]>
Commit | Line | Data |
---|---|---|
f67539c2 | 1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- |
7c673cae FG |
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 | * | |
8 | * This is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU Lesser General Public | |
f67539c2 | 10 | * License version 2.1, as published by the Free Software |
7c673cae | 11 | * Foundation. See file COPYING. |
f67539c2 | 12 | * |
7c673cae FG |
13 | */ |
14 | ||
15 | #ifndef CEPH_OSD_H | |
16 | #define CEPH_OSD_H | |
17 | ||
18 | #include "PG.h" | |
19 | ||
20 | #include "msg/Dispatcher.h" | |
21 | ||
f67539c2 | 22 | #include "common/async/context_pool.h" |
7c673cae FG |
23 | #include "common/Timer.h" |
24 | #include "common/WorkQueue.h" | |
25 | #include "common/AsyncReserver.h" | |
26 | #include "common/ceph_context.h" | |
11fdf7f2 | 27 | #include "common/config_cacher.h" |
7c673cae | 28 | #include "common/zipkin_trace.h" |
9f95a23c | 29 | #include "common/ceph_timer.h" |
7c673cae FG |
30 | |
31 | #include "mgr/MgrClient.h" | |
32 | ||
33 | #include "os/ObjectStore.h" | |
7c673cae FG |
34 | |
35 | #include "include/CompatSet.h" | |
9f95a23c | 36 | #include "include/common_fwd.h" |
7c673cae FG |
37 | |
38 | #include "OpRequest.h" | |
39 | #include "Session.h" | |
40 | ||
9f95a23c | 41 | #include "osd/scheduler/OpScheduler.h" |
224ce89b | 42 | |
7c673cae FG |
43 | #include <atomic> |
44 | #include <map> | |
45 | #include <memory> | |
11fdf7f2 | 46 | #include <string> |
7c673cae FG |
47 | |
48 | #include "include/unordered_map.h" | |
49 | ||
50 | #include "common/shared_cache.hpp" | |
51 | #include "common/simple_cache.hpp" | |
7c673cae | 52 | #include "messages/MOSDOp.h" |
7c673cae | 53 | #include "common/EventTrace.h" |
9f95a23c | 54 | #include "osd/osd_perf_counters.h" |
f67539c2 | 55 | #include "common/Finisher.h" |
7c673cae FG |
56 | |
57 | #define CEPH_OSD_PROTOCOL 10 /* cluster internal */ | |
58 | ||
11fdf7f2 TL |
59 | /* |
60 | ||
61 | lock ordering for pg map | |
62 | ||
63 | PG::lock | |
64 | ShardData::lock | |
65 | OSD::pg_map_lock | |
66 | ||
67 | */ | |
7c673cae | 68 | |
7c673cae FG |
69 | class Messenger; |
70 | class Message; | |
71 | class MonClient; | |
7c673cae FG |
72 | class ObjectStore; |
73 | class FuseStore; | |
74 | class OSDMap; | |
75 | class MLog; | |
76 | class Objecter; | |
11fdf7f2 | 77 | class KeyStore; |
7c673cae FG |
78 | |
79 | class Watch; | |
80 | class PrimaryLogPG; | |
81 | ||
7c673cae | 82 | class TestOpsSocketHook; |
11fdf7f2 | 83 | struct C_FinishSplits; |
7c673cae FG |
84 | struct C_OpenPGs; |
85 | class LogChannel; | |
7c673cae | 86 | |
11fdf7f2 TL |
87 | class MOSDPGCreate2; |
88 | class MOSDPGQuery; | |
89 | class MOSDPGNotify; | |
90 | class MOSDPGInfo; | |
91 | class MOSDPGRemove; | |
92 | class MOSDForceRecovery; | |
9f95a23c | 93 | class MMonGetPurgedSnapsReply; |
7c673cae FG |
94 | |
95 | class OSD; | |
96 | ||
7c673cae | 97 | class OSDService { |
9f95a23c | 98 | using OpSchedulerItem = ceph::osd::scheduler::OpSchedulerItem; |
7c673cae FG |
99 | public: |
100 | OSD *osd; | |
101 | CephContext *cct; | |
11fdf7f2 | 102 | ObjectStore::CollectionHandle meta_ch; |
7c673cae FG |
103 | const int whoami; |
104 | ObjectStore *&store; | |
105 | LogClient &log_client; | |
106 | LogChannelRef clog; | |
107 | PGRecoveryStats &pg_recovery_stats; | |
108 | private: | |
109 | Messenger *&cluster_messenger; | |
110 | Messenger *&client_messenger; | |
111 | public: | |
112 | PerfCounters *&logger; | |
113 | PerfCounters *&recoverystate_perf; | |
114 | MonClient *&monc; | |
7c673cae | 115 | |
11fdf7f2 TL |
116 | md_config_cacher_t<Option::size_t> osd_max_object_size; |
117 | md_config_cacher_t<bool> osd_skip_data_digest; | |
118 | ||
9f95a23c TL |
119 | void enqueue_back(OpSchedulerItem&& qi); |
120 | void enqueue_front(OpSchedulerItem&& qi); | |
7c673cae FG |
121 | |
122 | void maybe_inject_dispatch_delay() { | |
11fdf7f2 | 123 | if (g_conf()->osd_debug_inject_dispatch_delay_probability > 0) { |
7c673cae | 124 | if (rand() % 10000 < |
11fdf7f2 | 125 | g_conf()->osd_debug_inject_dispatch_delay_probability * 10000) { |
7c673cae | 126 | utime_t t; |
11fdf7f2 | 127 | t.set_from_double(g_conf()->osd_debug_inject_dispatch_delay_duration); |
7c673cae FG |
128 | t.sleep(); |
129 | } | |
130 | } | |
131 | } | |
132 | ||
9f95a23c TL |
133 | ceph::signedspan get_mnow(); |
134 | ||
7c673cae FG |
135 | private: |
136 | // -- superblock -- | |
11fdf7f2 | 137 | ceph::mutex publish_lock, pre_publish_lock; // pre-publish orders before publish |
7c673cae FG |
138 | OSDSuperblock superblock; |
139 | ||
140 | public: | |
141 | OSDSuperblock get_superblock() { | |
11fdf7f2 | 142 | std::lock_guard l(publish_lock); |
7c673cae FG |
143 | return superblock; |
144 | } | |
145 | void publish_superblock(const OSDSuperblock &block) { | |
11fdf7f2 | 146 | std::lock_guard l(publish_lock); |
7c673cae FG |
147 | superblock = block; |
148 | } | |
149 | ||
150 | int get_nodeid() const { return whoami; } | |
151 | ||
152 | std::atomic<epoch_t> max_oldest_map; | |
153 | private: | |
154 | OSDMapRef osdmap; | |
155 | ||
156 | public: | |
157 | OSDMapRef get_osdmap() { | |
11fdf7f2 | 158 | std::lock_guard l(publish_lock); |
7c673cae FG |
159 | return osdmap; |
160 | } | |
161 | epoch_t get_osdmap_epoch() { | |
11fdf7f2 | 162 | std::lock_guard l(publish_lock); |
7c673cae FG |
163 | return osdmap ? osdmap->get_epoch() : 0; |
164 | } | |
165 | void publish_map(OSDMapRef map) { | |
11fdf7f2 | 166 | std::lock_guard l(publish_lock); |
7c673cae FG |
167 | osdmap = map; |
168 | } | |
169 | ||
170 | /* | |
f67539c2 TL |
171 | * osdmap - current published std::map |
172 | * next_osdmap - pre_published std::map that is about to be published. | |
7c673cae FG |
173 | * |
174 | * We use the next_osdmap to send messages and initiate connections, | |
f67539c2 | 175 | * but only if the target is the same instance as the one in the std::map |
7c673cae FG |
176 | * epoch the current user is working from (i.e., the result is |
177 | * equivalent to what is in next_osdmap). | |
178 | * | |
179 | * This allows the helpers to start ignoring osds that are about to | |
180 | * go down, and let OSD::handle_osd_map()/note_down_osd() mark them | |
181 | * down, without worrying about reopening connections from threads | |
182 | * working from old maps. | |
183 | */ | |
184 | private: | |
185 | OSDMapRef next_osdmap; | |
11fdf7f2 | 186 | ceph::condition_variable pre_publish_cond; |
9f95a23c | 187 | int pre_publish_waiter = 0; |
7c673cae FG |
188 | |
189 | public: | |
190 | void pre_publish_map(OSDMapRef map) { | |
11fdf7f2 | 191 | std::lock_guard l(pre_publish_lock); |
7c673cae FG |
192 | next_osdmap = std::move(map); |
193 | } | |
194 | ||
195 | void activate_map(); | |
196 | /// map epochs reserved below | |
f67539c2 | 197 | std::map<epoch_t, unsigned> map_reservations; |
7c673cae FG |
198 | |
199 | /// gets ref to next_osdmap and registers the epoch as reserved | |
200 | OSDMapRef get_nextmap_reserved() { | |
11fdf7f2 | 201 | std::lock_guard l(pre_publish_lock); |
7c673cae | 202 | epoch_t e = next_osdmap->get_epoch(); |
f67539c2 TL |
203 | std::map<epoch_t, unsigned>::iterator i = |
204 | map_reservations.insert(std::make_pair(e, 0)).first; | |
7c673cae FG |
205 | i->second++; |
206 | return next_osdmap; | |
207 | } | |
208 | /// releases reservation on map | |
209 | void release_map(OSDMapRef osdmap) { | |
11fdf7f2 | 210 | std::lock_guard l(pre_publish_lock); |
f67539c2 | 211 | std::map<epoch_t, unsigned>::iterator i = |
7c673cae | 212 | map_reservations.find(osdmap->get_epoch()); |
11fdf7f2 TL |
213 | ceph_assert(i != map_reservations.end()); |
214 | ceph_assert(i->second > 0); | |
7c673cae FG |
215 | if (--(i->second) == 0) { |
216 | map_reservations.erase(i); | |
217 | } | |
9f95a23c TL |
218 | if (pre_publish_waiter) { |
219 | pre_publish_cond.notify_all(); | |
220 | } | |
7c673cae FG |
221 | } |
222 | /// blocks until there are no reserved maps prior to next_osdmap | |
223 | void await_reserved_maps() { | |
11fdf7f2 TL |
224 | std::unique_lock l{pre_publish_lock}; |
225 | ceph_assert(next_osdmap); | |
9f95a23c | 226 | pre_publish_waiter++; |
11fdf7f2 TL |
227 | pre_publish_cond.wait(l, [this] { |
228 | auto i = map_reservations.cbegin(); | |
229 | return (i == map_reservations.cend() || | |
230 | i->first >= next_osdmap->get_epoch()); | |
231 | }); | |
9f95a23c | 232 | pre_publish_waiter--; |
11fdf7f2 TL |
233 | } |
234 | OSDMapRef get_next_osdmap() { | |
235 | std::lock_guard l(pre_publish_lock); | |
11fdf7f2 | 236 | return next_osdmap; |
7c673cae FG |
237 | } |
238 | ||
9f95a23c TL |
239 | void maybe_share_map(Connection *con, |
240 | const OSDMapRef& osdmap, | |
241 | epoch_t peer_epoch_lb=0); | |
7c673cae FG |
242 | |
243 | void send_map(class MOSDMap *m, Connection *con); | |
9f95a23c TL |
244 | void send_incremental_map(epoch_t since, Connection *con, |
245 | const OSDMapRef& osdmap); | |
7c673cae FG |
246 | MOSDMap *build_incremental_map_msg(epoch_t from, epoch_t to, |
247 | OSDSuperblock& superblock); | |
7c673cae FG |
248 | |
249 | ConnectionRef get_con_osd_cluster(int peer, epoch_t from_epoch); | |
f67539c2 | 250 | std::pair<ConnectionRef,ConnectionRef> get_con_osd_hb(int peer, epoch_t from_epoch); // (back, front) |
7c673cae | 251 | void send_message_osd_cluster(int peer, Message *m, epoch_t from_epoch); |
9f95a23c | 252 | void send_message_osd_cluster(std::vector<std::pair<int, Message*>>& messages, epoch_t from_epoch); |
f67539c2 TL |
253 | void send_message_osd_cluster(MessageRef m, Connection *con) { |
254 | con->send_message2(std::move(m)); | |
7c673cae FG |
255 | } |
256 | void send_message_osd_cluster(Message *m, const ConnectionRef& con) { | |
257 | con->send_message(m); | |
258 | } | |
7c673cae FG |
259 | void send_message_osd_client(Message *m, const ConnectionRef& con) { |
260 | con->send_message(m); | |
261 | } | |
11fdf7f2 | 262 | entity_name_t get_cluster_msgr_name() const; |
7c673cae FG |
263 | |
264 | private: | |
265 | // -- scrub scheduling -- | |
9f95a23c | 266 | ceph::mutex sched_scrub_lock = ceph::make_mutex("OSDService::sched_scrub_lock"); |
eafe8130 TL |
267 | int scrubs_local; |
268 | int scrubs_remote; | |
7c673cae FG |
269 | |
270 | public: | |
271 | struct ScrubJob { | |
272 | CephContext* cct; | |
273 | /// pg to be scrubbed | |
274 | spg_t pgid; | |
275 | /// a time scheduled for scrub. but the scrub could be delayed if system | |
276 | /// load is too high or it fails to fall in the scrub hours | |
277 | utime_t sched_time; | |
278 | /// the hard upper bound of scrub time | |
279 | utime_t deadline; | |
280 | ScrubJob() : cct(nullptr) {} | |
281 | explicit ScrubJob(CephContext* cct, const spg_t& pg, | |
282 | const utime_t& timestamp, | |
283 | double pool_scrub_min_interval = 0, | |
284 | double pool_scrub_max_interval = 0, bool must = true); | |
285 | /// order the jobs by sched_time | |
286 | bool operator<(const ScrubJob& rhs) const; | |
287 | }; | |
