]>
Commit | Line | Data |
---|---|---|
f67539c2 TL |
1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- |
2 | // vim: ts=2 sw=2 smarttab | |
3 | ||
20effc67 | 4 | #include "./pg_scrubber.h" // the '.' notation used to affect clang-format order |
f67539c2 | 5 | |
20effc67 | 6 | #include <cmath> |
f67539c2 TL |
7 | #include <iostream> |
8 | #include <vector> | |
9 | ||
10 | #include "debug.h" | |
11 | ||
12 | #include "common/errno.h" | |
13 | #include "messages/MOSDOp.h" | |
14 | #include "messages/MOSDRepScrub.h" | |
15 | #include "messages/MOSDRepScrubMap.h" | |
16 | #include "messages/MOSDScrub.h" | |
17 | #include "messages/MOSDScrubReserve.h" | |
18 | ||
20effc67 TL |
19 | #include "osd/OSD.h" |
20 | #include "osd/osd_types_fmt.h" | |
f67539c2 TL |
21 | #include "ScrubStore.h" |
22 | #include "scrub_machine.h" | |
23 | ||
20effc67 TL |
24 | using std::list; |
25 | using std::map; | |
26 | using std::pair; | |
27 | using std::set; | |
28 | using std::stringstream; | |
29 | using std::vector; | |
f67539c2 TL |
30 | using namespace Scrub; |
31 | using namespace std::chrono; | |
32 | using namespace std::chrono_literals; | |
20effc67 | 33 | using namespace std::literals; |
f67539c2 | 34 | |
20effc67 | 35 | #define dout_context (m_osds->cct) |
f67539c2 TL |
36 | #define dout_subsys ceph_subsys_osd |
37 | #undef dout_prefix | |
20effc67 | 38 | #define dout_prefix _prefix(_dout, this) |
f67539c2 | 39 | |
20effc67 TL |
40 | template <class T> |
41 | static ostream& _prefix(std::ostream* _dout, T* t) | |
f67539c2 | 42 | { |
20effc67 | 43 | return t->gen_prefix(*_dout); |
f67539c2 TL |
44 | } |
45 | ||
46 | ostream& operator<<(ostream& out, const scrub_flags_t& sf) | |
47 | { | |
48 | if (sf.auto_repair) | |
49 | out << " AUTO_REPAIR"; | |
50 | if (sf.check_repair) | |
51 | out << " CHECK_REPAIR"; | |
52 | if (sf.deep_scrub_on_error) | |
53 | out << " DEEP_SCRUB_ON_ERROR"; | |
54 | if (sf.required) | |
55 | out << " REQ_SCRUB"; | |
56 | ||
57 | return out; | |
58 | } | |
59 | ||
60 | ostream& operator<<(ostream& out, const requested_scrub_t& sf) | |
61 | { | |
62 | if (sf.must_repair) | |
63 | out << " MUST_REPAIR"; | |
64 | if (sf.auto_repair) | |
65 | out << " planned AUTO_REPAIR"; | |
66 | if (sf.check_repair) | |
67 | out << " planned CHECK_REPAIR"; | |
68 | if (sf.deep_scrub_on_error) | |
69 | out << " planned DEEP_SCRUB_ON_ERROR"; | |
70 | if (sf.must_deep_scrub) | |
71 | out << " MUST_DEEP_SCRUB"; | |
72 | if (sf.must_scrub) | |
73 | out << " MUST_SCRUB"; | |
74 | if (sf.time_for_deep) | |
75 | out << " TIME_FOR_DEEP"; | |
76 | if (sf.need_auto) | |
77 | out << " NEED_AUTO"; | |
78 | if (sf.req_scrub) | |
79 | out << " planned REQ_SCRUB"; | |
80 | ||
81 | return out; | |
82 | } | |
83 | ||
84 | /* | |
85 | * if the incoming message is from a previous interval, it must mean | |
86 | * PrimaryLogPG::on_change() was called when that interval ended. We can safely discard | |
87 | * the stale message. | |
88 | */ | |
89 | bool PgScrubber::check_interval(epoch_t epoch_to_verify) | |
90 | { | |
91 | return epoch_to_verify >= m_pg->get_same_interval_since(); | |
92 | } | |
93 | ||
94 | bool PgScrubber::is_message_relevant(epoch_t epoch_to_verify) | |
95 | { | |
96 | if (!m_active) { | |
97 | // not scrubbing. We can assume that the scrub was already terminated, and we | |
98 | // can silently discard the incoming event. | |
99 | return false; | |
100 | } | |
101 | ||
102 | // is this a message from before we started this scrub? | |
