, either removing or reverting to a prior version if one is available", | |
11fdf7f2 TL |
6383 | "osd", "rw") |
6384 | COMMAND("list_unfound " \ | |
7c673cae | 6385 | "name=offset,type=CephString,req=false", |
11fdf7f2 TL |
6386 | "list unfound objects on this pg, perhaps starting at an offset given in JSON", |
6387 | "osd", "r") | |
31f18b77 FG |
6388 | COMMAND("perf histogram dump " |
6389 | "name=logger,type=CephString,req=false " | |
6390 | "name=counter,type=CephString,req=false", | |
6391 | "Get histogram data", | |
11fdf7f2 | 6392 | "osd", "r") |
7c673cae FG |
6393 | |
6394 | // tell <osd.n> commands. Validation of osd.n must be special-cased in client | |
11fdf7f2 TL |
6395 | COMMAND("version", "report version of OSD", "osd", "r") |
6396 | COMMAND("get_command_descriptions", "list commands descriptions", "osd", "r") | |
7c673cae FG |
6397 | COMMAND("injectargs " \ |
6398 | "name=injected_args,type=CephString,n=N", | |
6399 | "inject configuration arguments into running OSD", | |
11fdf7f2 | 6400 | "osd", "rw") |
c07f9fc5 FG |
6401 | COMMAND("config set " \ |
6402 | "name=key,type=CephString name=value,type=CephString", | |
6403 | "Set a configuration option at runtime (not persistent)", | |
11fdf7f2 TL |
6404 | "osd", "rw") |
6405 | COMMAND("config get " \ | |
6406 | "name=key,type=CephString", | |
6407 | "Get a configuration option at runtime", | |
6408 | "osd", "r") | |
6409 | COMMAND("config unset " \ | |
6410 | "name=key,type=CephString", | |
6411 | "Unset a configuration option at runtime (not persistent)", | |
6412 | "osd", "rw") | |
7c673cae FG |
6413 | COMMAND("cluster_log " \ |
6414 | "name=level,type=CephChoices,strings=error,warning,info,debug " \ | |
6415 | "name=message,type=CephString,n=N", | |
6416 | "log a message to the cluster log", | |
11fdf7f2 | 6417 | "osd", "rw") |
7c673cae FG |
6418 | COMMAND("bench " \ |
6419 | "name=count,type=CephInt,req=false " \ | |
6420 | "name=size,type=CephInt,req=false " \ | |
6421 | "name=object_size,type=CephInt,req=false " \ | |
6422 | "name=object_num,type=CephInt,req=false ", \ | |
6423 | "OSD benchmark: write <count> <size>-byte objects, " \ | |
6424 | "(default 1G size 4MB). Results in log.", | |
11fdf7f2 TL |
6425 | "osd", "rw") |
6426 | COMMAND("flush_pg_stats", "flush pg stats", "osd", "rw") | |
7c673cae | 6427 | COMMAND("heap " \ |
11fdf7f2 TL |
6428 | "name=heapcmd,type=CephChoices,strings="\ |
6429 | "dump|start_profiler|stop_profiler|release|get_release_rate|set_release_rate|stats " \ | |
6430 | "name=value,type=CephString,req=false", | |
6431 | "show heap usage info (available only if compiled with tcmalloc)", | |
6432 | "osd", "rw") | |
7c673cae FG |
6433 | COMMAND("debug dump_missing " \ |
6434 | "name=filename,type=CephFilepath", | |
11fdf7f2 | 6435 | "dump missing objects to a named file", "osd", "r") |
7c673cae FG |
6436 | COMMAND("debug kick_recovery_wq " \ |
6437 | "name=delay,type=CephInt,range=0", | |
11fdf7f2 | 6438 | "set osd_recovery_delay_start to <val>", "osd", "rw") |
7c673cae FG |
6439 | COMMAND("cpu_profiler " \ |
6440 | "name=arg,type=CephChoices,strings=status|flush", | |
11fdf7f2 | 6441 | "run cpu profiling on daemon", "osd", "rw") |
7c673cae | 6442 | COMMAND("dump_pg_recovery_stats", "dump pg recovery statistics", |
11fdf7f2 | 6443 | "osd", "r") |
7c673cae | 6444 | COMMAND("reset_pg_recovery_stats", "reset pg recovery statistics", |
11fdf7f2 | 6445 | "osd", "rw") |
224ce89b WB |
6446 | COMMAND("compact", |
6447 | "compact object store's omap. " | |
6448 | "WARNING: Compaction probably slows your requests", | |
11fdf7f2 TL |
6449 | "osd", "rw") |
6450 | COMMAND("smart name=devid,type=CephString,req=False", | |
6451 | "runs smartctl on this osd devices. ", | |
6452 | "osd", "rw") | |
6453 | COMMAND("cache drop", | |
6454 | "Drop all OSD caches", | |
6455 | "osd", "rwx") | |
6456 | COMMAND("cache status", | |
6457 | "Get OSD caches statistics", | |
6458 | "osd", "r") | |
6459 | COMMAND("send_beacon", | |
6460 | "Send OSD beacon to mon immediately", | |
6461 | "osd", "r") | |
7c673cae FG |
6462 | }; |
6463 | ||
11fdf7f2 TL |
6464 | void OSD::do_command( |
6465 | Connection *con, ceph_tid_t tid, vector<string>& cmd, bufferlist& data) | |
6466 | { | |
6467 | dout(20) << "do_command tid " << tid << " " << cmd << dendl; | |
6468 | ||
6469 | int r = 0; | |
6470 | stringstream ss, ds; | |
6471 | bufferlist odata; | |
6472 | cmdmap_t cmdmap; | |
6473 | if (cmd.empty()) { | |
6474 | ss << "no command given"; | |
6475 | goto out; | |
6476 | } | |
6477 | if (!cmdmap_from_json(cmd, &cmdmap, ss)) { | |
6478 | r = -EINVAL; | |
6479 | goto out; | |
6480 | } | |
6481 | ||
6482 | try { | |
6483 | r = _do_command(con, cmdmap, tid, data, odata, ss, ds); | |
6484 | } catch (const bad_cmd_get& e) { | |
6485 | r = -EINVAL; | |
6486 | ss << e.what(); | |
6487 | } | |
6488 | if (r == -EAGAIN) { | |
6489 | return; | |
6490 | } | |
6491 | out: | |
6492 | string rs = ss.str(); | |
6493 | odata.append(ds); | |
6494 | dout(0) << "do_command r=" << r << " " << rs << dendl; | |
6495 | clog->info() << rs; | |
6496 | if (con) { | |
6497 | MCommandReply *reply = new MCommandReply(r, rs); | |
6498 | reply->set_tid(tid); | |
6499 | reply->set_data(odata); | |
6500 | con->send_message(reply); | |
6501 | } | |
6502 | } | |
6503 | ||
f64942e4 AA |
6504 | namespace { |
6505 | class unlock_guard { | |
6506 | Mutex& m; | |
6507 | public: | |
6508 | explicit unlock_guard(Mutex& mutex) | |
6509 | : m(mutex) | |
6510 | { | |
11fdf7f2 | 6511 | m.unlock(); |
f64942e4 AA |
6512 | } |
6513 | unlock_guard(unlock_guard&) = delete; | |
6514 | ~unlock_guard() { | |
11fdf7f2 | 6515 | m.lock(); |
f64942e4 AA |
6516 | } |
6517 | }; | |
6518 | } | |
6519 | ||
11fdf7f2 TL |
6520 | int OSD::_do_command( |
6521 | Connection *con, cmdmap_t& cmdmap, ceph_tid_t tid, bufferlist& data, | |
6522 | bufferlist& odata, stringstream& ss, stringstream& ds) | |
7c673cae FG |
6523 | { |
6524 | int r = 0; | |
7c673cae FG |
6525 | string prefix; |
6526 | string format; | |
6527 | string pgidstr; | |
6528 | boost::scoped_ptr<Formatter> f; | |
6529 | ||
7c673cae FG |
6530 | cmd_getval(cct, cmdmap, "prefix", prefix); |
6531 | ||
6532 | if (prefix == "get_command_descriptions") { | |
6533 | int cmdnum = 0; | |
6534 | JSONFormatter *f = new JSONFormatter(); | |
6535 | f->open_object_section("command_descriptions"); | |
6536 | for (OSDCommand *cp = osd_commands; | |
11fdf7f2 | 6537 | cp < &osd_commands[std::size(osd_commands)]; cp++) { |
7c673cae FG |
6538 | |
6539 | ostringstream secname; | |
6540 | secname << "cmd" << setfill('0') << std::setw(3) << cmdnum; | |
11fdf7f2 TL |
6541 | dump_cmddesc_to_json(f, con->get_features(), |
6542 | secname.str(), cp->cmdstring, cp->helpstring, | |
6543 | cp->module, cp->perm, 0); | |
7c673cae FG |
6544 | cmdnum++; |
6545 | } | |
6546 | f->close_section(); // command_descriptions | |
6547 | ||
6548 | f->flush(ds); | |
6549 | delete f; | |
6550 | goto out; | |
6551 | } | |
6552 | ||
6553 | cmd_getval(cct, cmdmap, "format", format); | |
6554 | f.reset(Formatter::create(format)); | |
6555 | ||
6556 | if (prefix == "version") { | |
6557 | if (f) { | |
6558 | f->open_object_section("version"); | |
6559 | f->dump_string("version", pretty_version_to_str()); | |
6560 | f->close_section(); | |
6561 | f->flush(ds); | |
6562 | } else { | |
6563 | ds << pretty_version_to_str(); | |
6564 | } | |
6565 | goto out; | |
6566 | } | |
6567 | else if (prefix == "injectargs") { | |
6568 | vector<string> argsvec; | |
6569 | cmd_getval(cct, cmdmap, "injected_args", argsvec); | |
6570 | ||
6571 | if (argsvec.empty()) { | |
6572 | r = -EINVAL; | |
6573 | ss << "ignoring empty injectargs"; | |
6574 | goto out; | |
6575 | } | |
6576 | string args = argsvec.front(); | |
6577 | for (vector<string>::iterator a = ++argsvec.begin(); a != argsvec.end(); ++a) | |
6578 | args += " " + *a; | |
f64942e4 | 6579 | unlock_guard unlock{osd_lock}; |
11fdf7f2 | 6580 | r = cct->_conf.injectargs(args, &ss); |
7c673cae | 6581 | } |
c07f9fc5 FG |
6582 | else if (prefix == "config set") { |
6583 | std::string key; | |
6584 | std::string val; | |
6585 | cmd_getval(cct, cmdmap, "key", key); | |
6586 | cmd_getval(cct, cmdmap, "value", val); | |
f64942e4 | 6587 | unlock_guard unlock{osd_lock}; |
11fdf7f2 TL |
6588 | r = cct->_conf.set_val(key, val, &ss); |
6589 | if (r == 0) { | |
6590 | cct->_conf.apply_changes(nullptr); | |
6591 | } | |
6592 | } | |
6593 | else if (prefix == "config get") { | |
6594 | std::string key; | |
6595 | cmd_getval(cct, cmdmap, "key", key); | |
6596 | unlock_guard unlock{osd_lock}; | |
6597 | std::string val; | |
6598 | r = cct->_conf.get_val(key, &val); | |
6599 | if (r == 0) { | |
6600 | ds << val; | |
6601 | } | |
6602 | } | |
6603 | else if (prefix == "config unset") { | |
6604 | std::string key; | |
6605 | cmd_getval(cct, cmdmap, "key", key); | |
6606 | unlock_guard unlock{osd_lock}; | |
6607 | r = cct->_conf.rm_val(key); | |
d2e6a577 | 6608 | if (r == 0) { |
11fdf7f2 TL |
6609 | cct->_conf.apply_changes(nullptr); |
6610 | } | |
6611 | if (r == -ENOENT) { | |
6612 | r = 0; // make command idempotent | |
d2e6a577 | 6613 | } |
c07f9fc5 | 6614 | } |
7c673cae FG |
6615 | else if (prefix == "cluster_log") { |
6616 | vector<string> msg; | |
6617 | cmd_getval(cct, cmdmap, "message", msg); | |
6618 | if (msg.empty()) { | |
6619 | r = -EINVAL; | |
6620 | ss << "ignoring empty log message"; | |
6621 | goto out; | |
6622 | } | |
6623 | string message = msg.front(); | |
6624 | for (vector<string>::iterator a = ++msg.begin(); a != msg.end(); ++a) | |
6625 | message += " " + *a; | |
6626 | string lvl; | |
6627 | cmd_getval(cct, cmdmap, "level", lvl); | |
6628 | clog_type level = string_to_clog_type(lvl); | |
6629 | if (level < 0) { | |
6630 | r = -EINVAL; | |
6631 | ss << "unknown level '" << lvl << "'"; | |
6632 | goto out; | |
6633 | } | |
6634 | clog->do_log(level, message); | |
6635 | } | |
6636 | ||
6637 | // either 'pg <pgid> <command>' or | |
6638 | // 'tell <pgid>' (which comes in without any of that prefix)? | |
6639 | ||
6640 | else if (prefix == "pg" || | |
6641 | prefix == "query" || | |
6642 | prefix == "mark_unfound_lost" || | |
11fdf7f2 | 6643 | prefix == "list_unfound" |
7c673cae FG |
6644 | ) { |
6645 | pg_t pgid; | |
6646 | ||
6647 | if (!cmd_getval(cct, cmdmap, "pgid", pgidstr)) { | |
6648 | ss << "no pgid specified"; | |
6649 | r = -EINVAL; | |
6650 | } else if (!pgid.parse(pgidstr.c_str())) { | |
6651 | ss << "couldn't parse pgid '" << pgidstr << "'"; | |
6652 | r = -EINVAL; | |
6653 | } else { | |
6654 | spg_t pcand; | |
11fdf7f2 | 6655 | PGRef pg; |
7c673cae FG |
6656 | if (osdmap->get_primary_shard(pgid, &pcand) && |
6657 | (pg = _lookup_lock_pg(pcand))) { | |
6658 | if (pg->is_primary()) { | |
6659 | // simulate pg <pgid> cmd= for pg->do-command | |
6660 | if (prefix != "pg") | |
6661 | cmd_putval(cct, cmdmap, "cmd", prefix); | |
11fdf7f2 TL |
6662 | try { |
6663 | r = pg->do_command(cmdmap, ss, data, odata, con, tid); | |
6664 | } catch (const bad_cmd_get& e) { | |
6665 | pg->unlock(); | |
6666 | ss << e.what(); | |
6667 | return -EINVAL; | |
6668 | } | |
7c673cae FG |
6669 | if (r == -EAGAIN) { |
6670 | pg->unlock(); | |
6671 | // don't reply, pg will do so async | |
11fdf7f2 | 6672 | return -EAGAIN; |
7c673cae FG |
6673 | } |
6674 | } else { | |
6675 | ss << "not primary for pgid " << pgid; | |
6676 | ||
6677 | // send them the latest diff to ensure they realize the mapping | |
6678 | // has changed. | |
6679 | service.send_incremental_map(osdmap->get_epoch() - 1, con, osdmap); | |
6680 | ||
6681 | // do not reply; they will get newer maps and realize they | |
6682 | // need to resend. | |
6683 | pg->unlock(); | |
11fdf7f2 | 6684 | return -EAGAIN; |
7c673cae FG |
6685 | } |
6686 | pg->unlock(); | |
6687 | } else { | |
6688 | ss << "i don't have pgid " << pgid; | |
6689 | r = -ENOENT; | |
6690 | } | |
6691 | } | |
6692 | } | |
6693 | ||
6694 | else if (prefix == "bench") { | |
6695 | int64_t count; | |
6696 | int64_t bsize; | |
6697 | int64_t osize, onum; | |
6698 | // default count 1G, size 4MB | |
6699 | cmd_getval(cct, cmdmap, "count", count, (int64_t)1 << 30); | |
6700 | cmd_getval(cct, cmdmap, "size", bsize, (int64_t)4 << 20); | |
6701 | cmd_getval(cct, cmdmap, "object_size", osize, (int64_t)0); | |
6702 | cmd_getval(cct, cmdmap, "object_num", onum, (int64_t)0); | |
6703 | ||
7c673cae FG |
6704 | uint32_t duration = cct->_conf->osd_bench_duration; |
6705 | ||
6706 | if (bsize > (int64_t) cct->_conf->osd_bench_max_block_size) { | |
6707 | // let us limit the block size because the next checks rely on it | |
6708 | // having a sane value. If we allow any block size to be set things | |
6709 | // can still go sideways. | |
6710 | ss << "block 'size' values are capped at " | |
1adf2230 | 6711 | << byte_u_t(cct->_conf->osd_bench_max_block_size) << ". If you wish to use" |
7c673cae FG |
6712 | << " a higher value, please adjust 'osd_bench_max_block_size'"; |
6713 | r = -EINVAL; | |
6714 | goto out; | |
6715 | } else if (bsize < (int64_t) (1 << 20)) { | |
6716 | // entering the realm of small block sizes. | |
6717 | // limit the count to a sane value, assuming a configurable amount of | |
6718 | // IOPS and duration, so that the OSD doesn't get hung up on this, | |
6719 | // preventing timeouts from going off | |
6720 | int64_t max_count = | |
6721 | bsize * duration * cct->_conf->osd_bench_small_size_max_iops; | |
6722 | if (count > max_count) { | |
6723 | ss << "'count' values greater than " << max_count | |
1adf2230 | 6724 | << " for a block size of " << byte_u_t(bsize) << ", assuming " |
7c673cae FG |
6725 | << cct->_conf->osd_bench_small_size_max_iops << " IOPS," |
6726 | << " for " << duration << " seconds," | |
6727 | << " can cause ill effects on osd. " | |
6728 | << " Please adjust 'osd_bench_small_size_max_iops' with a higher" | |
6729 | << " value if you wish to use a higher 'count'."; | |
6730 | r = -EINVAL; | |
6731 | goto out; | |
6732 | } | |
6733 | } else { | |
6734 | // 1MB block sizes are big enough so that we get more stuff done. | |
6735 | // However, to avoid the osd from getting hung on this and having | |
6736 | // timers being triggered, we are going to limit the count assuming | |
6737 | // a configurable throughput and duration. | |
6738 | // NOTE: max_count is the total amount of bytes that we believe we | |
6739 | // will be able to write during 'duration' for the given | |
6740 | // throughput. The block size hardly impacts this unless it's | |
6741 | // way too big. Given we already check how big the block size | |
6742 | // is, it's safe to assume everything will check out. | |
6743 | int64_t max_count = | |
6744 | cct->_conf->osd_bench_large_size_max_throughput * duration; | |
6745 | if (count > max_count) { | |
6746 | ss << "'count' values greater than " << max_count | |
1adf2230 AA |
6747 | << " for a block size of " << byte_u_t(bsize) << ", assuming " |
6748 | << byte_u_t(cct->_conf->osd_bench_large_size_max_throughput) << "/s," | |
7c673cae FG |
6749 | << " for " << duration << " seconds," |
6750 | << " can cause ill effects on osd. " | |
6751 | << " Please adjust 'osd_bench_large_size_max_throughput'" | |
6752 | << " with a higher value if you wish to use a higher 'count'."; | |
6753 | r = -EINVAL; | |
6754 | goto out; | |
6755 | } | |
6756 | } | |
6757 | ||
6758 | if (osize && bsize > osize) | |
6759 | bsize = osize; | |
6760 | ||
6761 | dout(1) << " bench count " << count | |
1adf2230 | 6762 | << " bsize " << byte_u_t(bsize) << dendl; |
7c673cae FG |
6763 | |
6764 | ObjectStore::Transaction cleanupt; | |
6765 | ||
6766 | if (osize && onum) { | |
6767 | bufferlist bl; | |
6768 | bufferptr bp(osize); | |
6769 | bp.zero(); | |
6770 | bl.push_back(std::move(bp)); | |
6771 | bl.rebuild_page_aligned(); | |
6772 | for (int i=0; i<onum; ++i) { | |
6773 | char nm[30]; | |
6774 | snprintf(nm, sizeof(nm), "disk_bw_test_%d", i); | |
6775 | object_t oid(nm); | |
6776 | hobject_t soid(sobject_t(oid, 0)); | |
6777 | ObjectStore::Transaction t; | |
6778 | t.write(coll_t(), ghobject_t(soid), 0, osize, bl); | |
11fdf7f2 | 6779 | store->queue_transaction(service.meta_ch, std::move(t), NULL); |
7c673cae FG |
6780 | cleanupt.remove(coll_t(), ghobject_t(soid)); |
6781 | } | |
6782 | } | |
6783 | ||
6784 | bufferlist bl; | |
6785 | bufferptr bp(bsize); | |
6786 | bp.zero(); | |
6787 | bl.push_back(std::move(bp)); | |
6788 | bl.rebuild_page_aligned(); | |
6789 | ||
6790 | { | |
6791 | C_SaferCond waiter; | |
11fdf7f2 | 6792 | if (!service.meta_ch->flush_commit(&waiter)) { |
7c673cae FG |
6793 | waiter.wait(); |
6794 | } | |
6795 | } | |
6796 | ||
6797 | utime_t start = ceph_clock_now(); | |
6798 | for (int64_t pos = 0; pos < count; pos += bsize) { | |
6799 | char nm[30]; | |
6800 | unsigned offset = 0; | |
6801 | if (onum && osize) { | |
6802 | snprintf(nm, sizeof(nm), "disk_bw_test_%d", (int)(rand() % onum)); | |
6803 | offset = rand() % (osize / bsize) * bsize; | |
6804 | } else { | |
6805 | snprintf(nm, sizeof(nm), "disk_bw_test_%lld", (long long)pos); | |
6806 | } | |
6807 | object_t oid(nm); | |
6808 | hobject_t soid(sobject_t(oid, 0)); | |
6809 | ObjectStore::Transaction t; | |
6810 | t.write(coll_t::meta(), ghobject_t(soid), offset, bsize, bl); | |
11fdf7f2 | 6811 | store->queue_transaction(service.meta_ch, std::move(t), NULL); |
7c673cae FG |
6812 | if (!onum || !osize) |
6813 | cleanupt.remove(coll_t::meta(), ghobject_t(soid)); | |
6814 | } | |
6815 | ||
6816 | { | |
6817 | C_SaferCond waiter; | |
11fdf7f2 | 6818 | if (!service.meta_ch->flush_commit(&waiter)) { |
7c673cae FG |
6819 | waiter.wait(); |
6820 | } | |
6821 | } | |
6822 | utime_t end = ceph_clock_now(); | |
6823 | ||
6824 | // clean up | |
11fdf7f2 | 6825 | store->queue_transaction(service.meta_ch, std::move(cleanupt), NULL); |
7c673cae FG |
6826 | { |
6827 | C_SaferCond waiter; | |
11fdf7f2 | 6828 | if (!service.meta_ch->flush_commit(&waiter)) { |
7c673cae FG |
6829 | waiter.wait(); |
6830 | } | |
6831 | } | |
6832 | ||
91327a77 AA |
6833 | double elapsed = end - start; |
6834 | double rate = count / elapsed; | |
6835 | double iops = rate / bsize; | |
7c673cae FG |
6836 | if (f) { |
6837 | f->open_object_section("osd_bench_results"); | |
6838 | f->dump_int("bytes_written", count); | |
6839 | f->dump_int("blocksize", bsize); | |
91327a77 AA |
6840 | f->dump_float("elapsed_sec", elapsed); |
6841 | f->dump_float("bytes_per_sec", rate); | |
6842 | f->dump_float("iops", iops); | |
7c673cae | 6843 | f->close_section(); |
91327a77 | 6844 | f->flush(ds); |
7c673cae | 6845 | } else { |
91327a77 | 6846 | ds << "bench: wrote " << byte_u_t(count) |
1adf2230 | 6847 | << " in blocks of " << byte_u_t(bsize) << " in " |
91327a77 AA |
6848 | << elapsed << " sec at " << byte_u_t(rate) << "/sec " |
6849 | << si_u_t(iops) << " IOPS"; | |
7c673cae FG |
6850 | } |
6851 | } | |
6852 | ||
6853 | else if (prefix == "flush_pg_stats") { | |
11fdf7f2 TL |
6854 | mgrc.send_pgstats(); |
6855 | ds << service.get_osd_stat_seq() << "\n"; | |
7c673cae FG |
6856 | } |
6857 | ||
6858 | else if (prefix == "heap") { | |
6859 | r = ceph::osd_cmds::heap(*cct, cmdmap, *f, ds); | |
6860 | } | |
6861 | ||
6862 | else if (prefix == "debug dump_missing") { | |
11fdf7f2 TL |
6863 | if (!f) { |
6864 | f.reset(new JSONFormatter(true)); | |
7c673cae | 6865 | } |
11fdf7f2 TL |
6866 | f->open_array_section("pgs"); |
6867 | vector<PGRef> pgs; | |
6868 | _get_pgs(&pgs); | |
6869 | for (auto& pg : pgs) { | |
6870 | string s = stringify(pg->pg_id); | |
6871 | f->open_array_section(s.c_str()); | |
7c673cae | 6872 | pg->lock(); |
11fdf7f2 | 6873 | pg->dump_missing(f.get()); |
7c673cae | 6874 | pg->unlock(); |
11fdf7f2 | 6875 | f->close_section(); |
7c673cae | 6876 | } |
11fdf7f2 TL |
6877 | f->close_section(); |
6878 | f->flush(ds); | |
7c673cae FG |
6879 | } |
6880 | else if (prefix == "debug kick_recovery_wq") { | |
6881 | int64_t delay; | |
6882 | cmd_getval(cct, cmdmap, "delay", delay); | |
6883 | ostringstream oss; | |
6884 | oss << delay; | |
f64942e4 | 6885 | unlock_guard unlock{osd_lock}; |
11fdf7f2 | 6886 | r = cct->_conf.set_val("osd_recovery_delay_start", oss.str().c_str()); |
7c673cae FG |
6887 | if (r != 0) { |
6888 | ss << "kick_recovery_wq: error setting " | |
6889 | << "osd_recovery_delay_start to '" << delay << "': error " | |
6890 | << r; | |
6891 | goto out; | |
6892 | } | |
11fdf7f2 | 6893 | cct->_conf.apply_changes(nullptr); |
7c673cae FG |
6894 | ss << "kicking recovery queue. set osd_recovery_delay_start " |
6895 | << "to " << cct->_conf->osd_recovery_delay_start; | |
6896 | } | |
6897 | ||
6898 | else if (prefix == "cpu_profiler") { | |
6899 | string arg; | |
6900 | cmd_getval(cct, cmdmap, "arg", arg); | |
6901 | vector<string> argvec; | |
6902 | get_str_vec(arg, argvec); | |
6903 | cpu_profiler_handle_command(argvec, ds); | |
6904 | } | |
6905 | ||
6906 | else if (prefix == "dump_pg_recovery_stats") { | |
6907 | stringstream s; | |
6908 | if (f) { | |
6909 | pg_recovery_stats.dump_formatted(f.get()); | |
6910 | f->flush(ds); | |
6911 | } else { | |
6912 | pg_recovery_stats.dump(s); | |
6913 | ds << "dump pg recovery stats: " << s.str(); | |
6914 | } | |
6915 | } | |
6916 | ||
6917 | else if (prefix == "reset_pg_recovery_stats") { | |
6918 | ss << "reset pg recovery stats"; | |
6919 | pg_recovery_stats.reset(); | |
6920 | } | |
6921 | ||
31f18b77 FG |
6922 | else if (prefix == "perf histogram dump") { |
6923 | std::string logger; | |
6924 | std::string counter; | |
6925 | cmd_getval(cct, cmdmap, "logger", logger); | |
6926 | cmd_getval(cct, cmdmap, "counter", counter); | |
6927 | if (f) { | |
6928 | cct->get_perfcounters_collection()->dump_formatted_histograms( | |
6929 | f.get(), false, logger, counter); | |
6930 | f->flush(ds); | |
6931 | } | |
6932 | } | |
6933 | ||
224ce89b WB |
6934 | else if (prefix == "compact") { |
6935 | dout(1) << "triggering manual compaction" << dendl; | |
6936 | auto start = ceph::coarse_mono_clock::now(); | |
6937 | store->compact(); | |
6938 | auto end = ceph::coarse_mono_clock::now(); | |
11fdf7f2 | 6939 | double duration = std::chrono::duration<double>(end-start).count(); |
224ce89b | 6940 | dout(1) << "finished manual compaction in " |
11fdf7f2 | 6941 | << duration |
224ce89b | 6942 | << " seconds" << dendl; |
11fdf7f2 TL |
6943 | ss << "compacted omap in " << duration << " seconds"; |
6944 | } | |
6945 | ||
6946 | else if (prefix == "smart") { | |
6947 | string devid; | |
6948 | cmd_getval(cct, cmdmap, "devid", devid); | |
6949 | probe_smart(devid, ds); | |
6950 | } | |
6951 | ||
6952 | else if (prefix == "cache drop") { | |
6953 | dout(20) << "clearing all caches" << dendl; | |
6954 | // Clear the objectstore's cache - onode and buffer for Bluestore, | |
6955 | // system's pagecache for Filestore | |
6956 | r = store->flush_cache(&ss); | |
6957 | if (r < 0) { | |
6958 | ds << "Error flushing objectstore cache: " << cpp_strerror(r); | |
6959 | goto out; | |
6960 | } | |
6961 | // Clear the objectcontext cache (per PG) | |
6962 | vector<PGRef> pgs; | |
6963 | _get_pgs(&pgs); | |
6964 | for (auto& pg: pgs) { | |
6965 | pg->clear_cache(); | |
6966 | } | |
224ce89b WB |
6967 | } |
6968 | ||