f67539c2 | 288 | std::set<ScrubJob> sched_scrub_pg; |
7c673cae | 289 | |
f67539c2 TL |
290 | /// @returns the scrub_reg_stamp used for unregistering the scrub job |
291 | utime_t reg_pg_scrub(spg_t pgid, | |
292 | utime_t t, | |
293 | double pool_scrub_min_interval, | |
294 | double pool_scrub_max_interval, | |
295 | bool must) { | |
296 | ScrubJob scrub_job(cct, pgid, t, pool_scrub_min_interval, pool_scrub_max_interval, | |
297 | must); | |
298 | std::lock_guard l(OSDService::sched_scrub_lock); | |
299 | sched_scrub_pg.insert(scrub_job); | |
300 | return scrub_job.sched_time; | |
7c673cae | 301 | } |
f67539c2 | 302 | |
7c673cae | 303 | void unreg_pg_scrub(spg_t pgid, utime_t t) { |
11fdf7f2 | 304 | std::lock_guard l(sched_scrub_lock); |
7c673cae | 305 | size_t removed = sched_scrub_pg.erase(ScrubJob(cct, pgid, t)); |
11fdf7f2 | 306 | ceph_assert(removed); |
7c673cae | 307 | } |
f67539c2 | 308 | |
7c673cae | 309 | bool first_scrub_stamp(ScrubJob *out) { |
11fdf7f2 | 310 | std::lock_guard l(sched_scrub_lock); |
7c673cae FG |
311 | if (sched_scrub_pg.empty()) |
312 | return false; | |
f67539c2 | 313 | std::set<ScrubJob>::iterator iter = sched_scrub_pg.begin(); |
7c673cae FG |
314 | *out = *iter; |
315 | return true; | |
316 | } | |
317 | bool next_scrub_stamp(const ScrubJob& next, | |
318 | ScrubJob *out) { | |
11fdf7f2 | 319 | std::lock_guard l(sched_scrub_lock); |
7c673cae FG |
320 | if (sched_scrub_pg.empty()) |
321 | return false; | |
b3b6e05e | 322 | std::set<ScrubJob>::const_iterator iter = sched_scrub_pg.upper_bound(next); |
7c673cae FG |
323 | if (iter == sched_scrub_pg.cend()) |
324 | return false; | |
325 | *out = *iter; | |
326 | return true; | |
327 | } | |
328 | ||
f67539c2 | 329 | void dumps_scrub(ceph::Formatter* f); |
7c673cae | 330 | |
eafe8130 TL |
331 | bool can_inc_scrubs(); |
332 | bool inc_scrubs_local(); | |
333 | void dec_scrubs_local(); | |
334 | bool inc_scrubs_remote(); | |
335 | void dec_scrubs_remote(); | |
f67539c2 | 336 | void dump_scrub_reservations(ceph::Formatter *f); |
7c673cae FG |
337 | |
338 | void reply_op_error(OpRequestRef op, int err); | |
9f95a23c | 339 | void reply_op_error(OpRequestRef op, int err, eversion_t v, version_t uv, |
f67539c2 | 340 | std::vector<pg_log_op_return_item_t> op_returns); |
7c673cae FG |
341 | void handle_misdirected_op(PG *pg, OpRequestRef op); |
342 | ||
343 | ||
344 | private: | |
345 | // -- agent shared state -- | |
9f95a23c TL |
346 | ceph::mutex agent_lock = ceph::make_mutex("OSDService::agent_lock"); |
347 | ceph::condition_variable agent_cond; | |
f67539c2 TL |
348 | std::map<uint64_t, std::set<PGRef> > agent_queue; |
349 | std::set<PGRef>::iterator agent_queue_pos; | |
7c673cae FG |
350 | bool agent_valid_iterator; |
351 | int agent_ops; | |
352 | int flush_mode_high_count; //once have one pg with FLUSH_MODE_HIGH then flush objects with high speed | |
f67539c2 | 353 | std::set<hobject_t> agent_oids; |
7c673cae FG |
354 | bool agent_active; |
355 | struct AgentThread : public Thread { | |
356 | OSDService *osd; | |
357 | explicit AgentThread(OSDService *o) : osd(o) {} | |
358 | void *entry() override { | |
359 | osd->agent_entry(); | |
360 | return NULL; | |
361 | } | |
362 | } agent_thread; | |
363 | bool agent_stop_flag; | |
9f95a23c | 364 | ceph::mutex agent_timer_lock = ceph::make_mutex("OSDService::agent_timer_lock"); |
7c673cae FG |
365 | SafeTimer agent_timer; |
366 | ||
367 | public: | |
368 | void agent_entry(); | |
369 | void agent_stop(); | |
370 | ||
371 | void _enqueue(PG *pg, uint64_t priority) { | |
372 | if (!agent_queue.empty() && | |
373 | agent_queue.rbegin()->first < priority) | |
374 | agent_valid_iterator = false; // inserting higher-priority queue | |
f67539c2 | 375 | std::set<PGRef>& nq = agent_queue[priority]; |
7c673cae | 376 | if (nq.empty()) |
9f95a23c | 377 | agent_cond.notify_all(); |
7c673cae FG |
378 | nq.insert(pg); |
379 | } | |
380 | ||
381 | void _dequeue(PG *pg, uint64_t old_priority) { | |
f67539c2 TL |
382 | std::set<PGRef>& oq = agent_queue[old_priority]; |
383 | std::set<PGRef>::iterator p = oq.find(pg); | |
11fdf7f2 | 384 | ceph_assert(p != oq.end()); |
7c673cae FG |
385 | if (p == agent_queue_pos) |
386 | ++agent_queue_pos; | |
387 | oq.erase(p); | |
388 | if (oq.empty()) { | |
389 | if (agent_queue.rbegin()->first == old_priority) | |
390 | agent_valid_iterator = false; | |
391 | agent_queue.erase(old_priority); | |
392 | } | |
393 | } | |
394 | ||
395 | /// enable agent for a pg | |
396 | void agent_enable_pg(PG *pg, uint64_t priority) { | |
11fdf7f2 | 397 | std::lock_guard l(agent_lock); |
7c673cae FG |
398 | _enqueue(pg, priority); |
399 | } | |
400 | ||
401 | /// adjust priority for an enagled pg | |
402 | void agent_adjust_pg(PG *pg, uint64_t old_priority, uint64_t new_priority) { | |
11fdf7f2 TL |
403 | std::lock_guard l(agent_lock); |
404 | ceph_assert(new_priority != old_priority); | |
7c673cae FG |
405 | _enqueue(pg, new_priority); |
406 | _dequeue(pg, old_priority); | |
407 | } | |
408 | ||
409 | /// disable agent for a pg | |
410 | void agent_disable_pg(PG *pg, uint64_t old_priority) { | |
11fdf7f2 | 411 | std::lock_guard l(agent_lock); |
7c673cae FG |
412 | _dequeue(pg, old_priority); |
413 | } | |
414 | ||
415 | /// note start of an async (evict) op | |
416 | void agent_start_evict_op() { | |
11fdf7f2 | 417 | std::lock_guard l(agent_lock); |
7c673cae FG |
418 | ++agent_ops; |
419 | } | |
420 | ||
421 | /// note finish or cancellation of an async (evict) op | |
422 | void agent_finish_evict_op() { | |
11fdf7f2 TL |
423 | std::lock_guard l(agent_lock); |
424 | ceph_assert(agent_ops > 0); | |
7c673cae | 425 | --agent_ops; |
9f95a23c | 426 | agent_cond.notify_all(); |
7c673cae FG |
427 | } |
428 | ||
429 | /// note start of an async (flush) op | |
430 | void agent_start_op(const hobject_t& oid) { | |
11fdf7f2 | 431 | std::lock_guard l(agent_lock); |
7c673cae | 432 | ++agent_ops; |
11fdf7f2 | 433 | ceph_assert(agent_oids.count(oid) == 0); |
7c673cae FG |
434 | agent_oids.insert(oid); |
435 | } | |
436 | ||
437 | /// note finish or cancellation of an async (flush) op | |
438 | void agent_finish_op(const hobject_t& oid) { | |
11fdf7f2 TL |
439 | std::lock_guard l(agent_lock); |
440 | ceph_assert(agent_ops > 0); | |
7c673cae | 441 | --agent_ops; |
11fdf7f2 | 442 | ceph_assert(agent_oids.count(oid) == 1); |
7c673cae | 443 | agent_oids.erase(oid); |
9f95a23c | 444 | agent_cond.notify_all(); |
7c673cae FG |
445 | } |
446 | ||
447 | /// check if we are operating on an object | |
448 | bool agent_is_active_oid(const hobject_t& oid) { | |
11fdf7f2 | 449 | std::lock_guard l(agent_lock); |
7c673cae FG |
450 | return agent_oids.count(oid); |
451 | } | |
452 | ||
453 | /// get count of active agent ops | |
454 | int agent_get_num_ops() { | |
11fdf7f2 | 455 | std::lock_guard l(agent_lock); |
7c673cae FG |
456 | return agent_ops; |
457 | } | |
458 | ||
459 | void agent_inc_high_count() { | |
11fdf7f2 | 460 | std::lock_guard l(agent_lock); |
7c673cae FG |
461 | flush_mode_high_count ++; |
462 | } | |
463 | ||
464 | void agent_dec_high_count() { | |
11fdf7f2 | 465 | std::lock_guard l(agent_lock); |
7c673cae FG |
466 | flush_mode_high_count --; |
467 | } | |
468 | ||
469 | private: | |
470 | /// throttle promotion attempts | |
11fdf7f2 | 471 | std::atomic<unsigned int> promote_probability_millis{1000}; ///< probability thousands. one word. |
7c673cae FG |
472 | PromoteCounter promote_counter; |
473 | utime_t last_recalibrate; | |
474 | unsigned long promote_max_objects, promote_max_bytes; | |
475 | ||
476 | public: | |
477 | bool promote_throttle() { | |
478 | // NOTE: lockless! we rely on the probability being a single word. | |
479 | promote_counter.attempt(); | |
480 | if ((unsigned)rand() % 1000 > promote_probability_millis) | |
481 | return true; // yes throttle (no promote) | |
482 | if (promote_max_objects && | |
483 | promote_counter.objects > promote_max_objects) | |
484 | return true; // yes throttle | |
485 | if (promote_max_bytes && | |
486 | promote_counter.bytes > promote_max_bytes) | |
487 | return true; // yes throttle | |
488 | return false; // no throttle (promote) | |
489 | } | |
490 | void promote_finish(uint64_t bytes) { | |
491 | promote_counter.finish(bytes); | |
492 | } | |
493 | void promote_throttle_recalibrate(); | |
9f95a23c TL |
494 | unsigned get_num_shards() const { |
495 | return m_objecter_finishers; | |
496 | } | |
497 | Finisher* get_objecter_finisher(int shard) { | |
498 | return objecter_finishers[shard].get(); | |
499 | } | |
7c673cae FG |
500 | |
501 | // -- Objecter, for tiering reads/writes from/to other OSDs -- | |
f67539c2 | 502 | ceph::async::io_context_pool& poolctx; |
9f95a23c | 503 | std::unique_ptr<Objecter> objecter; |
11fdf7f2 | 504 | int m_objecter_finishers; |
9f95a23c | 505 | std::vector<std::unique_ptr<Finisher>> objecter_finishers; |
7c673cae FG |
506 | |
507 | // -- Watch -- | |
9f95a23c | 508 | ceph::mutex watch_lock = ceph::make_mutex("OSDService::watch_lock"); |
7c673cae FG |
509 | SafeTimer watch_timer; |
510 | uint64_t next_notif_id; | |
511 | uint64_t get_next_id(epoch_t cur_epoch) { | |
11fdf7f2 | 512 | std::lock_guard l(watch_lock); |
7c673cae FG |
513 | return (((uint64_t)cur_epoch) << 32) | ((uint64_t)(next_notif_id++)); |
514 | } | |
515 | ||
516 | // -- Recovery/Backfill Request Scheduling -- | |
9f95a23c | 517 | ceph::mutex recovery_request_lock = ceph::make_mutex("OSDService::recovery_request_lock"); |
7c673cae FG |
518 | SafeTimer recovery_request_timer; |
519 | ||
31f18b77 FG |
520 | // For async recovery sleep |
521 | bool recovery_needs_sleep = true; | |
9f95a23c | 522 | ceph::real_clock::time_point recovery_schedule_time; |
31f18b77 | 523 | |
11fdf7f2 | 524 | // For recovery & scrub & snap |
9f95a23c | 525 | ceph::mutex sleep_lock = ceph::make_mutex("OSDService::sleep_lock"); |
11fdf7f2 | 526 | SafeTimer sleep_timer; |
31f18b77 | 527 | |
7c673cae FG |
528 | // -- tids -- |
529 | // for ops i issue | |
11fdf7f2 | 530 | std::atomic<unsigned int> last_tid{0}; |
7c673cae FG |
531 | ceph_tid_t get_tid() { |
532 | return (ceph_tid_t)last_tid++; | |
533 | } | |
534 | ||
535 | // -- backfill_reservation -- | |
536 | Finisher reserver_finisher; | |
f67539c2 TL |
537 | AsyncReserver<spg_t, Finisher> local_reserver; |
538 | AsyncReserver<spg_t, Finisher> remote_reserver; | |
7c673cae | 539 | |
11fdf7f2 | 540 | // -- pg merge -- |
9f95a23c | 541 | ceph::mutex merge_lock = ceph::make_mutex("OSD::merge_lock"); |
f67539c2 TL |
542 | std::map<pg_t,eversion_t> ready_to_merge_source; // pg -> version |
543 | std::map<pg_t,std::tuple<eversion_t,epoch_t,epoch_t>> ready_to_merge_target; // pg -> (version,les,lec) | |
544 | std::set<pg_t> not_ready_to_merge_source; | |
545 | std::map<pg_t,pg_t> not_ready_to_merge_target; | |
546 | std::set<pg_t> sent_ready_to_merge_source; | |
11fdf7f2 TL |
547 | |
548 | void set_ready_to_merge_source(PG *pg, | |
549 | eversion_t version); | |
550 | void set_ready_to_merge_target(PG *pg, | |