103 | if (epoch_to_verify < m_epoch_start) { | |
104 | return false; | |
105 | } | |
106 | ||
107 | // has a new interval started? | |
108 | if (!check_interval(epoch_to_verify)) { | |
109 | // if this is a new interval, on_change() has already terminated that | |
110 | // old scrub. | |
111 | return false; | |
112 | } | |
113 | ||
114 | ceph_assert(is_primary()); | |
115 | ||
116 | // were we instructed to abort? | |
117 | return verify_against_abort(epoch_to_verify); | |
118 | } | |
119 | ||
120 | bool PgScrubber::verify_against_abort(epoch_t epoch_to_verify) | |
121 | { | |
122 | if (!should_abort()) { | |
123 | return true; | |
124 | } | |
125 | ||
126 | dout(10) << __func__ << " aborting. incoming epoch: " << epoch_to_verify | |
127 | << " vs last-aborted: " << m_last_aborted << dendl; | |
128 | ||
129 | // if we were not aware of the abort before - kill the scrub. | |
2a845540 | 130 | if (epoch_to_verify >= m_last_aborted) { |
f67539c2 TL |
131 | scrub_clear_state(); |
132 | m_last_aborted = std::max(epoch_to_verify, m_epoch_start); | |
133 | } | |
134 | return false; | |
135 | } | |
136 | ||
137 | bool PgScrubber::should_abort() const | |
138 | { | |
139 | if (m_flags.required) { | |
140 | return false; // not stopping 'required' scrubs for configuration changes | |
141 | } | |
142 | ||
143 | if (m_is_deep) { | |
144 | if (get_osdmap()->test_flag(CEPH_OSDMAP_NODEEP_SCRUB) || | |
145 | m_pg->pool.info.has_flag(pg_pool_t::FLAG_NODEEP_SCRUB)) { | |
146 | dout(10) << "nodeep_scrub set, aborting" << dendl; | |
147 | return true; | |
148 | } | |
2a845540 | 149 | } else if (get_osdmap()->test_flag(CEPH_OSDMAP_NOSCRUB) || |
f67539c2 TL |
150 | m_pg->pool.info.has_flag(pg_pool_t::FLAG_NOSCRUB)) { |
151 | dout(10) << "noscrub set, aborting" << dendl; | |
152 | return true; | |
153 | } | |
154 | ||
155 | return false; | |
156 | } | |
157 | ||
158 | // initiating state-machine events -------------------------------- | |
159 | ||
160 | /* | |
161 | * a note re the checks performed before sending scrub-initiating messages: | |
162 | * | |
163 | * For those ('StartScrub', 'AfterRepairScrub') scrub-initiation messages that | |
164 | * possibly were in the queue while the PG changed state and became unavailable for | |
165 | * scrubbing: | |
166 | * | |
167 | * The check_interval() catches all major changes to the PG. As for the other conditions | |
168 | * we may check (and see is_message_relevant() above): | |
169 | * | |
170 | * - we are not 'active' yet, so must not check against is_active(), and: | |
171 | * | |
172 | * - the 'abort' flags were just verified (when the triggering message was queued). As | |
173 | * those are only modified in human speeds - they need not be queried again. | |
174 | * | |
175 | * Some of the considerations above are also relevant to the replica-side initiation | |
176 | * ('StartReplica' & 'StartReplicaNoWait'). | |
177 | */ | |
178 | ||
179 | void PgScrubber::initiate_regular_scrub(epoch_t epoch_queued) | |
180 | { | |
181 | dout(15) << __func__ << " epoch: " << epoch_queued << dendl; | |
182 | // we may have lost our Primary status while the message languished in the queue | |
183 | if (check_interval(epoch_queued)) { | |
184 | dout(10) << "scrubber event -->> StartScrub epoch: " << epoch_queued << dendl; | |
185 | reset_epoch(epoch_queued); | |
f67539c2 TL |
186 | m_fsm->process_event(StartScrub{}); |
187 | dout(10) << "scrubber event --<< StartScrub" << dendl; | |
20effc67 | 188 | } else { |