11fdf7f2 TL |
6969 | else if (prefix == "cache status") { |
6970 | int obj_ctx_count = 0; | |
6971 | vector<PGRef> pgs; | |
6972 | _get_pgs(&pgs); | |
6973 | for (auto& pg: pgs) { | |
6974 | obj_ctx_count += pg->get_cache_obj_count(); | |
6975 | } | |
6976 | if (f) { | |
6977 | f->open_object_section("cache_status"); | |
6978 | f->dump_int("object_ctx", obj_ctx_count); | |
6979 | store->dump_cache_stats(f.get()); | |
6980 | f->close_section(); | |
6981 | f->flush(ds); | |
6982 | } else { | |
6983 | ds << "object_ctx: " << obj_ctx_count; | |
6984 | store->dump_cache_stats(ds); | |
6985 | } | |
6986 | } | |
6987 | else if (prefix == "send_beacon") { | |
6988 | if (is_active()) { | |
6989 | send_beacon(ceph::coarse_mono_clock::now()); | |
6990 | } | |
6991 | } else { | |
6992 | ss << "unrecognized command '" << prefix << "'"; | |
7c673cae FG |
6993 | r = -EINVAL; |
6994 | } | |
6995 | ||
6996 | out: | |
11fdf7f2 TL |
6997 | return r; |
6998 | } | |
6999 | ||
7000 | void OSD::probe_smart(const string& only_devid, ostream& ss) | |
7001 | { | |
7002 | set<string> devnames; | |
7003 | store->get_devices(&devnames); | |
7004 | uint64_t smart_timeout = cct->_conf.get_val<uint64_t>( | |
7005 | "osd_smart_report_timeout"); | |
7006 | ||
7007 | // == typedef std::map<std::string, mValue> mObject; | |
7008 | json_spirit::mObject json_map; | |
7009 | ||
7010 | for (auto dev : devnames) { | |
7011 | // smartctl works only on physical devices; filter out any logical device | |
7012 | if (dev.find("dm-") == 0) { | |
7013 | continue; | |
7014 | } | |
7015 | ||
7016 | string err; | |
7017 | string devid = get_device_id(dev, &err); | |
7018 | if (devid.size() == 0) { | |
7019 | dout(10) << __func__ << " no unique id for dev " << dev << " (" | |
7020 | << err << "), skipping" << dendl; | |
7021 | continue; | |
7022 | } | |
7023 | if (only_devid.size() && devid != only_devid) { | |
7024 | continue; | |
7025 | } | |
7026 | ||
7027 | json_spirit::mValue smart_json; | |
7028 | if (block_device_get_metrics(dev, smart_timeout, | |
7029 | &smart_json)) { | |
7030 | dout(10) << "block_device_get_metrics failed for /dev/" << dev << dendl; | |
7031 | continue; | |
7032 | } | |
7033 | json_map[devid] = smart_json; | |
7c673cae | 7034 | } |
11fdf7f2 | 7035 | json_spirit::write(json_map, ss, json_spirit::pretty_print); |
7c673cae FG |
7036 | } |
7037 | ||
7038 | bool OSD::heartbeat_dispatch(Message *m) | |
7039 | { | |
7040 | dout(30) << "heartbeat_dispatch " << m << dendl; | |
7041 | switch (m->get_type()) { | |
7042 | ||
7043 | case CEPH_MSG_PING: | |
7044 | dout(10) << "ping from " << m->get_source_inst() << dendl; | |
7045 | m->put(); | |
7046 | break; | |
7047 | ||
7048 | case MSG_OSD_PING: | |
7049 | handle_osd_ping(static_cast<MOSDPing*>(m)); | |
7050 | break; | |
7051 | ||
7052 | default: | |
7053 | dout(0) << "dropping unexpected message " << *m << " from " << m->get_source_inst() << dendl; | |
7054 | m->put(); | |
7055 | } | |
7056 | ||
7057 | return true; | |
7058 | } | |
7059 | ||
7060 | bool OSD::ms_dispatch(Message *m) | |
7061 | { | |
7062 | dout(20) << "OSD::ms_dispatch: " << *m << dendl; | |
7063 | if (m->get_type() == MSG_OSD_MARK_ME_DOWN) { | |
7064 | service.got_stop_ack(); | |
7065 | m->put(); | |
7066 | return true; | |
7067 | } | |
7068 | ||
7069 | // lock! | |
7070 | ||
7071 | osd_lock.Lock(); | |
7072 | if (is_stopping()) { | |
7073 | osd_lock.Unlock(); | |
7074 | m->put(); | |
7075 | return true; | |
7076 | } | |
7077 | ||
7078 | do_waiters(); | |
7079 | _dispatch(m); | |
7080 | ||
7081 | osd_lock.Unlock(); | |
7082 | ||
7083 | return true; | |
7084 | } | |
7085 | ||
7086 | void OSD::maybe_share_map( | |
7087 | Session *session, | |
7088 | OpRequestRef op, | |
7089 | OSDMapRef osdmap) | |
7090 | { | |
7091 | if (!op->check_send_map) { | |
7092 | return; | |
7093 | } | |
7094 | epoch_t last_sent_epoch = 0; | |
7095 | ||
7096 | session->sent_epoch_lock.lock(); | |
7097 | last_sent_epoch = session->last_sent_epoch; | |
7098 | session->sent_epoch_lock.unlock(); | |
7099 | ||
11fdf7f2 TL |
7100 | // assume the peer has the newer of the op's sent_epoch and what |
7101 | // we think we sent them. | |
7102 | epoch_t from = std::max(last_sent_epoch, op->sent_epoch); | |
7103 | ||
7c673cae FG |
7104 | const Message *m = op->get_req(); |
7105 | service.share_map( | |
7106 | m->get_source(), | |
7107 | m->get_connection().get(), | |
11fdf7f2 | 7108 | from, |
7c673cae FG |
7109 | osdmap, |
7110 | session ? &last_sent_epoch : NULL); | |
7111 | ||
7112 | session->sent_epoch_lock.lock(); | |
7113 | if (session->last_sent_epoch < last_sent_epoch) { | |
7114 | session->last_sent_epoch = last_sent_epoch; | |
7115 | } | |
7116 | session->sent_epoch_lock.unlock(); | |
7117 | ||
7118 | op->check_send_map = false; | |
7119 | } | |
7120 | ||
11fdf7f2 | 7121 | void OSD::dispatch_session_waiting(SessionRef session, OSDMapRef osdmap) |
7c673cae | 7122 | { |
11fdf7f2 | 7123 | ceph_assert(session->session_dispatch_lock.is_locked()); |
7c673cae FG |
7124 | |
7125 | auto i = session->waiting_on_map.begin(); | |
7126 | while (i != session->waiting_on_map.end()) { | |
7127 | OpRequestRef op = &(*i); | |
11fdf7f2 | 7128 | ceph_assert(ms_can_fast_dispatch(op->get_req())); |
7c673cae FG |
7129 | const MOSDFastDispatchOp *m = static_cast<const MOSDFastDispatchOp*>( |
7130 | op->get_req()); | |
7131 | if (m->get_min_epoch() > osdmap->get_epoch()) { | |
7132 | break; | |
7133 | } | |
7134 | session->waiting_on_map.erase(i++); | |
7135 | op->put(); | |
7136 | ||
7137 | spg_t pgid; | |
7138 | if (m->get_type() == CEPH_MSG_OSD_OP) { | |
7139 | pg_t actual_pgid = osdmap->raw_pg_to_pg( | |
7140 | static_cast<const MOSDOp*>(m)->get_pg()); | |
7141 | if (!osdmap->get_primary_shard(actual_pgid, &pgid)) { | |
7142 | continue; | |
7143 | } | |
7144 | } else { | |
7145 | pgid = m->get_spg(); | |
7146 | } | |
11fdf7f2 | 7147 | enqueue_op(pgid, std::move(op), m->get_map_epoch()); |
7c673cae FG |
7148 | } |
7149 | ||
7150 | if (session->waiting_on_map.empty()) { | |
7151 | clear_session_waiting_on_map(session); | |
7152 | } else { | |
7153 | register_session_waiting_on_map(session); | |
7154 | } | |
7155 | } | |
7156 | ||
7157 | void OSD::ms_fast_dispatch(Message *m) | |
7158 | { | |
11fdf7f2 | 7159 | FUNCTRACE(cct); |
7c673cae FG |
7160 | if (service.is_stopping()) { |
7161 | m->put(); | |
7162 | return; | |
7163 | } | |
11fdf7f2 TL |
7164 | |
7165 | // peering event? | |
7166 | switch (m->get_type()) { | |
7167 | case CEPH_MSG_PING: | |
7168 | dout(10) << "ping from " << m->get_source() << dendl; | |
7169 | m->put(); | |
7170 | return; | |
7171 | case MSG_MON_COMMAND: | |
7172 | handle_command(static_cast<MMonCommand*>(m)); | |
7173 | return; | |
7174 | case MSG_OSD_FORCE_RECOVERY: | |
7175 | handle_fast_force_recovery(static_cast<MOSDForceRecovery*>(m)); | |
7176 | return; | |
7177 | case MSG_OSD_SCRUB2: | |
7178 | handle_fast_scrub(static_cast<MOSDScrub2*>(m)); | |
7179 | return; | |
7180 | ||
7181 | case MSG_OSD_PG_CREATE2: | |
7182 | return handle_fast_pg_create(static_cast<MOSDPGCreate2*>(m)); | |
7183 | case MSG_OSD_PG_QUERY: | |
7184 | return handle_fast_pg_query(static_cast<MOSDPGQuery*>(m)); | |
7185 | case MSG_OSD_PG_NOTIFY: | |
7186 | return handle_fast_pg_notify(static_cast<MOSDPGNotify*>(m)); | |
7187 | case MSG_OSD_PG_INFO: | |
7188 | return handle_fast_pg_info(static_cast<MOSDPGInfo*>(m)); | |
7189 | case MSG_OSD_PG_REMOVE: | |
7190 | return handle_fast_pg_remove(static_cast<MOSDPGRemove*>(m)); | |
7191 | ||
7192 | // these are single-pg messages that handle themselves | |
7193 | case MSG_OSD_PG_LOG: | |
7194 | case MSG_OSD_PG_TRIM: | |
7195 | case MSG_OSD_BACKFILL_RESERVE: | |
7196 | case MSG_OSD_RECOVERY_RESERVE: | |
7197 | { | |
7198 | MOSDPeeringOp *pm = static_cast<MOSDPeeringOp*>(m); | |
7199 | if (require_osd_peer(pm)) { | |
7200 | enqueue_peering_evt( | |
7201 | pm->get_spg(), | |
7202 | PGPeeringEventRef(pm->get_event())); | |
7203 | } | |
7204 | pm->put(); | |
7205 | return; | |
7206 | } | |
7207 | } | |
7208 | ||
7c673cae FG |
7209 | OpRequestRef op = op_tracker.create_request<OpRequest, Message*>(m); |
7210 | { | |
7211 | #ifdef WITH_LTTNG | |
7212 | osd_reqid_t reqid = op->get_reqid(); | |
7213 | #endif | |
7214 | tracepoint(osd, ms_fast_dispatch, reqid.name._type, | |
7215 | reqid.name._num, reqid.tid, reqid.inc); | |
7216 | } | |
7217 | ||
7218 | if (m->trace) | |
7219 | op->osd_trace.init("osd op", &trace_endpoint, &m->trace); | |
7220 | ||
11fdf7f2 | 7221 | // note sender epoch, min req's epoch |
7c673cae FG |
7222 | op->sent_epoch = static_cast<MOSDFastDispatchOp*>(m)->get_map_epoch(); |
7223 | op->min_epoch = static_cast<MOSDFastDispatchOp*>(m)->get_min_epoch(); | |
11fdf7f2 | 7224 | ceph_assert(op->min_epoch <= op->sent_epoch); // sanity check! |
7c673cae FG |
7225 | |
7226 | service.maybe_inject_dispatch_delay(); | |
7227 | ||
7228 | if (m->get_connection()->has_features(CEPH_FEATUREMASK_RESEND_ON_SPLIT) || | |
7229 | m->get_type() != CEPH_MSG_OSD_OP) { | |
7230 | // queue it directly | |
7231 | enqueue_op( | |
7232 | static_cast<MOSDFastDispatchOp*>(m)->get_spg(), | |
11fdf7f2 | 7233 | std::move(op), |
7c673cae FG |
7234 | static_cast<MOSDFastDispatchOp*>(m)->get_map_epoch()); |
7235 | } else { | |
7236 | // legacy client, and this is an MOSDOp (the *only* fast dispatch | |
7237 | // message that didn't have an explicit spg_t); we need to map | |
7238 | // them to an spg_t while preserving delivery order. | |
11fdf7f2 TL |
7239 | auto priv = m->get_connection()->get_priv(); |
7240 | if (auto session = static_cast<Session*>(priv.get()); session) { | |
7241 | std::lock_guard l{session->session_dispatch_lock}; | |
7242 | op->get(); | |
7243 | session->waiting_on_map.push_back(*op); | |
7244 | OSDMapRef nextmap = service.get_nextmap_reserved(); | |
7245 | dispatch_session_waiting(session, nextmap); | |
7246 | service.release_map(nextmap); | |
7c673cae FG |
7247 | } |
7248 | } | |
7249 | OID_EVENT_TRACE_WITH_MSG(m, "MS_FAST_DISPATCH_END", false); | |
7250 | } | |
7251 | ||
11fdf7f2 | 7252 | bool OSD::ms_get_authorizer(int dest_type, AuthAuthorizer **authorizer) |
7c673cae FG |
7253 | { |
7254 | dout(10) << "OSD::ms_get_authorizer type=" << ceph_entity_type_name(dest_type) << dendl; | |
7255 | ||
31f18b77 FG |
7256 | if (is_stopping()) { |
7257 | dout(10) << __func__ << " bailing, we are shutting down" << dendl; | |
7258 | return false; | |
7259 | } | |
7260 | ||
7c673cae FG |
7261 | if (dest_type == CEPH_ENTITY_TYPE_MON) |
7262 | return true; | |
7263 | ||
7c673cae FG |
7264 | *authorizer = monc->build_authorizer(dest_type); |
7265 | return *authorizer != NULL; | |
7266 | } | |
7267 | ||
11fdf7f2 TL |
7268 | KeyStore *OSD::ms_get_auth1_authorizer_keystore() |
7269 | { | |
7270 | return monc->rotating_secrets.get(); | |
7271 | } | |
7c673cae | 7272 | |
11fdf7f2 | 7273 | int OSD::ms_handle_authentication(Connection *con) |
7c673cae | 7274 | { |
11fdf7f2 TL |
7275 | int ret = 0; |
7276 | auto priv = con->get_priv(); | |
7277 | Session *s = static_cast<Session*>(priv.get()); | |
7278 | if (!s) { | |
7279 | s = new Session(cct, con); | |
7280 | con->set_priv(RefCountedPtr{s, false}); | |
7281 | s->entity_name = con->get_peer_entity_name(); | |
7282 | dout(10) << __func__ << " new session " << s << " con " << s->con | |
7283 | << " entity " << s->entity_name | |
7284 | << " addr " << con->get_peer_addrs() << dendl; | |
7285 | } else { | |
7286 | dout(10) << __func__ << " existing session " << s << " con " << s->con | |
7287 | << " entity " << s->entity_name | |
7288 | << " addr " << con->get_peer_addrs() << dendl; | |
7c673cae FG |
7289 | } |
7290 | ||
11fdf7f2 TL |
7291 | AuthCapsInfo &caps_info = con->get_peer_caps_info(); |
7292 | if (caps_info.allow_all) | |
7293 | s->caps.set_allow_all(); | |
7c673cae | 7294 | |
11fdf7f2 TL |
7295 | if (caps_info.caps.length() > 0) { |
7296 | bufferlist::const_iterator p = caps_info.caps.cbegin(); | |
7297 | string str; | |
7298 | try { | |
7299 | decode(str, p); | |
7300 | } | |
7301 | catch (buffer::error& e) { | |
7302 | dout(10) << __func__ << " session " << s << " " << s->entity_name | |
7303 | << " failed to decode caps string" << dendl; | |
7304 | ret = -EPERM; | |
7305 | } | |
7306 | if (!ret) { | |
7c673cae | 7307 | bool success = s->caps.parse(str); |
11fdf7f2 TL |
7308 | if (success) { |
7309 | dout(10) << __func__ << " session " << s | |
7310 | << " " << s->entity_name | |
7311 | << " has caps " << s->caps << " '" << str << "'" << dendl; | |
7312 | ret = 1; | |
7313 | } else { | |
7314 | dout(10) << __func__ << " session " << s << " " << s->entity_name | |
7315 | << " failed to parse caps '" << str << "'" << dendl; | |
7316 | ret = -EPERM; | |
7317 | } | |
7c673cae | 7318 | } |
7c673cae | 7319 | } |
11fdf7f2 | 7320 | return ret; |
7c673cae FG |
7321 | } |
7322 | ||
7323 | void OSD::do_waiters() | |
7324 | { | |
11fdf7f2 | 7325 | ceph_assert(osd_lock.is_locked()); |
7c673cae FG |
7326 | |
7327 | dout(10) << "do_waiters -- start" << dendl; | |
7328 | while (!finished.empty()) { | |
7329 | OpRequestRef next = finished.front(); | |
7330 | finished.pop_front(); | |
7331 | dispatch_op(next); | |
7332 | } | |
7333 | dout(10) << "do_waiters -- finish" << dendl; | |
7334 | } | |
7335 | ||
7336 | void OSD::dispatch_op(OpRequestRef op) | |
7337 | { | |
7338 | switch (op->get_req()->get_type()) { | |
7339 | ||
7340 | case MSG_OSD_PG_CREATE: | |
7341 | handle_pg_create(op); | |
7342 | break; | |
7c673cae FG |
7343 | } |
7344 | } | |
7345 | ||
7346 | void OSD::_dispatch(Message *m) | |
7347 | { | |
11fdf7f2 | 7348 | ceph_assert(osd_lock.is_locked()); |
7c673cae FG |
7349 | dout(20) << "_dispatch " << m << " " << *m << dendl; |
7350 | ||
7351 | switch (m->get_type()) { | |
7c673cae FG |
7352 | // -- don't need OSDMap -- |
7353 | ||
7354 | // map and replication | |
7355 | case CEPH_MSG_OSD_MAP: | |
7356 | handle_osd_map(static_cast<MOSDMap*>(m)); | |
7357 | break; | |
7358 | ||
7359 | // osd | |
7c673cae FG |
7360 | case MSG_OSD_SCRUB: |
7361 | handle_scrub(static_cast<MOSDScrub*>(m)); | |
7362 | break; | |
7363 | ||
11fdf7f2 TL |
7364 | case MSG_COMMAND: |
7365 | handle_command(static_cast<MCommand*>(m)); | |
7366 | return; | |
c07f9fc5 | 7367 | |
7c673cae FG |
7368 | // -- need OSDMap -- |
7369 | ||
7370 | case MSG_OSD_PG_CREATE: | |
7c673cae FG |
7371 | { |
7372 | OpRequestRef op = op_tracker.create_request<OpRequest, Message*>(m); | |
7373 | if (m->trace) | |
7374 | op->osd_trace.init("osd op", &trace_endpoint, &m->trace); | |
7375 | // no map? starting up? | |
7376 | if (!osdmap) { | |
7377 | dout(7) << "no OSDMap, not booted" << dendl; | |
7378 | logger->inc(l_osd_waiting_for_map); | |
7379 | waiting_for_osdmap.push_back(op); | |
7380 | op->mark_delayed("no osdmap"); | |
7381 | break; | |
7382 | } | |
7383 | ||
7384 | // need OSDMap | |
7385 | dispatch_op(op); | |
7386 | } | |
7387 | } | |
7388 | } | |
7389 | ||
11fdf7f2 | 7390 | // remove me post-nautilus |
7c673cae FG |
7391 | void OSD::handle_scrub(MOSDScrub *m) |
7392 | { | |
7393 | dout(10) << "handle_scrub " << *m << dendl; | |
7394 | if (!require_mon_or_mgr_peer(m)) { | |
7395 | m->put(); | |
7396 | return; | |
7397 | } | |
7398 | if (m->fsid != monc->get_fsid()) { | |
11fdf7f2 TL |
7399 | dout(0) << "handle_scrub fsid " << m->fsid << " != " << monc->get_fsid() |
7400 | << dendl; | |
7c673cae FG |
7401 | m->put(); |
7402 | return; | |
7403 | } | |
7404 | ||
11fdf7f2 TL |
7405 | vector<spg_t> spgs; |
7406 | _get_pgids(&spgs); | |
7407 | ||
7408 | if (!m->scrub_pgs.empty()) { | |
7409 | vector<spg_t> v; | |
7410 | for (auto pgid : m->scrub_pgs) { | |
7c673cae | 7411 | spg_t pcand; |
11fdf7f2 TL |
7412 | if (osdmap->get_primary_shard(pgid, &pcand) && |
7413 | std::find(spgs.begin(), spgs.end(), pcand) != spgs.end()) { | |
7414 | v.push_back(pcand); | |
7c673cae FG |
7415 | } |
7416 | } | |
11fdf7f2 TL |
7417 | spgs.swap(v); |
7418 | } | |
7419 | ||
7420 | for (auto pgid : spgs) { | |
7421 | enqueue_peering_evt( | |
7422 | pgid, | |
7423 | PGPeeringEventRef( | |
7424 | std::make_shared<PGPeeringEvent>( | |
7425 | get_osdmap_epoch(), | |
7426 | get_osdmap_epoch(), | |
7427 | PG::RequestScrub(m->deep, m->repair)))); | |
7c673cae FG |
7428 | } |
7429 | ||
7430 | m->put(); | |
7431 | } | |
7432 | ||
11fdf7f2 TL |
7433 | void OSD::handle_fast_scrub(MOSDScrub2 *m) |
7434 | { | |
7435 | dout(10) << __func__ << " " << *m << dendl; | |
7436 | if (!require_mon_or_mgr_peer(m)) { | |
7437 | m->put(); | |
7438 | return; | |
7439 | } | |
7440 | if (m->fsid != monc->get_fsid()) { | |
7441 | dout(0) << __func__ << " fsid " << m->fsid << " != " << monc->get_fsid() | |
7442 | << dendl; | |
7443 | m->put(); | |
7444 | return; | |
7445 | } | |
7446 | for (auto pgid : m->scrub_pgs) { | |
7447 | enqueue_peering_evt( | |
7448 | pgid, | |
7449 | PGPeeringEventRef( | |
7450 | std::make_shared<PGPeeringEvent>( | |
7451 | m->epoch, | |
7452 | m->epoch, | |
7453 | PG::RequestScrub(m->deep, m->repair)))); | |
7454 | } | |
7455 | m->put(); | |
7456 | } | |
7457 | ||
7c673cae FG |
7458 | bool OSD::scrub_random_backoff() |
7459 | { | |
7460 | bool coin_flip = (rand() / (double)RAND_MAX >= | |
7461 | cct->_conf->osd_scrub_backoff_ratio); | |
7462 | if (!coin_flip) { | |
7463 | dout(20) << "scrub_random_backoff lost coin flip, randomly backing off" << dendl; | |
7464 | return true; | |
7465 | } | |
7466 | return false; | |
7467 | } | |
7468 | ||
7469 | OSDService::ScrubJob::ScrubJob(CephContext* cct, | |
7470 | const spg_t& pg, const utime_t& timestamp, | |
7471 | double pool_scrub_min_interval, | |
7472 | double pool_scrub_max_interval, bool must) | |
7473 | : cct(cct), | |
7474 | pgid(pg), | |
7475 | sched_time(timestamp), | |
7476 | deadline(timestamp) | |
7477 | { | |
7478 | // if not explicitly requested, postpone the scrub with a random delay | |
7479 | if (!must) { | |
7480 | double scrub_min_interval = pool_scrub_min_interval > 0 ? | |
7481 | pool_scrub_min_interval : cct->_conf->osd_scrub_min_interval; | |
7482 | double scrub_max_interval = pool_scrub_max_interval > 0 ? | |
7483 | pool_scrub_max_interval : cct->_conf->osd_scrub_max_interval; | |
7484 | ||
7485 | sched_time += scrub_min_interval; | |
7486 | double r = rand() / (double)RAND_MAX; | |
7487 | sched_time += | |
7488 | scrub_min_interval * cct->_conf->osd_scrub_interval_randomize_ratio * r; | |
11fdf7f2 TL |
7489 | if (scrub_max_interval == 0) { |
7490 | deadline = utime_t(); | |
7491 | } else { | |
7492 | deadline += scrub_max_interval; | |
7493 | } | |
7494 | ||
7c673cae FG |
7495 | } |
7496 | } | |
7497 | ||
7498 | bool OSDService::ScrubJob::ScrubJob::operator<(const OSDService::ScrubJob& rhs) const { | |
7499 | if (sched_time < rhs.sched_time) | |
7500 | return true; | |
7501 | if (sched_time > rhs.sched_time) | |
7502 | return false; | |
7503 | return pgid < rhs.pgid; | |
7504 | } | |
7505 | ||
7506 | bool OSD::scrub_time_permit(utime_t now) | |
7507 | { | |
7508 | struct tm bdt; | |
7509 | time_t tt = now.sec(); | |
7510 | localtime_r(&tt, &bdt); | |
28e407b8 AA |
7511 | |
7512 | bool day_permit = false; | |
7513 | if (cct->_conf->osd_scrub_begin_week_day < cct->_conf->osd_scrub_end_week_day) { | |
7514 | if (bdt.tm_wday >= cct->_conf->osd_scrub_begin_week_day && bdt.tm_wday < cct->_conf->osd_scrub_end_week_day) { | |
7515 | day_permit = true; | |
7516 | } | |
7517 | } else { | |
7518 | if (bdt.tm_wday >= cct->_conf->osd_scrub_begin_week_day || bdt.tm_wday < cct->_conf->osd_scrub_end_week_day) { | |
7519 | day_permit = true; | |
7520 | } | |
7521 | } | |
7522 | ||
7523 | if (!day_permit) { | |
7524 | dout(20) << __func__ << " should run between week day " << cct->_conf->osd_scrub_begin_week_day | |
7525 | << " - " << cct->_conf->osd_scrub_end_week_day | |
7526 | << " now " << bdt.tm_wday << " = no" << dendl; | |
7527 | return false; | |
7528 | } | |
7529 | ||
7c673cae FG |
7530 | bool time_permit = false; |
7531 | if (cct->_conf->osd_scrub_begin_hour < cct->_conf->osd_scrub_end_hour) { | |
7532 | if (bdt.tm_hour >= cct->_conf->osd_scrub_begin_hour && bdt.tm_hour < cct->_conf->osd_scrub_end_hour) { | |
7533 | time_permit = true; | |
7534 | } | |
7535 | } else { | |
7536 | if (bdt.tm_hour >= cct->_conf->osd_scrub_begin_hour || bdt.tm_hour < cct->_conf->osd_scrub_end_hour) { | |
7537 | time_permit = true; | |
7538 | } | |
7539 | } | |
7540 | if (!time_permit) { | |
7541 | dout(20) << __func__ << " should run between " << cct->_conf->osd_scrub_begin_hour | |
7542 | << " - " << cct->_conf->osd_scrub_end_hour | |
7543 | << " now " << bdt.tm_hour << " = no" << dendl; | |
7544 | } else { | |
7545 | dout(20) << __func__ << " should run between " << cct->_conf->osd_scrub_begin_hour | |
7546 | << " - " << cct->_conf->osd_scrub_end_hour | |
7547 | << " now " << bdt.tm_hour << " = yes" << dendl; | |
7548 | } | |
7549 | return time_permit; | |
7550 | } | |
7551 | ||
7552 | bool OSD::scrub_load_below_threshold() | |
7553 | { | |
7554 | double loadavgs[3]; | |
7555 | if (getloadavg(loadavgs, 3) != 3) { | |
7556 | dout(10) << __func__ << " couldn't read loadavgs\n" << dendl; | |
7557 | return false; | |
7558 | } | |
7559 | ||
7560 | // allow scrub if below configured threshold | |
91327a77 AA |
7561 | long cpus = sysconf(_SC_NPROCESSORS_ONLN); |
7562 | double loadavg_per_cpu = cpus > 0 ? loadavgs[0] / cpus : loadavgs[0]; | |
7563 | if (loadavg_per_cpu < cct->_conf->osd_scrub_load_threshold) { | |
7564 | dout(20) << __func__ << " loadavg per cpu " << loadavg_per_cpu | |
7c673cae FG |
7565 | << " < max " << cct->_conf->osd_scrub_load_threshold |
7566 | << " = yes" << dendl; | |
7567 | return true; | |
7568 | } | |
7569 | ||
7570 | // allow scrub if below daily avg and currently decreasing | |
7571 | if (loadavgs[0] < daily_loadavg && loadavgs[0] < loadavgs[2]) { | |
7572 | dout(20) << __func__ << " loadavg " << loadavgs[0] | |
7573 | << " < daily_loadavg " << daily_loadavg | |
7574 | << " and < 15m avg " << loadavgs[2] | |
7575 | << " = yes" << dendl; | |
7576 | return true; | |
7577 | } | |
7578 | ||
7579 | dout(20) << __func__ << " loadavg " << loadavgs[0] | |
7580 | << " >= max " << cct->_conf->osd_scrub_load_threshold | |
7581 | << " and ( >= daily_loadavg " << daily_loadavg | |
7582 | << " or >= 15m avg " << loadavgs[2] | |
7583 | << ") = no" << dendl; | |
7584 | return false; | |
7585 | } | |
7586 | ||
7587 | void OSD::sched_scrub() | |
7588 | { | |
7589 | // if not permitted, fail fast | |
7590 | if (!service.can_inc_scrubs_pending()) { | |
7591 | return; | |
7592 | } | |
b5b8bbf5 FG |
7593 | if (!cct->_conf->osd_scrub_during_recovery && service.is_recovery_active()) { |
7594 | dout(20) << __func__ << " not scheduling scrubs due to active recovery" << dendl; | |
7595 | return; | |
7596 | } | |
7597 | ||
7c673cae FG |
7598 | |
7599 | utime_t now = ceph_clock_now(); | |
7600 | bool time_permit = scrub_time_permit(now); | |
7601 | bool load_is_low = scrub_load_below_threshold(); | |
7602 | dout(20) << "sched_scrub load_is_low=" << (int)load_is_low << dendl; | |
7603 | ||
7604 | OSDService::ScrubJob scrub; | |
7605 | if (service.first_scrub_stamp(&scrub)) { | |
7606 | do { | |