551 | eversion_t version, | |
552 | epoch_t last_epoch_started, | |
553 | epoch_t last_epoch_clean); | |
554 | void set_not_ready_to_merge_source(pg_t source); | |
555 | void set_not_ready_to_merge_target(pg_t target, pg_t source); | |
556 | void clear_ready_to_merge(PG *pg); | |
557 | void send_ready_to_merge(); | |
558 | void _send_ready_to_merge(); | |
559 | void clear_sent_ready_to_merge(); | |
9f95a23c | 560 | void prune_sent_ready_to_merge(const OSDMapRef& osdmap); |
11fdf7f2 | 561 | |
7c673cae FG |
562 | // -- pg_temp -- |
563 | private: | |
9f95a23c | 564 | ceph::mutex pg_temp_lock = ceph::make_mutex("OSDService::pg_temp_lock"); |
94b18763 | 565 | struct pg_temp_t { |
f67539c2 | 566 | std::vector<int> acting; |
94b18763 FG |
567 | bool forced = false; |
568 | }; | |
f67539c2 TL |
569 | std::map<pg_t, pg_temp_t> pg_temp_wanted; |
570 | std::map<pg_t, pg_temp_t> pg_temp_pending; | |
7c673cae | 571 | void _sent_pg_temp(); |
94b18763 | 572 | friend std::ostream& operator<<(std::ostream&, const pg_temp_t&); |
7c673cae | 573 | public: |
f67539c2 | 574 | void queue_want_pg_temp(pg_t pgid, const std::vector<int>& want, |
94b18763 | 575 | bool forced = false); |
7c673cae FG |
576 | void remove_want_pg_temp(pg_t pgid); |
577 | void requeue_pg_temp(); | |
578 | void send_pg_temp(); | |
579 | ||
11fdf7f2 | 580 | ceph::mutex pg_created_lock = ceph::make_mutex("OSDService::pg_created_lock"); |
f67539c2 | 581 | std::set<pg_t> pg_created; |
7c673cae | 582 | void send_pg_created(pg_t pgid); |
11fdf7f2 TL |
583 | void prune_pg_created(); |
584 | void send_pg_created(); | |
31f18b77 | 585 | |
f67539c2 | 586 | AsyncReserver<spg_t, Finisher> snap_reserver; |
11fdf7f2 | 587 | void queue_recovery_context(PG *pg, GenContext<ThreadPool::TPHandle&> *c); |
7c673cae | 588 | void queue_for_snap_trim(PG *pg); |
f67539c2 TL |
589 | void queue_for_scrub(PG* pg, Scrub::scrub_prio_t with_priority); |
590 | void queue_scrub_after_repair(PG* pg, Scrub::scrub_prio_t with_priority); | |
591 | ||
592 | /// queue the message (-> event) that all replicas reserved scrub resources for us | |
593 | void queue_for_scrub_granted(PG* pg, Scrub::scrub_prio_t with_priority); | |
594 | ||
595 | /// queue the message (-> event) that some replicas denied our scrub resources request | |
596 | void queue_for_scrub_denied(PG* pg, Scrub::scrub_prio_t with_priority); | |
597 | ||
598 | /// Signals either (a) the end of a sleep period, or (b) a recheck of the availability | |
599 | /// of the primary map being created by the backend. | |
600 | void queue_for_scrub_resched(PG* pg, Scrub::scrub_prio_t with_priority); | |
601 | ||
602 | /// Signals a change in the number of in-flight recovery writes | |
603 | void queue_scrub_pushes_update(PG* pg, Scrub::scrub_prio_t with_priority); | |
604 | ||
605 | /// Signals that all pending updates were applied | |
606 | void queue_scrub_applied_update(PG* pg, Scrub::scrub_prio_t with_priority); | |
607 | ||
608 | /// The block-range that was locked and prevented the scrubbing - is freed | |
609 | void queue_scrub_unblocking(PG* pg, Scrub::scrub_prio_t with_priority); | |
610 | ||
611 | /// Signals that all write OPs are done | |
612 | void queue_scrub_digest_update(PG* pg, Scrub::scrub_prio_t with_priority); | |
613 | ||
614 | /// Signals that we (the Primary) got all waited-for scrub-maps from our replicas | |
615 | void queue_scrub_got_repl_maps(PG* pg, Scrub::scrub_prio_t with_priority); | |
616 | ||
617 | void queue_for_rep_scrub(PG* pg, | |
618 | Scrub::scrub_prio_t with_high_priority, | |
619 | unsigned int qu_priority); | |
620 | ||
621 | /// Signals a change in the number of in-flight recovery writes | |
622 | void queue_scrub_replica_pushes(PG *pg, Scrub::scrub_prio_t with_priority); | |
623 | ||
624 | void queue_for_rep_scrub_resched(PG* pg, | |
625 | Scrub::scrub_prio_t with_high_priority, | |
626 | unsigned int qu_priority); | |
627 | ||
11fdf7f2 TL |
628 | void queue_for_pg_delete(spg_t pgid, epoch_t e); |
629 | bool try_finish_pg_delete(PG *pg, unsigned old_pg_num); | |
7c673cae FG |
630 | |
631 | private: | |
632 | // -- pg recovery and associated throttling -- | |
9f95a23c | 633 | ceph::mutex recovery_lock = ceph::make_mutex("OSDService::recovery_lock"); |
f67539c2 TL |
634 | std::list<std::pair<epoch_t, PGRef> > awaiting_throttle; |
635 | ||
636 | /// queue a scrub-related message for a PG | |
637 | template <class MSG_TYPE> | |
638 | void queue_scrub_event_msg(PG* pg, | |
639 | Scrub::scrub_prio_t with_priority, | |
640 | unsigned int qu_priority); | |
641 | ||
642 | /// An alternative version of queue_scrub_event_msg(), in which the queuing priority is | |
643 | /// provided by the executing scrub (i.e. taken from PgScrubber::m_flags) | |
644 | template <class MSG_TYPE> | |
645 | void queue_scrub_event_msg(PG* pg, Scrub::scrub_prio_t with_priority); | |
7c673cae FG |
646 | |
647 | utime_t defer_recovery_until; | |
648 | uint64_t recovery_ops_active; | |
649 | uint64_t recovery_ops_reserved; | |
650 | bool recovery_paused; | |
651 | #ifdef DEBUG_RECOVERY_OIDS | |
f67539c2 | 652 | std::map<spg_t, std::set<hobject_t> > recovery_oids; |
7c673cae FG |
653 | #endif |
654 | bool _recover_now(uint64_t *available_pushes); | |
655 | void _maybe_queue_recovery(); | |
656 | void _queue_for_recovery( | |
f67539c2 | 657 | std::pair<epoch_t, PGRef> p, uint64_t reserved_pushes); |
7c673cae FG |
658 | public: |
659 | void start_recovery_op(PG *pg, const hobject_t& soid); | |
660 | void finish_recovery_op(PG *pg, const hobject_t& soid, bool dequeue); | |
661 | bool is_recovery_active(); | |
11fdf7f2 | 662 | void release_reserved_pushes(uint64_t pushes); |
7c673cae FG |
663 | void defer_recovery(float defer_for) { |
664 | defer_recovery_until = ceph_clock_now(); | |
665 | defer_recovery_until += defer_for; | |
666 | } | |
667 | void pause_recovery() { | |
11fdf7f2 | 668 | std::lock_guard l(recovery_lock); |
7c673cae FG |
669 | recovery_paused = true; |
670 | } | |
671 | bool recovery_is_paused() { | |
11fdf7f2 | 672 | std::lock_guard l(recovery_lock); |
7c673cae FG |
673 | return recovery_paused; |
674 | } | |
675 | void unpause_recovery() { | |
11fdf7f2 | 676 | std::lock_guard l(recovery_lock); |
7c673cae FG |
677 | recovery_paused = false; |
678 | _maybe_queue_recovery(); | |
679 | } | |
680 | void kick_recovery_queue() { | |
11fdf7f2 | 681 | std::lock_guard l(recovery_lock); |
7c673cae FG |
682 | _maybe_queue_recovery(); |
683 | } | |
684 | void clear_queued_recovery(PG *pg) { | |
11fdf7f2 TL |
685 | std::lock_guard l(recovery_lock); |
686 | awaiting_throttle.remove_if( | |
687 | [pg](decltype(awaiting_throttle)::const_reference awaiting ) { | |
688 | return awaiting.second.get() == pg; | |
689 | }); | |
7c673cae | 690 | } |
9f95a23c TL |
691 | |
692 | unsigned get_target_pg_log_entries() const; | |
f67539c2 | 693 | |
7c673cae | 694 | // delayed pg activation |
c07f9fc5 | 695 | void queue_for_recovery(PG *pg) { |
11fdf7f2 | 696 | std::lock_guard l(recovery_lock); |
c07f9fc5 | 697 | |
11fdf7f2 | 698 | if (pg->is_forced_recovery_or_backfill()) { |
f67539c2 | 699 | awaiting_throttle.push_front(std::make_pair(pg->get_osdmap()->get_epoch(), pg)); |
7c673cae | 700 | } else { |
f67539c2 | 701 | awaiting_throttle.push_back(std::make_pair(pg->get_osdmap()->get_epoch(), pg)); |
7c673cae FG |
702 | } |
703 | _maybe_queue_recovery(); | |
704 | } | |
31f18b77 | 705 | void queue_recovery_after_sleep(PG *pg, epoch_t queued, uint64_t reserved_pushes) { |
11fdf7f2 | 706 | std::lock_guard l(recovery_lock); |
f67539c2 | 707 | _queue_for_recovery(std::make_pair(queued, pg), reserved_pushes); |
31f18b77 | 708 | } |
7c673cae | 709 | |
9f95a23c TL |
710 | void queue_check_readable(spg_t spgid, |
711 | epoch_t lpr, | |
712 | ceph::signedspan delay = ceph::signedspan::zero()); | |
713 | ||
7c673cae | 714 | // osd map cache (past osd maps) |
9f95a23c | 715 | ceph::mutex map_cache_lock = ceph::make_mutex("OSDService::map_cache_lock"); |
7c673cae | 716 | SharedLRU<epoch_t, const OSDMap> map_cache; |
f67539c2 TL |
717 | SimpleLRU<epoch_t, ceph::buffer::list> map_bl_cache; |
718 | SimpleLRU<epoch_t, ceph::buffer::list> map_bl_inc_cache; | |
7c673cae FG |
719 | |
720 | OSDMapRef try_get_map(epoch_t e); | |
721 | OSDMapRef get_map(epoch_t e) { | |
722 | OSDMapRef ret(try_get_map(e)); | |
11fdf7f2 | 723 | ceph_assert(ret); |
7c673cae FG |
724 | return ret; |
725 | } | |
726 | OSDMapRef add_map(OSDMap *o) { | |
11fdf7f2 | 727 | std::lock_guard l(map_cache_lock); |
7c673cae FG |
728 | return _add_map(o); |
729 | } | |
730 | OSDMapRef _add_map(OSDMap *o); | |
731 | ||
f67539c2 TL |
732 | void _add_map_bl(epoch_t e, ceph::buffer::list& bl); |
733 | bool get_map_bl(epoch_t e, ceph::buffer::list& bl) { | |
11fdf7f2 | 734 | std::lock_guard l(map_cache_lock); |
7c673cae FG |
735 | return _get_map_bl(e, bl); |
736 | } | |
f67539c2 | 737 | bool _get_map_bl(epoch_t e, ceph::buffer::list& bl); |
7c673cae | 738 | |
f67539c2 TL |
739 | void _add_map_inc_bl(epoch_t e, ceph::buffer::list& bl); |
740 | bool get_inc_map_bl(epoch_t e, ceph::buffer::list& bl); | |
7c673cae | 741 | |
11fdf7f2 TL |
742 | /// identify split child pgids over a osdmap interval |
743 | void identify_splits_and_merges( | |
744 | OSDMapRef old_map, | |
745 | OSDMapRef new_map, | |
746 | spg_t pgid, | |
f67539c2 TL |
747 | std::set<std::pair<spg_t,epoch_t>> *new_children, |
748 | std::set<std::pair<spg_t,epoch_t>> *merge_pgs); | |
11fdf7f2 TL |
749 | |
750 | void need_heartbeat_peer_update(); | |
7c673cae FG |
751 | |
752 | void init(); | |
f67539c2 | 753 | void final_init(); |
7c673cae | 754 | void start_shutdown(); |
31f18b77 | 755 | void shutdown_reserver(); |
7c673cae FG |
756 | void shutdown(); |
757 | ||
7c673cae | 758 | // -- stats -- |
9f95a23c | 759 | ceph::mutex stat_lock = ceph::make_mutex("OSDService::stat_lock"); |
7c673cae | 760 | osd_stat_t osd_stat; |
31f18b77 | 761 | uint32_t seq = 0; |
7c673cae | 762 | |
11fdf7f2 TL |
763 | void set_statfs(const struct store_statfs_t &stbuf, |
764 | osd_alert_list_t& alerts); | |
f67539c2 | 765 | osd_stat_t set_osd_stat(std::vector<int>& hb_peers, int num_pgs); |
11fdf7f2 TL |
766 | void inc_osd_stat_repaired(void); |
767 | float compute_adjusted_ratio(osd_stat_t new_stat, float *pratio, uint64_t adjust_used = 0); | |
7c673cae | 768 | osd_stat_t get_osd_stat() { |
11fdf7f2 | 769 | std::lock_guard l(stat_lock); |
31f18b77 FG |
770 | ++seq; |
771 | osd_stat.up_from = up_epoch; | |
772 | osd_stat.seq = ((uint64_t)osd_stat.up_from << 32) + seq; | |
7c673cae FG |
773 | return osd_stat; |
774 | } | |
31f18b77 | 775 | uint64_t get_osd_stat_seq() { |
11fdf7f2 | 776 | std::lock_guard l(stat_lock); |
31f18b77 FG |
777 | return osd_stat.seq; |
778 | } | |
f67539c2 | 779 | void get_hb_pingtime(std::map<int, osd_stat_t::Interfaces> *pp) |
eafe8130 TL |
780 | { |
781 | std::lock_guard l(stat_lock); | |
782 | *pp = osd_stat.hb_pingtime; | |
783 | return; | |
784 | } | |
7c673cae FG |
785 | |
786 | // -- OSD Full Status -- | |
787 | private: | |