33c7a0ef | 189 | clear_queued_or_active(); // also restarts snap trimming |
f67539c2 TL |
190 | } |
191 | } | |
192 | ||
193 | void PgScrubber::initiate_scrub_after_repair(epoch_t epoch_queued) | |
194 | { | |
195 | dout(15) << __func__ << " epoch: " << epoch_queued << dendl; | |
196 | // we may have lost our Primary status while the message languished in the queue | |
197 | if (check_interval(epoch_queued)) { | |
198 | dout(10) << "scrubber event -->> AfterRepairScrub epoch: " << epoch_queued << dendl; | |
199 | reset_epoch(epoch_queued); | |
f67539c2 TL |
200 | m_fsm->process_event(AfterRepairScrub{}); |
201 | dout(10) << "scrubber event --<< AfterRepairScrub" << dendl; | |
20effc67 | 202 | } else { |
33c7a0ef | 203 | clear_queued_or_active(); // also restarts snap trimming |
f67539c2 TL |
204 | } |
205 | } | |
206 | ||
207 | void PgScrubber::send_scrub_unblock(epoch_t epoch_queued) | |
208 | { | |
209 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
210 | if (is_message_relevant(epoch_queued)) { | |
f67539c2 TL |
211 | m_fsm->process_event(Unblocked{}); |
212 | } | |
213 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
214 | } | |
215 | ||
216 | void PgScrubber::send_scrub_resched(epoch_t epoch_queued) | |
217 | { | |
218 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
219 | if (is_message_relevant(epoch_queued)) { | |
f67539c2 TL |
220 | m_fsm->process_event(InternalSchedScrub{}); |
221 | } | |
222 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
223 | } | |
224 | ||
20effc67 | 225 | void PgScrubber::send_start_replica(epoch_t epoch_queued, Scrub::act_token_t token) |
f67539c2 | 226 | { |
20effc67 TL |
227 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued |
228 | << " token: " << token << dendl; | |
f67539c2 TL |
229 | if (is_primary()) { |
230 | // shouldn't happen. Ignore | |
231 | dout(1) << "got a replica scrub request while Primary!" << dendl; | |
232 | return; | |
233 | } | |
20effc67 TL |
234 | |
235 | if (check_interval(epoch_queued) && is_token_current(token)) { | |
f67539c2 TL |
236 | // save us some time by not waiting for updates if there are none |
237 | // to wait for. Affects the transition from NotActive into either | |
238 | // ReplicaWaitUpdates or ActiveReplica. | |
239 | if (pending_active_pushes()) | |
240 | m_fsm->process_event(StartReplica{}); | |
241 | else | |
242 | m_fsm->process_event(StartReplicaNoWait{}); | |
243 | } | |
244 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
245 | } | |
246 | ||
20effc67 | 247 | void PgScrubber::send_sched_replica(epoch_t epoch_queued, Scrub::act_token_t token) |
f67539c2 | 248 | { |
20effc67 TL |
249 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued |
250 | << " token: " << token << dendl; | |
251 | if (check_interval(epoch_queued) && is_token_current(token)) { | |
f67539c2 TL |
252 | m_fsm->process_event(SchedReplica{}); // retest for map availability |
253 | } | |
254 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
255 | } | |
256 | ||
257 | void PgScrubber::active_pushes_notification(epoch_t epoch_queued) | |
258 | { | |
259 | // note: Primary only | |
260 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
261 | if (is_message_relevant(epoch_queued)) { | |
f67539c2 TL |
262 | m_fsm->process_event(ActivePushesUpd{}); |
263 | } | |
264 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
265 | } | |
266 | ||
267 | void PgScrubber::update_applied_notification(epoch_t epoch_queued) | |
268 | { | |
269 | // note: Primary only | |