7607 | dout(30) << "sched_scrub examine " << scrub.pgid << " at " << scrub.sched_time << dendl; | |
7608 | ||
7609 | if (scrub.sched_time > now) { | |
7610 | // save ourselves some effort | |
7611 | dout(10) << "sched_scrub " << scrub.pgid << " scheduled at " << scrub.sched_time | |
7612 | << " > " << now << dendl; | |
7613 | break; | |
7614 | } | |
7615 | ||
11fdf7f2 | 7616 | if ((scrub.deadline.is_zero() || scrub.deadline >= now) && !(time_permit && load_is_low)) { |
7c673cae FG |
7617 | dout(10) << __func__ << " not scheduling scrub for " << scrub.pgid << " due to " |
7618 | << (!time_permit ? "time not permit" : "high load") << dendl; | |
7619 | continue; | |
7620 | } | |
7621 | ||
11fdf7f2 | 7622 | PGRef pg = _lookup_lock_pg(scrub.pgid); |
7c673cae FG |
7623 | if (!pg) |
7624 | continue; | |
11fdf7f2 TL |
7625 | dout(10) << "sched_scrub scrubbing " << scrub.pgid << " at " << scrub.sched_time |
7626 | << (pg->get_must_scrub() ? ", explicitly requested" : | |
7627 | (load_is_low ? ", load_is_low" : " deadline < now")) | |
7628 | << dendl; | |
7629 | if (pg->sched_scrub()) { | |
7630 | pg->unlock(); | |
7631 | break; | |
7c673cae FG |
7632 | } |
7633 | pg->unlock(); | |
7634 | } while (service.next_scrub_stamp(scrub, &scrub)); | |
7635 | } | |
7636 | dout(20) << "sched_scrub done" << dendl; | |
7637 | } | |
7638 | ||
11fdf7f2 TL |
7639 | MPGStats* OSD::collect_pg_stats() |
7640 | { | |
7641 | // This implementation unconditionally sends every is_primary PG's | |
7642 | // stats every time we're called. This has equivalent cost to the | |
7643 | // previous implementation's worst case where all PGs are busy and | |
7644 | // their stats are always enqueued for sending. | |
7645 | RWLock::RLocker l(map_lock); | |
7646 | ||
7647 | utime_t had_for = ceph_clock_now() - had_map_since; | |
7648 | osd_stat_t cur_stat = service.get_osd_stat(); | |
7649 | cur_stat.os_perf_stat = store->get_cur_stats(); | |
7650 | ||
7651 | auto m = new MPGStats(monc->get_fsid(), osdmap->get_epoch(), had_for); | |
7652 | m->osd_stat = cur_stat; | |
7653 | ||
7654 | std::lock_guard lec{min_last_epoch_clean_lock}; | |
7655 | min_last_epoch_clean = osdmap->get_epoch(); | |
7656 | min_last_epoch_clean_pgs.clear(); | |
7657 | ||
7658 | std::set<int64_t> pool_set; | |
7659 | vector<PGRef> pgs; | |
7660 | _get_pgs(&pgs); | |
7661 | for (auto& pg : pgs) { | |
7662 | auto pool = pg->pg_id.pgid.pool(); | |
7663 | pool_set.emplace((int64_t)pool); | |
7664 | if (!pg->is_primary()) { | |
7665 | continue; | |
7666 | } | |
7667 | pg->get_pg_stats([&](const pg_stat_t& s, epoch_t lec) { | |
7668 | m->pg_stat[pg->pg_id.pgid] = s; | |
7669 | min_last_epoch_clean = min(min_last_epoch_clean, lec); | |
7670 | min_last_epoch_clean_pgs.push_back(pg->pg_id.pgid); | |
7671 | }); | |
7672 | } | |
7673 | store_statfs_t st; | |
7674 | for (auto p : pool_set) { | |
7675 | int r = store->pool_statfs(p, &st); | |
7676 | if (r == -ENOTSUP) { | |
7677 | break; | |
7678 | } else { | |
7679 | assert(r >= 0); | |
7680 | m->pool_stat[p] = st; | |
7681 | } | |
7682 | } | |
7c673cae | 7683 | |
11fdf7f2 TL |
7684 | return m; |
7685 | } | |
7c673cae | 7686 | |
11fdf7f2 | 7687 | vector<DaemonHealthMetric> OSD::get_health_metrics() |
b32b8144 | 7688 | { |
11fdf7f2 TL |
7689 | vector<DaemonHealthMetric> metrics; |
7690 | { | |
7691 | utime_t oldest_secs; | |
7692 | const utime_t now = ceph_clock_now(); | |
7693 | auto too_old = now; | |
7694 | too_old -= cct->_conf.get_val<double>("osd_op_complaint_time"); | |
7695 | int slow = 0; | |
7696 | TrackedOpRef oldest_op; | |
7697 | auto count_slow_ops = [&](TrackedOp& op) { | |
7698 | if (op.get_initiated() < too_old) { | |
7699 | lgeneric_subdout(cct,osd,20) << "slow op " << op.get_desc() | |
7700 | << " initiated " | |
7701 | << op.get_initiated() << dendl; | |
7702 | slow++; | |
7703 | if (!oldest_op || op.get_initiated() < oldest_op->get_initiated()) { | |
7704 | oldest_op = &op; | |
7705 | } | |
7706 | return true; | |
7707 | } else { | |
7708 | return false; | |
7709 | } | |
7710 | }; | |
7711 | if (op_tracker.visit_ops_in_flight(&oldest_secs, count_slow_ops)) { | |
7712 | if (slow) { | |
7713 | derr << __func__ << " reporting " << slow << " slow ops, oldest is " | |
7714 | << oldest_op->get_desc() << dendl; | |
7715 | } | |
7716 | metrics.emplace_back(daemon_metric::SLOW_OPS, slow, oldest_secs); | |
7717 | } else { | |
7718 | // no news is not good news. | |
7719 | metrics.emplace_back(daemon_metric::SLOW_OPS, 0, 0); | |
7720 | } | |
7721 | } | |
7722 | { | |
7723 | std::lock_guard l(pending_creates_lock); | |
7724 | auto n_primaries = pending_creates_from_mon; | |
7725 | for (const auto& create : pending_creates_from_osd) { | |
7726 | if (create.second) { | |
7727 | n_primaries++; | |
7728 | } | |
b32b8144 | 7729 | } |
11fdf7f2 | 7730 | metrics.emplace_back(daemon_metric::PENDING_CREATING_PGS, n_primaries); |
b32b8144 | 7731 | } |
b32b8144 FG |
7732 | return metrics; |
7733 | } | |
7734 | ||
7c673cae FG |
7735 | // ===================================================== |
7736 | // MAP | |
7737 | ||
7738 | void OSD::wait_for_new_map(OpRequestRef op) | |
7739 | { | |
7740 | // ask? | |
7741 | if (waiting_for_osdmap.empty()) { | |
7742 | osdmap_subscribe(osdmap->get_epoch() + 1, false); | |
7743 | } | |
7744 | ||
7745 | logger->inc(l_osd_waiting_for_map); | |
7746 | waiting_for_osdmap.push_back(op); | |
7747 | op->mark_delayed("wait for new map"); | |
7748 | } | |
7749 | ||
7750 | ||
7751 | /** update_map | |
7752 | * assimilate new OSDMap(s). scan pgs, etc. | |
7753 | */ | |
7754 | ||
7755 | void OSD::note_down_osd(int peer) | |
7756 | { | |
11fdf7f2 TL |
7757 | ceph_assert(osd_lock.is_locked()); |
7758 | cluster_messenger->mark_down_addrs(osdmap->get_cluster_addrs(peer)); | |
7c673cae FG |
7759 | |
7760 | heartbeat_lock.Lock(); | |
7761 | failure_queue.erase(peer); | |
7762 | failure_pending.erase(peer); | |
7763 | map<int,HeartbeatInfo>::iterator p = heartbeat_peers.find(peer); | |
7764 | if (p != heartbeat_peers.end()) { | |
7765 | p->second.con_back->mark_down(); | |
7766 | if (p->second.con_front) { | |
7767 | p->second.con_front->mark_down(); | |
7768 | } | |
7769 | heartbeat_peers.erase(p); | |
7770 | } | |
7771 | heartbeat_lock.Unlock(); | |
7772 | } | |
7773 | ||
7774 | void OSD::note_up_osd(int peer) | |
7775 | { | |
7776 | service.forget_peer_epoch(peer, osdmap->get_epoch() - 1); | |
7777 | heartbeat_set_peers_need_update(); | |
7778 | } | |
7779 | ||
7780 | struct C_OnMapCommit : public Context { | |
7781 | OSD *osd; | |
7782 | epoch_t first, last; | |
7783 | MOSDMap *msg; | |
7784 | C_OnMapCommit(OSD *o, epoch_t f, epoch_t l, MOSDMap *m) | |
7785 | : osd(o), first(f), last(l), msg(m) {} | |
7786 | void finish(int r) override { | |
7787 | osd->_committed_osd_maps(first, last, msg); | |
7788 | msg->put(); | |
7789 | } | |
7790 | }; | |
7791 | ||
7c673cae FG |
7792 | void OSD::osdmap_subscribe(version_t epoch, bool force_request) |
7793 | { | |
11fdf7f2 | 7794 | std::lock_guard l(osdmap_subscribe_lock); |
181888fb | 7795 | if (latest_subscribed_epoch >= epoch && !force_request) |
7c673cae FG |
7796 | return; |
7797 | ||
11fdf7f2 | 7798 | latest_subscribed_epoch = std::max<uint64_t>(epoch, latest_subscribed_epoch); |
181888fb | 7799 | |
7c673cae FG |
7800 | if (monc->sub_want_increment("osdmap", epoch, CEPH_SUBSCRIBE_ONETIME) || |
7801 | force_request) { | |
7802 | monc->renew_subs(); | |
7803 | } | |
7804 | } | |
7805 | ||
7806 | void OSD::trim_maps(epoch_t oldest, int nreceived, bool skip_maps) | |
7807 | { | |
7808 | epoch_t min = std::min(oldest, service.map_cache.cached_key_lower_bound()); | |
7809 | if (min <= superblock.oldest_map) | |
7810 | return; | |
7811 | ||
7812 | int num = 0; | |
7813 | ObjectStore::Transaction t; | |
7814 | for (epoch_t e = superblock.oldest_map; e < min; ++e) { | |
7815 | dout(20) << " removing old osdmap epoch " << e << dendl; | |
7816 | t.remove(coll_t::meta(), get_osdmap_pobject_name(e)); | |
7817 | t.remove(coll_t::meta(), get_inc_osdmap_pobject_name(e)); | |
7818 | superblock.oldest_map = e + 1; | |
7819 | num++; | |
7820 | if (num >= cct->_conf->osd_target_transaction_size && num >= nreceived) { | |
7821 | service.publish_superblock(superblock); | |
7822 | write_superblock(t); | |
11fdf7f2 TL |
7823 | int tr = store->queue_transaction(service.meta_ch, std::move(t), nullptr); |
7824 | ceph_assert(tr == 0); | |
7c673cae FG |
7825 | num = 0; |
7826 | if (!skip_maps) { | |
7827 | // skip_maps leaves us with a range of old maps if we fail to remove all | |
7828 | // of them before moving superblock.oldest_map forward to the first map | |
7829 | // in the incoming MOSDMap msg. so we should continue removing them in | |
7830 | // this case, even we could do huge series of delete transactions all at | |
7831 | // once. | |
7832 | break; | |
7833 | } | |
7834 | } | |
7835 | } | |
7836 | if (num > 0) { | |
7837 | service.publish_superblock(superblock); | |
7838 | write_superblock(t); | |
11fdf7f2 TL |
7839 | int tr = store->queue_transaction(service.meta_ch, std::move(t), nullptr); |
7840 | ceph_assert(tr == 0); | |
7c673cae FG |
7841 | } |
7842 | // we should not remove the cached maps | |
11fdf7f2 | 7843 | ceph_assert(min <= service.map_cache.cached_key_lower_bound()); |
7c673cae FG |
7844 | } |
7845 | ||
7846 | void OSD::handle_osd_map(MOSDMap *m) | |
7847 | { | |
11fdf7f2 TL |
7848 | // wait for pgs to catch up |
7849 | { | |
7850 | // we extend the map cache pins to accomodate pgs slow to consume maps | |
7851 | // for some period, until we hit the max_lag_factor bound, at which point | |
7852 | // we block here to stop injesting more maps than they are able to keep | |
7853 | // up with. | |
7854 | epoch_t max_lag = cct->_conf->osd_map_cache_size * | |
7855 | m_osd_pg_epoch_max_lag_factor; | |
7856 | ceph_assert(max_lag > 0); | |
7857 | epoch_t osd_min = 0; | |
7858 | for (auto shard : shards) { | |
7859 | epoch_t min = shard->get_min_pg_epoch(); | |
7860 | if (osd_min == 0 || min < osd_min) { | |
7861 | osd_min = min; | |
7862 | } | |
7863 | } | |
7864 | if (osd_min > 0 && | |
7865 | osdmap->get_epoch() > max_lag && | |
7866 | osdmap->get_epoch() - max_lag > osd_min) { | |
7867 | epoch_t need = osdmap->get_epoch() - max_lag; | |
7868 | dout(10) << __func__ << " waiting for pgs to catch up (need " << need | |
7869 | << " max_lag " << max_lag << ")" << dendl; | |
7870 | for (auto shard : shards) { | |
7871 | epoch_t min = shard->get_min_pg_epoch(); | |
7872 | if (need > min) { | |
7873 | dout(10) << __func__ << " waiting for pgs to consume " << need | |
7874 | << " (shard " << shard->shard_id << " min " << min | |
7875 | << ", map cache is " << cct->_conf->osd_map_cache_size | |
7876 | << ", max_lag_factor " << m_osd_pg_epoch_max_lag_factor | |
7877 | << ")" << dendl; | |
7878 | unlock_guard unlock{osd_lock}; | |
7879 | shard->wait_min_pg_epoch(need); | |
7880 | } | |
7881 | } | |
7882 | } | |
7883 | } | |
7884 | ||
7885 | ceph_assert(osd_lock.is_locked()); | |
7886 | map<epoch_t,OSDMapRef> added_maps; | |
7887 | map<epoch_t,bufferlist> added_maps_bl; | |
7c673cae FG |
7888 | if (m->fsid != monc->get_fsid()) { |
7889 | dout(0) << "handle_osd_map fsid " << m->fsid << " != " | |
7890 | << monc->get_fsid() << dendl; | |
7891 | m->put(); | |
7892 | return; | |
7893 | } | |
7894 | if (is_initializing()) { | |
7895 | dout(0) << "ignoring osdmap until we have initialized" << dendl; | |
7896 | m->put(); | |
7897 | return; | |
7898 | } | |
7899 | ||
11fdf7f2 TL |
7900 | auto priv = m->get_connection()->get_priv(); |
7901 | if (auto session = static_cast<Session *>(priv.get()); | |
7902 | session && !(session->entity_name.is_mon() || | |
7c673cae FG |
7903 | session->entity_name.is_osd())) { |
7904 | //not enough perms! | |
7905 | dout(10) << "got osd map from Session " << session | |
7906 | << " which we can't take maps from (not a mon or osd)" << dendl; | |
7907 | m->put(); | |
7c673cae FG |
7908 | return; |
7909 | } | |
7c673cae FG |
7910 | |
7911 | // share with the objecter | |
7912 | if (!is_preboot()) | |
7913 | service.objecter->handle_osd_map(m); | |
7914 | ||
7915 | epoch_t first = m->get_first(); | |
7916 | epoch_t last = m->get_last(); | |
7917 | dout(3) << "handle_osd_map epochs [" << first << "," << last << "], i have " | |
7918 | << superblock.newest_map | |
7919 | << ", src has [" << m->oldest_map << "," << m->newest_map << "]" | |
7920 | << dendl; | |
7921 | ||
7922 | logger->inc(l_osd_map); | |
7923 | logger->inc(l_osd_mape, last - first + 1); | |
7924 | if (first <= superblock.newest_map) | |
7925 | logger->inc(l_osd_mape_dup, superblock.newest_map - first + 1); | |
7926 | if (service.max_oldest_map < m->oldest_map) { | |
7927 | service.max_oldest_map = m->oldest_map; | |
11fdf7f2 | 7928 | ceph_assert(service.max_oldest_map >= superblock.oldest_map); |
7c673cae FG |
7929 | } |
7930 | ||
7931 | // make sure there is something new, here, before we bother flushing | |
7932 | // the queues and such | |
7933 | if (last <= superblock.newest_map) { | |
7934 | dout(10) << " no new maps here, dropping" << dendl; | |
7935 | m->put(); | |
7936 | return; | |
7937 | } | |
7938 | ||
7939 | // missing some? | |
7940 | bool skip_maps = false; | |
7941 | if (first > superblock.newest_map + 1) { | |
7942 | dout(10) << "handle_osd_map message skips epochs " | |
7943 | << superblock.newest_map + 1 << ".." << (first-1) << dendl; | |
7944 | if (m->oldest_map <= superblock.newest_map + 1) { | |
7945 | osdmap_subscribe(superblock.newest_map + 1, false); | |
7946 | m->put(); | |
7947 | return; | |
7948 | } | |
7949 | // always try to get the full range of maps--as many as we can. this | |
7950 | // 1- is good to have | |
7951 | // 2- is at present the only way to ensure that we get a *full* map as | |
7952 | // the first map! | |
7953 | if (m->oldest_map < first) { | |
7954 | osdmap_subscribe(m->oldest_map - 1, true); | |
7955 | m->put(); | |
7956 | return; | |
7957 | } | |
7958 | skip_maps = true; | |
7959 | } | |
7960 | ||
7961 | ObjectStore::Transaction t; | |
7962 | uint64_t txn_size = 0; | |
7963 | ||
7964 | // store new maps: queue for disk and put in the osdmap cache | |
11fdf7f2 | 7965 | epoch_t start = std::max(superblock.newest_map + 1, first); |
7c673cae FG |
7966 | for (epoch_t e = start; e <= last; e++) { |
7967 | if (txn_size >= t.get_num_bytes()) { | |
7968 | derr << __func__ << " transaction size overflowed" << dendl; | |
11fdf7f2 | 7969 | ceph_assert(txn_size < t.get_num_bytes()); |
7c673cae FG |
7970 | } |
7971 | txn_size = t.get_num_bytes(); | |
7972 | map<epoch_t,bufferlist>::iterator p; | |
7973 | p = m->maps.find(e); | |
7974 | if (p != m->maps.end()) { | |
7975 | dout(10) << "handle_osd_map got full map for epoch " << e << dendl; | |
7976 | OSDMap *o = new OSDMap; | |
7977 | bufferlist& bl = p->second; | |
7978 | ||
7979 | o->decode(bl); | |
7980 | ||
7981 | ghobject_t fulloid = get_osdmap_pobject_name(e); | |
7982 | t.write(coll_t::meta(), fulloid, 0, bl.length(), bl); | |
11fdf7f2 TL |
7983 | added_maps[e] = add_map(o); |
7984 | added_maps_bl[e] = bl; | |
7c673cae FG |
7985 | got_full_map(e); |
7986 | continue; | |
7987 | } | |
7988 | ||
7989 | p = m->incremental_maps.find(e); | |
7990 | if (p != m->incremental_maps.end()) { | |
7991 | dout(10) << "handle_osd_map got inc map for epoch " << e << dendl; | |
7992 | bufferlist& bl = p->second; | |
7993 | ghobject_t oid = get_inc_osdmap_pobject_name(e); | |
7994 | t.write(coll_t::meta(), oid, 0, bl.length(), bl); | |
7c673cae FG |
7995 | |
7996 | OSDMap *o = new OSDMap; | |
7997 | if (e > 1) { | |
7998 | bufferlist obl; | |
7999 | bool got = get_map_bl(e - 1, obl); | |
11fdf7f2 TL |
8000 | if (!got) { |
8001 | auto p = added_maps_bl.find(e - 1); | |
8002 | ceph_assert(p != added_maps_bl.end()); | |
8003 | obl = p->second; | |
8004 | } | |
7c673cae FG |
8005 | o->decode(obl); |
8006 | } | |
8007 | ||
8008 | OSDMap::Incremental inc; | |
11fdf7f2 | 8009 | auto p = bl.cbegin(); |
7c673cae FG |
8010 | inc.decode(p); |
8011 | if (o->apply_incremental(inc) < 0) { | |
8012 | derr << "ERROR: bad fsid? i have " << osdmap->get_fsid() << " and inc has " << inc.fsid << dendl; | |
11fdf7f2 | 8013 | ceph_abort_msg("bad fsid"); |
7c673cae FG |
8014 | } |
8015 | ||
8016 | bufferlist fbl; | |
8017 | o->encode(fbl, inc.encode_features | CEPH_FEATURE_RESERVED); | |
8018 | ||
8019 | bool injected_failure = false; | |
8020 | if (cct->_conf->osd_inject_bad_map_crc_probability > 0 && | |
8021 | (rand() % 10000) < cct->_conf->osd_inject_bad_map_crc_probability*10000.0) { | |
8022 | derr << __func__ << " injecting map crc failure" << dendl; | |
8023 | injected_failure = true; | |
8024 | } | |
8025 | ||
8026 | if ((inc.have_crc && o->get_crc() != inc.full_crc) || injected_failure) { | |
8027 | dout(2) << "got incremental " << e | |
8028 | << " but failed to encode full with correct crc; requesting" | |
8029 | << dendl; | |
8030 | clog->warn() << "failed to encode map e" << e << " with expected crc"; | |
8031 | dout(20) << "my encoded map was:\n"; | |
8032 | fbl.hexdump(*_dout); | |
8033 | *_dout << dendl; | |
8034 | delete o; | |
8035 | request_full_map(e, last); | |
8036 | last = e - 1; | |
8037 | break; | |
8038 | } | |
8039 | got_full_map(e); | |
8040 | ||
8041 | ghobject_t fulloid = get_osdmap_pobject_name(e); | |
8042 | t.write(coll_t::meta(), fulloid, 0, fbl.length(), fbl); | |
11fdf7f2 TL |
8043 | added_maps[e] = add_map(o); |
8044 | added_maps_bl[e] = fbl; | |
7c673cae FG |
8045 | continue; |
8046 | } | |
8047 | ||
11fdf7f2 | 8048 | ceph_abort_msg("MOSDMap lied about what maps it had?"); |
7c673cae FG |
8049 | } |
8050 | ||
8051 | // even if this map isn't from a mon, we may have satisfied our subscription | |
8052 | monc->sub_got("osdmap", last); | |
8053 | ||
8054 | if (!m->maps.empty() && requested_full_first) { | |
8055 | dout(10) << __func__ << " still missing full maps " << requested_full_first | |
8056 | << ".." << requested_full_last << dendl; | |
8057 | rerequest_full_maps(); | |
8058 | } | |
8059 | ||
7c673cae FG |
8060 | if (superblock.oldest_map) { |
8061 | // make sure we at least keep pace with incoming maps | |
8062 | trim_maps(m->oldest_map, last - first + 1, skip_maps); | |
11fdf7f2 | 8063 | pg_num_history.prune(superblock.oldest_map); |
7c673cae FG |
8064 | } |
8065 | ||
8066 | if (!superblock.oldest_map || skip_maps) | |
8067 | superblock.oldest_map = first; | |
8068 | superblock.newest_map = last; | |
8069 | superblock.current_epoch = last; | |
8070 | ||
8071 | // note in the superblock that we were clean thru the prior epoch | |
8072 | epoch_t boot_epoch = service.get_boot_epoch(); | |
8073 | if (boot_epoch && boot_epoch >= superblock.mounted) { | |
8074 | superblock.mounted = boot_epoch; | |
8075 | superblock.clean_thru = last; | |
8076 | } | |
8077 | ||
11fdf7f2 TL |
8078 | // check for pg_num changes and deleted pools |
8079 | OSDMapRef lastmap; | |
8080 | for (auto& i : added_maps) { | |
8081 | if (!lastmap) { | |
8082 | if (!(lastmap = service.try_get_map(i.first - 1))) { | |
8083 | dout(10) << __func__ << " can't get previous map " << i.first - 1 | |
8084 | << " probably first start of this osd" << dendl; | |
8085 | continue; | |
8086 | } | |
8087 | } | |
8088 | ceph_assert(lastmap->get_epoch() + 1 == i.second->get_epoch()); | |
8089 | for (auto& j : lastmap->get_pools()) { | |
8090 | if (!i.second->have_pg_pool(j.first)) { | |
8091 | pg_num_history.log_pool_delete(i.first, j.first); | |
8092 | dout(10) << __func__ << " recording final pg_pool_t for pool " | |
8093 | << j.first << dendl; | |
8094 | // this information is needed by _make_pg() if have to restart before | |
8095 | // the pool is deleted and need to instantiate a new (zombie) PG[Pool]. | |
8096 | ghobject_t obj = make_final_pool_info_oid(j.first); | |
8097 | bufferlist bl; | |
8098 | encode(j.second, bl, CEPH_FEATURES_ALL); | |
8099 | string name = lastmap->get_pool_name(j.first); | |
8100 | encode(name, bl); | |
8101 | map<string,string> profile; | |
8102 | if (lastmap->get_pg_pool(j.first)->is_erasure()) { | |
8103 | profile = lastmap->get_erasure_code_profile( | |
8104 | lastmap->get_pg_pool(j.first)->erasure_code_profile); | |
8105 | } | |
8106 | encode(profile, bl); | |
8107 | t.write(coll_t::meta(), obj, 0, bl.length(), bl); | |
8108 | service.store_deleted_pool_pg_num(j.first, j.second.get_pg_num()); | |
8109 | } else if (unsigned new_pg_num = i.second->get_pg_num(j.first); | |
8110 | new_pg_num != j.second.get_pg_num()) { | |
8111 | dout(10) << __func__ << " recording pool " << j.first << " pg_num " | |
8112 | << j.second.get_pg_num() << " -> " << new_pg_num << dendl; | |
8113 | pg_num_history.log_pg_num_change(i.first, j.first, new_pg_num); | |
8114 | } | |
8115 | } | |
8116 | for (auto& j : i.second->get_pools()) { | |
8117 | if (!lastmap->have_pg_pool(j.first)) { | |
8118 | dout(10) << __func__ << " recording new pool " << j.first << " pg_num " | |
8119 | << j.second.get_pg_num() << dendl; | |
8120 | pg_num_history.log_pg_num_change(i.first, j.first, | |
8121 | j.second.get_pg_num()); | |
8122 | } | |
8123 | } | |
8124 | lastmap = i.second; | |
8125 | } | |
8126 | pg_num_history.epoch = last; | |
8127 | { | |
8128 | bufferlist bl; | |
8129 | ::encode(pg_num_history, bl); | |
8130 | t.write(coll_t::meta(), make_pg_num_history_oid(), 0, bl.length(), bl); | |
8131 | dout(20) << __func__ << " pg_num_history " << pg_num_history << dendl; | |
8132 | } | |
8133 | ||
7c673cae FG |
8134 | // superblock and commit |
8135 | write_superblock(t); | |
11fdf7f2 | 8136 | t.register_on_commit(new C_OnMapCommit(this, start, last, m)); |
7c673cae | 8137 | store->queue_transaction( |
11fdf7f2 TL |
8138 | service.meta_ch, |
8139 | std::move(t)); | |
7c673cae FG |
8140 | service.publish_superblock(superblock); |
8141 | } | |
8142 | ||
8143 | void OSD::_committed_osd_maps(epoch_t first, epoch_t last, MOSDMap *m) | |
8144 | { | |
8145 | dout(10) << __func__ << " " << first << ".." << last << dendl; | |
8146 | if (is_stopping()) { | |
8147 | dout(10) << __func__ << " bailing, we are shutting down" << dendl; | |
8148 | return; | |
8149 | } | |
11fdf7f2 | 8150 | std::lock_guard l(osd_lock); |
31f18b77 FG |
8151 | if (is_stopping()) { |
8152 | dout(10) << __func__ << " bailing, we are shutting down" << dendl; | |
8153 | return; | |
8154 | } | |
7c673cae FG |
8155 | map_lock.get_write(); |
8156 | ||
8157 | bool do_shutdown = false; | |
8158 | bool do_restart = false; | |
8159 | bool network_error = false; | |
8160 | ||
8161 | // advance through the new maps | |
8162 | for (epoch_t cur = first; cur <= last; cur++) { | |
8163 | dout(10) << " advance to epoch " << cur | |
8164 | << " (<= last " << last | |
8165 | << " <= newest_map " << superblock.newest_map | |
8166 | << ")" << dendl; | |
8167 | ||
8168 | OSDMapRef newmap = get_map(cur); | |
11fdf7f2 | 8169 | ceph_assert(newmap); // we just cached it above! |
7c673cae FG |
8170 | |
8171 | // start blacklisting messages sent to peers that go down. | |
8172 | service.pre_publish_map(newmap); | |
8173 | ||
8174 | // kill connections to newly down osds | |
8175 | bool waited_for_reservations = false; | |
8176 | set<int> old; | |
8177 | osdmap->get_all_osds(old); | |