788 | friend TestOpsSocketHook; | |
9f95a23c | 789 | mutable ceph::mutex full_status_lock = ceph::make_mutex("OSDService::full_status_lock"); |
7c673cae FG |
790 | enum s_names { INVALID = -1, NONE, NEARFULL, BACKFILLFULL, FULL, FAILSAFE } cur_state; // ascending |
791 | const char *get_full_state_name(s_names s) const { | |
792 | switch (s) { | |
793 | case NONE: return "none"; | |
794 | case NEARFULL: return "nearfull"; | |
795 | case BACKFILLFULL: return "backfillfull"; | |
796 | case FULL: return "full"; | |
797 | case FAILSAFE: return "failsafe"; | |
798 | default: return "???"; | |
799 | } | |
800 | } | |
f67539c2 | 801 | s_names get_full_state(std::string type) const { |
7c673cae FG |
802 | if (type == "none") |
803 | return NONE; | |
804 | else if (type == "failsafe") | |
805 | return FAILSAFE; | |
806 | else if (type == "full") | |
807 | return FULL; | |
808 | else if (type == "backfillfull") | |
809 | return BACKFILLFULL; | |
810 | else if (type == "nearfull") | |
811 | return NEARFULL; | |
812 | else | |
813 | return INVALID; | |
814 | } | |
11fdf7f2 | 815 | double cur_ratio, physical_ratio; ///< current utilization |
7c673cae FG |
816 | mutable int64_t injectfull = 0; |
817 | s_names injectfull_state = NONE; | |
818 | float get_failsafe_full_ratio(); | |
11fdf7f2 TL |
819 | bool _check_inject_full(DoutPrefixProvider *dpp, s_names type) const; |
820 | bool _check_full(DoutPrefixProvider *dpp, s_names type) const; | |
7c673cae | 821 | public: |
11fdf7f2 | 822 | void check_full_status(float ratio, float pratio); |
f67539c2 | 823 | s_names recalc_full_state(float ratio, float pratio, std::string &inject); |
11fdf7f2 TL |
824 | bool _tentative_full(DoutPrefixProvider *dpp, s_names type, uint64_t adjust_used, osd_stat_t); |
825 | bool check_failsafe_full(DoutPrefixProvider *dpp) const; | |
826 | bool check_full(DoutPrefixProvider *dpp) const; | |
827 | bool tentative_backfill_full(DoutPrefixProvider *dpp, uint64_t adjust_used, osd_stat_t); | |
828 | bool check_backfill_full(DoutPrefixProvider *dpp) const; | |
829 | bool check_nearfull(DoutPrefixProvider *dpp) const; | |
7c673cae FG |
830 | bool is_failsafe_full() const; |
831 | bool is_full() const; | |
832 | bool is_backfillfull() const; | |
833 | bool is_nearfull() const; | |
834 | bool need_fullness_update(); ///< osdmap state needs update | |
835 | void set_injectfull(s_names type, int64_t count); | |
7c673cae FG |
836 | |
837 | ||
838 | // -- epochs -- | |
839 | private: | |
9f95a23c TL |
840 | // protects access to boot_epoch, up_epoch, bind_epoch |
841 | mutable ceph::mutex epoch_lock = ceph::make_mutex("OSDService::epoch_lock"); | |
7c673cae FG |
842 | epoch_t boot_epoch; // _first_ epoch we were marked up (after this process started) |
843 | epoch_t up_epoch; // _most_recent_ epoch we were marked up | |
844 | epoch_t bind_epoch; // epoch we last did a bind to new ip:ports | |
845 | public: | |
846 | /** | |
f67539c2 | 847 | * Retrieve the boot_, up_, and bind_ epochs the OSD has std::set. The params |
7c673cae FG |
848 | * can be NULL if you don't care about them. |
849 | */ | |
850 | void retrieve_epochs(epoch_t *_boot_epoch, epoch_t *_up_epoch, | |
851 | epoch_t *_bind_epoch) const; | |
852 | /** | |
f67539c2 | 853 | * Std::set the boot, up, and bind epochs. Any NULL params will not be std::set. |
7c673cae FG |
854 | */ |
855 | void set_epochs(const epoch_t *_boot_epoch, const epoch_t *_up_epoch, | |
856 | const epoch_t *_bind_epoch); | |
857 | epoch_t get_boot_epoch() const { | |
858 | epoch_t ret; | |
859 | retrieve_epochs(&ret, NULL, NULL); | |
860 | return ret; | |
861 | } | |
862 | epoch_t get_up_epoch() const { | |
863 | epoch_t ret; | |
864 | retrieve_epochs(NULL, &ret, NULL); | |
865 | return ret; | |
866 | } | |
867 | epoch_t get_bind_epoch() const { | |
868 | epoch_t ret; | |
869 | retrieve_epochs(NULL, NULL, &ret); | |
870 | return ret; | |
871 | } | |
872 | ||
181888fb FG |
873 | void request_osdmap_update(epoch_t e); |
874 | ||
9f95a23c TL |
875 | // -- heartbeats -- |
876 | ceph::mutex hb_stamp_lock = ceph::make_mutex("OSDServce::hb_stamp_lock"); | |
877 | ||
878 | /// osd -> heartbeat stamps | |
f67539c2 | 879 | std::vector<HeartbeatStampsRef> hb_stamps; |
9f95a23c TL |
880 | |
881 | /// get or create a ref for a peer's HeartbeatStamps | |
882 | HeartbeatStampsRef get_hb_stamps(unsigned osd); | |
883 | ||
884 | ||
885 | // Timer for readable leases | |
886 | ceph::timer<ceph::mono_clock> mono_timer = ceph::timer<ceph::mono_clock>{ceph::construct_suspended}; | |
887 | ||
888 | void queue_renew_lease(epoch_t epoch, spg_t spgid); | |
889 | ||
7c673cae | 890 | // -- stopping -- |
9f95a23c TL |
891 | ceph::mutex is_stopping_lock = ceph::make_mutex("OSDService::is_stopping_lock"); |
892 | ceph::condition_variable is_stopping_cond; | |
7c673cae FG |
893 | enum { |
894 | NOT_STOPPING, | |
895 | PREPARING_TO_STOP, | |
896 | STOPPING }; | |
11fdf7f2 TL |
897 | std::atomic<int> state{NOT_STOPPING}; |
898 | int get_state() const { | |
7c673cae FG |
899 | return state; |
900 | } | |
901 | void set_state(int s) { | |
902 | state = s; | |
903 | } | |
904 | bool is_stopping() const { | |
905 | return state == STOPPING; | |
906 | } | |
907 | bool is_preparing_to_stop() const { | |
908 | return state == PREPARING_TO_STOP; | |
909 | } | |
910 | bool prepare_to_stop(); | |
911 | void got_stop_ack(); | |
912 | ||
913 | ||
914 | #ifdef PG_DEBUG_REFS | |
9f95a23c | 915 | ceph::mutex pgid_lock = ceph::make_mutex("OSDService::pgid_lock"); |
f67539c2 TL |
916 | std::map<spg_t, int> pgid_tracker; |
917 | std::map<spg_t, PG*> live_pgs; | |
31f18b77 FG |
918 | void add_pgid(spg_t pgid, PG *pg); |
919 | void remove_pgid(spg_t pgid, PG *pg); | |
920 | void dump_live_pgids(); | |
7c673cae FG |
921 | #endif |
922 | ||
f67539c2 | 923 | explicit OSDService(OSD *osd, ceph::async::io_context_pool& poolctx); |
9f95a23c | 924 | ~OSDService() = default; |
7c673cae FG |
925 | }; |
926 | ||
11fdf7f2 TL |
927 | /* |
928 | ||
929 | Each PG slot includes queues for events that are processing and/or waiting | |
930 | for a PG to be materialized in the slot. | |
931 | ||
932 | These are the constraints: | |
933 | ||
f67539c2 | 934 | - client ops must remained ordered by client, regardless of std::map epoch |
11fdf7f2 TL |
935 | - peering messages/events from peers must remain ordered by peer |
936 | - peering messages and client ops need not be ordered relative to each other | |
937 | ||
938 | - some peering events can create a pg (e.g., notify) | |
939 | - the query peering event can proceed when a PG doesn't exist | |
940 | ||
941 | Implementation notes: | |
942 | ||
943 | - everybody waits for split. If the OSD has the parent PG it will instantiate | |
944 | the PGSlot early and mark it waiting_for_split. Everything will wait until | |
945 | the parent is able to commit the split operation and the child PG's are | |
946 | materialized in the child slots. | |
947 | ||
948 | - every event has an epoch property and will wait for the OSDShard to catch | |
949 | up to that epoch. For example, if we get a peering event from a future | |
950 | epoch, the event will wait in the slot until the local OSD has caught up. | |
951 | (We should be judicious in specifying the required epoch [by, e.g., setting | |
952 | it to the same_interval_since epoch] so that we don't wait for epochs that | |
953 | don't affect the given PG.) | |
954 | ||
955 | - we maintain two separate wait lists, *waiting* and *waiting_peering*. The | |
9f95a23c | 956 | OpSchedulerItem has an is_peering() bool to determine which we use. Waiting |
11fdf7f2 TL |
957 | peering events are queued up by epoch required. |
958 | ||
959 | - when we wake a PG slot (e.g., we finished split, or got a newer osdmap, or | |
960 | materialized the PG), we wake *all* waiting items. (This could be optimized, | |
961 | probably, but we don't bother.) We always requeue peering items ahead of | |
962 | client ops. | |
963 | ||
964 | - some peering events are marked !peering_requires_pg (PGQuery). if we do | |
965 | not have a PG these are processed immediately (under the shard lock). | |
966 | ||
967 | - we do not have a PG present, we check if the slot maps to the current host. | |
968 | if so, we either queue the item and wait for the PG to materialize, or | |
969 | (if the event is a pg creating event like PGNotify), we materialize the PG. | |
970 | ||
971 | - when we advance the osdmap on the OSDShard, we scan pg slots and | |
972 | discard any slots with no pg (and not waiting_for_split) that no | |
f67539c2 | 973 | longer std::map to the current host. |
11fdf7f2 TL |
974 | |
975 | */ | |
976 | ||
977 | struct OSDShardPGSlot { | |
9f95a23c | 978 | using OpSchedulerItem = ceph::osd::scheduler::OpSchedulerItem; |
11fdf7f2 | 979 | PGRef pg; ///< pg reference |
f67539c2 | 980 | std::deque<OpSchedulerItem> to_process; ///< order items for this slot |
11fdf7f2 TL |
981 | int num_running = 0; ///< _process threads doing pg lookup/lock |
982 | ||
f67539c2 | 983 | std::deque<OpSchedulerItem> waiting; ///< waiting for pg (or map + pg) |
11fdf7f2 TL |
984 | |
985 | /// waiting for map (peering evt) | |
f67539c2 | 986 | std::map<epoch_t,std::deque<OpSchedulerItem>> waiting_peering; |
11fdf7f2 TL |
987 | |
988 | /// incremented by wake_pg_waiters; indicates racing _process threads | |
989 | /// should bail out (their op has been requeued) | |
990 | uint64_t requeue_seq = 0; | |
991 | ||
992 | /// waiting for split child to materialize in these epoch(s) | |
f67539c2 | 993 | std::set<epoch_t> waiting_for_split; |
11fdf7f2 TL |
994 | |
995 | epoch_t epoch = 0; | |
996 | boost::intrusive::set_member_hook<> pg_epoch_item; | |
997 | ||
998 | /// waiting for a merge (source or target) by this epoch | |
999 | epoch_t waiting_for_merge_epoch = 0; | |
1000 | }; | |
1001 | ||
1002 | struct OSDShard { | |
1003 | const unsigned shard_id; | |
1004 | CephContext *cct; | |
1005 | OSD *osd; | |
1006 | ||
f67539c2 | 1007 | std::string shard_name; |
11fdf7f2 | 1008 | |
f67539c2 | 1009 | std::string sdata_wait_lock_name; |
11fdf7f2 TL |
1010 | ceph::mutex sdata_wait_lock; |
1011 | ceph::condition_variable sdata_cond; | |
f67539c2 | 1012 | int waiting_threads = 0; |
11fdf7f2 | 1013 | |
11fdf7f2 TL |
1014 | ceph::mutex osdmap_lock; ///< protect shard_osdmap updates vs users w/o shard_lock |
1015 | OSDMapRef shard_osdmap; | |
1016 | ||
1017 | OSDMapRef get_osdmap() { | |
1018 | std::lock_guard l(osdmap_lock); | |
1019 | return shard_osdmap; | |
1020 | } | |
1021 | ||
f67539c2 | 1022 | std::string shard_lock_name; |
11fdf7f2 TL |
1023 | ceph::mutex shard_lock; ///< protects remaining members below |
1024 | ||
1025 | /// map of slots for each spg_t. maintains ordering of items dequeued | |
9f95a23c | 1026 | /// from scheduler while _process thread drops shard lock to acquire the |
11fdf7f2 | 1027 | /// pg lock. stale slots are removed by consume_map. |
f67539c2 | 1028 | std::unordered_map<spg_t,std::unique_ptr<OSDShardPGSlot>> pg_slots; |