270 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
271 | if (is_message_relevant(epoch_queued)) { | |
f67539c2 TL |
272 | m_fsm->process_event(UpdatesApplied{}); |
273 | } | |
274 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
275 | } | |
276 | ||
277 | void PgScrubber::digest_update_notification(epoch_t epoch_queued) | |
278 | { | |
279 | // note: Primary only | |
280 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
281 | if (is_message_relevant(epoch_queued)) { | |
f67539c2 TL |
282 | m_fsm->process_event(DigestUpdate{}); |
283 | } | |
284 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
285 | } | |
286 | ||
20effc67 TL |
287 | void PgScrubber::send_local_map_done(epoch_t epoch_queued) |
288 | { | |
289 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
290 | if (is_message_relevant(epoch_queued)) { | |
291 | m_fsm->process_event(Scrub::IntLocalMapDone{}); | |
292 | } | |
293 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
294 | } | |
295 | ||
f67539c2 TL |
296 | void PgScrubber::send_replica_maps_ready(epoch_t epoch_queued) |
297 | { | |
298 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
299 | if (is_message_relevant(epoch_queued)) { | |
f67539c2 TL |
300 | m_fsm->process_event(GotReplicas{}); |
301 | } | |
302 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
303 | } | |
304 | ||
305 | void PgScrubber::send_replica_pushes_upd(epoch_t epoch_queued) | |
306 | { | |
307 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
308 | if (check_interval(epoch_queued)) { | |
f67539c2 TL |
309 | m_fsm->process_event(ReplicaPushesUpd{}); |
310 | } | |
311 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
312 | } | |
313 | ||
314 | void PgScrubber::send_remotes_reserved(epoch_t epoch_queued) | |
315 | { | |
316 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
317 | // note: scrub is not active yet | |
318 | if (check_interval(epoch_queued)) { | |
f67539c2 TL |
319 | m_fsm->process_event(RemotesReserved{}); |
320 | } | |
321 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
322 | } | |
323 | ||
324 | void PgScrubber::send_reservation_failure(epoch_t epoch_queued) | |
325 | { | |
326 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
327 | if (check_interval(epoch_queued)) { // do not check for 'active'! | |
f67539c2 TL |
328 | m_fsm->process_event(ReservationFailure{}); |
329 | } | |
330 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
331 | } | |
332 | ||
20effc67 TL |
333 | void PgScrubber::send_full_reset(epoch_t epoch_queued) |
334 | { | |
335 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
336 | ||
337 | m_fsm->process_event(Scrub::FullReset{}); | |
338 | ||
339 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
340 | } | |
341 | ||
342 | void PgScrubber::send_chunk_free(epoch_t epoch_queued) | |
343 | { | |
344 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
345 | if (check_interval(epoch_queued)) { | |
346 | m_fsm->process_event(Scrub::SelectedChunkFree{}); | |
347 | } | |
348 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
349 | } | |
350 | ||
351 | void PgScrubber::send_chunk_busy(epoch_t epoch_queued) | |
352 | { | |
353 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
354 | if (check_interval(epoch_queued)) { | |
355 | m_fsm->process_event(Scrub::ChunkIsBusy{}); | |
356 | } | |
357 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
358 | } | |
359 | ||
360 | void PgScrubber::send_get_next_chunk(epoch_t epoch_queued) | |
361 | { | |