8178 | for (set<int>::iterator p = old.begin(); p != old.end(); ++p) { | |
8179 | if (*p != whoami && | |
8180 | osdmap->is_up(*p) && // in old map | |
8181 | newmap->is_down(*p)) { // but not the new one | |
8182 | if (!waited_for_reservations) { | |
8183 | service.await_reserved_maps(); | |
8184 | waited_for_reservations = true; | |
8185 | } | |
8186 | note_down_osd(*p); | |
8187 | } else if (*p != whoami && | |
8188 | osdmap->is_down(*p) && | |
8189 | newmap->is_up(*p)) { | |
8190 | note_up_osd(*p); | |
8191 | } | |
8192 | } | |
8193 | ||
31f18b77 FG |
8194 | if ((osdmap->test_flag(CEPH_OSDMAP_NOUP) != |
8195 | newmap->test_flag(CEPH_OSDMAP_NOUP)) || | |
8196 | (osdmap->is_noup(whoami) != newmap->is_noup(whoami))) { | |
7c673cae FG |
8197 | dout(10) << __func__ << " NOUP flag changed in " << newmap->get_epoch() |
8198 | << dendl; | |
8199 | if (is_booting()) { | |
8200 | // this captures the case where we sent the boot message while | |
8201 | // NOUP was being set on the mon and our boot request was | |
8202 | // dropped, and then later it is cleared. it imperfectly | |
8203 | // handles the case where our original boot message was not | |
8204 | // dropped and we restart even though we might have booted, but | |
8205 | // that is harmless (boot will just take slightly longer). | |
8206 | do_restart = true; | |
8207 | } | |
8208 | } | |
8209 | ||
8210 | osdmap = newmap; | |
8211 | epoch_t up_epoch; | |
8212 | epoch_t boot_epoch; | |
8213 | service.retrieve_epochs(&boot_epoch, &up_epoch, NULL); | |
8214 | if (!up_epoch && | |
8215 | osdmap->is_up(whoami) && | |
11fdf7f2 | 8216 | osdmap->get_addrs(whoami) == client_messenger->get_myaddrs()) { |
7c673cae FG |
8217 | up_epoch = osdmap->get_epoch(); |
8218 | dout(10) << "up_epoch is " << up_epoch << dendl; | |
8219 | if (!boot_epoch) { | |
8220 | boot_epoch = osdmap->get_epoch(); | |
8221 | dout(10) << "boot_epoch is " << boot_epoch << dendl; | |
8222 | } | |
8223 | service.set_epochs(&boot_epoch, &up_epoch, NULL); | |
8224 | } | |
8225 | } | |
8226 | ||
8227 | had_map_since = ceph_clock_now(); | |
8228 | ||
8229 | epoch_t _bind_epoch = service.get_bind_epoch(); | |
8230 | if (osdmap->is_up(whoami) && | |
11fdf7f2 TL |
8231 | osdmap->get_addrs(whoami).legacy_equals( |
8232 | client_messenger->get_myaddrs()) && | |
7c673cae FG |
8233 | _bind_epoch < osdmap->get_up_from(whoami)) { |
8234 | ||
8235 | if (is_booting()) { | |
8236 | dout(1) << "state: booting -> active" << dendl; | |
8237 | set_state(STATE_ACTIVE); | |
11fdf7f2 | 8238 | do_restart = false; |
7c673cae FG |
8239 | |
8240 | // set incarnation so that osd_reqid_t's we generate for our | |
8241 | // objecter requests are unique across restarts. | |
8242 | service.objecter->set_client_incarnation(osdmap->get_epoch()); | |
11fdf7f2 | 8243 | cancel_pending_failures(); |
7c673cae FG |
8244 | } |
8245 | } | |
8246 | ||
8247 | if (osdmap->get_epoch() > 0 && | |
8248 | is_active()) { | |
8249 | if (!osdmap->exists(whoami)) { | |
8250 | dout(0) << "map says i do not exist. shutting down." << dendl; | |
8251 | do_shutdown = true; // don't call shutdown() while we have | |
8252 | // everything paused | |
8253 | } else if (!osdmap->is_up(whoami) || | |
11fdf7f2 TL |
8254 | !osdmap->get_addrs(whoami).legacy_equals( |
8255 | client_messenger->get_myaddrs()) || | |
8256 | !osdmap->get_cluster_addrs(whoami).legacy_equals( | |
8257 | cluster_messenger->get_myaddrs()) || | |
8258 | !osdmap->get_hb_back_addrs(whoami).legacy_equals( | |
8259 | hb_back_server_messenger->get_myaddrs()) || | |
8260 | !osdmap->get_hb_front_addrs(whoami).legacy_equals( | |
8261 | hb_front_server_messenger->get_myaddrs())) { | |
7c673cae FG |
8262 | if (!osdmap->is_up(whoami)) { |
8263 | if (service.is_preparing_to_stop() || service.is_stopping()) { | |
8264 | service.got_stop_ack(); | |
8265 | } else { | |
c07f9fc5 FG |
8266 | clog->warn() << "Monitor daemon marked osd." << whoami << " down, " |
8267 | "but it is still running"; | |
8268 | clog->debug() << "map e" << osdmap->get_epoch() | |
8269 | << " wrongly marked me down at e" | |
8270 | << osdmap->get_down_at(whoami); | |
7c673cae | 8271 | } |
11fdf7f2 TL |
8272 | } else if (!osdmap->get_addrs(whoami).legacy_equals( |
8273 | client_messenger->get_myaddrs())) { | |
7c673cae | 8274 | clog->error() << "map e" << osdmap->get_epoch() |
11fdf7f2 TL |
8275 | << " had wrong client addr (" << osdmap->get_addrs(whoami) |
8276 | << " != my " << client_messenger->get_myaddrs() << ")"; | |
8277 | } else if (!osdmap->get_cluster_addrs(whoami).legacy_equals( | |
8278 | cluster_messenger->get_myaddrs())) { | |
7c673cae FG |
8279 | clog->error() << "map e" << osdmap->get_epoch() |
8280 | << " had wrong cluster addr (" | |
11fdf7f2 TL |
8281 | << osdmap->get_cluster_addrs(whoami) |
8282 | << " != my " << cluster_messenger->get_myaddrs() << ")"; | |
8283 | } else if (!osdmap->get_hb_back_addrs(whoami).legacy_equals( | |
8284 | hb_back_server_messenger->get_myaddrs())) { | |
7c673cae | 8285 | clog->error() << "map e" << osdmap->get_epoch() |
c07f9fc5 | 8286 | << " had wrong heartbeat back addr (" |
11fdf7f2 TL |
8287 | << osdmap->get_hb_back_addrs(whoami) |
8288 | << " != my " << hb_back_server_messenger->get_myaddrs() | |
7c673cae | 8289 | << ")"; |
11fdf7f2 TL |
8290 | } else if (!osdmap->get_hb_front_addrs(whoami).legacy_equals( |
8291 | hb_front_server_messenger->get_myaddrs())) { | |
7c673cae | 8292 | clog->error() << "map e" << osdmap->get_epoch() |
c07f9fc5 | 8293 | << " had wrong heartbeat front addr (" |
11fdf7f2 TL |
8294 | << osdmap->get_hb_front_addrs(whoami) |
8295 | << " != my " << hb_front_server_messenger->get_myaddrs() | |
7c673cae FG |
8296 | << ")"; |
8297 | } | |
8298 | ||
8299 | if (!service.is_stopping()) { | |
8300 | epoch_t up_epoch = 0; | |
8301 | epoch_t bind_epoch = osdmap->get_epoch(); | |
8302 | service.set_epochs(NULL,&up_epoch, &bind_epoch); | |
8303 | do_restart = true; | |
8304 | ||
8305 | //add markdown log | |
8306 | utime_t now = ceph_clock_now(); | |
8307 | utime_t grace = utime_t(cct->_conf->osd_max_markdown_period, 0); | |
8308 | osd_markdown_log.push_back(now); | |
8309 | //clear all out-of-date log | |
8310 | while (!osd_markdown_log.empty() && | |
8311 | osd_markdown_log.front() + grace < now) | |
8312 | osd_markdown_log.pop_front(); | |
8313 | if ((int)osd_markdown_log.size() > cct->_conf->osd_max_markdown_count) { | |
8314 | dout(0) << __func__ << " marked down " | |
8315 | << osd_markdown_log.size() | |
8316 | << " > osd_max_markdown_count " | |
8317 | << cct->_conf->osd_max_markdown_count | |
8318 | << " in last " << grace << " seconds, shutting down" | |
8319 | << dendl; | |
8320 | do_restart = false; | |
8321 | do_shutdown = true; | |
8322 | } | |
8323 | ||
8324 | start_waiting_for_healthy(); | |
8325 | ||
8326 | set<int> avoid_ports; | |
8327 | #if defined(__FreeBSD__) | |
8328 | // prevent FreeBSD from grabbing the client_messenger port during | |
8329 | // rebinding. In which case a cluster_meesneger will connect also | |
8330 | // to the same port | |
11fdf7f2 | 8331 | client_messenger->get_myaddrs().get_ports(&avoid_ports); |
7c673cae | 8332 | #endif |
11fdf7f2 TL |
8333 | cluster_messenger->get_myaddrs().get_ports(&avoid_ports); |
8334 | hb_back_server_messenger->get_myaddrs().get_ports(&avoid_ports); | |
8335 | hb_front_server_messenger->get_myaddrs().get_ports(&avoid_ports); | |
7c673cae FG |
8336 | |
8337 | int r = cluster_messenger->rebind(avoid_ports); | |
8338 | if (r != 0) { | |
8339 | do_shutdown = true; // FIXME: do_restart? | |
8340 | network_error = true; | |
8341 | dout(0) << __func__ << " marked down:" | |
8342 | << " rebind cluster_messenger failed" << dendl; | |
8343 | } | |
8344 | ||
8345 | r = hb_back_server_messenger->rebind(avoid_ports); | |
8346 | if (r != 0) { | |
8347 | do_shutdown = true; // FIXME: do_restart? | |
8348 | network_error = true; | |
8349 | dout(0) << __func__ << " marked down:" | |
8350 | << " rebind hb_back_server_messenger failed" << dendl; | |
8351 | } | |
8352 | ||
8353 | r = hb_front_server_messenger->rebind(avoid_ports); | |
8354 | if (r != 0) { | |
8355 | do_shutdown = true; // FIXME: do_restart? | |
8356 | network_error = true; | |
8357 | dout(0) << __func__ << " marked down:" | |
8358 | << " rebind hb_front_server_messenger failed" << dendl; | |
8359 | } | |
8360 | ||
8361 | hb_front_client_messenger->mark_down_all(); | |
8362 | hb_back_client_messenger->mark_down_all(); | |
8363 | ||
8364 | reset_heartbeat_peers(); | |
8365 | } | |
8366 | } | |
8367 | } | |
8368 | ||
8369 | map_lock.put_write(); | |
8370 | ||
11fdf7f2 | 8371 | check_osdmap_features(); |
7c673cae FG |
8372 | |
8373 | // yay! | |
8374 | consume_map(); | |
8375 | ||
8376 | if (is_active() || is_waiting_for_healthy()) | |
8377 | maybe_update_heartbeat_peers(); | |
8378 | ||
11fdf7f2 | 8379 | if (is_active()) { |
7c673cae FG |
8380 | activate_map(); |
8381 | } | |
8382 | ||
31f18b77 | 8383 | if (do_shutdown) { |
7c673cae | 8384 | if (network_error) { |
11fdf7f2 | 8385 | cancel_pending_failures(); |
7c673cae FG |
8386 | } |
8387 | // trigger shutdown in a different thread | |
8388 | dout(0) << __func__ << " shutdown OSD via async signal" << dendl; | |
8389 | queue_async_signal(SIGINT); | |
8390 | } | |
31f18b77 FG |
8391 | else if (m->newest_map && m->newest_map > last) { |
8392 | dout(10) << " msg say newest map is " << m->newest_map | |
8393 | << ", requesting more" << dendl; | |
8394 | osdmap_subscribe(osdmap->get_epoch()+1, false); | |
8395 | } | |
7c673cae FG |
8396 | else if (is_preboot()) { |
8397 | if (m->get_source().is_mon()) | |
8398 | _preboot(m->oldest_map, m->newest_map); | |
8399 | else | |
8400 | start_boot(); | |
8401 | } | |
8402 | else if (do_restart) | |
8403 | start_boot(); | |
8404 | ||
8405 | } | |
8406 | ||
11fdf7f2 | 8407 | void OSD::check_osdmap_features() |
7c673cae FG |
8408 | { |
8409 | // adjust required feature bits? | |
8410 | ||
8411 | // we have to be a bit careful here, because we are accessing the | |
8412 | // Policy structures without taking any lock. in particular, only | |
8413 | // modify integer values that can safely be read by a racing CPU. | |
8414 | // since we are only accessing existing Policy structures a their | |
8415 | // current memory location, and setting or clearing bits in integer | |
8416 | // fields, and we are the only writer, this is not a problem. | |
8417 | ||
8418 | { | |
8419 | Messenger::Policy p = client_messenger->get_default_policy(); | |
8420 | uint64_t mask; | |
8421 | uint64_t features = osdmap->get_features(entity_name_t::TYPE_CLIENT, &mask); | |
8422 | if ((p.features_required & mask) != features) { | |
8423 | dout(0) << "crush map has features " << features | |
8424 | << ", adjusting msgr requires for clients" << dendl; | |
8425 | p.features_required = (p.features_required & ~mask) | features; | |
8426 | client_messenger->set_default_policy(p); | |
8427 | } | |
8428 | } | |
8429 | { | |
8430 | Messenger::Policy p = client_messenger->get_policy(entity_name_t::TYPE_MON); | |
8431 | uint64_t mask; | |
8432 | uint64_t features = osdmap->get_features(entity_name_t::TYPE_MON, &mask); | |
8433 | if ((p.features_required & mask) != features) { | |
8434 | dout(0) << "crush map has features " << features | |
8435 | << " was " << p.features_required | |
8436 | << ", adjusting msgr requires for mons" << dendl; | |
8437 | p.features_required = (p.features_required & ~mask) | features; | |
8438 | client_messenger->set_policy(entity_name_t::TYPE_MON, p); | |
8439 | } | |
8440 | } | |
8441 | { | |
8442 | Messenger::Policy p = cluster_messenger->get_policy(entity_name_t::TYPE_OSD); | |
8443 | uint64_t mask; | |
8444 | uint64_t features = osdmap->get_features(entity_name_t::TYPE_OSD, &mask); | |
8445 | ||
8446 | if ((p.features_required & mask) != features) { | |
8447 | dout(0) << "crush map has features " << features | |
8448 | << ", adjusting msgr requires for osds" << dendl; | |
8449 | p.features_required = (p.features_required & ~mask) | features; | |
8450 | cluster_messenger->set_policy(entity_name_t::TYPE_OSD, p); | |
8451 | } | |
8452 | ||
11fdf7f2 | 8453 | if (!superblock.compat_features.incompat.contains(CEPH_OSD_FEATURE_INCOMPAT_SHARDS)) { |
7c673cae FG |
8454 | dout(0) << __func__ << " enabling on-disk ERASURE CODES compat feature" << dendl; |
8455 | superblock.compat_features.incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_SHARDS); | |
8456 | ObjectStore::Transaction t; | |
8457 | write_superblock(t); | |
11fdf7f2 TL |
8458 | int err = store->queue_transaction(service.meta_ch, std::move(t), NULL); |
8459 | ceph_assert(err == 0); | |
7c673cae FG |
8460 | } |
8461 | } | |
11fdf7f2 TL |
8462 | |
8463 | if (osdmap->require_osd_release < CEPH_RELEASE_NAUTILUS) { | |
8464 | heartbeat_dispatcher.ms_set_require_authorizer(false); | |
8465 | } | |
8466 | ||
8467 | if (osdmap->require_osd_release != last_require_osd_release) { | |
8468 | dout(1) << __func__ << " require_osd_release " << last_require_osd_release | |
8469 | << " -> " << to_string(osdmap->require_osd_release) << dendl; | |
8470 | store->write_meta("require_osd_release", | |
8471 | stringify((int)osdmap->require_osd_release)); | |
8472 | last_require_osd_release = osdmap->require_osd_release; | |
8473 | } | |
7c673cae FG |
8474 | } |
8475 | ||
11fdf7f2 TL |
8476 | struct C_FinishSplits : public Context { |
8477 | OSD *osd; | |
8478 | set<PGRef> pgs; | |
8479 | C_FinishSplits(OSD *osd, const set<PGRef> &in) | |
8480 | : osd(osd), pgs(in) {} | |
8481 | void finish(int r) override { | |
8482 | osd->_finish_splits(pgs); | |
8483 | } | |
8484 | }; | |
8485 | ||
8486 | void OSD::_finish_splits(set<PGRef>& pgs) | |
7c673cae | 8487 | { |
11fdf7f2 TL |
8488 | dout(10) << __func__ << " " << pgs << dendl; |
8489 | if (is_stopping()) | |
8490 | return; | |
8491 | PG::RecoveryCtx rctx = create_context(); | |
8492 | for (set<PGRef>::iterator i = pgs.begin(); | |
8493 | i != pgs.end(); | |
8494 | ++i) { | |
8495 | PG *pg = i->get(); | |
7c673cae | 8496 | |
11fdf7f2 TL |
8497 | pg->lock(); |
8498 | dout(10) << __func__ << " " << *pg << dendl; | |
8499 | epoch_t e = pg->get_osdmap_epoch(); | |
8500 | pg->handle_initialize(&rctx); | |
8501 | pg->queue_null(e, e); | |
8502 | dispatch_context_transaction(rctx, pg); | |
8503 | pg->unlock(); | |
7c673cae | 8504 | |
11fdf7f2 TL |
8505 | unsigned shard_index = pg->pg_id.hash_to_shard(num_shards); |
8506 | shards[shard_index]->register_and_wake_split_child(pg); | |
7c673cae FG |
8507 | } |
8508 | ||
11fdf7f2 TL |
8509 | dispatch_context(rctx, 0, service.get_osdmap()); |
8510 | }; | |
8511 | ||
8512 | bool OSD::add_merge_waiter(OSDMapRef nextmap, spg_t target, PGRef src, | |
8513 | unsigned need) | |
8514 | { | |
8515 | std::lock_guard l(merge_lock); | |
8516 | auto& p = merge_waiters[nextmap->get_epoch()][target]; | |
8517 | p[src->pg_id] = src; | |
8518 | dout(10) << __func__ << " added merge_waiter " << src->pg_id | |
8519 | << " for " << target << ", have " << p.size() << "/" << need | |
8520 | << dendl; | |
8521 | return p.size() == need; | |
8522 | } | |
8523 | ||
8524 | bool OSD::advance_pg( | |
8525 | epoch_t osd_epoch, | |
8526 | PG *pg, | |
8527 | ThreadPool::TPHandle &handle, | |
8528 | PG::RecoveryCtx *rctx) | |
8529 | { | |
8530 | if (osd_epoch <= pg->get_osdmap_epoch()) { | |
8531 | return true; | |
8532 | } | |
8533 | ceph_assert(pg->is_locked()); | |
8534 | OSDMapRef lastmap = pg->get_osdmap(); | |
8535 | ceph_assert(lastmap->get_epoch() < osd_epoch); | |
8536 | set<PGRef> new_pgs; // any split children | |
8537 | bool ret = true; | |
8538 | ||
8539 | unsigned old_pg_num = lastmap->have_pg_pool(pg->pg_id.pool()) ? | |
8540 | lastmap->get_pg_num(pg->pg_id.pool()) : 0; | |
8541 | for (epoch_t next_epoch = pg->get_osdmap_epoch() + 1; | |
8542 | next_epoch <= osd_epoch; | |
7c673cae FG |
8543 | ++next_epoch) { |
8544 | OSDMapRef nextmap = service.try_get_map(next_epoch); | |
8545 | if (!nextmap) { | |
8546 | dout(20) << __func__ << " missing map " << next_epoch << dendl; | |
7c673cae FG |
8547 | continue; |
8548 | } | |
8549 | ||
11fdf7f2 TL |
8550 | unsigned new_pg_num = |
8551 | (old_pg_num && nextmap->have_pg_pool(pg->pg_id.pool())) ? | |
8552 | nextmap->get_pg_num(pg->pg_id.pool()) : 0; | |
8553 | if (old_pg_num && new_pg_num && old_pg_num != new_pg_num) { | |
8554 | // check for merge | |
8555 | if (nextmap->have_pg_pool(pg->pg_id.pool())) { | |
8556 | spg_t parent; | |
8557 | if (pg->pg_id.is_merge_source( | |
8558 | old_pg_num, | |
8559 | new_pg_num, | |
8560 | &parent)) { | |
8561 | // we are merge source | |
8562 | PGRef spg = pg; // carry a ref | |
8563 | dout(1) << __func__ << " " << pg->pg_id | |
8564 | << " is merge source, target is " << parent | |
8565 | << dendl; | |
8566 | pg->write_if_dirty(rctx); | |
8567 | dispatch_context_transaction(*rctx, pg, &handle); | |
8568 | pg->ch->flush(); | |
8569 | pg->on_shutdown(); | |
8570 | OSDShard *sdata = pg->osd_shard; | |
8571 | { | |
8572 | std::lock_guard l(sdata->shard_lock); | |
8573 | if (pg->pg_slot) { | |
8574 | sdata->_detach_pg(pg->pg_slot); | |
8575 | // update pg count now since we might not get an osdmap | |
8576 | // any time soon. | |
8577 | if (pg->is_primary()) | |
8578 | logger->dec(l_osd_pg_primary); | |
8579 | else if (pg->is_replica()) | |
8580 | logger->dec(l_osd_pg_replica); | |
8581 | else | |
8582 | logger->dec(l_osd_pg_stray); | |
8583 | } | |
8584 | } | |
8585 | pg->unlock(); | |
8586 | ||
8587 | set<spg_t> children; | |
8588 | parent.is_split(new_pg_num, old_pg_num, &children); | |
8589 | if (add_merge_waiter(nextmap, parent, pg, children.size())) { | |
8590 | enqueue_peering_evt( | |
8591 | parent, | |
8592 | PGPeeringEventRef( | |
8593 | std::make_shared<PGPeeringEvent>( | |
8594 | nextmap->get_epoch(), | |
8595 | nextmap->get_epoch(), | |
8596 | NullEvt()))); | |
8597 | } | |
8598 | ret = false; | |
8599 | goto out; | |
8600 | } else if (pg->pg_id.is_merge_target(old_pg_num, new_pg_num)) { | |
8601 | // we are merge target | |
8602 | set<spg_t> children; | |
8603 | pg->pg_id.is_split(new_pg_num, old_pg_num, &children); | |
8604 | dout(20) << __func__ << " " << pg->pg_id | |
8605 | << " is merge target, sources are " << children | |
8606 | << dendl; | |
8607 | map<spg_t,PGRef> sources; | |
8608 | { | |
8609 | std::lock_guard l(merge_lock); | |
8610 | auto& s = merge_waiters[nextmap->get_epoch()][pg->pg_id]; | |
8611 | unsigned need = children.size(); | |
8612 | dout(20) << __func__ << " have " << s.size() << "/" | |
8613 | << need << dendl; | |
8614 | if (s.size() == need) { | |
8615 | sources.swap(s); | |
8616 | merge_waiters[nextmap->get_epoch()].erase(pg->pg_id); | |
8617 | if (merge_waiters[nextmap->get_epoch()].empty()) { | |
8618 | merge_waiters.erase(nextmap->get_epoch()); | |
8619 | } | |
8620 | } | |
8621 | } | |
8622 | if (!sources.empty()) { | |
8623 | unsigned new_pg_num = nextmap->get_pg_num(pg->pg_id.pool()); | |
8624 | unsigned split_bits = pg->pg_id.get_split_bits(new_pg_num); | |
8625 | dout(1) << __func__ << " merging " << pg->pg_id << dendl; | |
8626 | pg->merge_from( | |
8627 | sources, rctx, split_bits, | |
8628 | nextmap->get_pg_pool( | |
8629 | pg->pg_id.pool())->last_pg_merge_meta); | |
8630 | pg->pg_slot->waiting_for_merge_epoch = 0; | |
8631 | } else { | |
8632 | dout(20) << __func__ << " not ready to merge yet" << dendl; | |
8633 | pg->write_if_dirty(rctx); | |
8634 | pg->unlock(); | |
8635 | // kick source(s) to get them ready | |
8636 | for (auto& i : children) { | |
8637 | dout(20) << __func__ << " kicking source " << i << dendl; | |
8638 | enqueue_peering_evt( | |
8639 | i, | |
8640 | PGPeeringEventRef( | |
8641 | std::make_shared<PGPeeringEvent>( | |
8642 | nextmap->get_epoch(), | |
8643 | nextmap->get_epoch(), | |
8644 | NullEvt()))); | |
8645 | } | |
8646 | ret = false; | |
8647 | goto out; | |
8648 | } | |
8649 | } | |
8650 | } | |
8651 | } | |
8652 | ||
7c673cae FG |
8653 | vector<int> newup, newacting; |
8654 | int up_primary, acting_primary; | |
8655 | nextmap->pg_to_up_acting_osds( | |
11fdf7f2 | 8656 | pg->pg_id.pgid, |
7c673cae FG |
8657 | &newup, &up_primary, |
8658 | &newacting, &acting_primary); | |
8659 | pg->handle_advance_map( | |
8660 | nextmap, lastmap, newup, up_primary, | |
8661 | newacting, acting_primary, rctx); | |
8662 | ||
11fdf7f2 TL |
8663 | if (new_pg_num && old_pg_num != new_pg_num) { |
8664 | // check for split | |
8665 | set<spg_t> children; | |
8666 | if (pg->pg_id.is_split( | |
8667 | old_pg_num, | |
8668 | new_pg_num, | |
8669 | &children)) { | |
8670 | split_pgs( | |
8671 | pg, children, &new_pgs, lastmap, nextmap, | |
8672 | rctx); | |
8673 | } | |
7c673cae FG |
8674 | } |
8675 | ||
8676 | lastmap = nextmap; | |
11fdf7f2 | 8677 | old_pg_num = new_pg_num; |
7c673cae FG |
8678 | handle.reset_tp_timeout(); |
8679 | } | |
7c673cae | 8680 | pg->handle_activate_map(rctx); |
11fdf7f2 TL |
8681 | |
8682 | ret = true; | |
8683 | out: | |
8684 | if (!new_pgs.empty()) { | |
8685 | rctx->transaction->register_on_applied(new C_FinishSplits(this, new_pgs)); | |
7c673cae | 8686 | } |
11fdf7f2 | 8687 | return ret; |
7c673cae FG |
8688 | } |
8689 | ||
8690 | void OSD::consume_map() | |
8691 | { | |
11fdf7f2 | 8692 | ceph_assert(osd_lock.is_locked()); |
7c673cae FG |
8693 | dout(7) << "consume_map version " << osdmap->get_epoch() << dendl; |
8694 | ||
3efd9988 FG |
8695 | /** make sure the cluster is speaking in SORTBITWISE, because we don't |
8696 | * speak the older sorting version any more. Be careful not to force | |
8697 | * a shutdown if we are merely processing old maps, though. | |
8698 | */ | |
8699 | if (!osdmap->test_flag(CEPH_OSDMAP_SORTBITWISE) && is_active()) { | |
8700 | derr << __func__ << " SORTBITWISE flag is not set" << dendl; | |
8701 | ceph_abort(); | |
8702 | } | |
8703 | ||
11fdf7f2 TL |
8704 | service.pre_publish_map(osdmap); |
8705 | service.await_reserved_maps(); | |
8706 | service.publish_map(osdmap); | |
7c673cae | 8707 | |
11fdf7f2 TL |
8708 | // prime splits and merges |
8709 | set<pair<spg_t,epoch_t>> newly_split; // splits, and when | |
8710 | set<pair<spg_t,epoch_t>> merge_pgs; // merge participants, and when | |
8711 | for (auto& shard : shards) { | |
8712 | shard->identify_splits_and_merges(osdmap, &newly_split, &merge_pgs); | |
8713 | } | |
8714 | if (!newly_split.empty()) { | |
8715 | for (auto& shard : shards) { | |
8716 | shard->prime_splits(osdmap, &newly_split); | |
8717 | } | |
8718 | ceph_assert(newly_split.empty()); | |
8719 | } | |
7c673cae | 8720 | |
11fdf7f2 TL |
8721 | // prune sent_ready_to_merge |
8722 | service.prune_sent_ready_to_merge(osdmap); | |
7c673cae | 8723 | |
11fdf7f2 TL |
8724 | // FIXME, maybe: We could race against an incoming peering message |
8725 | // that instantiates a merge PG after identify_merges() below and | |
8726 | // never set up its peer to complete the merge. An OSD restart | |
8727 | // would clear it up. This is a hard race to resolve, | |
8728 | // extraordinarily rare (we only merge PGs that are stable and | |
8729 | // clean, so it'd have to be an imported PG to an OSD with a | |
8730 | // slightly stale OSDMap...), so I'm ignoring it for now. We plan to | |