11fdf7f2 TL |
1029 | |
1030 | struct pg_slot_compare_by_epoch { | |
1031 | bool operator()(const OSDShardPGSlot& l, const OSDShardPGSlot& r) const { | |
1032 | return l.epoch < r.epoch; | |
1033 | } | |
1034 | }; | |
1035 | ||
1036 | /// maintain an ordering of pg slots by pg epoch | |
1037 | boost::intrusive::multiset< | |
1038 | OSDShardPGSlot, | |
1039 | boost::intrusive::member_hook< | |
1040 | OSDShardPGSlot, | |
1041 | boost::intrusive::set_member_hook<>, | |
1042 | &OSDShardPGSlot::pg_epoch_item>, | |
1043 | boost::intrusive::compare<pg_slot_compare_by_epoch>> pg_slots_by_epoch; | |
1044 | int waiting_for_min_pg_epoch = 0; | |
1045 | ceph::condition_variable min_pg_epoch_cond; | |
1046 | ||
1047 | /// priority queue | |
9f95a23c | 1048 | ceph::osd::scheduler::OpSchedulerRef scheduler; |
11fdf7f2 TL |
1049 | |
1050 | bool stop_waiting = false; | |
1051 | ||
1052 | ContextQueue context_queue; | |
1053 | ||
11fdf7f2 TL |
1054 | void _attach_pg(OSDShardPGSlot *slot, PG *pg); |
1055 | void _detach_pg(OSDShardPGSlot *slot); | |
1056 | ||
1057 | void update_pg_epoch(OSDShardPGSlot *slot, epoch_t epoch); | |
1058 | epoch_t get_min_pg_epoch(); | |
1059 | void wait_min_pg_epoch(epoch_t need); | |
1060 | ||
1061 | /// return newest epoch we are waiting for | |
1062 | epoch_t get_max_waiting_epoch(); | |
1063 | ||
1064 | /// push osdmap into shard | |
1065 | void consume_map( | |
9f95a23c | 1066 | const OSDMapRef& osdmap, |
11fdf7f2 TL |
1067 | unsigned *pushes_to_free); |
1068 | ||
1069 | void _wake_pg_slot(spg_t pgid, OSDShardPGSlot *slot); | |
1070 | ||
1071 | void identify_splits_and_merges( | |
1072 | const OSDMapRef& as_of_osdmap, | |
f67539c2 TL |
1073 | std::set<std::pair<spg_t,epoch_t>> *split_children, |
1074 | std::set<std::pair<spg_t,epoch_t>> *merge_pgs); | |
1075 | void _prime_splits(std::set<std::pair<spg_t,epoch_t>> *pgids); | |
11fdf7f2 | 1076 | void prime_splits(const OSDMapRef& as_of_osdmap, |
f67539c2 | 1077 | std::set<std::pair<spg_t,epoch_t>> *pgids); |
11fdf7f2 | 1078 | void prime_merges(const OSDMapRef& as_of_osdmap, |
f67539c2 | 1079 | std::set<std::pair<spg_t,epoch_t>> *merge_pgs); |
11fdf7f2 TL |
1080 | void register_and_wake_split_child(PG *pg); |
1081 | void unprime_split_children(spg_t parent, unsigned old_pg_num); | |
1082 | ||
1083 | OSDShard( | |
1084 | int id, | |
1085 | CephContext *cct, | |
9f95a23c | 1086 | OSD *osd); |
11fdf7f2 TL |
1087 | }; |
1088 | ||
7c673cae FG |
1089 | class OSD : public Dispatcher, |
1090 | public md_config_obs_t { | |
9f95a23c TL |
1091 | using OpSchedulerItem = ceph::osd::scheduler::OpSchedulerItem; |
1092 | ||
7c673cae | 1093 | /** OSD **/ |
9f95a23c TL |
1094 | // global lock |
1095 | ceph::mutex osd_lock = ceph::make_mutex("OSD::osd_lock"); | |
7c673cae FG |
1096 | SafeTimer tick_timer; // safe timer (osd_lock) |
1097 | ||
1098 | // Tick timer for those stuff that do not need osd_lock | |
9f95a23c | 1099 | ceph::mutex tick_timer_lock = ceph::make_mutex("OSD::tick_timer_lock"); |
7c673cae | 1100 | SafeTimer tick_timer_without_osd_lock; |
11fdf7f2 TL |
1101 | std::string gss_ktfile_client{}; |
1102 | ||
7c673cae FG |
1103 | public: |
1104 | // config observer bits | |
1105 | const char** get_tracked_conf_keys() const override; | |
11fdf7f2 | 1106 | void handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
1107 | const std::set <std::string> &changed) override; |
1108 | void update_log_config(); | |
1109 | void check_config(); | |
1110 | ||
1111 | protected: | |
1112 | ||
91327a77 AA |
1113 | const double OSD_TICK_INTERVAL = { 1.0 }; |
1114 | double get_tick_interval() const; | |
7c673cae | 1115 | |
7c673cae FG |
1116 | Messenger *cluster_messenger; |
1117 | Messenger *client_messenger; | |
1118 | Messenger *objecter_messenger; | |
1119 | MonClient *monc; // check the "monc helpers" list before accessing directly | |
1120 | MgrClient mgrc; | |
1121 | PerfCounters *logger; | |
1122 | PerfCounters *recoverystate_perf; | |
1123 | ObjectStore *store; | |
1124 | #ifdef HAVE_LIBFUSE | |
1125 | FuseStore *fuse_store = nullptr; | |
1126 | #endif | |
1127 | LogClient log_client; | |
1128 | LogChannelRef clog; | |
1129 | ||
1130 | int whoami; | |
1131 | std::string dev_path, journal_path; | |
1132 | ||
9f95a23c | 1133 | ceph_release_t last_require_osd_release{ceph_release_t::unknown}; |
11fdf7f2 TL |
1134 | |
1135 | int numa_node = -1; | |
1136 | size_t numa_cpu_set_size = 0; | |
1137 | cpu_set_t numa_cpu_set; | |
1138 | ||
31f18b77 | 1139 | bool store_is_rotational = true; |
d2e6a577 | 1140 | bool journal_is_rotational = true; |
31f18b77 | 1141 | |
7c673cae | 1142 | ZTracer::Endpoint trace_endpoint; |
f67539c2 TL |
1143 | PerfCounters* create_logger(); |
1144 | PerfCounters* create_recoverystate_perf(); | |
7c673cae FG |
1145 | void tick(); |
1146 | void tick_without_osd_lock(); | |
1147 | void _dispatch(Message *m); | |
1148 | void dispatch_op(OpRequestRef op); | |
1149 | ||
11fdf7f2 | 1150 | void check_osdmap_features(); |
7c673cae FG |
1151 | |
1152 | // asok | |
1153 | friend class OSDSocketHook; | |
1154 | class OSDSocketHook *asok_hook; | |
9f95a23c TL |
1155 | void asok_command( |
1156 | std::string_view prefix, | |
1157 | const cmdmap_t& cmdmap, | |
f67539c2 TL |
1158 | ceph::Formatter *f, |
1159 | const ceph::buffer::list& inbl, | |
1160 | std::function<void(int,const std::string&,ceph::buffer::list&)> on_finish); | |
7c673cae FG |
1161 | |
1162 | public: | |
7c673cae | 1163 | int get_nodeid() { return whoami; } |
f67539c2 | 1164 | |
7c673cae FG |
1165 | static ghobject_t get_osdmap_pobject_name(epoch_t epoch) { |
1166 | char foo[20]; | |
1167 | snprintf(foo, sizeof(foo), "osdmap.%d", epoch); | |
1168 | return ghobject_t(hobject_t(sobject_t(object_t(foo), 0))); | |
1169 | } | |
1170 | static ghobject_t get_inc_osdmap_pobject_name(epoch_t epoch) { | |
1171 | char foo[22]; | |
1172 | snprintf(foo, sizeof(foo), "inc_osdmap.%d", epoch); | |
1173 | return ghobject_t(hobject_t(sobject_t(object_t(foo), 0))); | |
1174 | } | |
1175 | ||
1176 | static ghobject_t make_snapmapper_oid() { | |
1177 | return ghobject_t(hobject_t( | |
1178 | sobject_t( | |
1179 | object_t("snapmapper"), | |
1180 | 0))); | |
1181 | } | |
9f95a23c TL |
1182 | static ghobject_t make_purged_snaps_oid() { |
1183 | return ghobject_t(hobject_t( | |
1184 | sobject_t( | |
1185 | object_t("purged_snaps"), | |
1186 | 0))); | |
1187 | } | |
7c673cae FG |
1188 | |
1189 | static ghobject_t make_pg_log_oid(spg_t pg) { | |
f67539c2 | 1190 | std::stringstream ss; |
7c673cae | 1191 | ss << "pglog_" << pg; |
f67539c2 | 1192 | std::string s; |
7c673cae FG |
1193 | getline(ss, s); |
1194 | return ghobject_t(hobject_t(sobject_t(object_t(s.c_str()), 0))); | |
1195 | } | |
f67539c2 | 1196 | |
7c673cae | 1197 | static ghobject_t make_pg_biginfo_oid(spg_t pg) { |
f67539c2 | 1198 | std::stringstream ss; |
7c673cae | 1199 | ss << "pginfo_" << pg; |
f67539c2 | 1200 | std::string s; |
7c673cae FG |
1201 | getline(ss, s); |
1202 | return ghobject_t(hobject_t(sobject_t(object_t(s.c_str()), 0))); | |
1203 | } | |
1204 | static ghobject_t make_infos_oid() { | |
1205 | hobject_t oid(sobject_t("infos", CEPH_NOSNAP)); | |
1206 | return ghobject_t(oid); | |
1207 | } | |
11fdf7f2 TL |
1208 | |
1209 | static ghobject_t make_final_pool_info_oid(int64_t pool) { | |
1210 | return ghobject_t( | |
1211 | hobject_t( | |
1212 | sobject_t( | |
f67539c2 | 1213 | object_t(std::string("final_pool_") + stringify(pool)), |
11fdf7f2 TL |
1214 | CEPH_NOSNAP))); |
1215 | } | |
1216 | ||
1217 | static ghobject_t make_pg_num_history_oid() { | |
1218 | return ghobject_t(hobject_t(sobject_t("pg_num_history", CEPH_NOSNAP))); | |
1219 | } | |
1220 | ||
7c673cae FG |
1221 | static void recursive_remove_collection(CephContext* cct, |
1222 | ObjectStore *store, | |
1223 | spg_t pgid, | |
1224 | coll_t tmp); | |
1225 | ||
1226 | /** | |
1227 | * get_osd_initial_compat_set() | |
1228 | * | |
f67539c2 | 1229 | * Get the initial feature std::set for this OSD. Features |
7c673cae FG |
1230 | * here are automatically upgraded. |
1231 | * | |
1232 | * Return value: Initial osd CompatSet | |
1233 | */ | |
1234 | static CompatSet get_osd_initial_compat_set(); | |
1235 | ||
1236 | /** | |
1237 | * get_osd_compat_set() | |
1238 | * | |
1239 | * Get all features supported by this OSD | |
1240 | * | |
1241 | * Return value: CompatSet of all supported features | |
1242 | */ | |
1243 | static CompatSet get_osd_compat_set(); | |
f67539c2 | 1244 | |
7c673cae FG |
1245 | |
1246 | private: | |
1247 | class C_Tick; | |
1248 | class C_Tick_WithoutOSDLock; | |
1249 | ||
11fdf7f2 TL |
1250 | // -- config settings -- |
1251 | float m_osd_pg_epoch_max_lag_factor; | |
1252 | ||
7c673cae FG |
1253 | // -- superblock -- |
1254 | OSDSuperblock superblock; | |
1255 | ||
1256 | void write_superblock(); | |
1257 | void write_superblock(ObjectStore::Transaction& t); | |
1258 | int read_superblock(); | |
1259 | ||
1260 | void clear_temp_objects(); | |
1261 | ||
1262 | CompatSet osd_compat; | |
1263 | ||
1264 | // -- state -- | |
1265 | public: | |
1266 | typedef enum { | |
1267 | STATE_INITIALIZING = 1, | |
1268 | STATE_PREBOOT, | |
1269 | STATE_BOOTING, | |
1270 | STATE_ACTIVE, | |
1271 | STATE_STOPPING, | |
1272 | STATE_WAITING_FOR_HEALTHY | |
1273 | } osd_state_t; | |
1274 | ||
1275 | static const char *get_state_name(int s) { | |
1276 | switch (s) { | |
1277 | case STATE_INITIALIZING: return "initializing"; | |
1278 | case STATE_PREBOOT: return "preboot"; | |
1279 | case STATE_BOOTING: return "booting"; | |
1280 | case STATE_ACTIVE: return "active"; | |
1281 | case STATE_STOPPING: return "stopping"; | |
1282 | case STATE_WAITING_FOR_HEALTHY: return "waiting_for_healthy"; | |
1283 | default: return "???"; | |
1284 | } | |
1285 | } | |
1286 | ||
1287 | private: | |
11fdf7f2 | 1288 | std::atomic<int> state{STATE_INITIALIZING}; |
7c673cae FG |
1289 | |
1290 | public: | |
1291 | int get_state() const { | |
1292 | return state; | |
1293 | } | |
1294 | void set_state(int s) { | |
1295 | state = s; | |
1296 | } | |
1297 | bool is_initializing() const { | |
1298 | return state == STATE_INITIALIZING; | |
1299 | } | |
1300 | bool is_preboot() const { | |
1301 | return state == STATE_PREBOOT; | |
1302 | } | |
1303 | bool is_booting() const { | |
1304 | return state == STATE_BOOTING; | |
1305 | } | |
1306 | bool is_active() const { | |
1307 | return state == STATE_ACTIVE; | |
1308 | } | |
1309 | bool is_stopping() const { | |
1310 | return state == STATE_STOPPING; | |
1311 | } | |
1312 | bool is_waiting_for_healthy() const { | |
1313 | return state == STATE_WAITING_FOR_HEALTHY; | |
1314 | } | |
1315 | ||
1316 | private: | |
1317 | ||
7c673cae | 1318 | ShardedThreadPool osd_op_tp; |
7c673cae | 1319 | |
7c673cae FG |
1320 | void get_latest_osdmap(); |
1321 | ||
1322 | // -- sessions -- | |
1323 | private: | |
9f95a23c | 1324 | void dispatch_session_waiting(const ceph::ref_t<Session>& session, OSDMapRef osdmap); |
7c673cae | 1325 | |