8731 | // replace all of this with a seastar-based code soon anyway. | |
8732 | if (!merge_pgs.empty()) { | |
8733 | // mark the pgs we already have, or create new and empty merge | |
8734 | // participants for those we are missing. do this all under the | |
8735 | // shard lock so we don't have to worry about racing pg creates | |
8736 | // via _process. | |
8737 | for (auto& shard : shards) { | |
8738 | shard->prime_merges(osdmap, &merge_pgs); | |
7c673cae | 8739 | } |
11fdf7f2 TL |
8740 | ceph_assert(merge_pgs.empty()); |
8741 | } | |
8742 | ||
8743 | service.prune_pg_created(); | |
8744 | ||
8745 | unsigned pushes_to_free = 0; | |
8746 | for (auto& shard : shards) { | |
8747 | shard->consume_map(osdmap, &pushes_to_free); | |
8748 | } | |
8749 | ||
8750 | vector<spg_t> pgids; | |
8751 | _get_pgids(&pgids); | |
8752 | ||
8753 | // count (FIXME, probably during seastar rewrite) | |
8754 | int num_pg_primary = 0, num_pg_replica = 0, num_pg_stray = 0; | |
8755 | vector<PGRef> pgs; | |
8756 | _get_pgs(&pgs); | |
8757 | for (auto& pg : pgs) { | |
8758 | // FIXME (probably during seastar rewrite): this is lockless and | |
8759 | // racy, but we don't want to take pg lock here. | |
8760 | if (pg->is_primary()) | |
8761 | num_pg_primary++; | |
8762 | else if (pg->is_replica()) | |
8763 | num_pg_replica++; | |
8764 | else | |
8765 | num_pg_stray++; | |
8766 | } | |
3efd9988 | 8767 | |
11fdf7f2 TL |
8768 | { |
8769 | // FIXME (as part of seastar rewrite): move to OSDShard | |
8770 | std::lock_guard l(pending_creates_lock); | |
8771 | for (auto pg = pending_creates_from_osd.begin(); | |
8772 | pg != pending_creates_from_osd.end();) { | |
b32b8144 | 8773 | if (osdmap->get_pg_acting_rank(pg->first, whoami) < 0) { |
11fdf7f2 TL |
8774 | dout(10) << __func__ << " pg " << pg->first << " doesn't map here, " |
8775 | << "discarding pending_create_from_osd" << dendl; | |
3efd9988 FG |
8776 | pg = pending_creates_from_osd.erase(pg); |
8777 | } else { | |
8778 | ++pg; | |
8779 | } | |
8780 | } | |
7c673cae FG |
8781 | } |
8782 | ||
7c673cae FG |
8783 | service.maybe_inject_dispatch_delay(); |
8784 | ||
8785 | dispatch_sessions_waiting_on_map(); | |
8786 | ||
8787 | service.maybe_inject_dispatch_delay(); | |
8788 | ||
11fdf7f2 | 8789 | service.release_reserved_pushes(pushes_to_free); |
7c673cae | 8790 | |
11fdf7f2 TL |
8791 | // queue null events to push maps down to individual PGs |
8792 | for (auto pgid : pgids) { | |
8793 | enqueue_peering_evt( | |
8794 | pgid, | |
8795 | PGPeeringEventRef( | |
8796 | std::make_shared<PGPeeringEvent>( | |
8797 | osdmap->get_epoch(), | |
8798 | osdmap->get_epoch(), | |
8799 | NullEvt()))); | |
7c673cae | 8800 | } |
11fdf7f2 | 8801 | logger->set(l_osd_pg, pgids.size()); |
7c673cae FG |
8802 | logger->set(l_osd_pg_primary, num_pg_primary); |
8803 | logger->set(l_osd_pg_replica, num_pg_replica); | |
8804 | logger->set(l_osd_pg_stray, num_pg_stray); | |
8805 | } | |
8806 | ||
8807 | void OSD::activate_map() | |
8808 | { | |
11fdf7f2 | 8809 | ceph_assert(osd_lock.is_locked()); |
7c673cae FG |
8810 | |
8811 | dout(7) << "activate_map version " << osdmap->get_epoch() << dendl; | |
8812 | ||
7c673cae FG |
8813 | if (osdmap->test_flag(CEPH_OSDMAP_FULL)) { |
8814 | dout(10) << " osdmap flagged full, doing onetime osdmap subscribe" << dendl; | |
8815 | osdmap_subscribe(osdmap->get_epoch() + 1, false); | |
8816 | } | |
8817 | ||
8818 | // norecover? | |
8819 | if (osdmap->test_flag(CEPH_OSDMAP_NORECOVER)) { | |
8820 | if (!service.recovery_is_paused()) { | |
8821 | dout(1) << "pausing recovery (NORECOVER flag set)" << dendl; | |
8822 | service.pause_recovery(); | |
8823 | } | |
8824 | } else { | |
8825 | if (service.recovery_is_paused()) { | |
8826 | dout(1) << "unpausing recovery (NORECOVER flag unset)" << dendl; | |
8827 | service.unpause_recovery(); | |
8828 | } | |
8829 | } | |
8830 | ||
8831 | service.activate_map(); | |
8832 | ||
8833 | // process waiters | |
8834 | take_waiters(waiting_for_osdmap); | |
8835 | } | |
8836 | ||
8837 | bool OSD::require_mon_peer(const Message *m) | |
8838 | { | |
8839 | if (!m->get_connection()->peer_is_mon()) { | |
8840 | dout(0) << "require_mon_peer received from non-mon " | |
8841 | << m->get_connection()->get_peer_addr() | |
8842 | << " " << *m << dendl; | |
8843 | return false; | |
8844 | } | |
8845 | return true; | |
8846 | } | |
8847 | ||
8848 | bool OSD::require_mon_or_mgr_peer(const Message *m) | |
8849 | { | |
8850 | if (!m->get_connection()->peer_is_mon() && | |
8851 | !m->get_connection()->peer_is_mgr()) { | |
8852 | dout(0) << "require_mon_or_mgr_peer received from non-mon, non-mgr " | |
8853 | << m->get_connection()->get_peer_addr() | |
8854 | << " " << *m << dendl; | |
8855 | return false; | |
8856 | } | |
8857 | return true; | |
8858 | } | |
8859 | ||
8860 | bool OSD::require_osd_peer(const Message *m) | |
8861 | { | |
8862 | if (!m->get_connection()->peer_is_osd()) { | |
8863 | dout(0) << "require_osd_peer received from non-osd " | |
8864 | << m->get_connection()->get_peer_addr() | |
8865 | << " " << *m << dendl; | |
8866 | return false; | |
8867 | } | |
8868 | return true; | |
8869 | } | |
8870 | ||
8871 | bool OSD::require_self_aliveness(const Message *m, epoch_t epoch) | |
8872 | { | |
8873 | epoch_t up_epoch = service.get_up_epoch(); | |
8874 | if (epoch < up_epoch) { | |
8875 | dout(7) << "from pre-up epoch " << epoch << " < " << up_epoch << dendl; | |
8876 | return false; | |
8877 | } | |
8878 | ||
8879 | if (!is_active()) { | |
8880 | dout(7) << "still in boot state, dropping message " << *m << dendl; | |
8881 | return false; | |
8882 | } | |
8883 | ||
8884 | return true; | |
8885 | } | |
8886 | ||
8887 | bool OSD::require_same_peer_instance(const Message *m, OSDMapRef& map, | |
8888 | bool is_fast_dispatch) | |
8889 | { | |
8890 | int from = m->get_source().num(); | |
8891 | ||
8892 | if (map->is_down(from) || | |
11fdf7f2 | 8893 | (map->get_cluster_addrs(from) != m->get_source_addrs())) { |
7c673cae FG |
8894 | dout(5) << "from dead osd." << from << ", marking down, " |
8895 | << " msg was " << m->get_source_inst().addr | |
11fdf7f2 TL |
8896 | << " expected " |
8897 | << (map->is_up(from) ? | |
8898 | map->get_cluster_addrs(from) : entity_addrvec_t()) | |
7c673cae FG |
8899 | << dendl; |
8900 | ConnectionRef con = m->get_connection(); | |
8901 | con->mark_down(); | |
11fdf7f2 TL |
8902 | auto priv = con->get_priv(); |
8903 | if (auto s = static_cast<Session*>(priv.get()); s) { | |
7c673cae FG |
8904 | if (!is_fast_dispatch) |
8905 | s->session_dispatch_lock.Lock(); | |
8906 | clear_session_waiting_on_map(s); | |
11fdf7f2 TL |
8907 | con->set_priv(nullptr); // break ref <-> session cycle, if any |
8908 | s->con.reset(); | |
7c673cae FG |
8909 | if (!is_fast_dispatch) |
8910 | s->session_dispatch_lock.Unlock(); | |
7c673cae FG |
8911 | } |
8912 | return false; | |
8913 | } | |
8914 | return true; | |
8915 | } | |
8916 | ||
8917 | ||
8918 | /* | |
8919 | * require that we have same (or newer) map, and that | |
8920 | * the source is the pg primary. | |
8921 | */ | |
8922 | bool OSD::require_same_or_newer_map(OpRequestRef& op, epoch_t epoch, | |
8923 | bool is_fast_dispatch) | |
8924 | { | |
8925 | const Message *m = op->get_req(); | |
8926 | dout(15) << "require_same_or_newer_map " << epoch | |
8927 | << " (i am " << osdmap->get_epoch() << ") " << m << dendl; | |
8928 | ||
11fdf7f2 | 8929 | ceph_assert(osd_lock.is_locked()); |
7c673cae FG |
8930 | |
8931 | // do they have a newer map? | |
8932 | if (epoch > osdmap->get_epoch()) { | |
8933 | dout(7) << "waiting for newer map epoch " << epoch | |
8934 | << " > my " << osdmap->get_epoch() << " with " << m << dendl; | |
8935 | wait_for_new_map(op); | |
8936 | return false; | |
8937 | } | |
8938 | ||
8939 | if (!require_self_aliveness(op->get_req(), epoch)) { | |
8940 | return false; | |
8941 | } | |
8942 | ||
8943 | // ok, our map is same or newer.. do they still exist? | |
8944 | if (m->get_connection()->get_messenger() == cluster_messenger && | |
8945 | !require_same_peer_instance(op->get_req(), osdmap, is_fast_dispatch)) { | |
8946 | return false; | |
8947 | } | |
8948 | ||
8949 | return true; | |
8950 | } | |
8951 | ||
8952 | ||
8953 | ||
8954 | ||
8955 | ||
8956 | // ---------------------------------------- | |
8957 | // pg creation | |
8958 | ||
8959 | void OSD::split_pgs( | |
8960 | PG *parent, | |
31f18b77 | 8961 | const set<spg_t> &childpgids, set<PGRef> *out_pgs, |
7c673cae FG |
8962 | OSDMapRef curmap, |
8963 | OSDMapRef nextmap, | |
8964 | PG::RecoveryCtx *rctx) | |
8965 | { | |
11fdf7f2 TL |
8966 | unsigned pg_num = nextmap->get_pg_num(parent->pg_id.pool()); |
8967 | parent->update_snap_mapper_bits(parent->get_pgid().get_split_bits(pg_num)); | |
7c673cae | 8968 | |
11fdf7f2 TL |
8969 | vector<object_stat_sum_t> updated_stats; |
8970 | parent->start_split_stats(childpgids, &updated_stats); | |
7c673cae FG |
8971 | |
8972 | vector<object_stat_sum_t>::iterator stat_iter = updated_stats.begin(); | |
8973 | for (set<spg_t>::const_iterator i = childpgids.begin(); | |
8974 | i != childpgids.end(); | |
8975 | ++i, ++stat_iter) { | |
11fdf7f2 TL |
8976 | ceph_assert(stat_iter != updated_stats.end()); |
8977 | dout(10) << __func__ << " splitting " << *parent << " into " << *i << dendl; | |
7c673cae FG |
8978 | PG* child = _make_pg(nextmap, *i); |
8979 | child->lock(true); | |
8980 | out_pgs->insert(child); | |
11fdf7f2 | 8981 | child->ch = store->create_new_collection(child->coll); |
7c673cae | 8982 | |
11fdf7f2 TL |
8983 | { |
8984 | uint32_t shard_index = i->hash_to_shard(shards.size()); | |
8985 | assert(NULL != shards[shard_index]); | |
8986 | store->set_collection_commit_queue(child->coll, &(shards[shard_index]->context_queue)); | |
8987 | } | |
7c673cae | 8988 | |
11fdf7f2 TL |
8989 | unsigned split_bits = i->get_split_bits(pg_num); |
8990 | dout(10) << " pg_num is " << pg_num | |
8991 | << ", m_seed " << i->ps() | |
8992 | << ", split_bits is " << split_bits << dendl; | |
7c673cae FG |
8993 | parent->split_colls( |
8994 | *i, | |
8995 | split_bits, | |
8996 | i->ps(), | |
11fdf7f2 | 8997 | &child->get_pool().info, |
7c673cae FG |
8998 | rctx->transaction); |
8999 | parent->split_into( | |
9000 | i->pgid, | |
9001 | child, | |
9002 | split_bits); | |
7c673cae | 9003 | |
11fdf7f2 | 9004 | child->finish_split_stats(*stat_iter, rctx->transaction); |
7c673cae FG |
9005 | child->unlock(); |
9006 | } | |
11fdf7f2 TL |
9007 | ceph_assert(stat_iter != updated_stats.end()); |
9008 | parent->finish_split_stats(*stat_iter, rctx->transaction); | |
7c673cae FG |
9009 | } |
9010 | ||
9011 | /* | |
9012 | * holding osd_lock | |
9013 | */ | |
9014 | void OSD::handle_pg_create(OpRequestRef op) | |
9015 | { | |
9016 | const MOSDPGCreate *m = static_cast<const MOSDPGCreate*>(op->get_req()); | |
11fdf7f2 | 9017 | ceph_assert(m->get_type() == MSG_OSD_PG_CREATE); |
7c673cae FG |
9018 | |
9019 | dout(10) << "handle_pg_create " << *m << dendl; | |
9020 | ||
9021 | if (!require_mon_peer(op->get_req())) { | |
9022 | return; | |
9023 | } | |
9024 | ||
9025 | if (!require_same_or_newer_map(op, m->epoch, false)) | |
9026 | return; | |
9027 | ||
9028 | op->mark_started(); | |
9029 | ||
9030 | map<pg_t,utime_t>::const_iterator ci = m->ctimes.begin(); | |
9031 | for (map<pg_t,pg_create_t>::const_iterator p = m->mkpg.begin(); | |
9032 | p != m->mkpg.end(); | |
9033 | ++p, ++ci) { | |
11fdf7f2 | 9034 | ceph_assert(ci != m->ctimes.end() && ci->first == p->first); |
7c673cae FG |
9035 | epoch_t created = p->second.created; |
9036 | if (p->second.split_bits) // Skip split pgs | |
9037 | continue; | |
9038 | pg_t on = p->first; | |
9039 | ||
7c673cae FG |
9040 | if (!osdmap->have_pg_pool(on.pool())) { |
9041 | dout(20) << "ignoring pg on deleted pool " << on << dendl; | |
9042 | continue; | |
9043 | } | |
9044 | ||
9045 | dout(20) << "mkpg " << on << " e" << created << "@" << ci->second << dendl; | |
9046 | ||
9047 | // is it still ours? | |
9048 | vector<int> up, acting; | |
9049 | int up_primary = -1; | |
9050 | int acting_primary = -1; | |
9051 | osdmap->pg_to_up_acting_osds(on, &up, &up_primary, &acting, &acting_primary); | |
9052 | int role = osdmap->calc_pg_role(whoami, acting, acting.size()); | |
9053 | ||
9054 | if (acting_primary != whoami) { | |
9055 | dout(10) << "mkpg " << on << " not acting_primary (" << acting_primary | |
9056 | << "), my role=" << role << ", skipping" << dendl; | |
9057 | continue; | |
9058 | } | |
9059 | ||
9060 | spg_t pgid; | |
9061 | bool mapped = osdmap->get_primary_shard(on, &pgid); | |
11fdf7f2 | 9062 | ceph_assert(mapped); |
7c673cae | 9063 | |
11fdf7f2 | 9064 | PastIntervals pi; |
7c673cae FG |
9065 | pg_history_t history; |
9066 | build_initial_pg_history(pgid, created, ci->second, &history, &pi); | |
9067 | ||
11fdf7f2 TL |
9068 | // The mon won't resend unless the primary changed, so we ignore |
9069 | // same_interval_since. We'll pass this history with the current | |
9070 | // epoch as the event. | |
7c673cae FG |
9071 | if (history.same_primary_since > m->epoch) { |
9072 | dout(10) << __func__ << ": got obsolete pg create on pgid " | |
9073 | << pgid << " from epoch " << m->epoch | |
9074 | << ", primary changed in " << history.same_primary_since | |
9075 | << dendl; | |
9076 | continue; | |
9077 | } | |
11fdf7f2 TL |
9078 | enqueue_peering_evt( |
9079 | pgid, | |
9080 | PGPeeringEventRef( | |
9081 | std::make_shared<PGPeeringEvent>( | |
9082 | osdmap->get_epoch(), | |
9083 | osdmap->get_epoch(), | |
9084 | NullEvt(), | |
9085 | true, | |
9086 | new PGCreateInfo( | |
9087 | pgid, | |
9088 | osdmap->get_epoch(), | |
9089 | history, | |
9090 | pi, | |
9091 | true) | |
9092 | ))); | |
7c673cae | 9093 | } |
7c673cae | 9094 | |
3efd9988 | 9095 | { |
11fdf7f2 | 9096 | std::lock_guard l(pending_creates_lock); |
3efd9988 FG |
9097 | if (pending_creates_from_mon == 0) { |
9098 | last_pg_create_epoch = m->epoch; | |
9099 | } | |
9100 | } | |
11fdf7f2 | 9101 | |
7c673cae FG |
9102 | maybe_update_heartbeat_peers(); |
9103 | } | |
9104 | ||
9105 | ||
9106 | // ---------------------------------------- | |
9107 | // peering and recovery | |
9108 | ||
9109 | PG::RecoveryCtx OSD::create_context() | |
9110 | { | |
9111 | ObjectStore::Transaction *t = new ObjectStore::Transaction; | |
7c673cae FG |
9112 | map<int, map<spg_t,pg_query_t> > *query_map = |
9113 | new map<int, map<spg_t, pg_query_t> >; | |
9114 | map<int,vector<pair<pg_notify_t, PastIntervals> > > *notify_list = | |
9115 | new map<int, vector<pair<pg_notify_t, PastIntervals> > >; | |
9116 | map<int,vector<pair<pg_notify_t, PastIntervals> > > *info_map = | |
9117 | new map<int,vector<pair<pg_notify_t, PastIntervals> > >; | |
11fdf7f2 | 9118 | PG::RecoveryCtx rctx(query_map, info_map, notify_list, t); |
7c673cae FG |
9119 | return rctx; |
9120 | } | |
9121 | ||
7c673cae FG |
9122 | void OSD::dispatch_context_transaction(PG::RecoveryCtx &ctx, PG *pg, |
9123 | ThreadPool::TPHandle *handle) | |
9124 | { | |
11fdf7f2 | 9125 | if (!ctx.transaction->empty() || ctx.transaction->has_contexts()) { |
7c673cae | 9126 | int tr = store->queue_transaction( |
11fdf7f2 TL |
9127 | pg->ch, |
9128 | std::move(*ctx.transaction), TrackedOpRef(), handle); | |
9129 | ceph_assert(tr == 0); | |
7c673cae | 9130 | delete (ctx.transaction); |
7c673cae | 9131 | ctx.transaction = new ObjectStore::Transaction; |
7c673cae FG |
9132 | } |
9133 | } | |
9134 | ||
9135 | void OSD::dispatch_context(PG::RecoveryCtx &ctx, PG *pg, OSDMapRef curmap, | |
9136 | ThreadPool::TPHandle *handle) | |
9137 | { | |
11fdf7f2 TL |
9138 | if (!service.get_osdmap()->is_up(whoami)) { |
9139 | dout(20) << __func__ << " not up in osdmap" << dendl; | |
9140 | } else if (!is_active()) { | |
9141 | dout(20) << __func__ << " not active" << dendl; | |
9142 | } else { | |
7c673cae FG |
9143 | do_notifies(*ctx.notify_list, curmap); |
9144 | do_queries(*ctx.query_map, curmap); | |
9145 | do_infos(*ctx.info_map, curmap); | |
9146 | } | |
11fdf7f2 | 9147 | if ((!ctx.transaction->empty() || ctx.transaction->has_contexts()) && pg) { |
7c673cae | 9148 | int tr = store->queue_transaction( |
11fdf7f2 TL |
9149 | pg->ch, |
9150 | std::move(*ctx.transaction), TrackedOpRef(), | |
7c673cae | 9151 | handle); |
11fdf7f2 | 9152 | ceph_assert(tr == 0); |
7c673cae | 9153 | } |
11fdf7f2 TL |
9154 | delete ctx.notify_list; |
9155 | delete ctx.query_map; | |
9156 | delete ctx.info_map; | |
9157 | delete ctx.transaction; | |
9158 | } | |
9159 | ||
9160 | void OSD::discard_context(PG::RecoveryCtx& ctx) | |
9161 | { | |
9162 | delete ctx.notify_list; | |
9163 | delete ctx.query_map; | |
9164 | delete ctx.info_map; | |
9165 | delete ctx.transaction; | |
7c673cae FG |
9166 | } |
9167 | ||
11fdf7f2 | 9168 | |
7c673cae FG |
9169 | /** do_notifies |
9170 | * Send an MOSDPGNotify to a primary, with a list of PGs that I have | |
9171 | * content for, and they are primary for. | |
9172 | */ | |
9173 | ||
9174 | void OSD::do_notifies( | |
9175 | map<int,vector<pair<pg_notify_t,PastIntervals> > >& notify_list, | |
9176 | OSDMapRef curmap) | |
9177 | { | |
9178 | for (map<int, | |
9179 | vector<pair<pg_notify_t,PastIntervals> > >::iterator it = | |
9180 | notify_list.begin(); | |
9181 | it != notify_list.end(); | |
9182 | ++it) { | |
9183 | if (!curmap->is_up(it->first)) { | |
9184 | dout(20) << __func__ << " skipping down osd." << it->first << dendl; | |
9185 | continue; | |
9186 | } | |
9187 | ConnectionRef con = service.get_con_osd_cluster( | |
9188 | it->first, curmap->get_epoch()); | |
9189 | if (!con) { | |
9190 | dout(20) << __func__ << " skipping osd." << it->first | |
9191 | << " (NULL con)" << dendl; | |
9192 | continue; | |
9193 | } | |
9194 | service.share_map_peer(it->first, con.get(), curmap); | |
3efd9988 | 9195 | dout(7) << __func__ << " osd." << it->first |
7c673cae FG |
9196 | << " on " << it->second.size() << " PGs" << dendl; |
9197 | MOSDPGNotify *m = new MOSDPGNotify(curmap->get_epoch(), | |
9198 | it->second); | |
9199 | con->send_message(m); | |
9200 | } | |
9201 | } | |
9202 | ||
9203 | ||
9204 | /** do_queries | |
9205 | * send out pending queries for info | summaries | |
9206 | */ | |
9207 | void OSD::do_queries(map<int, map<spg_t,pg_query_t> >& query_map, | |
9208 | OSDMapRef curmap) | |
9209 | { | |
9210 | for (map<int, map<spg_t,pg_query_t> >::iterator pit = query_map.begin(); | |
9211 | pit != query_map.end(); | |
9212 | ++pit) { | |
9213 | if (!curmap->is_up(pit->first)) { | |
9214 | dout(20) << __func__ << " skipping down osd." << pit->first << dendl; | |
9215 | continue; | |
9216 | } | |
9217 | int who = pit->first; | |
9218 | ConnectionRef con = service.get_con_osd_cluster(who, curmap->get_epoch()); | |
9219 | if (!con) { | |
9220 | dout(20) << __func__ << " skipping osd." << who | |
9221 | << " (NULL con)" << dendl; | |
9222 | continue; | |
9223 | } | |
9224 | service.share_map_peer(who, con.get(), curmap); | |
9225 | dout(7) << __func__ << " querying osd." << who | |
9226 | << " on " << pit->second.size() << " PGs" << dendl; | |
9227 | MOSDPGQuery *m = new MOSDPGQuery(curmap->get_epoch(), pit->second); | |
9228 | con->send_message(m); | |
9229 | } | |
9230 | } | |
9231 | ||
9232 | ||
9233 | void OSD::do_infos(map<int, | |
9234 | vector<pair<pg_notify_t, PastIntervals> > >& info_map, | |
9235 | OSDMapRef curmap) | |
9236 | { | |
9237 | for (map<int, | |
9238 | vector<pair<pg_notify_t, PastIntervals> > >::iterator p = | |
9239 | info_map.begin(); | |
9240 | p != info_map.end(); | |
9241 | ++p) { | |
9242 | if (!curmap->is_up(p->first)) { | |
9243 | dout(20) << __func__ << " skipping down osd." << p->first << dendl; | |
9244 | continue; | |
9245 | } | |
9246 | for (vector<pair<pg_notify_t,PastIntervals> >::iterator i = p->second.begin(); | |
9247 | i != p->second.end(); | |
9248 | ++i) { | |
9249 | dout(20) << __func__ << " sending info " << i->first.info | |
9250 | << " to shard " << p->first << dendl; | |
9251 | } | |
9252 | ConnectionRef con = service.get_con_osd_cluster( | |
9253 | p->first, curmap->get_epoch()); | |
9254 | if (!con) { | |
9255 | dout(20) << __func__ << " skipping osd." << p->first | |
9256 | << " (NULL con)" << dendl; | |
9257 | continue; | |
9258 | } | |
9259 | service.share_map_peer(p->first, con.get(), curmap); | |
9260 | MOSDPGInfo *m = new MOSDPGInfo(curmap->get_epoch()); | |
9261 | m->pg_list = p->second; | |
9262 | con->send_message(m); | |
9263 | } | |
9264 | info_map.clear(); | |
9265 | } | |
9266 | ||
11fdf7f2 | 9267 | void OSD::handle_fast_pg_create(MOSDPGCreate2 *m) |
7c673cae | 9268 | { |
11fdf7f2 TL |
9269 | dout(7) << __func__ << " " << *m << " from " << m->get_source() << dendl; |
9270 | if (!require_mon_peer(m)) { | |
9271 | m->put(); | |
7c673cae | 9272 | return; |
7c673cae | 9273 | } |
11fdf7f2 TL |
9274 | for (auto& p : m->pgs) { |
9275 | spg_t pgid = p.first; | |
9276 | epoch_t created = p.second.first; | |
9277 | utime_t created_stamp = p.second.second; | |
9278 | dout(20) << __func__ << " " << pgid << " e" << created | |
9279 | << "@" << created_stamp << dendl; | |
9280 | pg_history_t h; | |
9281 | h.epoch_created = created; | |
9282 | h.epoch_pool_created = created; | |
9283 | h.same_up_since = created; | |
9284 | h.same_interval_since = created; | |
9285 | h.same_primary_since = created; | |
9286 | h.last_scrub_stamp = created_stamp; | |
9287 | h.last_deep_scrub_stamp = created_stamp; | |
9288 | h.last_clean_scrub_stamp = created_stamp; | |
9289 | ||
9290 | enqueue_peering_evt( | |
9291 | pgid, | |
9292 | PGPeeringEventRef( | |
9293 | std::make_shared<PGPeeringEvent>( | |
9294 | m->epoch, | |
9295 | m->epoch, | |
9296 | NullEvt(), | |
9297 | true, | |
9298 | new PGCreateInfo( | |
9299 | pgid, | |
9300 | created, | |
9301 | h, | |
9302 | PastIntervals(), | |
9303 | true) | |
9304 | ))); | |
9305 | } | |
7c673cae | 9306 | |
11fdf7f2 TL |
9307 | { |
9308 | std::lock_guard l(pending_creates_lock); | |
9309 | if (pending_creates_from_mon == 0) { | |
9310 | last_pg_create_epoch = m->epoch; | |
9311 | } | |
7c673cae FG |
9312 | } |
9313 | ||
11fdf7f2 | 9314 | m->put(); |
7c673cae FG |
9315 | } |
9316 | ||
11fdf7f2 | 9317 | void OSD::handle_fast_pg_query(MOSDPGQuery *m) |
7c673cae | 9318 | { |
11fdf7f2 TL |
9319 | dout(7) << __func__ << " " << *m << " from " << m->get_source() << dendl; |
9320 | if (!require_osd_peer(m)) { | |
9321 | m->put(); | |
7c673cae | 9322 | return; |
11fdf7f2 | 9323 | } |
7c673cae | 9324 | int from = m->get_source().num(); |
11fdf7f2 TL |
9325 | for (auto& p : m->pg_list) { |
9326 | enqueue_peering_evt( | |
9327 | p.first, | |
9328 | PGPeeringEventRef( | |
9329 | std::make_shared<PGPeeringEvent>( | |
9330 | p.second.epoch_sent, p.second.epoch_sent, | |
9331 | MQuery( | |
9332 | p.first, | |
9333 | pg_shard_t(from, p.second.from), | |
9334 | p.second, | |
9335 | p.second.epoch_sent), | |
9336 | false)) | |
7c673cae FG |
9337 | ); |
9338 | } | |
11fdf7f2 | 9339 | m->put(); |
7c673cae FG |
9340 | } |
9341 | ||
11fdf7f2 | 9342 | void OSD::handle_fast_pg_notify(MOSDPGNotify* m) |
7c673cae | 9343 | { |
11fdf7f2 TL |
9344 | dout(7) << __func__ << " " << *m << " from " << m->get_source() << dendl; |
9345 | if (!require_osd_peer(m)) { | |
9346 | m->put(); | |
7c673cae FG |
9347 | return; |
9348 | } | |
11fdf7f2 TL |
9349 | int from = m->get_source().num(); |
9350 | for (auto& p : m->get_pg_list()) { | |
9351 | spg_t pgid(p.first.info.pgid.pgid, p.first.to); | |
9352 | enqueue_peering_evt( | |
9353 | pgid, | |
9354 | PGPeeringEventRef( | |
9355 | std::make_shared<PGPeeringEvent>( | |
9356 | p.first.epoch_sent, | |
9357 | p.first.query_epoch, | |
9358 | MNotifyRec( | |
9359 | pgid, pg_shard_t(from, p.first.from), | |
9360 | p.first, | |
9361 | m->get_connection()->get_features(), | |
9362 | p.second), | |
9363 | true, | |
9364 | new PGCreateInfo( | |
9365 | pgid, | |
9366 | p.first.query_epoch, | |
9367 | p.first.info.history, | |
9368 | p.second, | |
9369 | false) | |
9370 | ))); | |
7c673cae | 9371 | } |
11fdf7f2 | 9372 | m->put(); |
7c673cae FG |
9373 | } |
9374 | ||
11fdf7f2 | 9375 | void OSD::handle_fast_pg_info(MOSDPGInfo* m) |
7c673cae | 9376 | { |
11fdf7f2 TL |
9377 | dout(7) << __func__ << " " << *m << " from " << m->get_source() << dendl; |
9378 | if (!require_osd_peer(m)) { | |
9379 | m->put(); | |
7c673cae FG |
9380 | return; |
9381 | } | |
11fdf7f2 TL |
9382 | int from = m->get_source().num(); |
9383 | for (auto& p : m->pg_list) { | |
9384 | enqueue_peering_evt( | |
9385 | spg_t(p.first.info.pgid.pgid, p.first.to), | |
9386 | PGPeeringEventRef( | |
9387 | std::make_shared<PGPeeringEvent>( | |
9388 | p.first.epoch_sent, p.first.query_epoch, | |
9389 | MInfoRec( | |
9390 | pg_shard_t(from, p.first.from), | |
9391 | p.first.info, | |
9392 | p.first.epoch_sent))) | |
9393 | ); | |
7c673cae | 9394 | } |
11fdf7f2 | 9395 | m->put(); |
7c673cae FG |
9396 | } |
9397 | ||
11fdf7f2 | 9398 | void OSD::handle_fast_pg_remove(MOSDPGRemove *m) |
7c673cae | 9399 | { |
11fdf7f2 TL |
9400 | dout(7) << __func__ << " " << *m << " from " << m->get_source() << dendl; |
9401 | if (!require_osd_peer(m)) { | |
9402 | m->put(); | |
7c673cae FG |
9403 | return; |
9404 | } | |
11fdf7f2 TL |
9405 | for (auto& pgid : m->pg_list) { |
9406 | enqueue_peering_evt( | |
9407 | pgid, | |
9408 | PGPeeringEventRef( | |
9409 | std::make_shared<PGPeeringEvent>( | |
9410 | m->get_epoch(), m->get_epoch(), | |
9411 | PG::DeleteStart()))); | |
7c673cae | 9412 | } |
11fdf7f2 | 9413 | m->put(); |
7c673cae FG |
9414 | } |
9415 | ||
11fdf7f2 | 9416 | void OSD::handle_fast_force_recovery(MOSDForceRecovery *m) |
c07f9fc5 | 9417 | { |
11fdf7f2 TL |
9418 | dout(10) << __func__ << " " << *m << dendl; |
9419 | if (!require_mon_or_mgr_peer(m)) { | |
9420 | m->put(); | |
9421 | return; | |
9422 | } | |
9423 | epoch_t epoch = get_osdmap_epoch(); | |
9424 | for (auto pgid : m->forced_pgs) { | |
9425 | if (m->options & OFR_BACKFILL) { | |
9426 | if (m->options & OFR_CANCEL) { | |
9427 | enqueue_peering_evt( | |
9428 | pgid, | |
9429 | PGPeeringEventRef( | |
9430 | std::make_shared<PGPeeringEvent>( | |
9431 | epoch, epoch, | |
9432 | PG::UnsetForceBackfill()))); | |
9433 | } else { | |
9434 | enqueue_peering_evt( | |
9435 | pgid, | |
9436 | PGPeeringEventRef( | |
9437 | std::make_shared<PGPeeringEvent>( | |
9438 | epoch, epoch, | |
9439 | PG::SetForceBackfill()))); | |
9440 | } | |
9441 | } else if (m->options & OFR_RECOVERY) { | |
9442 | if (m->options & OFR_CANCEL) { | |
9443 | enqueue_peering_evt( | |
9444 | pgid, | |
9445 | PGPeeringEventRef( | |
9446 | std::make_shared<PGPeeringEvent>( | |
9447 | epoch, epoch, | |
9448 | PG::UnsetForceRecovery()))); | |
9449 | } else { | |
9450 | enqueue_peering_evt( | |
9451 | pgid, | |
9452 | PGPeeringEventRef( | |
9453 | std::make_shared<PGPeeringEvent>( | |
9454 | epoch, epoch, | |
9455 | PG::SetForceRecovery()))); | |
c07f9fc5 FG |
9456 | } |
9457 | } | |
9458 | } | |
11fdf7f2 | 9459 | m->put(); |
c07f9fc5 | 9460 | } |
7c673cae | 9461 | |
11fdf7f2 | 9462 | void OSD::handle_pg_query_nopg(const MQuery& q) |
7c673cae | 9463 | { |
11fdf7f2 TL |
9464 | spg_t pgid = q.pgid; |
9465 | dout(10) << __func__ << " " << pgid << dendl; | |
7c673cae | 9466 | |
11fdf7f2 TL |
9467 | OSDMapRef osdmap = get_osdmap(); |
9468 | if (!osdmap->have_pg_pool(pgid.pool())) | |
7c673cae FG |
9469 | return; |
9470 | ||
11fdf7f2 TL |
9471 | dout(10) << " pg " << pgid << " dne" << dendl; |
9472 | pg_info_t empty(spg_t(pgid.pgid, q.query.to)); | |
9473 | ConnectionRef con = service.get_con_osd_cluster(q.from.osd, osdmap->get_epoch()); | |
9474 | if (con) { | |
9475 | Message *m; | |
9476 | if (q.query.type == pg_query_t::LOG || | |
9477 | q.query.type == pg_query_t::FULLLOG) { | |
9478 | m = new MOSDPGLog( | |
9479 | q.query.from, q.query.to, | |
9480 | osdmap->get_epoch(), empty, | |
9481 | q.query.epoch_sent); | |
7c673cae | 9482 | } else { |
11fdf7f2 TL |
9483 | vector<pair<pg_notify_t,PastIntervals>> ls; |
9484 | ls.push_back( | |
7c673cae FG |
9485 | make_pair( |
9486 | pg_notify_t( | |
11fdf7f2 TL |
9487 | q.query.from, q.query.to, |
9488 | q.query.epoch_sent, | |
7c673cae FG |
9489 | osdmap->get_epoch(), |
9490 | empty), | |
11fdf7f2 TL |
9491 | PastIntervals())); |
9492 | m = new MOSDPGNotify(osdmap->get_epoch(), ls); | |
7c673cae | 9493 | } |
11fdf7f2 TL |
9494 | service.share_map_peer(q.from.osd, con.get(), osdmap); |
9495 | con->send_message(m); | |
7c673cae FG |
9496 | } |
9497 | } | |
9498 | ||
7c673cae | 9499 | |
7c673cae FG |
9500 | // ========================================================= |
9501 | // RECOVERY | |
9502 | ||
9503 | void OSDService::_maybe_queue_recovery() { | |
11fdf7f2 | 9504 | ceph_assert(recovery_lock.is_locked_by_me()); |
7c673cae FG |
9505 | uint64_t available_pushes; |
9506 | while (!awaiting_throttle.empty() && | |
9507 | _recover_now(&available_pushes)) { | |
11fdf7f2 | 9508 | uint64_t to_start = std::min( |
7c673cae FG |
9509 | available_pushes, |
9510 | cct->_conf->osd_recovery_max_single_start); | |
9511 | _queue_for_recovery(awaiting_throttle.front(), to_start); | |
9512 | awaiting_throttle.pop_front(); | |
11fdf7f2 TL |
9513 | dout(10) << __func__ << " starting " << to_start |
9514 | << ", recovery_ops_reserved " << recovery_ops_reserved | |
9515 | << " -> " << (recovery_ops_reserved + to_start) << dendl; | |
7c673cae FG |
9516 | recovery_ops_reserved += to_start; |
9517 | } | |
9518 | } | |
9519 | ||
9520 | bool OSDService::_recover_now(uint64_t *available_pushes) | |
9521 | { | |
9522 | if (available_pushes) | |
9523 | *available_pushes = 0; | |
9524 | ||
9525 | if (ceph_clock_now() < defer_recovery_until) { | |
9526 | dout(15) << __func__ << " defer until " << defer_recovery_until << dendl; | |
9527 | return false; | |
9528 | } | |
9529 | ||
9530 | if (recovery_paused) { | |
9531 | dout(15) << __func__ << " paused" << dendl; | |
9532 | return false; | |
9533 | } | |
9534 | ||
9535 | uint64_t max = cct->_conf->osd_recovery_max_active; | |
9536 | if (max <= recovery_ops_active + recovery_ops_reserved) { | |
9537 | dout(15) << __func__ << " active " << recovery_ops_active | |
9538 | << " + reserved " << recovery_ops_reserved | |
9539 | << " >= max " << max << dendl; | |
9540 | return false; | |
9541 | } | |
9542 | ||
9543 | if (available_pushes) | |
9544 | *available_pushes = max - recovery_ops_active - recovery_ops_reserved; | |
9545 | ||
9546 | return true; | |
9547 | } | |
9548 | ||
9549 | void OSD::do_recovery( | |
9550 | PG *pg, epoch_t queued, uint64_t reserved_pushes, | |
9551 | ThreadPool::TPHandle &handle) | |
9552 | { | |
9553 | uint64_t started = 0; | |
31f18b77 FG |
9554 | |
9555 | /* | |
9556 | * When the value of osd_recovery_sleep is set greater than zero, recovery | |
9557 | * ops are scheduled after osd_recovery_sleep amount of time from the previous | |
9558 | * recovery event's schedule time. This is done by adding a | |
9559 | * recovery_requeue_callback event, which re-queues the recovery op using | |
9560 | * queue_recovery_after_sleep. | |
9561 | */ | |
c07f9fc5 | 9562 | float recovery_sleep = get_osd_recovery_sleep(); |
b32b8144 | 9563 | { |
11fdf7f2 | 9564 | std::lock_guard l(service.sleep_lock); |
b32b8144 FG |
9565 | if (recovery_sleep > 0 && service.recovery_needs_sleep) { |
9566 | PGRef pgref(pg); | |
9567 | auto recovery_requeue_callback = new FunctionContext([this, pgref, queued, reserved_pushes](int r) { | |
9568 | dout(20) << "do_recovery wake up at " | |
9569 | << ceph_clock_now() | |
9570 | << ", re-queuing recovery" << dendl; | |
11fdf7f2 | 9571 | std::lock_guard l(service.sleep_lock); |
b32b8144 FG |
9572 | service.recovery_needs_sleep = false; |
9573 | service.queue_recovery_after_sleep(pgref.get(), queued, reserved_pushes); | |
9574 | }); | |
9575 | ||
9576 | // This is true for the first recovery op and when the previous recovery op | |
9577 | // has been scheduled in the past. The next recovery op is scheduled after | |
9578 | // completing the sleep from now. | |
9579 | if (service.recovery_schedule_time < ceph_clock_now()) { | |
9580 | service.recovery_schedule_time = ceph_clock_now(); | |
9581 | } | |
9582 | service.recovery_schedule_time += recovery_sleep; | |
11fdf7f2 | 9583 | service.sleep_timer.add_event_at(service.recovery_schedule_time, |
b32b8144 FG |
9584 | recovery_requeue_callback); |
9585 | dout(20) << "Recovery event scheduled at " | |
9586 | << service.recovery_schedule_time << dendl; | |
9587 | return; | |
9588 | } | |
7c673cae FG |
9589 | } |
9590 | ||
9591 | { | |
b32b8144 | 9592 | { |
11fdf7f2 | 9593 | std::lock_guard l(service.sleep_lock); |
b32b8144 FG |
9594 | service.recovery_needs_sleep = true; |
9595 | } | |
9596 | ||
7c673cae FG |
9597 | if (pg->pg_has_reset_since(queued)) { |
9598 | goto out; | |
9599 | } | |
9600 | ||
7c673cae FG |
9601 | dout(10) << "do_recovery starting " << reserved_pushes << " " << *pg << dendl; |
9602 | #ifdef DEBUG_RECOVERY_OIDS | |
11fdf7f2 | 9603 | dout(20) << " active was " << service.recovery_oids[pg->pg_id] << dendl; |
7c673cae FG |
9604 | #endif |
9605 | ||
11fdf7f2 | 9606 | bool do_unfound = pg->start_recovery_ops(reserved_pushes, handle, &started); |
7c673cae FG |
9607 | dout(10) << "do_recovery started " << started << "/" << reserved_pushes |
9608 | << " on " << *pg << dendl; | |
9609 | ||
11fdf7f2 TL |
9610 | if (do_unfound) { |
9611 | PG::RecoveryCtx rctx = create_context(); | |
9612 | rctx.handle = &handle; | |
9613 | pg->find_unfound(queued, &rctx); | |
9614 | dispatch_context(rctx, pg, pg->get_osdmap()); | |
7c673cae | 9615 | } |
7c673cae FG |
9616 | } |
9617 | ||
9618 | out: | |
11fdf7f2 | 9619 | ceph_assert(started <= reserved_pushes); |
7c673cae FG |
9620 | service.release_reserved_pushes(reserved_pushes); |
9621 | } | |
9622 | ||
9623 | void OSDService::start_recovery_op(PG *pg, const hobject_t& soid) | |
9624 | { | |
11fdf7f2 | 9625 | std::lock_guard l(recovery_lock); |
7c673cae FG |
9626 | dout(10) << "start_recovery_op " << *pg << " " << soid |
9627 | << " (" << recovery_ops_active << "/" | |
9628 | << cct->_conf->osd_recovery_max_active << " rops)" | |
9629 | << dendl; | |
9630 | recovery_ops_active++; | |
9631 | ||
9632 | #ifdef DEBUG_RECOVERY_OIDS | |
11fdf7f2 TL |
9633 | dout(20) << " active was " << recovery_oids[pg->pg_id] << dendl; |
9634 | ceph_assert(recovery_oids[pg->pg_id].count(soid) == 0); | |
9635 | recovery_oids[pg->pg_id].insert(soid); | |
7c673cae FG |
9636 | #endif |
9637 | } | |
9638 | ||
9639 | void OSDService::finish_recovery_op(PG *pg, const hobject_t& soid, bool dequeue) | |
9640 | { | |
11fdf7f2 | 9641 | std::lock_guard l(recovery_lock); |
7c673cae FG |
9642 | dout(10) << "finish_recovery_op " << *pg << " " << soid |
9643 | << " dequeue=" << dequeue | |
9644 | << " (" << recovery_ops_active << "/" << cct->_conf->osd_recovery_max_active << " rops)" | |
9645 | << dendl; | |
9646 | ||
9647 | // adjust count | |
11fdf7f2 | 9648 | ceph_assert(recovery_ops_active > 0); |
7c673cae FG |
9649 | recovery_ops_active--; |
9650 | ||
9651 | #ifdef DEBUG_RECOVERY_OIDS | |
11fdf7f2 TL |
9652 | dout(20) << " active oids was " << recovery_oids[pg->pg_id] << dendl; |
9653 | ceph_assert(recovery_oids[pg->pg_id].count(soid)); | |
9654 | recovery_oids[pg->pg_id].erase(soid); | |
7c673cae FG |
9655 | #endif |
9656 | ||
9657 | _maybe_queue_recovery(); | |
9658 | } | |
9659 | ||
9660 | bool OSDService::is_recovery_active() | |
9661 | { | |
b5b8bbf5 | 9662 | return local_reserver.has_reservation() || remote_reserver.has_reservation(); |
7c673cae FG |
9663 | } |
9664 | ||
11fdf7f2 TL |
9665 | void OSDService::release_reserved_pushes(uint64_t pushes) |
9666 | { | |
9667 | std::lock_guard l(recovery_lock); | |
9668 | dout(10) << __func__ << "(" << pushes << "), recovery_ops_reserved " | |
9669 | << recovery_ops_reserved << " -> " << (recovery_ops_reserved-pushes) | |
9670 | << dendl; | |
9671 | ceph_assert(recovery_ops_reserved >= pushes); | |
9672 | recovery_ops_reserved -= pushes; | |
9673 | _maybe_queue_recovery(); | |
9674 | } | |
9675 | ||
7c673cae FG |
9676 | // ========================================================= |
9677 | // OPS | |
9678 | ||
9679 | bool OSD::op_is_discardable(const MOSDOp *op) | |
9680 | { | |
9681 | // drop client request if they are not connected and can't get the | |
9682 | // reply anyway. | |
9683 | if (!op->get_connection()->is_connected()) { | |
9684 | return true; | |
9685 | } | |
9686 | return false; | |
9687 | } | |
9688 | ||
11fdf7f2 | 9689 | void OSD::enqueue_op(spg_t pg, OpRequestRef&& op, epoch_t epoch) |
7c673cae | 9690 | { |
11fdf7f2 TL |
9691 | const utime_t stamp = op->get_req()->get_recv_stamp(); |
9692 | const utime_t latency = ceph_clock_now() - stamp; | |
9693 | const unsigned priority = op->get_req()->get_priority(); | |
9694 | const int cost = op->get_req()->get_cost(); | |
9695 | const uint64_t owner = op->get_req()->get_source().num(); | |
9696 | ||
9697 | dout(15) << "enqueue_op " << op << " prio " << priority | |
9698 | << " cost " << cost | |
7c673cae FG |
9699 | << " latency " << latency |
9700 | << " epoch " << epoch | |
9701 | << " " << *(op->get_req()) << dendl; | |
9702 | op->osd_trace.event("enqueue op"); | |
11fdf7f2 TL |
9703 | op->osd_trace.keyval("priority", priority); |
9704 | op->osd_trace.keyval("cost", cost); | |
7c673cae | 9705 | op->mark_queued_for_pg(); |
224ce89b | 9706 | logger->tinc(l_osd_op_before_queue_op_lat, latency); |
11fdf7f2 TL |
9707 | op_shardedwq.queue( |
9708 | OpQueueItem( | |
9709 | unique_ptr<OpQueueItem::OpQueueable>(new PGOpItem(pg, std::move(op))), | |
9710 | cost, priority, stamp, owner, epoch)); | |
7c673cae FG |
9711 | } |
9712 | ||
11fdf7f2 TL |
9713 | void OSD::enqueue_peering_evt(spg_t pgid, PGPeeringEventRef evt) |
9714 | { | |
9715 | dout(15) << __func__ << " " << pgid << " " << evt->get_desc() << dendl; | |
9716 | op_shardedwq.queue( | |
9717 | OpQueueItem( | |
9718 | unique_ptr<OpQueueItem::OpQueueable>(new PGPeeringItem(pgid, evt)), | |
9719 | 10, | |
9720 | cct->_conf->osd_peering_op_priority, | |
9721 | utime_t(), | |
9722 | 0, | |
9723 | evt->get_epoch_sent())); | |
9724 | } | |
7c673cae | 9725 | |
11fdf7f2 TL |
9726 | void OSD::enqueue_peering_evt_front(spg_t pgid, PGPeeringEventRef evt) |
9727 | { | |
9728 | dout(15) << __func__ << " " << pgid << " " << evt->get_desc() << dendl; | |
9729 | op_shardedwq.queue_front( | |
9730 | OpQueueItem( | |
9731 | unique_ptr<OpQueueItem::OpQueueable>(new PGPeeringItem(pgid, evt)), | |
9732 | 10, | |
9733 | cct->_conf->osd_peering_op_priority, | |
9734 | utime_t(), | |
9735 | 0, | |
9736 | evt->get_epoch_sent())); | |
9737 | } | |
7c673cae FG |
9738 | |
9739 | /* | |
9740 | * NOTE: dequeue called in worker thread, with pg lock | |
9741 | */ | |
9742 | void OSD::dequeue_op( | |
9743 | PGRef pg, OpRequestRef op, | |
9744 | ThreadPool::TPHandle &handle) | |
9745 | { | |
11fdf7f2 | 9746 | FUNCTRACE(cct); |
7c673cae FG |
9747 | OID_EVENT_TRACE_WITH_MSG(op->get_req(), "DEQUEUE_OP_BEGIN", false); |
9748 | ||
9749 | utime_t now = ceph_clock_now(); | |
9750 | op->set_dequeued_time(now); | |
9751 | utime_t latency = now - op->get_req()->get_recv_stamp(); | |
9752 | dout(10) << "dequeue_op " << op << " prio " << op->get_req()->get_priority() | |
9753 | << " cost " << op->get_req()->get_cost() | |
9754 | << " latency " << latency | |
9755 | << " " << *(op->get_req()) | |
9756 | << " pg " << *pg << dendl; | |
9757 | ||
224ce89b WB |
9758 | logger->tinc(l_osd_op_before_dequeue_op_lat, latency); |
9759 | ||
11fdf7f2 TL |
9760 | auto priv = op->get_req()->get_connection()->get_priv(); |
9761 | if (auto session = static_cast<Session *>(priv.get()); session) { | |
7c673cae | 9762 | maybe_share_map(session, op, pg->get_osdmap()); |
7c673cae FG |
9763 | } |
9764 | ||
11fdf7f2 | 9765 | if (pg->is_deleting()) |
7c673cae FG |
9766 | return; |
9767 | ||
9768 | op->mark_reached_pg(); | |
9769 | op->osd_trace.event("dequeue_op"); | |
9770 | ||
9771 | pg->do_request(op, handle); | |
9772 | ||
9773 | // finish | |
9774 | dout(10) << "dequeue_op " << op << " finish" << dendl; | |
9775 | OID_EVENT_TRACE_WITH_MSG(op->get_req(), "DEQUEUE_OP_END", false); | |
9776 | } | |
9777 | ||
9778 | ||
11fdf7f2 TL |
9779 | void OSD::dequeue_peering_evt( |
9780 | OSDShard *sdata, | |
9781 | PG *pg, | |
9782 | PGPeeringEventRef evt, | |
9783 | ThreadPool::TPHandle& handle) | |
7c673cae | 9784 | { |
7c673cae | 9785 | PG::RecoveryCtx rctx = create_context(); |
11fdf7f2 TL |
9786 | auto curmap = sdata->get_osdmap(); |
9787 | epoch_t need_up_thru = 0, same_interval_since = 0; | |
9788 | if (!pg) { | |
9789 | if (const MQuery *q = dynamic_cast<const MQuery*>(evt->evt.get())) { | |
9790 | handle_pg_query_nopg(*q); | |
7c673cae | 9791 | } else { |
11fdf7f2 TL |
9792 | derr << __func__ << " unrecognized pg-less event " << evt->get_desc() << dendl; |
9793 | ceph_abort(); | |
9794 | } | |
9795 | } else if (advance_pg(curmap->get_epoch(), pg, handle, &rctx)) { | |
9796 | pg->do_peering_event(evt, &rctx); | |
9797 | if (pg->is_deleted()) { | |
9798 | // do not dispatch rctx; the final _delete_some already did it. | |
9799 | discard_context(rctx); | |
9800 | pg->unlock(); | |
9801 | return; | |
7c673cae FG |
9802 | } |
9803 | dispatch_context_transaction(rctx, pg, &handle); | |
11fdf7f2 TL |
9804 | need_up_thru = pg->get_need_up_thru(); |
9805 | same_interval_since = pg->get_same_interval_since(); | |
7c673cae FG |
9806 | pg->unlock(); |
9807 | } | |
11fdf7f2 TL |
9808 | |
9809 | if (need_up_thru) { | |
7c673cae | 9810 | queue_want_up_thru(same_interval_since); |
11fdf7f2 TL |
9811 | } |
9812 | dispatch_context(rctx, pg, curmap, &handle); | |
7c673cae FG |
9813 | |
9814 | service.send_pg_temp(); | |
9815 | } | |
9816 | ||
11fdf7f2 TL |
9817 | void OSD::dequeue_delete( |
9818 | OSDShard *sdata, | |
9819 | PG *pg, | |
9820 | epoch_t e, | |
9821 | ThreadPool::TPHandle& handle) | |
9822 | { | |
9823 | dequeue_peering_evt( | |
9824 | sdata, | |
9825 | pg, | |
9826 | PGPeeringEventRef( | |
9827 | std::make_shared<PGPeeringEvent>( | |
9828 | e, e, | |
9829 | PG::DeleteSome())), | |
9830 | handle); | |
9831 | } | |
9832 | ||
9833 | ||
9834 | ||
7c673cae FG |
9835 | // -------------------------------- |
9836 | ||
9837 | const char** OSD::get_tracked_conf_keys() const | |
9838 | { | |
9839 | static const char* KEYS[] = { | |
9840 | "osd_max_backfills", | |
9841 | "osd_min_recovery_priority", | |
224ce89b WB |
9842 | "osd_max_trimming_pgs", |
9843 | "osd_op_complaint_time", | |
9844 | "osd_op_log_threshold", | |
9845 | "osd_op_history_size", | |
9846 | "osd_op_history_duration", | |
9847 | "osd_op_history_slow_op_size", | |
9848 | "osd_op_history_slow_op_threshold", | |
7c673cae FG |
9849 | "osd_enable_op_tracker", |
9850 | "osd_map_cache_size", | |
11fdf7f2 | 9851 | "osd_pg_epoch_max_lag_factor", |
7c673cae | 9852 | "osd_pg_epoch_persisted_max_stale", |
7c673cae FG |
9853 | // clog & admin clog |
9854 | "clog_to_monitors", | |
9855 | "clog_to_syslog", | |
9856 | "clog_to_syslog_facility", | |
9857 | "clog_to_syslog_level", | |
9858 | "osd_objectstore_fuse", | |
9859 | "clog_to_graylog", | |
9860 | "clog_to_graylog_host", | |
9861 | "clog_to_graylog_port", | |
9862 | "host", | |
9863 | "fsid", | |
9864 | "osd_recovery_delay_start", | |
9865 | "osd_client_message_size_cap", | |
9866 | "osd_client_message_cap", | |
31f18b77 FG |
9867 | "osd_heartbeat_min_size", |
9868 | "osd_heartbeat_interval", | |
7c673cae FG |
9869 | NULL |
9870 | }; | |
9871 | return KEYS; | |
9872 | } | |
9873 | ||
11fdf7f2 | 9874 | void OSD::handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
9875 | const std::set <std::string> &changed) |
9876 | { | |
f64942e4 | 9877 | Mutex::Locker l(osd_lock); |
7c673cae FG |
9878 | if (changed.count("osd_max_backfills")) { |
9879 | service.local_reserver.set_max(cct->_conf->osd_max_backfills); | |
9880 | service.remote_reserver.set_max(cct->_conf->osd_max_backfills); | |
9881 | } | |
9882 | if (changed.count("osd_min_recovery_priority")) { | |
9883 | service.local_reserver.set_min_priority(cct->_conf->osd_min_recovery_priority); | |
9884 | service.remote_reserver.set_min_priority(cct->_conf->osd_min_recovery_priority); | |
9885 | } | |
9886 | if (changed.count("osd_max_trimming_pgs")) { | |
9887 | service.snap_reserver.set_max(cct->_conf->osd_max_trimming_pgs); | |
9888 | } | |
9889 | if (changed.count("osd_op_complaint_time") || | |
9890 | changed.count("osd_op_log_threshold")) { | |
9891 | op_tracker.set_complaint_and_threshold(cct->_conf->osd_op_complaint_time, | |
9892 | cct->_conf->osd_op_log_threshold); | |
9893 | } | |
9894 | if (changed.count("osd_op_history_size") || | |
9895 | changed.count("osd_op_history_duration")) { | |
9896 | op_tracker.set_history_size_and_duration(cct->_conf->osd_op_history_size, | |
9897 | cct->_conf->osd_op_history_duration); | |
9898 | } | |
9899 | if (changed.count("osd_op_history_slow_op_size") || | |
9900 | changed.count("osd_op_history_slow_op_threshold")) { | |
9901 | op_tracker.set_history_slow_op_size_and_threshold(cct->_conf->osd_op_history_slow_op_size, | |
9902 | cct->_conf->osd_op_history_slow_op_threshold); | |
9903 | } | |
9904 | if (changed.count("osd_enable_op_tracker")) { | |
9905 | op_tracker.set_tracking(cct->_conf->osd_enable_op_tracker); | |
9906 | } | |
7c673cae FG |
9907 | if (changed.count("osd_map_cache_size")) { |
9908 | service.map_cache.set_size(cct->_conf->osd_map_cache_size); | |
9909 | service.map_bl_cache.set_size(cct->_conf->osd_map_cache_size); | |
9910 | service.map_bl_inc_cache.set_size(cct->_conf->osd_map_cache_size); | |
9911 | } | |
9912 | if (changed.count("clog_to_monitors") || | |
9913 | changed.count("clog_to_syslog") || | |
9914 | changed.count("clog_to_syslog_level") || | |
9915 | changed.count("clog_to_syslog_facility") || | |
9916 | changed.count("clog_to_graylog") || | |
9917 | changed.count("clog_to_graylog_host") || | |
9918 | changed.count("clog_to_graylog_port") || | |
9919 | changed.count("host") || | |
9920 | changed.count("fsid")) { | |
9921 | update_log_config(); | |
9922 | } | |
11fdf7f2 TL |
9923 | if (changed.count("osd_pg_epoch_max_lag_factor")) { |
9924 | m_osd_pg_epoch_max_lag_factor = conf.get_val<double>( | |
9925 | "osd_pg_epoch_max_lag_factor"); | |
9926 | } | |
7c673cae FG |
9927 | |
9928 | #ifdef HAVE_LIBFUSE | |
9929 | if (changed.count("osd_objectstore_fuse")) { | |
9930 | if (store) { | |
9931 | enable_disable_fuse(false); | |
9932 | } | |
9933 | } | |