9f95a23c | 1326 | ceph::mutex session_waiting_lock = ceph::make_mutex("OSD::session_waiting_lock"); |
f67539c2 | 1327 | std::set<ceph::ref_t<Session>> session_waiting_for_map; |
7c673cae FG |
1328 | |
1329 | /// Caller assumes refs for included Sessions | |
f67539c2 | 1330 | void get_sessions_waiting_for_map(std::set<ceph::ref_t<Session>> *out) { |
11fdf7f2 | 1331 | std::lock_guard l(session_waiting_lock); |
7c673cae FG |
1332 | out->swap(session_waiting_for_map); |
1333 | } | |
9f95a23c | 1334 | void register_session_waiting_on_map(const ceph::ref_t<Session>& session) { |
11fdf7f2 TL |
1335 | std::lock_guard l(session_waiting_lock); |
1336 | session_waiting_for_map.insert(session); | |
7c673cae | 1337 | } |
9f95a23c | 1338 | void clear_session_waiting_on_map(const ceph::ref_t<Session>& session) { |
11fdf7f2 TL |
1339 | std::lock_guard l(session_waiting_lock); |
1340 | session_waiting_for_map.erase(session); | |
7c673cae FG |
1341 | } |
1342 | void dispatch_sessions_waiting_on_map() { | |
f67539c2 | 1343 | std::set<ceph::ref_t<Session>> sessions_to_check; |
7c673cae | 1344 | get_sessions_waiting_for_map(&sessions_to_check); |
11fdf7f2 | 1345 | for (auto i = sessions_to_check.begin(); |
7c673cae FG |
1346 | i != sessions_to_check.end(); |
1347 | sessions_to_check.erase(i++)) { | |
11fdf7f2 | 1348 | std::lock_guard l{(*i)->session_dispatch_lock}; |
9f95a23c | 1349 | dispatch_session_waiting(*i, get_osdmap()); |
7c673cae FG |
1350 | } |
1351 | } | |
9f95a23c | 1352 | void session_handle_reset(const ceph::ref_t<Session>& session) { |
11fdf7f2 | 1353 | std::lock_guard l(session->session_dispatch_lock); |
7c673cae FG |
1354 | clear_session_waiting_on_map(session); |
1355 | ||
1356 | session->clear_backoffs(); | |
1357 | ||
1358 | /* Messages have connection refs, we need to clear the | |
1359 | * connection->session->message->connection | |
1360 | * cycles which result. | |
1361 | * Bug #12338 | |
1362 | */ | |
1363 | session->waiting_on_map.clear_and_dispose(TrackedOp::Putter()); | |
1364 | } | |
1365 | ||
1366 | private: | |
1367 | /** | |
1368 | * @defgroup monc helpers | |
1369 | * @{ | |
1370 | * Right now we only have the one | |
1371 | */ | |
1372 | ||
1373 | /** | |
1374 | * Ask the Monitors for a sequence of OSDMaps. | |
1375 | * | |
1376 | * @param epoch The epoch to start with when replying | |
1377 | * @param force_request True if this request forces a new subscription to | |
1378 | * the monitors; false if an outstanding request that encompasses it is | |
1379 | * sufficient. | |
1380 | */ | |
1381 | void osdmap_subscribe(version_t epoch, bool force_request); | |
1382 | /** @} monc helpers */ | |
1383 | ||
9f95a23c | 1384 | ceph::mutex osdmap_subscribe_lock = ceph::make_mutex("OSD::osdmap_subscribe_lock"); |
181888fb FG |
1385 | epoch_t latest_subscribed_epoch{0}; |
1386 | ||
7c673cae FG |
1387 | // -- heartbeat -- |
1388 | /// information about a heartbeat peer | |
1389 | struct HeartbeatInfo { | |
1390 | int peer; ///< peer | |
1391 | ConnectionRef con_front; ///< peer connection (front) | |
1392 | ConnectionRef con_back; ///< peer connection (back) | |
1393 | utime_t first_tx; ///< time we sent our first ping request | |
1394 | utime_t last_tx; ///< last time we sent a ping request | |
1395 | utime_t last_rx_front; ///< last time we got a ping reply on the front side | |
1396 | utime_t last_rx_back; ///< last time we got a ping reply on the back side | |
1397 | epoch_t epoch; ///< most recent epoch we wanted this peer | |
11fdf7f2 TL |
1398 | /// number of connections we send and receive heartbeat pings/replies |
1399 | static constexpr int HEARTBEAT_MAX_CONN = 2; | |
1400 | /// history of inflight pings, arranging by timestamp we sent | |
1401 | /// send time -> deadline -> remaining replies | |
f67539c2 | 1402 | std::map<utime_t, std::pair<utime_t, int>> ping_history; |
11fdf7f2 | 1403 | |
eafe8130 TL |
1404 | utime_t hb_interval_start; |
1405 | uint32_t hb_average_count = 0; | |
1406 | uint32_t hb_index = 0; | |
1407 | ||
1408 | uint32_t hb_total_back = 0; | |
1409 | uint32_t hb_min_back = UINT_MAX; | |
1410 | uint32_t hb_max_back = 0; | |
f67539c2 TL |
1411 | std::vector<uint32_t> hb_back_pingtime; |
1412 | std::vector<uint32_t> hb_back_min; | |
1413 | std::vector<uint32_t> hb_back_max; | |
eafe8130 TL |
1414 | |
1415 | uint32_t hb_total_front = 0; | |
1416 | uint32_t hb_min_front = UINT_MAX; | |
1417 | uint32_t hb_max_front = 0; | |
f67539c2 TL |
1418 | std::vector<uint32_t> hb_front_pingtime; |
1419 | std::vector<uint32_t> hb_front_min; | |
1420 | std::vector<uint32_t> hb_front_max; | |
eafe8130 | 1421 | |
b3b6e05e | 1422 | bool is_stale(utime_t stale) const { |
494da23a TL |
1423 | if (ping_history.empty()) { |
1424 | return false; | |
1425 | } | |
1426 | utime_t oldest_deadline = ping_history.begin()->second.first; | |
1427 | return oldest_deadline <= stale; | |
1428 | } | |
1429 | ||
b3b6e05e | 1430 | bool is_unhealthy(utime_t now) const { |
11fdf7f2 TL |
1431 | if (ping_history.empty()) { |
1432 | /// we haven't sent a ping yet or we have got all replies, | |
1433 | /// in either way we are safe and healthy for now | |
1434 | return false; | |
1435 | } | |
7c673cae | 1436 | |
11fdf7f2 TL |
1437 | utime_t oldest_deadline = ping_history.begin()->second.first; |
1438 | return now > oldest_deadline; | |
7c673cae FG |
1439 | } |
1440 | ||
b3b6e05e | 1441 | bool is_healthy(utime_t now) const { |
11fdf7f2 TL |
1442 | if (last_rx_front == utime_t() || last_rx_back == utime_t()) { |
1443 | // only declare to be healthy until we have received the first | |
1444 | // replies from both front/back connections | |
1445 | return false; | |
1446 | } | |
1447 | return !is_unhealthy(now); | |
1448 | } | |
9f95a23c TL |
1449 | |
1450 | void clear_mark_down(Connection *except = nullptr) { | |
1451 | if (con_back && con_back != except) { | |
1452 | con_back->mark_down(); | |
1453 | con_back->clear_priv(); | |
1454 | con_back.reset(nullptr); | |
1455 | } | |
1456 | if (con_front && con_front != except) { | |
1457 | con_front->mark_down(); | |
1458 | con_front->clear_priv(); | |
1459 | con_front.reset(nullptr); | |
1460 | } | |
1461 | } | |
7c673cae | 1462 | }; |
9f95a23c TL |
1463 | |
1464 | ceph::mutex heartbeat_lock = ceph::make_mutex("OSD::heartbeat_lock"); | |
f67539c2 | 1465 | std::map<int, int> debug_heartbeat_drops_remaining; |
9f95a23c | 1466 | ceph::condition_variable heartbeat_cond; |
7c673cae | 1467 | bool heartbeat_stop; |
f67539c2 TL |
1468 | std::atomic<bool> heartbeat_need_update; |
1469 | std::map<int,HeartbeatInfo> heartbeat_peers; ///< map of osd id to HeartbeatInfo | |
7c673cae FG |
1470 | utime_t last_mon_heartbeat; |
1471 | Messenger *hb_front_client_messenger; | |
1472 | Messenger *hb_back_client_messenger; | |
1473 | Messenger *hb_front_server_messenger; | |
1474 | Messenger *hb_back_server_messenger; | |
1475 | utime_t last_heartbeat_resample; ///< last time we chose random peers in waiting-for-healthy state | |
1476 | double daily_loadavg; | |
9f95a23c | 1477 | ceph::mono_time startup_time; |
eafe8130 TL |
1478 | |
1479 | // Track ping repsonse times using vector as a circular buffer | |
1480 | // MUST BE A POWER OF 2 | |
1481 | const uint32_t hb_vector_size = 16; | |
1482 | ||
7c673cae FG |
1483 | void _add_heartbeat_peer(int p); |
1484 | void _remove_heartbeat_peer(int p); | |
1485 | bool heartbeat_reset(Connection *con); | |
1486 | void maybe_update_heartbeat_peers(); | |
494da23a | 1487 | void reset_heartbeat_peers(bool all); |
7c673cae FG |
1488 | bool heartbeat_peers_need_update() { |
1489 | return heartbeat_need_update.load(); | |
1490 | } | |
1491 | void heartbeat_set_peers_need_update() { | |
1492 | heartbeat_need_update.store(true); | |
1493 | } | |
1494 | void heartbeat_clear_peers_need_update() { | |
1495 | heartbeat_need_update.store(false); | |
1496 | } | |
1497 | void heartbeat(); | |
1498 | void heartbeat_check(); | |
1499 | void heartbeat_entry(); | |
1500 | void need_heartbeat_peer_update(); | |
1501 | ||
1502 | void heartbeat_kick() { | |
11fdf7f2 | 1503 | std::lock_guard l(heartbeat_lock); |
9f95a23c | 1504 | heartbeat_cond.notify_all(); |
7c673cae FG |
1505 | } |
1506 | ||
1507 | struct T_Heartbeat : public Thread { | |
1508 | OSD *osd; | |
1509 | explicit T_Heartbeat(OSD *o) : osd(o) {} | |
1510 | void *entry() override { | |
1511 | osd->heartbeat_entry(); | |
1512 | return 0; | |
1513 | } | |
1514 | } heartbeat_thread; | |
1515 | ||
1516 | public: | |
1517 | bool heartbeat_dispatch(Message *m); | |
1518 | ||
1519 | struct HeartbeatDispatcher : public Dispatcher { | |
1520 | OSD *osd; | |
1521 | explicit HeartbeatDispatcher(OSD *o) : Dispatcher(o->cct), osd(o) {} | |
1522 | ||
1523 | bool ms_can_fast_dispatch_any() const override { return true; } | |
1524 | bool ms_can_fast_dispatch(const Message *m) const override { | |
1525 | switch (m->get_type()) { | |
11fdf7f2 TL |
1526 | case CEPH_MSG_PING: |
1527 | case MSG_OSD_PING: | |
1528 | return true; | |
1529 | default: | |
1530 | return false; | |
1531 | } | |
7c673cae FG |
1532 | } |
1533 | void ms_fast_dispatch(Message *m) override { | |
1534 | osd->heartbeat_dispatch(m); | |
1535 | } | |
1536 | bool ms_dispatch(Message *m) override { | |
1537 | return osd->heartbeat_dispatch(m); | |
1538 | } | |
1539 | bool ms_handle_reset(Connection *con) override { | |
1540 | return osd->heartbeat_reset(con); | |
1541 | } | |
1542 | void ms_handle_remote_reset(Connection *con) override {} | |
1543 | bool ms_handle_refused(Connection *con) override { | |
1544 | return osd->ms_handle_refused(con); | |
1545 | } | |
11fdf7f2 | 1546 | int ms_handle_authentication(Connection *con) override { |
7c673cae FG |
1547 | return true; |
1548 | } | |
1549 | } heartbeat_dispatcher; | |
1550 | ||
1551 | private: | |
1552 | // -- waiters -- | |
f67539c2 TL |
1553 | std::list<OpRequestRef> finished; |
1554 | ||
1555 | void take_waiters(std::list<OpRequestRef>& ls) { | |
9f95a23c | 1556 | ceph_assert(ceph_mutex_is_locked(osd_lock)); |
7c673cae FG |
1557 | finished.splice(finished.end(), ls); |
1558 | } | |
1559 | void do_waiters(); | |
f67539c2 | 1560 | |
7c673cae FG |
1561 | // -- op tracking -- |
1562 | OpTracker op_tracker; | |
f67539c2 | 1563 | void test_ops(std::string command, std::string args, std::ostream& ss); |
7c673cae FG |
1564 | friend class TestOpsSocketHook; |
1565 | TestOpsSocketHook *test_ops_hook; | |
11fdf7f2 | 1566 | friend struct C_FinishSplits; |
7c673cae FG |
1567 | friend struct C_OpenPGs; |
1568 | ||
11fdf7f2 | 1569 | protected: |
7c673cae FG |
1570 | |
1571 | /* | |
1572 | * The ordered op delivery chain is: | |
1573 | * | |
9f95a23c TL |
1574 | * fast dispatch -> scheduler back |
1575 | * scheduler front <-> to_process back | |
7c673cae FG |
1576 | * to_process front -> RunVis(item) |
1577 | * <- queue_front() | |
1578 | * | |
9f95a23c TL |
1579 | * The scheduler is per-shard, and to_process is per pg_slot. Items can be |
1580 | * pushed back up into to_process and/or scheduler while order is preserved. | |
7c673cae FG |
1581 | * |
1582 | * Multiple worker threads can operate on each shard. | |
1583 | * | |
11fdf7f2 | 1584 | * Under normal circumstances, num_running == to_process.size(). There are |
7c673cae FG |
1585 | * two times when that is not true: (1) when waiting_for_pg == true and |
1586 | * to_process is accumulating requests that are waiting for the pg to be | |
1587 | * instantiated; in that case they will all get requeued together by | |
1588 | * wake_pg_waiters, and (2) when wake_pg_waiters just ran, waiting_for_pg | |
1589 | * and already requeued the items. | |
1590 | */ | |
9f95a23c TL |
1591 | friend class ceph::osd::scheduler::PGOpItem; |
1592 | friend class ceph::osd::scheduler::PGPeeringItem; | |
1593 | friend class ceph::osd::scheduler::PGRecovery; | |
f67539c2 | 1594 | friend class ceph::osd::scheduler::PGRecoveryMsg; |
9f95a23c | 1595 | friend class ceph::osd::scheduler::PGDelete; |
224ce89b | 1596 | |
7c673cae | 1597 | class ShardedOpWQ |
9f95a23c | 1598 | : public ShardedThreadPool::ShardedWQ<OpSchedulerItem> |
7c673cae | 1599 | { |
7c673cae | 1600 | OSD *osd; |
7c673cae FG |
1601 | |
1602 | public: | |
11fdf7f2 | 1603 | ShardedOpWQ(OSD *o, |
f67539c2 TL |
1604 | ceph::timespan ti, |
1605 | ceph::timespan si, | |
7c673cae | 1606 | ShardedThreadPool* tp) |
9f95a23c | 1607 | : ShardedThreadPool::ShardedWQ<OpSchedulerItem>(ti, si, tp), |
11fdf7f2 | 1608 | osd(o) { |
7c673cae FG |
1609 | } |
1610 | ||
11fdf7f2 TL |
1611 | void _add_slot_waiter( |
1612 | spg_t token, | |
1613 | OSDShardPGSlot *slot, | |
9f95a23c | 1614 | OpSchedulerItem&& qi); |
7c673cae FG |
1615 | |
1616 | /// try to do some work | |
f67539c2 | 1617 | void _process(uint32_t thread_index, ceph::heartbeat_handle_d *hb) override; |
7c673cae FG |
1618 | |
1619 | /// enqueue a new item | |
9f95a23c | 1620 | void _enqueue(OpSchedulerItem&& item) override; |
7c673cae FG |
1621 | |
1622 | /// requeue an old item (at the front of the line) | |
9f95a23c | 1623 | void _enqueue_front(OpSchedulerItem&& item) override; |
f67539c2 | 1624 | |
7c673cae | 1625 | void return_waiting_threads() override { |
11fdf7f2 TL |
1626 | for(uint32_t i = 0; i < osd->num_shards; i++) { |
1627 | OSDShard* sdata = osd->shards[i]; | |
1628 | assert (NULL != sdata); | |
1629 | std::scoped_lock l{sdata->sdata_wait_lock}; | |
1630 | sdata->stop_waiting = true; | |
1631 | sdata->sdata_cond.notify_all(); | |
7c673cae FG |
1632 | } |
1633 | } | |
1634 | ||
11fdf7f2 TL |
1635 | void stop_return_waiting_threads() override { |
1636 | for(uint32_t i = 0; i < osd->num_shards; i++) { | |
1637 | OSDShard* sdata = osd->shards[i]; | |
7c673cae | 1638 | assert (NULL != sdata); |
11fdf7f2 TL |
1639 | std::scoped_lock l{sdata->sdata_wait_lock}; |
1640 | sdata->stop_waiting = false; | |
1641 | } | |
1642 | } | |
1643 | ||
f67539c2 | 1644 | void dump(ceph::Formatter *f) { |
11fdf7f2 TL |
1645 | for(uint32_t i = 0; i < osd->num_shards; i++) { |
1646 | auto &&sdata = osd->shards[i]; | |
1647 | ||
1648 | char queue_name[32] = {0}; | |
1649 | snprintf(queue_name, sizeof(queue_name), "%s%" PRIu32, "OSD:ShardedOpWQ:", i); | |
1650 | ceph_assert(NULL != sdata); | |
1651 | ||
1652 | std::scoped_lock l{sdata->shard_lock}; | |
1653 | f->open_object_section(queue_name); | |
9f95a23c | 1654 | sdata->scheduler->dump(*f); |
7c673cae | 1655 | f->close_section(); |
7c673cae FG |
1656 | } |
1657 | } | |
1658 | ||
11fdf7f2 TL |
1659 | bool is_shard_empty(uint32_t thread_index) override { |
1660 | uint32_t shard_index = thread_index % osd->num_shards; | |
1661 | auto &&sdata = osd->shards[shard_index]; | |
1662 | ceph_assert(sdata); | |
1663 | std::lock_guard l(sdata->shard_lock); | |
1664 | if (thread_index < osd->num_shards) { | |
9f95a23c | 1665 | return sdata->scheduler->empty() && sdata->context_queue.empty(); |
11fdf7f2 | 1666 | } else { |
9f95a23c | 1667 | return sdata->scheduler->empty(); |
7c673cae | 1668 | } |
11fdf7f2 | 1669 | } |
7c673cae | 1670 | |
f67539c2 | 1671 | void handle_oncommits(std::list<Context*>& oncommits) { |
11fdf7f2 TL |
1672 | for (auto p : oncommits) { |
1673 | p->complete(0); | |
1674 | } | |
7c673cae FG |
1675 | } |
1676 | } op_shardedwq; | |
1677 | ||
1678 | ||
11fdf7f2 | 1679 | void enqueue_op(spg_t pg, OpRequestRef&& op, epoch_t epoch); |
7c673cae FG |
1680 | void dequeue_op( |
1681 | PGRef pg, OpRequestRef op, | |
1682 | ThreadPool::TPHandle &handle); | |
1683 | ||
11fdf7f2 TL |
1684 | void enqueue_peering_evt( |
1685 | spg_t pgid, | |
1686 | PGPeeringEventRef ref); | |
11fdf7f2 TL |
1687 | void dequeue_peering_evt( |
1688 | OSDShard *sdata, | |
1689 | PG *pg, | |
1690 | PGPeeringEventRef ref, | |
1691 | ThreadPool::TPHandle& handle); | |
1692 | ||
1693 | void dequeue_delete( | |
1694 | OSDShard *sdata, | |
1695 | PG *pg, | |
1696 | epoch_t epoch, | |
1697 | ThreadPool::TPHandle& handle); | |
7c673cae FG |
1698 | |
1699 | friend class PG; | |
f67539c2 | 1700 | friend struct OSDShard; |
7c673cae | 1701 | friend class PrimaryLogPG; |
f67539c2 | 1702 | friend class PgScrubber; |
7c673cae FG |
1703 | |
1704 | ||
1705 | protected: | |
1706 | ||
1707 | // -- osd map -- | |
9f95a23c TL |
1708 | // TODO: switch to std::atomic<OSDMapRef> when C++20 will be available. |
1709 | OSDMapRef _osdmap; | |
1710 | void set_osdmap(OSDMapRef osdmap) { | |
1711 | std::atomic_store(&_osdmap, osdmap); | |
1712 | } | |
1713 | OSDMapRef get_osdmap() const { | |
1714 | return std::atomic_load(&_osdmap); | |
7c673cae | 1715 | } |
224ce89b | 1716 | epoch_t get_osdmap_epoch() const { |
9f95a23c TL |
1717 | // XXX: performance? |
1718 | auto osdmap = get_osdmap(); | |
7c673cae FG |
1719 | return osdmap ? osdmap->get_epoch() : 0; |
1720 | } | |
1721 | ||
11fdf7f2 TL |
1722 | pool_pg_num_history_t pg_num_history; |
1723 | ||
9f95a23c | 1724 | ceph::shared_mutex map_lock = ceph::make_shared_mutex("OSD::map_lock"); |
f67539c2 TL |
1725 | std::list<OpRequestRef> waiting_for_osdmap; |
1726 | std::deque<utime_t> osd_markdown_log; | |
7c673cae FG |
1727 | |
1728 | friend struct send_map_on_destruct; | |
1729 | ||
1730 | void wait_for_new_map(OpRequestRef op); | |
1731 | void handle_osd_map(class MOSDMap *m); | |
1732 | void _committed_osd_maps(epoch_t first, epoch_t last, class MOSDMap *m); | |
1733 | void trim_maps(epoch_t oldest, int nreceived, bool skip_maps); | |
1734 | void note_down_osd(int osd); | |
1735 | void note_up_osd(int osd); | |
f67539c2 | 1736 | friend struct C_OnMapCommit; |
7c673cae FG |
1737 | |
1738 | bool advance_pg( | |
11fdf7f2 TL |
1739 | epoch_t advance_to, |
1740 | PG *pg, | |
7c673cae | 1741 | ThreadPool::TPHandle &handle, |
9f95a23c | 1742 | PeeringCtx &rctx); |
7c673cae FG |
1743 | void consume_map(); |
1744 | void activate_map(); | |
1745 | ||
1746 | // osd map cache (past osd maps) | |
1747 | OSDMapRef get_map(epoch_t e) { | |
1748 | return service.get_map(e); | |
1749 | } | |
1750 | OSDMapRef add_map(OSDMap *o) { | |
1751 | return service.add_map(o); | |
1752 | } | |
f67539c2 | 1753 | bool get_map_bl(epoch_t e, ceph::buffer::list& bl) { |
7c673cae FG |
1754 | return service.get_map_bl(e, bl); |
1755 | } | |
11fdf7f2 TL |
1756 | |
1757 | public: | |
1758 | // -- shards -- | |
f67539c2 | 1759 | std::vector<OSDShard*> shards; |
11fdf7f2 TL |
1760 | uint32_t num_shards = 0; |
1761 | ||
1762 | void inc_num_pgs() { | |
1763 | ++num_pgs; | |
1764 | } | |
1765 | void dec_num_pgs() { | |
1766 | --num_pgs; | |
1767 | } | |
1768 | int get_num_pgs() const { | |
1769 | return num_pgs; | |
7c673cae FG |
1770 | } |
1771 | ||
1772 | protected: | |
9f95a23c | 1773 | ceph::mutex merge_lock = ceph::make_mutex("OSD::merge_lock"); |
11fdf7f2 | 1774 | /// merge epoch -> target pgid -> source pgid -> pg |
f67539c2 | 1775 | std::map<epoch_t,std::map<spg_t,std::map<spg_t,PGRef>>> merge_waiters; |
11fdf7f2 TL |
1776 | |
1777 | bool add_merge_waiter(OSDMapRef nextmap, spg_t target, PGRef source, | |
1778 | unsigned need); | |
1779 | ||
7c673cae | 1780 | // -- placement groups -- |
11fdf7f2 | 1781 | std::atomic<size_t> num_pgs = {0}; |
7c673cae | 1782 | |
3efd9988 | 1783 | std::mutex pending_creates_lock; |
9f95a23c | 1784 | using create_from_osd_t = std::pair<spg_t, bool /* is primary*/>; |
b32b8144 | 1785 | std::set<create_from_osd_t> pending_creates_from_osd; |
3efd9988 FG |
1786 | unsigned pending_creates_from_mon = 0; |
1787 | ||
7c673cae FG |
1788 | PGRecoveryStats pg_recovery_stats; |
1789 | ||
11fdf7f2 TL |
1790 | PGRef _lookup_pg(spg_t pgid); |
1791 | PGRef _lookup_lock_pg(spg_t pgid); | |
1792 | void register_pg(PGRef pg); | |
1793 | bool try_finish_pg_delete(PG *pg, unsigned old_pg_num); | |
7c673cae | 1794 | |
f67539c2 TL |
1795 | void _get_pgs(std::vector<PGRef> *v, bool clear_too=false); |
1796 | void _get_pgids(std::vector<spg_t> *v); | |
31f18b77 FG |
1797 | |
1798 | public: | |
11fdf7f2 | 1799 | PGRef lookup_lock_pg(spg_t pgid); |
31f18b77 | 1800 | |
11fdf7f2 | 1801 | std::set<int64_t> get_mapped_pools(); |
35e4c445 | 1802 | |
31f18b77 | 1803 | protected: |
7c673cae | 1804 | PG* _make_pg(OSDMapRef createmap, spg_t pgid); |
7c673cae | 1805 | |
11fdf7f2 TL |
1806 | bool maybe_wait_for_max_pg(const OSDMapRef& osdmap, |
1807 | spg_t pgid, bool is_mon_create); | |
3efd9988 FG |
1808 | void resume_creating_pg(); |
1809 | ||
7c673cae | 1810 | void load_pgs(); |
7c673cae FG |
1811 | |
1812 | /// build initial pg history and intervals on create | |
1813 | void build_initial_pg_history( | |
1814 | spg_t pgid, | |
1815 | epoch_t created, | |
1816 | utime_t created_stamp, | |
1817 | pg_history_t *h, | |
1818 | PastIntervals *pi); | |
1819 | ||
7c673cae FG |
1820 | epoch_t last_pg_create_epoch; |
1821 | ||
1822 | void handle_pg_create(OpRequestRef op); | |
1823 | ||
1824 | void split_pgs( | |
1825 | PG *parent, | |
f67539c2 | 1826 | const std::set<spg_t> &childpgids, std::set<PGRef> *out_pgs, |
7c673cae FG |
1827 | OSDMapRef curmap, |
1828 | OSDMapRef nextmap, | |
9f95a23c | 1829 | PeeringCtx &rctx); |
f67539c2 | 1830 | void _finish_splits(std::set<PGRef>& pgs); |
7c673cae FG |
1831 | |
1832 | // == monitor interaction == | |
9f95a23c | 1833 | ceph::mutex mon_report_lock = ceph::make_mutex("OSD::mon_report_lock"); |
7c673cae | 1834 | utime_t last_mon_report; |
11fdf7f2 | 1835 | Finisher boot_finisher; |
7c673cae FG |
1836 | |
1837 | // -- boot -- | |
1838 | void start_boot(); | |
1839 | void _got_mon_epochs(epoch_t oldest, epoch_t newest); | |
1840 | void _preboot(epoch_t oldest, epoch_t newest); | |
1841 | void _send_boot(); | |
f67539c2 | 1842 | void _collect_metadata(std::map<std::string,std::string> *pmeta); |
9f95a23c TL |
1843 | void _get_purged_snaps(); |
1844 | void handle_get_purged_snaps_reply(MMonGetPurgedSnapsReply *r); | |
7c673cae FG |
1845 | |
1846 | void start_waiting_for_healthy(); | |
1847 | bool _is_healthy(); | |
1848 | ||
1849 | void send_full_update(); | |
f67539c2 TL |
1850 | |
1851 | friend struct CB_OSD_GetVersion; | |
7c673cae FG |
1852 | |
1853 | // -- alive -- | |
1854 | epoch_t up_thru_wanted; | |
1855 | ||