9934 | #endif | |
9935 | ||
9936 | if (changed.count("osd_recovery_delay_start")) { | |
9937 | service.defer_recovery(cct->_conf->osd_recovery_delay_start); | |
9938 | service.kick_recovery_queue(); | |
9939 | } | |
9940 | ||
9941 | if (changed.count("osd_client_message_cap")) { | |
9942 | uint64_t newval = cct->_conf->osd_client_message_cap; | |
9943 | Messenger::Policy pol = client_messenger->get_policy(entity_name_t::TYPE_CLIENT); | |
9944 | if (pol.throttler_messages && newval > 0) { | |
9945 | pol.throttler_messages->reset_max(newval); | |
9946 | } | |
9947 | } | |
9948 | if (changed.count("osd_client_message_size_cap")) { | |
9949 | uint64_t newval = cct->_conf->osd_client_message_size_cap; | |
9950 | Messenger::Policy pol = client_messenger->get_policy(entity_name_t::TYPE_CLIENT); | |
9951 | if (pol.throttler_bytes && newval > 0) { | |
9952 | pol.throttler_bytes->reset_max(newval); | |
9953 | } | |
9954 | } | |
9955 | ||
9956 | check_config(); | |
9957 | } | |
9958 | ||
9959 | void OSD::update_log_config() | |
9960 | { | |
9961 | map<string,string> log_to_monitors; | |
9962 | map<string,string> log_to_syslog; | |
9963 | map<string,string> log_channel; | |
9964 | map<string,string> log_prio; | |
9965 | map<string,string> log_to_graylog; | |
9966 | map<string,string> log_to_graylog_host; | |
9967 | map<string,string> log_to_graylog_port; | |
9968 | uuid_d fsid; | |
9969 | string host; | |
9970 | ||
9971 | if (parse_log_client_options(cct, log_to_monitors, log_to_syslog, | |
9972 | log_channel, log_prio, log_to_graylog, | |
9973 | log_to_graylog_host, log_to_graylog_port, | |
9974 | fsid, host) == 0) | |
9975 | clog->update_config(log_to_monitors, log_to_syslog, | |
9976 | log_channel, log_prio, log_to_graylog, | |
9977 | log_to_graylog_host, log_to_graylog_port, | |
9978 | fsid, host); | |
9979 | derr << "log_to_monitors " << log_to_monitors << dendl; | |
9980 | } | |
9981 | ||
9982 | void OSD::check_config() | |
9983 | { | |
9984 | // some sanity checks | |
7c673cae FG |
9985 | if (cct->_conf->osd_map_cache_size <= (int)cct->_conf->osd_pg_epoch_persisted_max_stale + 2) { |
9986 | clog->warn() << "osd_map_cache_size (" << cct->_conf->osd_map_cache_size << ")" | |
9987 | << " is not > osd_pg_epoch_persisted_max_stale (" | |
9988 | << cct->_conf->osd_pg_epoch_persisted_max_stale << ")"; | |
9989 | } | |
9990 | } | |
9991 | ||
7c673cae FG |
9992 | // -------------------------------- |
9993 | ||
9994 | void OSD::get_latest_osdmap() | |
9995 | { | |
9996 | dout(10) << __func__ << " -- start" << dendl; | |
9997 | ||
9998 | C_SaferCond cond; | |
9999 | service.objecter->wait_for_latest_osdmap(&cond); | |
10000 | cond.wait(); | |
10001 | ||
10002 | dout(10) << __func__ << " -- finish" << dendl; | |
10003 | } | |
10004 | ||
10005 | // -------------------------------- | |
10006 | ||
10007 | int OSD::init_op_flags(OpRequestRef& op) | |
10008 | { | |
10009 | const MOSDOp *m = static_cast<const MOSDOp*>(op->get_req()); | |
10010 | vector<OSDOp>::const_iterator iter; | |
10011 | ||
10012 | // client flags have no bearing on whether an op is a read, write, etc. | |
10013 | op->rmw_flags = 0; | |
10014 | ||
10015 | if (m->has_flag(CEPH_OSD_FLAG_RWORDERED)) { | |
10016 | op->set_force_rwordered(); | |
10017 | } | |
10018 | ||
10019 | // set bits based on op codes, called methods. | |
10020 | for (iter = m->ops.begin(); iter != m->ops.end(); ++iter) { | |
10021 | if ((iter->op.op == CEPH_OSD_OP_WATCH && | |
10022 | iter->op.watch.op == CEPH_OSD_WATCH_OP_PING)) { | |
10023 | /* This a bit odd. PING isn't actually a write. It can't | |
11fdf7f2 TL |
10024 | * result in an update to the object_info. PINGs also aren't |
10025 | * resent, so there's no reason to write out a log entry. | |
7c673cae FG |
10026 | * |
10027 | * However, we pipeline them behind writes, so let's force | |
10028 | * the write_ordered flag. | |
10029 | */ | |
10030 | op->set_force_rwordered(); | |
10031 | } else { | |
10032 | if (ceph_osd_op_mode_modify(iter->op.op)) | |
10033 | op->set_write(); | |
10034 | } | |
10035 | if (ceph_osd_op_mode_read(iter->op.op)) | |
10036 | op->set_read(); | |
10037 | ||
10038 | // set READ flag if there are src_oids | |
10039 | if (iter->soid.oid.name.length()) | |
10040 | op->set_read(); | |
10041 | ||
10042 | // set PGOP flag if there are PG ops | |
10043 | if (ceph_osd_op_type_pg(iter->op.op)) | |
10044 | op->set_pg_op(); | |
10045 | ||
10046 | if (ceph_osd_op_mode_cache(iter->op.op)) | |
10047 | op->set_cache(); | |
10048 | ||
10049 | // check for ec base pool | |
10050 | int64_t poolid = m->get_pg().pool(); | |
10051 | const pg_pool_t *pool = osdmap->get_pg_pool(poolid); | |
10052 | if (pool && pool->is_tier()) { | |
10053 | const pg_pool_t *base_pool = osdmap->get_pg_pool(pool->tier_of); | |
10054 | if (base_pool && base_pool->require_rollback()) { | |
10055 | if ((iter->op.op != CEPH_OSD_OP_READ) && | |
10056 | (iter->op.op != CEPH_OSD_OP_CHECKSUM) && | |
c07f9fc5 | 10057 | (iter->op.op != CEPH_OSD_OP_CMPEXT) && |
7c673cae FG |
10058 | (iter->op.op != CEPH_OSD_OP_STAT) && |
10059 | (iter->op.op != CEPH_OSD_OP_ISDIRTY) && | |
10060 | (iter->op.op != CEPH_OSD_OP_UNDIRTY) && | |
10061 | (iter->op.op != CEPH_OSD_OP_GETXATTR) && | |
10062 | (iter->op.op != CEPH_OSD_OP_GETXATTRS) && | |
10063 | (iter->op.op != CEPH_OSD_OP_CMPXATTR) && | |
10064 | (iter->op.op != CEPH_OSD_OP_ASSERT_VER) && | |
10065 | (iter->op.op != CEPH_OSD_OP_LIST_WATCHERS) && | |
10066 | (iter->op.op != CEPH_OSD_OP_LIST_SNAPS) && | |
10067 | (iter->op.op != CEPH_OSD_OP_SETALLOCHINT) && | |
10068 | (iter->op.op != CEPH_OSD_OP_WRITEFULL) && | |
10069 | (iter->op.op != CEPH_OSD_OP_ROLLBACK) && | |
10070 | (iter->op.op != CEPH_OSD_OP_CREATE) && | |
10071 | (iter->op.op != CEPH_OSD_OP_DELETE) && | |
10072 | (iter->op.op != CEPH_OSD_OP_SETXATTR) && | |
10073 | (iter->op.op != CEPH_OSD_OP_RMXATTR) && | |
10074 | (iter->op.op != CEPH_OSD_OP_STARTSYNC) && | |
10075 | (iter->op.op != CEPH_OSD_OP_COPY_GET) && | |
10076 | (iter->op.op != CEPH_OSD_OP_COPY_FROM)) { | |
10077 | op->set_promote(); | |
10078 | } | |
10079 | } | |
10080 | } | |
10081 | ||
10082 | switch (iter->op.op) { | |
10083 | case CEPH_OSD_OP_CALL: | |
10084 | { | |
10085 | bufferlist::iterator bp = const_cast<bufferlist&>(iter->indata).begin(); | |
10086 | int is_write, is_read; | |
10087 | string cname, mname; | |
10088 | bp.copy(iter->op.cls.class_len, cname); | |
10089 | bp.copy(iter->op.cls.method_len, mname); | |
10090 | ||
10091 | ClassHandler::ClassData *cls; | |
10092 | int r = class_handler->open_class(cname, &cls); | |
10093 | if (r) { | |
10094 | derr << "class " << cname << " open got " << cpp_strerror(r) << dendl; | |
10095 | if (r == -ENOENT) | |
10096 | r = -EOPNOTSUPP; | |
10097 | else if (r != -EPERM) // propagate permission errors | |
10098 | r = -EIO; | |
10099 | return r; | |
10100 | } | |
10101 | int flags = cls->get_method_flags(mname.c_str()); | |
10102 | if (flags < 0) { | |
10103 | if (flags == -ENOENT) | |
10104 | r = -EOPNOTSUPP; | |
10105 | else | |
10106 | r = flags; | |
10107 | return r; | |
10108 | } | |
10109 | is_read = flags & CLS_METHOD_RD; | |
10110 | is_write = flags & CLS_METHOD_WR; | |
10111 | bool is_promote = flags & CLS_METHOD_PROMOTE; | |
10112 | ||
10113 | dout(10) << "class " << cname << " method " << mname << " " | |
10114 | << "flags=" << (is_read ? "r" : "") | |
10115 | << (is_write ? "w" : "") | |
10116 | << (is_promote ? "p" : "") | |
10117 | << dendl; | |
10118 | if (is_read) | |
10119 | op->set_class_read(); | |
10120 | if (is_write) | |
10121 | op->set_class_write(); | |
10122 | if (is_promote) | |
10123 | op->set_promote(); | |
11fdf7f2 TL |
10124 | op->add_class(std::move(cname), std::move(mname), is_read, is_write, |
10125 | cls->whitelisted); | |
7c673cae FG |
10126 | break; |
10127 | } | |
10128 | ||
10129 | case CEPH_OSD_OP_WATCH: | |
10130 | // force the read bit for watch since it is depends on previous | |
10131 | // watch state (and may return early if the watch exists) or, in | |
10132 | // the case of ping, is simply a read op. | |
10133 | op->set_read(); | |
10134 | // fall through | |
10135 | case CEPH_OSD_OP_NOTIFY: | |
10136 | case CEPH_OSD_OP_NOTIFY_ACK: | |
10137 | { | |
10138 | op->set_promote(); | |
10139 | break; | |
10140 | } | |
10141 | ||
10142 | case CEPH_OSD_OP_DELETE: | |
10143 | // if we get a delete with FAILOK we can skip handle cache. without | |
10144 | // FAILOK we still need to promote (or do something smarter) to | |
10145 | // determine whether to return ENOENT or 0. | |
10146 | if (iter == m->ops.begin() && | |
10147 | iter->op.flags == CEPH_OSD_OP_FLAG_FAILOK) { | |
10148 | op->set_skip_handle_cache(); | |
10149 | } | |
10150 | // skip promotion when proxying a delete op | |
10151 | if (m->ops.size() == 1) { | |
10152 | op->set_skip_promote(); | |
10153 | } | |
10154 | break; | |
10155 | ||
10156 | case CEPH_OSD_OP_CACHE_TRY_FLUSH: | |
10157 | case CEPH_OSD_OP_CACHE_FLUSH: | |
10158 | case CEPH_OSD_OP_CACHE_EVICT: | |
10159 | // If try_flush/flush/evict is the only op, can skip handle cache. | |
10160 | if (m->ops.size() == 1) { | |
10161 | op->set_skip_handle_cache(); | |
10162 | } | |
10163 | break; | |
10164 | ||
10165 | case CEPH_OSD_OP_READ: | |
10166 | case CEPH_OSD_OP_SYNC_READ: | |
10167 | case CEPH_OSD_OP_SPARSE_READ: | |
10168 | case CEPH_OSD_OP_CHECKSUM: | |
10169 | case CEPH_OSD_OP_WRITEFULL: | |
10170 | if (m->ops.size() == 1 && | |
10171 | (iter->op.flags & CEPH_OSD_OP_FLAG_FADVISE_NOCACHE || | |
10172 | iter->op.flags & CEPH_OSD_OP_FLAG_FADVISE_DONTNEED)) { | |
10173 | op->set_skip_promote(); | |
10174 | } | |
10175 | break; | |
10176 | ||
10177 | // force promotion when pin an object in cache tier | |
10178 | case CEPH_OSD_OP_CACHE_PIN: | |
10179 | op->set_promote(); | |
10180 | break; | |
10181 | ||
10182 | default: | |
10183 | break; | |
10184 | } | |
10185 | } | |
10186 | ||
10187 | if (op->rmw_flags == 0) | |
10188 | return -EINVAL; | |
10189 | ||
10190 | return 0; | |
10191 | } | |
10192 | ||
11fdf7f2 TL |
10193 | void OSD::set_perf_queries( |
10194 | const std::map<OSDPerfMetricQuery, OSDPerfMetricLimits> &queries) { | |
10195 | dout(10) << "setting " << queries.size() << " queries" << dendl; | |
10196 | ||
10197 | std::list<OSDPerfMetricQuery> supported_queries; | |
10198 | for (auto &it : queries) { | |
10199 | auto &query = it.first; | |
10200 | if (!query.key_descriptor.empty()) { | |
10201 | supported_queries.push_back(query); | |
10202 | } | |
10203 | } | |
10204 | if (supported_queries.size() < queries.size()) { | |
10205 | dout(1) << queries.size() - supported_queries.size() | |
10206 | << " unsupported queries" << dendl; | |
10207 | } | |
10208 | ||
10209 | { | |
10210 | Mutex::Locker locker(m_perf_queries_lock); | |
10211 | m_perf_queries = supported_queries; | |
10212 | m_perf_limits = queries; | |
10213 | } | |
10214 | ||
10215 | std::vector<PGRef> pgs; | |
10216 | _get_pgs(&pgs); | |
10217 | for (auto& pg : pgs) { | |
10218 | if (pg->is_primary()) { | |
10219 | pg->lock(); | |
10220 | pg->set_dynamic_perf_stats_queries(supported_queries); | |
10221 | pg->unlock(); | |
10222 | } | |
7c673cae | 10223 | } |
7c673cae FG |
10224 | } |
10225 | ||
11fdf7f2 TL |
10226 | void OSD::get_perf_reports( |
10227 | std::map<OSDPerfMetricQuery, OSDPerfMetricReport> *reports) { | |
10228 | std::vector<PGRef> pgs; | |
10229 | _get_pgs(&pgs); | |
10230 | DynamicPerfStats dps; | |
10231 | for (auto& pg : pgs) { | |
10232 | if (pg->is_primary()) { | |
10233 | // m_perf_queries can be modified only in set_perf_queries by mgr client | |
10234 | // request, and it is protected by by mgr client's lock, which is held | |
10235 | // when set_perf_queries/get_perf_reports are called, so we may not hold | |
10236 | // m_perf_queries_lock here. | |
10237 | DynamicPerfStats pg_dps(m_perf_queries); | |
10238 | pg->lock(); | |
10239 | pg->get_dynamic_perf_stats(&pg_dps); | |
10240 | pg->unlock(); | |
10241 | dps.merge(pg_dps); | |
10242 | } | |
10243 | } | |
10244 | dps.add_to_reports(m_perf_limits, reports); | |
10245 | dout(20) << "reports for " << reports->size() << " queries" << dendl; | |
10246 | } | |
224ce89b | 10247 | |
7c673cae FG |
10248 | // ============================================================= |
10249 | ||
10250 | #undef dout_context | |
11fdf7f2 | 10251 | #define dout_context cct |
7c673cae | 10252 | #undef dout_prefix |
11fdf7f2 | 10253 | #define dout_prefix *_dout << "osd." << osd->get_nodeid() << ":" << shard_id << "." << __func__ << " " |
7c673cae | 10254 | |
11fdf7f2 | 10255 | void OSDShard::_attach_pg(OSDShardPGSlot *slot, PG *pg) |
7c673cae | 10256 | { |
11fdf7f2 TL |
10257 | dout(10) << pg->pg_id << " " << pg << dendl; |
10258 | slot->pg = pg; | |
10259 | pg->osd_shard = this; | |
10260 | pg->pg_slot = slot; | |
10261 | osd->inc_num_pgs(); | |
10262 | ||
10263 | slot->epoch = pg->get_osdmap_epoch(); | |
10264 | pg_slots_by_epoch.insert(*slot); | |
10265 | } | |
10266 | ||
10267 | void OSDShard::_detach_pg(OSDShardPGSlot *slot) | |
10268 | { | |
10269 | dout(10) << slot->pg->pg_id << " " << slot->pg << dendl; | |
10270 | slot->pg->osd_shard = nullptr; | |
10271 | slot->pg->pg_slot = nullptr; | |
10272 | slot->pg = nullptr; | |
10273 | osd->dec_num_pgs(); | |
10274 | ||
10275 | pg_slots_by_epoch.erase(pg_slots_by_epoch.iterator_to(*slot)); | |
10276 | slot->epoch = 0; | |
10277 | if (waiting_for_min_pg_epoch) { | |
10278 | min_pg_epoch_cond.notify_all(); | |
10279 | } | |
10280 | } | |
10281 | ||
10282 | void OSDShard::update_pg_epoch(OSDShardPGSlot *slot, epoch_t e) | |
10283 | { | |
10284 | std::lock_guard l(shard_lock); | |
10285 | dout(30) << "min was " << pg_slots_by_epoch.begin()->epoch | |
10286 | << " on " << pg_slots_by_epoch.begin()->pg->pg_id << dendl; | |
10287 | pg_slots_by_epoch.erase(pg_slots_by_epoch.iterator_to(*slot)); | |
10288 | dout(20) << slot->pg->pg_id << " " << slot->epoch << " -> " << e << dendl; | |
10289 | slot->epoch = e; | |
10290 | pg_slots_by_epoch.insert(*slot); | |
10291 | dout(30) << "min is now " << pg_slots_by_epoch.begin()->epoch | |
10292 | << " on " << pg_slots_by_epoch.begin()->pg->pg_id << dendl; | |
10293 | if (waiting_for_min_pg_epoch) { | |
10294 | min_pg_epoch_cond.notify_all(); | |
10295 | } | |
10296 | } | |
10297 | ||
10298 | epoch_t OSDShard::get_min_pg_epoch() | |
10299 | { | |
10300 | std::lock_guard l(shard_lock); | |
10301 | auto p = pg_slots_by_epoch.begin(); | |
10302 | if (p == pg_slots_by_epoch.end()) { | |
10303 | return 0; | |
10304 | } | |
10305 | return p->epoch; | |
10306 | } | |
10307 | ||
10308 | void OSDShard::wait_min_pg_epoch(epoch_t need) | |
10309 | { | |
10310 | std::unique_lock l{shard_lock}; | |
10311 | ++waiting_for_min_pg_epoch; | |
10312 | min_pg_epoch_cond.wait(l, [need, this] { | |
10313 | if (pg_slots_by_epoch.empty()) { | |
10314 | return true; | |
10315 | } else if (pg_slots_by_epoch.begin()->epoch >= need) { | |
10316 | return true; | |
10317 | } else { | |
10318 | dout(10) << need << " waiting on " | |
10319 | << pg_slots_by_epoch.begin()->epoch << dendl; | |
10320 | return false; | |
10321 | } | |
10322 | }); | |
10323 | --waiting_for_min_pg_epoch; | |
10324 | } | |
10325 | ||
10326 | epoch_t OSDShard::get_max_waiting_epoch() | |
10327 | { | |
10328 | std::lock_guard l(shard_lock); | |
10329 | epoch_t r = 0; | |
10330 | for (auto& i : pg_slots) { | |
10331 | if (!i.second->waiting_peering.empty()) { | |
10332 | r = std::max(r, i.second->waiting_peering.rbegin()->first); | |
10333 | } | |
10334 | } | |
10335 | return r; | |
10336 | } | |
10337 | ||
10338 | void OSDShard::consume_map( | |
10339 | OSDMapRef& new_osdmap, | |
10340 | unsigned *pushes_to_free) | |
10341 | { | |
10342 | std::lock_guard l(shard_lock); | |
10343 | OSDMapRef old_osdmap; | |
7c673cae | 10344 | { |
11fdf7f2 TL |
10345 | std::lock_guard l(osdmap_lock); |
10346 | old_osdmap = std::move(shard_osdmap); | |
10347 | shard_osdmap = new_osdmap; | |
10348 | } | |
10349 | dout(10) << new_osdmap->get_epoch() | |
10350 | << " (was " << (old_osdmap ? old_osdmap->get_epoch() : 0) << ")" | |
10351 | << dendl; | |
10352 | bool queued = false; | |
10353 | ||
10354 | // check slots | |
10355 | auto p = pg_slots.begin(); | |
10356 | while (p != pg_slots.end()) { | |
10357 | OSDShardPGSlot *slot = p->second.get(); | |
10358 | const spg_t& pgid = p->first; | |
10359 | dout(20) << __func__ << " " << pgid << dendl; | |
10360 | if (!slot->waiting_for_split.empty()) { | |
10361 | dout(20) << __func__ << " " << pgid | |
10362 | << " waiting for split " << slot->waiting_for_split << dendl; | |
10363 | ++p; | |
10364 | continue; | |
10365 | } | |
10366 | if (slot->waiting_for_merge_epoch > new_osdmap->get_epoch()) { | |
10367 | dout(20) << __func__ << " " << pgid | |
10368 | << " waiting for merge by epoch " << slot->waiting_for_merge_epoch | |
10369 | << dendl; | |
10370 | ++p; | |
10371 | continue; | |
10372 | } | |
10373 | if (!slot->waiting_peering.empty()) { | |
10374 | epoch_t first = slot->waiting_peering.begin()->first; | |
10375 | if (first <= new_osdmap->get_epoch()) { | |
10376 | dout(20) << __func__ << " " << pgid | |
10377 | << " pending_peering first epoch " << first | |
10378 | << " <= " << new_osdmap->get_epoch() << ", requeueing" << dendl; | |
10379 | _wake_pg_slot(pgid, slot); | |
10380 | queued = true; | |
10381 | } | |
10382 | ++p; | |
10383 | continue; | |
10384 | } | |
10385 | if (!slot->waiting.empty()) { | |
10386 | if (new_osdmap->is_up_acting_osd_shard(pgid, osd->get_nodeid())) { | |
10387 | dout(20) << __func__ << " " << pgid << " maps to us, keeping" | |
10388 | << dendl; | |
10389 | ++p; | |
10390 | continue; | |
7c673cae | 10391 | } |
11fdf7f2 TL |
10392 | while (!slot->waiting.empty() && |
10393 | slot->waiting.front().get_map_epoch() <= new_osdmap->get_epoch()) { | |
10394 | auto& qi = slot->waiting.front(); | |
10395 | dout(20) << __func__ << " " << pgid | |
10396 | << " waiting item " << qi | |
10397 | << " epoch " << qi.get_map_epoch() | |
10398 | << " <= " << new_osdmap->get_epoch() | |
10399 | << ", " | |
10400 | << (qi.get_map_epoch() < new_osdmap->get_epoch() ? "stale" : | |
10401 | "misdirected") | |
10402 | << ", dropping" << dendl; | |
10403 | *pushes_to_free += qi.get_reserved_pushes(); | |
10404 | slot->waiting.pop_front(); | |
10405 | } | |
10406 | } | |
10407 | if (slot->waiting.empty() && | |
10408 | slot->num_running == 0 && | |
10409 | slot->waiting_for_split.empty() && | |
10410 | !slot->pg) { | |
10411 | dout(20) << __func__ << " " << pgid << " empty, pruning" << dendl; | |
10412 | p = pg_slots.erase(p); | |
10413 | continue; | |
7c673cae | 10414 | } |
11fdf7f2 TL |
10415 | |
10416 | ++p; | |
7c673cae | 10417 | } |
7c673cae | 10418 | if (queued) { |
11fdf7f2 TL |
10419 | std::lock_guard l{sdata_wait_lock}; |
10420 | sdata_cond.notify_one(); | |
7c673cae FG |
10421 | } |
10422 | } | |
10423 | ||
11fdf7f2 TL |
10424 | void OSDShard::_wake_pg_slot( |
10425 | spg_t pgid, | |
10426 | OSDShardPGSlot *slot) | |
10427 | { | |
10428 | dout(20) << __func__ << " " << pgid | |
10429 | << " to_process " << slot->to_process | |
10430 | << " waiting " << slot->waiting | |
10431 | << " waiting_peering " << slot->waiting_peering << dendl; | |
10432 | for (auto i = slot->to_process.rbegin(); | |
10433 | i != slot->to_process.rend(); | |
10434 | ++i) { | |
10435 | _enqueue_front(std::move(*i), osd->op_prio_cutoff); | |
10436 | } | |
10437 | slot->to_process.clear(); | |
10438 | for (auto i = slot->waiting.rbegin(); | |
10439 | i != slot->waiting.rend(); | |
10440 | ++i) { | |
10441 | _enqueue_front(std::move(*i), osd->op_prio_cutoff); | |
10442 | } | |
10443 | slot->waiting.clear(); | |
10444 | for (auto i = slot->waiting_peering.rbegin(); | |
10445 | i != slot->waiting_peering.rend(); | |
10446 | ++i) { | |
10447 | // this is overkill; we requeue everything, even if some of these | |
10448 | // items are waiting for maps we don't have yet. FIXME, maybe, | |
10449 | // someday, if we decide this inefficiency matters | |
10450 | for (auto j = i->second.rbegin(); j != i->second.rend(); ++j) { | |
10451 | _enqueue_front(std::move(*j), osd->op_prio_cutoff); | |
10452 | } | |
10453 | } | |
10454 | slot->waiting_peering.clear(); | |
10455 | ++slot->requeue_seq; | |
10456 | } | |
10457 | ||
10458 | void OSDShard::identify_splits_and_merges( | |
10459 | const OSDMapRef& as_of_osdmap, | |
10460 | set<pair<spg_t,epoch_t>> *split_pgs, | |
10461 | set<pair<spg_t,epoch_t>> *merge_pgs) | |
10462 | { | |
10463 | std::lock_guard l(shard_lock); | |
10464 | if (shard_osdmap) { | |
10465 | for (auto& i : pg_slots) { | |
10466 | const spg_t& pgid = i.first; | |
10467 | auto *slot = i.second.get(); | |
10468 | if (slot->pg) { | |
10469 | osd->service.identify_splits_and_merges( | |
10470 | shard_osdmap, as_of_osdmap, pgid, | |
10471 | split_pgs, merge_pgs); | |
10472 | } else if (!slot->waiting_for_split.empty()) { | |
10473 | osd->service.identify_splits_and_merges( | |
10474 | shard_osdmap, as_of_osdmap, pgid, | |
10475 | split_pgs, nullptr); | |
10476 | } else { | |
10477 | dout(20) << __func__ << " slot " << pgid | |
10478 | << " has no pg and waiting_for_split " | |
10479 | << slot->waiting_for_split << dendl; | |
7c673cae | 10480 | } |
11fdf7f2 TL |
10481 | } |
10482 | } | |
10483 | } | |
10484 | ||
10485 | void OSDShard::prime_splits(const OSDMapRef& as_of_osdmap, | |
10486 | set<pair<spg_t,epoch_t>> *pgids) | |
10487 | { | |
10488 | std::lock_guard l(shard_lock); | |
10489 | _prime_splits(pgids); | |
10490 | if (shard_osdmap->get_epoch() > as_of_osdmap->get_epoch()) { | |
10491 | set<pair<spg_t,epoch_t>> newer_children; | |
10492 | for (auto i : *pgids) { | |
10493 | osd->service.identify_splits_and_merges( | |
10494 | as_of_osdmap, shard_osdmap, i.first, | |
10495 | &newer_children, nullptr); | |
10496 | } | |
10497 | newer_children.insert(pgids->begin(), pgids->end()); | |
10498 | dout(10) << "as_of_osdmap " << as_of_osdmap->get_epoch() << " < shard " | |
10499 | << shard_osdmap->get_epoch() << ", new children " << newer_children | |
10500 | << dendl; | |
10501 | _prime_splits(&newer_children); | |
10502 | // note: we don't care what is left over here for other shards. | |
10503 | // if this shard is ahead of us and one isn't, e.g., one thread is | |
10504 | // calling into prime_splits via _process (due to a newly created | |
10505 | // pg) and this shard has a newer map due to a racing consume_map, | |
10506 | // then any grandchildren left here will be identified (or were | |
10507 | // identified) when the slower shard's osdmap is advanced. | |
10508 | // _prime_splits() will tolerate the case where the pgid is | |
10509 | // already primed. | |
10510 | } | |
10511 | } | |
10512 | ||
10513 | void OSDShard::_prime_splits(set<pair<spg_t,epoch_t>> *pgids) | |
10514 | { | |
10515 | dout(10) << *pgids << dendl; | |
10516 | auto p = pgids->begin(); | |
10517 | while (p != pgids->end()) { | |
10518 | unsigned shard_index = p->first.hash_to_shard(osd->num_shards); | |
10519 | if (shard_index == shard_id) { | |
10520 | auto r = pg_slots.emplace(p->first, nullptr); | |
10521 | if (r.second) { | |
10522 | dout(10) << "priming slot " << p->first << " e" << p->second << dendl; | |