1856 | void queue_want_up_thru(epoch_t want); | |
1857 | void send_alive(); | |
1858 | ||
1859 | // -- full map requests -- | |
1860 | epoch_t requested_full_first, requested_full_last; | |
1861 | ||
1862 | void request_full_map(epoch_t first, epoch_t last); | |
1863 | void rerequest_full_maps() { | |
1864 | epoch_t first = requested_full_first; | |
1865 | epoch_t last = requested_full_last; | |
1866 | requested_full_first = 0; | |
1867 | requested_full_last = 0; | |
1868 | request_full_map(first, last); | |
1869 | } | |
1870 | void got_full_map(epoch_t e); | |
1871 | ||
1872 | // -- failures -- | |
f67539c2 TL |
1873 | std::map<int,utime_t> failure_queue; |
1874 | std::map<int,std::pair<utime_t,entity_addrvec_t> > failure_pending; | |
7c673cae FG |
1875 | |
1876 | void requeue_failures(); | |
1877 | void send_failures(); | |
11fdf7f2 TL |
1878 | void send_still_alive(epoch_t epoch, int osd, const entity_addrvec_t &addrs); |
1879 | void cancel_pending_failures(); | |
7c673cae FG |
1880 | |
1881 | ceph::coarse_mono_clock::time_point last_sent_beacon; | |
9f95a23c | 1882 | ceph::mutex min_last_epoch_clean_lock = ceph::make_mutex("OSD::min_last_epoch_clean_lock"); |
7c673cae FG |
1883 | epoch_t min_last_epoch_clean = 0; |
1884 | // which pgs were scanned for min_lec | |
1885 | std::vector<pg_t> min_last_epoch_clean_pgs; | |
1886 | void send_beacon(const ceph::coarse_mono_clock::time_point& now); | |
1887 | ||
7c673cae FG |
1888 | ceph_tid_t get_tid() { |
1889 | return service.get_tid(); | |
1890 | } | |
1891 | ||
9f95a23c TL |
1892 | double scrub_sleep_time(bool must_scrub); |
1893 | ||
7c673cae | 1894 | // -- generic pg peering -- |
9f95a23c TL |
1895 | PeeringCtx create_context(); |
1896 | void dispatch_context(PeeringCtx &ctx, PG *pg, OSDMapRef curmap, | |
7c673cae | 1897 | ThreadPool::TPHandle *handle = NULL); |
7c673cae FG |
1898 | |
1899 | bool require_mon_peer(const Message *m); | |
1900 | bool require_mon_or_mgr_peer(const Message *m); | |
1901 | bool require_osd_peer(const Message *m); | |
1902 | /*** | |
1903 | * Verifies that we were alive in the given epoch, and that | |
1904 | * still are. | |
1905 | */ | |
1906 | bool require_self_aliveness(const Message *m, epoch_t alive_since); | |
1907 | /** | |
1908 | * Verifies that the OSD who sent the given op has the same | |
f67539c2 | 1909 | * address as in the given std::map. |
7c673cae FG |
1910 | * @pre op was sent by an OSD using the cluster messenger |
1911 | */ | |
9f95a23c | 1912 | bool require_same_peer_instance(const Message *m, const OSDMapRef& map, |
7c673cae FG |
1913 | bool is_fast_dispatch); |
1914 | ||
1915 | bool require_same_or_newer_map(OpRequestRef& op, epoch_t e, | |
1916 | bool is_fast_dispatch); | |
1917 | ||
11fdf7f2 TL |
1918 | void handle_fast_pg_create(MOSDPGCreate2 *m); |
1919 | void handle_fast_pg_query(MOSDPGQuery *m); | |
1920 | void handle_pg_query_nopg(const MQuery& q); | |
1921 | void handle_fast_pg_notify(MOSDPGNotify *m); | |
1922 | void handle_pg_notify_nopg(const MNotifyRec& q); | |
1923 | void handle_fast_pg_info(MOSDPGInfo *m); | |
1924 | void handle_fast_pg_remove(MOSDPGRemove *m); | |
7c673cae | 1925 | |
11fdf7f2 TL |
1926 | public: |
1927 | // used by OSDShard | |
1928 | PGRef handle_pg_create_info(const OSDMapRef& osdmap, const PGCreateInfo *info); | |
1929 | protected: | |
c07f9fc5 | 1930 | |
11fdf7f2 | 1931 | void handle_fast_force_recovery(MOSDForceRecovery *m); |
7c673cae FG |
1932 | |
1933 | // -- commands -- | |
7c673cae | 1934 | void handle_command(class MCommand *m); |
11fdf7f2 | 1935 | |
7c673cae FG |
1936 | |
1937 | // -- pg recovery -- | |
1938 | void do_recovery(PG *pg, epoch_t epoch_queued, uint64_t pushes_reserved, | |
1939 | ThreadPool::TPHandle &handle); | |
1940 | ||
1941 | ||
1942 | // -- scrubbing -- | |
1943 | void sched_scrub(); | |
494da23a | 1944 | void resched_all_scrubs(); |
7c673cae FG |
1945 | bool scrub_random_backoff(); |
1946 | bool scrub_load_below_threshold(); | |
1947 | bool scrub_time_permit(utime_t now); | |
1948 | ||
b32b8144 FG |
1949 | // -- status reporting -- |
1950 | MPGStats *collect_pg_stats(); | |
11fdf7f2 TL |
1951 | std::vector<DaemonHealthMetric> get_health_metrics(); |
1952 | ||
b32b8144 | 1953 | |
224ce89b | 1954 | private: |
7c673cae FG |
1955 | bool ms_can_fast_dispatch_any() const override { return true; } |
1956 | bool ms_can_fast_dispatch(const Message *m) const override { | |
1957 | switch (m->get_type()) { | |
11fdf7f2 | 1958 | case CEPH_MSG_PING: |
7c673cae FG |
1959 | case CEPH_MSG_OSD_OP: |
1960 | case CEPH_MSG_OSD_BACKOFF: | |
11fdf7f2 TL |
1961 | case MSG_OSD_SCRUB2: |
1962 | case MSG_OSD_FORCE_RECOVERY: | |
1963 | case MSG_MON_COMMAND: | |
1964 | case MSG_OSD_PG_CREATE2: | |
1965 | case MSG_OSD_PG_QUERY: | |
9f95a23c | 1966 | case MSG_OSD_PG_QUERY2: |
11fdf7f2 | 1967 | case MSG_OSD_PG_INFO: |
9f95a23c | 1968 | case MSG_OSD_PG_INFO2: |
11fdf7f2 | 1969 | case MSG_OSD_PG_NOTIFY: |
9f95a23c | 1970 | case MSG_OSD_PG_NOTIFY2: |
11fdf7f2 TL |
1971 | case MSG_OSD_PG_LOG: |
1972 | case MSG_OSD_PG_TRIM: | |
1973 | case MSG_OSD_PG_REMOVE: | |
1974 | case MSG_OSD_BACKFILL_RESERVE: | |
1975 | case MSG_OSD_RECOVERY_RESERVE: | |
7c673cae | 1976 | case MSG_OSD_REPOP: |
7c673cae FG |
1977 | case MSG_OSD_REPOPREPLY: |
1978 | case MSG_OSD_PG_PUSH: | |
1979 | case MSG_OSD_PG_PULL: | |
1980 | case MSG_OSD_PG_PUSH_REPLY: | |
1981 | case MSG_OSD_PG_SCAN: | |
1982 | case MSG_OSD_PG_BACKFILL: | |
1983 | case MSG_OSD_PG_BACKFILL_REMOVE: | |
1984 | case MSG_OSD_EC_WRITE: | |
1985 | case MSG_OSD_EC_WRITE_REPLY: | |
1986 | case MSG_OSD_EC_READ: | |
1987 | case MSG_OSD_EC_READ_REPLY: | |
1988 | case MSG_OSD_SCRUB_RESERVE: | |
1989 | case MSG_OSD_REP_SCRUB: | |
1990 | case MSG_OSD_REP_SCRUBMAP: | |
1991 | case MSG_OSD_PG_UPDATE_LOG_MISSING: | |
1992 | case MSG_OSD_PG_UPDATE_LOG_MISSING_REPLY: | |
c07f9fc5 FG |
1993 | case MSG_OSD_PG_RECOVERY_DELETE: |
1994 | case MSG_OSD_PG_RECOVERY_DELETE_REPLY: | |
9f95a23c TL |
1995 | case MSG_OSD_PG_LEASE: |
1996 | case MSG_OSD_PG_LEASE_ACK: | |
7c673cae FG |
1997 | return true; |
1998 | default: | |
1999 | return false; | |
2000 | } | |
2001 | } | |
2002 | void ms_fast_dispatch(Message *m) override; | |
7c673cae | 2003 | bool ms_dispatch(Message *m) override; |
7c673cae FG |
2004 | void ms_handle_connect(Connection *con) override; |
2005 | void ms_handle_fast_connect(Connection *con) override; | |
2006 | void ms_handle_fast_accept(Connection *con) override; | |
11fdf7f2 | 2007 | int ms_handle_authentication(Connection *con) override; |
7c673cae FG |
2008 | bool ms_handle_reset(Connection *con) override; |
2009 | void ms_handle_remote_reset(Connection *con) override {} | |
2010 | bool ms_handle_refused(Connection *con) override; | |
2011 | ||
7c673cae FG |
2012 | public: |
2013 | /* internal and external can point to the same messenger, they will still | |
2014 | * be cleaned up properly*/ | |
2015 | OSD(CephContext *cct_, | |
2016 | ObjectStore *store_, | |
2017 | int id, | |
2018 | Messenger *internal, | |
2019 | Messenger *external, | |
2020 | Messenger *hb_front_client, | |
2021 | Messenger *hb_back_client, | |
2022 | Messenger *hb_front_server, | |
2023 | Messenger *hb_back_server, | |
2024 | Messenger *osdc_messenger, | |
f67539c2 TL |
2025 | MonClient *mc, const std::string &dev, const std::string &jdev, |
2026 | ceph::async::io_context_pool& poolctx); | |
7c673cae FG |
2027 | ~OSD() override; |
2028 | ||
2029 | // static bits | |
e306af50 | 2030 | static int mkfs(CephContext *cct, ObjectStore *store, uuid_d fsid, int whoami, std::string osdspec_affinity); |
11fdf7f2 | 2031 | |
f67539c2 TL |
2032 | /* remove any non-user xattrs from a std::map of them */ |
2033 | void filter_xattrs(std::map<std::string, ceph::buffer::ptr>& attrs) { | |
2034 | for (std::map<std::string, ceph::buffer::ptr>::iterator iter = attrs.begin(); | |
7c673cae FG |
2035 | iter != attrs.end(); |
2036 | ) { | |
2037 | if (('_' != iter->first.at(0)) || (iter->first.size() == 1)) | |
2038 | attrs.erase(iter++); | |
2039 | else ++iter; | |
2040 | } | |
2041 | } | |
2042 | ||
2043 | private: | |
f67539c2 | 2044 | int mon_cmd_maybe_osd_create(std::string &cmd); |
7c673cae FG |
2045 | int update_crush_device_class(); |
2046 | int update_crush_location(); | |
2047 | ||
3efd9988 FG |
2048 | static int write_meta(CephContext *cct, |
2049 | ObjectStore *store, | |
e306af50 | 2050 | uuid_d& cluster_fsid, uuid_d& osd_fsid, int whoami, std::string& osdspec_affinity); |
7c673cae | 2051 | |
f67539c2 TL |
2052 | void handle_scrub(class MOSDScrub *m); |
2053 | void handle_fast_scrub(class MOSDScrub2 *m); | |
7c673cae FG |
2054 | void handle_osd_ping(class MOSDPing *m); |
2055 | ||
9f95a23c | 2056 | size_t get_num_cache_shards(); |
31f18b77 FG |
2057 | int get_num_op_shards(); |
2058 | int get_num_op_threads(); | |
2059 | ||
c07f9fc5 | 2060 | float get_osd_recovery_sleep(); |
11fdf7f2 | 2061 | float get_osd_delete_sleep(); |
494da23a | 2062 | float get_osd_snap_trim_sleep(); |
11fdf7f2 | 2063 | |
9f95a23c | 2064 | int get_recovery_max_active(); |
b3b6e05e | 2065 | bool maybe_override_options_for_qos(); |
9f95a23c TL |
2066 | |
2067 | void scrub_purged_snaps(); | |
f67539c2 | 2068 | void probe_smart(const std::string& devid, std::ostream& ss); |
c07f9fc5 | 2069 | |
7c673cae | 2070 | public: |
11fdf7f2 | 2071 | static int peek_meta(ObjectStore *store, |
f67539c2 | 2072 | std::string *magic, |
11fdf7f2 TL |
2073 | uuid_d *cluster_fsid, |
2074 | uuid_d *osd_fsid, | |
2075 | int *whoami, | |
9f95a23c | 2076 | ceph_release_t *min_osd_release); |
f67539c2 | 2077 | |
7c673cae FG |
2078 | |
2079 | // startup/shutdown | |
2080 | int pre_init(); | |
2081 | int init(); | |
2082 | void final_init(); | |
2083 | ||
2084 | int enable_disable_fuse(bool stop); | |
11fdf7f2 | 2085 | int set_numa_affinity(); |
7c673cae FG |
2086 | |
2087 | void suicide(int exitcode); | |
2088 | int shutdown(); | |
2089 | ||
2090 | void handle_signal(int signum); | |
2091 | ||
2092 | /// check if we can throw out op from a disconnected client | |
2093 | static bool op_is_discardable(const MOSDOp *m); | |
2094 | ||
2095 | public: | |
2096 | OSDService service; | |
2097 | friend class OSDService; | |
11fdf7f2 TL |
2098 | |
2099 | private: | |
9f95a23c TL |
2100 | void set_perf_queries(const ConfigPayload &config_payload); |
2101 | MetricPayload get_perf_reports(); | |
11fdf7f2 | 2102 | |
9f95a23c | 2103 | ceph::mutex m_perf_queries_lock = ceph::make_mutex("OSD::m_perf_queries_lock"); |
11fdf7f2 TL |
2104 | std::list<OSDPerfMetricQuery> m_perf_queries; |
2105 | std::map<OSDPerfMetricQuery, OSDPerfMetricLimits> m_perf_limits; | |
7c673cae FG |
2106 | }; |
2107 | ||
224ce89b | 2108 | |
7c673cae FG |
2109 | //compatibility of the executable |
2110 | extern const CompatSet::Feature ceph_osd_feature_compat[]; | |
2111 | extern const CompatSet::Feature ceph_osd_feature_ro_compat[]; | |
2112 | extern const CompatSet::Feature ceph_osd_feature_incompat[]; | |
2113 | ||
224ce89b | 2114 | #endif // CEPH_OSD_H |