10523 | r.first->second = make_unique<OSDShardPGSlot>(); | |
10524 | r.first->second->waiting_for_split.insert(p->second); | |
7c673cae | 10525 | } else { |
11fdf7f2 TL |
10526 | auto q = r.first; |
10527 | ceph_assert(q != pg_slots.end()); | |
10528 | dout(10) << "priming (existing) slot " << p->first << " e" << p->second | |
10529 | << dendl; | |
10530 | q->second->waiting_for_split.insert(p->second); | |
7c673cae | 10531 | } |
11fdf7f2 TL |
10532 | p = pgids->erase(p); |
10533 | } else { | |
10534 | ++p; | |
7c673cae FG |
10535 | } |
10536 | } | |
11fdf7f2 TL |
10537 | } |
10538 | ||
10539 | void OSDShard::prime_merges(const OSDMapRef& as_of_osdmap, | |
10540 | set<pair<spg_t,epoch_t>> *merge_pgs) | |
10541 | { | |
10542 | std::lock_guard l(shard_lock); | |
10543 | dout(20) << __func__ << " checking shard " << shard_id | |
10544 | << " for remaining merge pgs " << merge_pgs << dendl; | |
10545 | auto p = merge_pgs->begin(); | |
10546 | while (p != merge_pgs->end()) { | |
10547 | spg_t pgid = p->first; | |
10548 | epoch_t epoch = p->second; | |
10549 | unsigned shard_index = pgid.hash_to_shard(osd->num_shards); | |
10550 | if (shard_index != shard_id) { | |
10551 | ++p; | |
10552 | continue; | |
10553 | } | |
10554 | OSDShardPGSlot *slot; | |
10555 | auto r = pg_slots.emplace(pgid, nullptr); | |
10556 | if (r.second) { | |
10557 | r.first->second = make_unique<OSDShardPGSlot>(); | |
10558 | } | |
10559 | slot = r.first->second.get(); | |
10560 | if (slot->pg) { | |
10561 | // already have pg | |
10562 | dout(20) << __func__ << " have merge participant pg " << pgid | |
10563 | << " " << slot->pg << dendl; | |
10564 | } else if (!slot->waiting_for_split.empty() && | |
10565 | *slot->waiting_for_split.begin() < epoch) { | |
10566 | dout(20) << __func__ << " pending split on merge participant pg " << pgid | |
10567 | << " " << slot->waiting_for_split << dendl; | |
10568 | } else { | |
10569 | dout(20) << __func__ << " creating empty merge participant " << pgid | |
10570 | << " for merge in " << epoch << dendl; | |
10571 | // leave history zeroed; PG::merge_from() will fill it in. | |
10572 | pg_history_t history; | |
10573 | PGCreateInfo cinfo(pgid, epoch - 1, | |
10574 | history, PastIntervals(), false); | |
10575 | PGRef pg = osd->handle_pg_create_info(shard_osdmap, &cinfo); | |
10576 | _attach_pg(r.first->second.get(), pg.get()); | |
10577 | _wake_pg_slot(pgid, slot); | |
10578 | pg->unlock(); | |
10579 | } | |
10580 | // mark slot for merge | |
10581 | dout(20) << __func__ << " marking merge participant " << pgid << dendl; | |
10582 | slot->waiting_for_merge_epoch = epoch; | |
10583 | p = merge_pgs->erase(p); | |
7c673cae FG |
10584 | } |
10585 | } | |
10586 | ||
11fdf7f2 | 10587 | void OSDShard::register_and_wake_split_child(PG *pg) |
7c673cae | 10588 | { |
11fdf7f2 TL |
10589 | epoch_t epoch; |
10590 | { | |
10591 | std::lock_guard l(shard_lock); | |
10592 | dout(10) << pg->pg_id << " " << pg << dendl; | |
10593 | auto p = pg_slots.find(pg->pg_id); | |
10594 | ceph_assert(p != pg_slots.end()); | |
10595 | auto *slot = p->second.get(); | |
10596 | dout(20) << pg->pg_id << " waiting_for_split " << slot->waiting_for_split | |
10597 | << dendl; | |
10598 | ceph_assert(!slot->pg); | |
10599 | ceph_assert(!slot->waiting_for_split.empty()); | |
10600 | _attach_pg(slot, pg); | |
10601 | ||
10602 | epoch = pg->get_osdmap_epoch(); | |
10603 | ceph_assert(slot->waiting_for_split.count(epoch)); | |
10604 | slot->waiting_for_split.erase(epoch); | |
10605 | if (slot->waiting_for_split.empty()) { | |
10606 | _wake_pg_slot(pg->pg_id, slot); | |
10607 | } else { | |
10608 | dout(10) << __func__ << " still waiting for split on " | |
10609 | << slot->waiting_for_split << dendl; | |
10610 | } | |
7c673cae | 10611 | } |
11fdf7f2 TL |
10612 | |
10613 | // kick child to ensure it pulls up to the latest osdmap | |
10614 | osd->enqueue_peering_evt( | |
10615 | pg->pg_id, | |
10616 | PGPeeringEventRef( | |
10617 | std::make_shared<PGPeeringEvent>( | |
10618 | epoch, | |
10619 | epoch, | |
10620 | NullEvt()))); | |
10621 | ||
10622 | std::lock_guard l{sdata_wait_lock}; | |
10623 | sdata_cond.notify_one(); | |
7c673cae FG |
10624 | } |
10625 | ||
11fdf7f2 | 10626 | void OSDShard::unprime_split_children(spg_t parent, unsigned old_pg_num) |
7c673cae | 10627 | { |
11fdf7f2 TL |
10628 | std::lock_guard l(shard_lock); |
10629 | vector<spg_t> to_delete; | |
10630 | for (auto& i : pg_slots) { | |
10631 | if (i.first != parent && | |
10632 | i.first.get_ancestor(old_pg_num) == parent) { | |
10633 | dout(10) << __func__ << " parent " << parent << " clearing " << i.first | |
10634 | << dendl; | |
10635 | _wake_pg_slot(i.first, i.second.get()); | |
10636 | to_delete.push_back(i.first); | |
10637 | } | |
10638 | } | |
10639 | for (auto pgid : to_delete) { | |
10640 | pg_slots.erase(pgid); | |
10641 | } | |
10642 | } | |
10643 | ||
10644 | ||
10645 | // ============================================================= | |
10646 | ||
10647 | #undef dout_context | |
10648 | #define dout_context osd->cct | |
10649 | #undef dout_prefix | |
10650 | #define dout_prefix *_dout << "osd." << osd->whoami << " op_wq " | |
10651 | ||
10652 | void OSD::ShardedOpWQ::_add_slot_waiter( | |
10653 | spg_t pgid, | |
10654 | OSDShardPGSlot *slot, | |
10655 | OpQueueItem&& qi) | |
10656 | { | |
10657 | if (qi.is_peering()) { | |
10658 | dout(20) << __func__ << " " << pgid | |
10659 | << " peering, item epoch is " | |
10660 | << qi.get_map_epoch() | |
10661 | << ", will wait on " << qi << dendl; | |
10662 | slot->waiting_peering[qi.get_map_epoch()].push_back(std::move(qi)); | |
10663 | } else { | |
10664 | dout(20) << __func__ << " " << pgid | |
10665 | << " item epoch is " | |
10666 | << qi.get_map_epoch() | |
10667 | << ", will wait on " << qi << dendl; | |
10668 | slot->waiting.push_back(std::move(qi)); | |
7c673cae FG |
10669 | } |
10670 | } | |
10671 | ||
10672 | #undef dout_prefix | |
10673 | #define dout_prefix *_dout << "osd." << osd->whoami << " op_wq(" << shard_index << ") " | |
10674 | ||
10675 | void OSD::ShardedOpWQ::_process(uint32_t thread_index, heartbeat_handle_d *hb) | |
10676 | { | |
11fdf7f2 TL |
10677 | uint32_t shard_index = thread_index % osd->num_shards; |
10678 | auto& sdata = osd->shards[shard_index]; | |
10679 | ceph_assert(sdata); | |
10680 | ||
10681 | // If all threads of shards do oncommits, there is a out-of-order | |
10682 | // problem. So we choose the thread which has the smallest | |
10683 | // thread_index(thread_index < num_shards) of shard to do oncommit | |
10684 | // callback. | |
10685 | bool is_smallest_thread_index = thread_index < osd->num_shards; | |
7c673cae FG |
10686 | |
10687 | // peek at spg_t | |
11fdf7f2 TL |
10688 | sdata->shard_lock.lock(); |
10689 | if (sdata->pqueue->empty() && | |
10690 | (!is_smallest_thread_index || sdata->context_queue.empty())) { | |
10691 | std::unique_lock wait_lock{sdata->sdata_wait_lock}; | |
10692 | if (is_smallest_thread_index && !sdata->context_queue.empty()) { | |
10693 | // we raced with a context_queue addition, don't wait | |
10694 | wait_lock.unlock(); | |
10695 | } else if (!sdata->stop_waiting) { | |
10696 | dout(20) << __func__ << " empty q, waiting" << dendl; | |
10697 | osd->cct->get_heartbeat_map()->clear_timeout(hb); | |
10698 | sdata->shard_lock.unlock(); | |
10699 | sdata->sdata_cond.wait(wait_lock); | |
10700 | wait_lock.unlock(); | |
10701 | sdata->shard_lock.lock(); | |
10702 | if (sdata->pqueue->empty() && | |
10703 | !(is_smallest_thread_index && !sdata->context_queue.empty())) { | |
10704 | sdata->shard_lock.unlock(); | |
10705 | return; | |
10706 | } | |
10707 | osd->cct->get_heartbeat_map()->reset_timeout(hb, | |
10708 | osd->cct->_conf->threadpool_default_timeout, 0); | |
10709 | } else { | |
10710 | dout(20) << __func__ << " need return immediately" << dendl; | |
10711 | wait_lock.unlock(); | |
10712 | sdata->shard_lock.unlock(); | |
7c673cae FG |
10713 | return; |
10714 | } | |
10715 | } | |
11fdf7f2 TL |
10716 | |
10717 | list<Context *> oncommits; | |
10718 | if (is_smallest_thread_index && !sdata->context_queue.empty()) { | |
10719 | sdata->context_queue.swap(oncommits); | |
7c673cae | 10720 | } |
11fdf7f2 TL |
10721 | |
10722 | if (sdata->pqueue->empty()) { | |
10723 | if (osd->is_stopping()) { | |
10724 | sdata->shard_lock.unlock(); | |
10725 | for (auto c : oncommits) { | |
10726 | dout(10) << __func__ << " discarding in-flight oncommit " << c << dendl; | |
10727 | delete c; | |
10728 | } | |
10729 | return; // OSD shutdown, discard. | |
7c673cae | 10730 | } |
11fdf7f2 TL |
10731 | sdata->shard_lock.unlock(); |
10732 | handle_oncommits(oncommits); | |
10733 | return; | |
7c673cae | 10734 | } |
7c673cae | 10735 | |
11fdf7f2 TL |
10736 | OpQueueItem item = sdata->pqueue->dequeue(); |
10737 | if (osd->is_stopping()) { | |
10738 | sdata->shard_lock.unlock(); | |
10739 | for (auto c : oncommits) { | |
10740 | dout(10) << __func__ << " discarding in-flight oncommit " << c << dendl; | |
10741 | delete c; | |
10742 | } | |
10743 | return; // OSD shutdown, discard. | |
10744 | } | |
7c673cae | 10745 | |
11fdf7f2 TL |
10746 | const auto token = item.get_ordering_token(); |
10747 | auto r = sdata->pg_slots.emplace(token, nullptr); | |
10748 | if (r.second) { | |
10749 | r.first->second = make_unique<OSDShardPGSlot>(); | |
7c673cae | 10750 | } |
11fdf7f2 TL |
10751 | OSDShardPGSlot *slot = r.first->second.get(); |
10752 | dout(20) << __func__ << " " << token | |
10753 | << (r.second ? " (new)" : "") | |
10754 | << " to_process " << slot->to_process | |
10755 | << " waiting " << slot->waiting | |
10756 | << " waiting_peering " << slot->waiting_peering | |
10757 | << dendl; | |
10758 | slot->to_process.push_back(std::move(item)); | |
10759 | dout(20) << __func__ << " " << slot->to_process.back() | |
10760 | << " queued" << dendl; | |
7c673cae | 10761 | |
11fdf7f2 TL |
10762 | retry_pg: |
10763 | PGRef pg = slot->pg; | |
7c673cae | 10764 | |
11fdf7f2 TL |
10765 | // lock pg (if we have it) |
10766 | if (pg) { | |
10767 | // note the requeue seq now... | |
10768 | uint64_t requeue_seq = slot->requeue_seq; | |
10769 | ++slot->num_running; | |
7c673cae | 10770 | |
11fdf7f2 TL |
10771 | sdata->shard_lock.unlock(); |
10772 | osd->service.maybe_inject_dispatch_delay(); | |
10773 | pg->lock(); | |
10774 | osd->service.maybe_inject_dispatch_delay(); | |
10775 | sdata->shard_lock.lock(); | |
7c673cae | 10776 | |
11fdf7f2 TL |
10777 | auto q = sdata->pg_slots.find(token); |
10778 | if (q == sdata->pg_slots.end()) { | |
10779 | // this can happen if we race with pg removal. | |
10780 | dout(20) << __func__ << " slot " << token << " no longer there" << dendl; | |
10781 | pg->unlock(); | |
10782 | sdata->shard_lock.unlock(); | |
10783 | handle_oncommits(oncommits); | |
10784 | return; | |
10785 | } | |
10786 | slot = q->second.get(); | |
10787 | --slot->num_running; | |
7c673cae | 10788 | |
11fdf7f2 TL |
10789 | if (slot->to_process.empty()) { |
10790 | // raced with _wake_pg_slot or consume_map | |
10791 | dout(20) << __func__ << " " << token | |
10792 | << " nothing queued" << dendl; | |
7c673cae | 10793 | pg->unlock(); |
11fdf7f2 TL |
10794 | sdata->shard_lock.unlock(); |
10795 | handle_oncommits(oncommits); | |
10796 | return; | |
7c673cae | 10797 | } |
11fdf7f2 TL |
10798 | if (requeue_seq != slot->requeue_seq) { |
10799 | dout(20) << __func__ << " " << token | |
10800 | << " requeue_seq " << slot->requeue_seq << " > our " | |
10801 | << requeue_seq << ", we raced with _wake_pg_slot" | |
10802 | << dendl; | |
7c673cae | 10803 | pg->unlock(); |
11fdf7f2 TL |
10804 | sdata->shard_lock.unlock(); |
10805 | handle_oncommits(oncommits); | |
10806 | return; | |
7c673cae | 10807 | } |
11fdf7f2 TL |
10808 | if (slot->pg != pg) { |
10809 | // this can happen if we race with pg removal. | |
10810 | dout(20) << __func__ << " slot " << token << " no longer attached to " | |
10811 | << pg << dendl; | |
7c673cae | 10812 | pg->unlock(); |
11fdf7f2 | 10813 | goto retry_pg; |
7c673cae | 10814 | } |
7c673cae FG |
10815 | } |
10816 | ||
11fdf7f2 TL |
10817 | dout(20) << __func__ << " " << token |
10818 | << " to_process " << slot->to_process | |
10819 | << " waiting " << slot->waiting | |
10820 | << " waiting_peering " << slot->waiting_peering << dendl; | |
10821 | ||
10822 | ThreadPool::TPHandle tp_handle(osd->cct, hb, timeout_interval, | |
10823 | suicide_interval); | |
10824 | ||
7c673cae | 10825 | // take next item |
11fdf7f2 TL |
10826 | auto qi = std::move(slot->to_process.front()); |
10827 | slot->to_process.pop_front(); | |
10828 | dout(20) << __func__ << " " << qi << " pg " << pg << dendl; | |
10829 | set<pair<spg_t,epoch_t>> new_children; | |
10830 | OSDMapRef osdmap; | |
7c673cae | 10831 | |
11fdf7f2 | 10832 | while (!pg) { |
7c673cae | 10833 | // should this pg shard exist on this osd in this (or a later) epoch? |
11fdf7f2 TL |
10834 | osdmap = sdata->shard_osdmap; |
10835 | const PGCreateInfo *create_info = qi.creates_pg(); | |
10836 | if (!slot->waiting_for_split.empty()) { | |
10837 | dout(20) << __func__ << " " << token | |
10838 | << " splitting " << slot->waiting_for_split << dendl; | |
10839 | _add_slot_waiter(token, slot, std::move(qi)); | |
10840 | } else if (qi.get_map_epoch() > osdmap->get_epoch()) { | |
10841 | dout(20) << __func__ << " " << token | |
10842 | << " map " << qi.get_map_epoch() << " > " | |
10843 | << osdmap->get_epoch() << dendl; | |
10844 | _add_slot_waiter(token, slot, std::move(qi)); | |
10845 | } else if (qi.is_peering()) { | |
10846 | if (!qi.peering_requires_pg()) { | |
10847 | // for pg-less events, we run them under the ordering lock, since | |
10848 | // we don't have the pg lock to keep them ordered. | |
10849 | qi.run(osd, sdata, pg, tp_handle); | |
10850 | } else if (osdmap->is_up_acting_osd_shard(token, osd->whoami)) { | |
10851 | if (create_info) { | |
10852 | if (create_info->by_mon && | |
10853 | osdmap->get_pg_acting_primary(token.pgid) != osd->whoami) { | |
10854 | dout(20) << __func__ << " " << token | |
10855 | << " no pg, no longer primary, ignoring mon create on " | |
10856 | << qi << dendl; | |
10857 | } else { | |
10858 | dout(20) << __func__ << " " << token | |
10859 | << " no pg, should create on " << qi << dendl; | |
10860 | pg = osd->handle_pg_create_info(osdmap, create_info); | |
10861 | if (pg) { | |
10862 | // we created the pg! drop out and continue "normally"! | |
10863 | sdata->_attach_pg(slot, pg.get()); | |
10864 | sdata->_wake_pg_slot(token, slot); | |
10865 | ||
10866 | // identify split children between create epoch and shard epoch. | |
10867 | osd->service.identify_splits_and_merges( | |
10868 | pg->get_osdmap(), osdmap, pg->pg_id, &new_children, nullptr); | |
10869 | sdata->_prime_splits(&new_children); | |
10870 | // distribute remaining split children to other shards below! | |
10871 | break; | |
10872 | } | |
10873 | dout(20) << __func__ << " ignored create on " << qi << dendl; | |
10874 | } | |
10875 | } else { | |
10876 | dout(20) << __func__ << " " << token | |
10877 | << " no pg, peering, !create, discarding " << qi << dendl; | |
10878 | } | |
10879 | } else { | |
10880 | dout(20) << __func__ << " " << token | |
10881 | << " no pg, peering, doesn't map here e" << osdmap->get_epoch() | |
10882 | << ", discarding " << qi | |
10883 | << dendl; | |
10884 | } | |
10885 | } else if (osdmap->is_up_acting_osd_shard(token, osd->whoami)) { | |
10886 | dout(20) << __func__ << " " << token | |
10887 | << " no pg, should exist e" << osdmap->get_epoch() | |
10888 | << ", will wait on " << qi << dendl; | |
10889 | _add_slot_waiter(token, slot, std::move(qi)); | |
7c673cae | 10890 | } else { |
11fdf7f2 TL |
10891 | dout(20) << __func__ << " " << token |
10892 | << " no pg, shouldn't exist e" << osdmap->get_epoch() | |
10893 | << ", dropping " << qi << dendl; | |
7c673cae | 10894 | // share map with client? |
11fdf7f2 TL |
10895 | if (boost::optional<OpRequestRef> _op = qi.maybe_get_op()) { |
10896 | auto priv = (*_op)->get_req()->get_connection()->get_priv(); | |
10897 | if (auto session = static_cast<Session *>(priv.get()); session) { | |
10898 | osd->maybe_share_map(session, *_op, sdata->shard_osdmap); | |
7c673cae FG |
10899 | } |
10900 | } | |
11fdf7f2 | 10901 | unsigned pushes_to_free = qi.get_reserved_pushes(); |
7c673cae | 10902 | if (pushes_to_free > 0) { |
11fdf7f2 | 10903 | sdata->shard_lock.unlock(); |
7c673cae | 10904 | osd->service.release_reserved_pushes(pushes_to_free); |
11fdf7f2 | 10905 | handle_oncommits(oncommits); |
7c673cae FG |
10906 | return; |
10907 | } | |
10908 | } | |
11fdf7f2 TL |
10909 | sdata->shard_lock.unlock(); |
10910 | handle_oncommits(oncommits); | |
7c673cae FG |
10911 | return; |
10912 | } | |
11fdf7f2 TL |
10913 | if (qi.is_peering()) { |
10914 | OSDMapRef osdmap = sdata->shard_osdmap; | |
10915 | if (qi.get_map_epoch() > osdmap->get_epoch()) { | |
10916 | _add_slot_waiter(token, slot, std::move(qi)); | |
10917 | sdata->shard_lock.unlock(); | |
10918 | pg->unlock(); | |
10919 | handle_oncommits(oncommits); | |
10920 | return; | |
10921 | } | |
10922 | } | |
10923 | sdata->shard_lock.unlock(); | |
7c673cae | 10924 | |
11fdf7f2 TL |
10925 | if (!new_children.empty()) { |
10926 | for (auto shard : osd->shards) { | |
10927 | shard->prime_splits(osdmap, &new_children); | |
10928 | } | |
10929 | ceph_assert(new_children.empty()); | |
10930 | } | |
7c673cae FG |
10931 | |
10932 | // osd_opwq_process marks the point at which an operation has been dequeued | |
10933 | // and will begin to be handled by a worker thread. | |
10934 | { | |
10935 | #ifdef WITH_LTTNG | |
10936 | osd_reqid_t reqid; | |
11fdf7f2 | 10937 | if (boost::optional<OpRequestRef> _op = qi.maybe_get_op()) { |
7c673cae FG |
10938 | reqid = (*_op)->get_reqid(); |
10939 | } | |
10940 | #endif | |
10941 | tracepoint(osd, opwq_process_start, reqid.name._type, | |
10942 | reqid.name._num, reqid.tid, reqid.inc); | |
10943 | } | |
10944 | ||
10945 | lgeneric_subdout(osd->cct, osd, 30) << "dequeue status: "; | |
10946 | Formatter *f = Formatter::create("json"); | |
10947 | f->open_object_section("q"); | |
10948 | dump(f); | |
10949 | f->close_section(); | |
10950 | f->flush(*_dout); | |
10951 | delete f; | |
10952 | *_dout << dendl; | |
10953 | ||
11fdf7f2 | 10954 | qi.run(osd, sdata, pg, tp_handle); |
7c673cae FG |
10955 | |
10956 | { | |
10957 | #ifdef WITH_LTTNG | |
10958 | osd_reqid_t reqid; | |
11fdf7f2 | 10959 | if (boost::optional<OpRequestRef> _op = qi.maybe_get_op()) { |
7c673cae FG |
10960 | reqid = (*_op)->get_reqid(); |
10961 | } | |
10962 | #endif | |
10963 | tracepoint(osd, opwq_process_finish, reqid.name._type, | |
10964 | reqid.name._num, reqid.tid, reqid.inc); | |
10965 | } | |
10966 | ||
11fdf7f2 | 10967 | handle_oncommits(oncommits); |
7c673cae FG |
10968 | } |
10969 | ||
11fdf7f2 | 10970 | void OSD::ShardedOpWQ::_enqueue(OpQueueItem&& item) { |
7c673cae | 10971 | uint32_t shard_index = |
11fdf7f2 | 10972 | item.get_ordering_token().hash_to_shard(osd->shards.size()); |
7c673cae | 10973 | |
11fdf7f2 | 10974 | OSDShard* sdata = osd->shards[shard_index]; |
7c673cae | 10975 | assert (NULL != sdata); |
11fdf7f2 TL |
10976 | unsigned priority = item.get_priority(); |
10977 | unsigned cost = item.get_cost(); | |
10978 | sdata->shard_lock.lock(); | |
7c673cae | 10979 | |
11fdf7f2 | 10980 | dout(20) << __func__ << " " << item << dendl; |
7c673cae FG |
10981 | if (priority >= osd->op_prio_cutoff) |
10982 | sdata->pqueue->enqueue_strict( | |
11fdf7f2 | 10983 | item.get_owner(), priority, std::move(item)); |
7c673cae FG |
10984 | else |
10985 | sdata->pqueue->enqueue( | |
11fdf7f2 TL |
10986 | item.get_owner(), priority, cost, std::move(item)); |
10987 | sdata->shard_lock.unlock(); | |
7c673cae | 10988 | |
11fdf7f2 TL |
10989 | std::lock_guard l{sdata->sdata_wait_lock}; |
10990 | sdata->sdata_cond.notify_one(); | |
7c673cae FG |
10991 | } |
10992 | ||
11fdf7f2 | 10993 | void OSD::ShardedOpWQ::_enqueue_front(OpQueueItem&& item) |
7c673cae | 10994 | { |
11fdf7f2 TL |
10995 | auto shard_index = item.get_ordering_token().hash_to_shard(osd->shards.size()); |
10996 | auto& sdata = osd->shards[shard_index]; | |
10997 | ceph_assert(sdata); | |
10998 | sdata->shard_lock.lock(); | |
10999 | auto p = sdata->pg_slots.find(item.get_ordering_token()); | |
11000 | if (p != sdata->pg_slots.end() && | |
11001 | !p->second->to_process.empty()) { | |
7c673cae FG |
11002 | // we may be racing with _process, which has dequeued a new item |
11003 | // from pqueue, put it on to_process, and is now busy taking the | |
11004 | // pg lock. ensure this old requeued item is ordered before any | |
11005 | // such newer item in to_process. | |
11fdf7f2 TL |
11006 | p->second->to_process.push_front(std::move(item)); |
11007 | item = std::move(p->second->to_process.back()); | |
11008 | p->second->to_process.pop_back(); | |
11009 | dout(20) << __func__ | |
11010 | << " " << p->second->to_process.front() | |
11011 | << " shuffled w/ " << item << dendl; | |
7c673cae | 11012 | } else { |
11fdf7f2 | 11013 | dout(20) << __func__ << " " << item << dendl; |
7c673cae | 11014 | } |
11fdf7f2 TL |
11015 | sdata->_enqueue_front(std::move(item), osd->op_prio_cutoff); |
11016 | sdata->shard_lock.unlock(); | |
11017 | std::lock_guard l{sdata->sdata_wait_lock}; | |
11018 | sdata->sdata_cond.notify_one(); | |
7c673cae FG |
11019 | } |
11020 | ||
11021 | namespace ceph { | |
11022 | namespace osd_cmds { | |
11023 | ||
11fdf7f2 TL |
11024 | int heap(CephContext& cct, const cmdmap_t& cmdmap, Formatter& f, |
11025 | std::ostream& os) | |
7c673cae FG |
11026 | { |
11027 | if (!ceph_using_tcmalloc()) { | |
11028 | os << "could not issue heap profiler command -- not using tcmalloc!"; | |
11029 | return -EOPNOTSUPP; | |
11030 | } | |
11031 | ||
11032 | string cmd; | |
11033 | if (!cmd_getval(&cct, cmdmap, "heapcmd", cmd)) { | |
11034 | os << "unable to get value for command \"" << cmd << "\""; | |
11035 | return -EINVAL; | |
11fdf7f2 | 11036 | } |
7c673cae FG |
11037 | |
11038 | std::vector<std::string> cmd_vec; | |
11039 | get_str_vec(cmd, cmd_vec); | |
11fdf7f2 TL |
11040 | |
11041 | string val; | |
11042 | if (cmd_getval(&cct, cmdmap, "value", val)) { | |
11043 | cmd_vec.push_back(val); | |
11044 | } | |
7c673cae FG |
11045 | |
11046 | ceph_heap_profiler_handle_command(cmd_vec, os); | |
11047 | ||
11048 | return 0; | |
11049 | } | |
11050 | ||
11051 | }} // namespace ceph::osd_cmds | |
11052 | ||
224ce89b | 11053 | |
11fdf7f2 | 11054 | std::ostream& operator<<(std::ostream& out, const io_queue& q) { |
224ce89b | 11055 | switch(q) { |
11fdf7f2 | 11056 | case io_queue::prioritized: |
224ce89b WB |
11057 | out << "prioritized"; |
11058 | break; | |
11fdf7f2 | 11059 | case io_queue::weightedpriority: |
224ce89b WB |
11060 | out << "weightedpriority"; |
11061 | break; | |
11fdf7f2 | 11062 | case io_queue::mclock_opclass: |
224ce89b WB |
11063 | out << "mclock_opclass"; |
11064 | break; | |
11fdf7f2 | 11065 | case io_queue::mclock_client: |
224ce89b WB |
11066 | out << "mclock_client"; |
11067 | break; | |
11068 | } | |
11069 | return out; | |
11070 | } |