362 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
363 | if (is_message_relevant(epoch_queued)) { | |
364 | m_fsm->process_event(Scrub::NextChunk{}); | |
365 | } | |
366 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
367 | } | |
368 | ||
369 | void PgScrubber::send_scrub_is_finished(epoch_t epoch_queued) | |
370 | { | |
371 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
372 | ||
373 | // can't check for "active" | |
374 | ||
375 | m_fsm->process_event(Scrub::ScrubFinished{}); | |
376 | ||
377 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
378 | } | |
379 | ||
380 | void PgScrubber::send_maps_compared(epoch_t epoch_queued) | |
381 | { | |
382 | dout(10) << "scrubber event -->> " << __func__ << " epoch: " << epoch_queued << dendl; | |
383 | ||
384 | m_fsm->process_event(Scrub::MapsCompared{}); | |
385 | ||
386 | dout(10) << "scrubber event --<< " << __func__ << dendl; | |
387 | } | |
388 | ||
389 | // ----------------- | |
390 | ||
f67539c2 TL |
391 | bool PgScrubber::is_reserving() const |
392 | { | |
393 | return m_fsm->is_reserving(); | |
394 | } | |
395 | ||
396 | void PgScrubber::reset_epoch(epoch_t epoch_queued) | |
397 | { | |
398 | dout(10) << __func__ << " state deep? " << state_test(PG_STATE_DEEP_SCRUB) << dendl; | |
399 | m_fsm->assert_not_active(); | |
400 | ||
401 | m_epoch_start = epoch_queued; | |
402 | m_needs_sleep = true; | |
403 | m_is_deep = state_test(PG_STATE_DEEP_SCRUB); | |
522d829b | 404 | update_op_mode_text(); |
f67539c2 TL |
405 | } |
406 | ||
407 | unsigned int PgScrubber::scrub_requeue_priority(Scrub::scrub_prio_t with_priority) const | |
408 | { | |
409 | unsigned int qu_priority = m_flags.priority; | |
410 | ||
411 | if (with_priority == Scrub::scrub_prio_t::high_priority) { | |
412 | qu_priority = | |
20effc67 | 413 | std::max(qu_priority, (unsigned int)m_pg->get_cct()->_conf->osd_client_op_priority); |
f67539c2 TL |
414 | } |
415 | return qu_priority; | |
416 | } | |
417 | ||
418 | unsigned int PgScrubber::scrub_requeue_priority(Scrub::scrub_prio_t with_priority, | |
419 | unsigned int suggested_priority) const | |
420 | { | |
421 | if (with_priority == Scrub::scrub_prio_t::high_priority) { | |
422 | suggested_priority = std::max(suggested_priority, | |
423 | (unsigned int)m_pg->cct->_conf->osd_client_op_priority); | |
424 | } | |
425 | return suggested_priority; | |
426 | } | |
427 | ||
428 | // ///////////////////////////////////////////////////////////////////// // | |
429 | // scrub-op registration handling | |
430 | ||
20effc67 TL |
431 | void PgScrubber::unregister_from_osd() |
432 | { | |
433 | if (m_scrub_job) { | |
434 | dout(15) << __func__ << " prev. state: " << registration_state() << dendl; | |
435 | m_osds->get_scrub_services().remove_from_osd_queue(m_scrub_job); | |
436 | } | |
437 | } | |
438 | ||
f67539c2 TL |
439 | bool PgScrubber::is_scrub_registered() const |
440 | { | |
20effc67 | 441 | return m_scrub_job && m_scrub_job->in_queues; |
f67539c2 TL |
442 | } |
443 | ||
20effc67 | 444 | std::string_view PgScrubber::registration_state() const |
f67539c2 | 445 | { |
20effc67 TL |
446 | if (m_scrub_job) { |
447 | return m_scrub_job->registration_state(); | |
f67539c2 | 448 | } |
20effc67 TL |
449 | return "(no sched job)"sv; |
450 | } | |
f67539c2 | 451 | |
20effc67 TL |
452 | void PgScrubber::rm_from_osd_scrubbing() |
453 | { | |
454 | // make sure the OSD won't try to scrub this one just now | |
455 | unregister_from_osd(); | |
456 | } | |
457 | ||
458 | void PgScrubber::on_primary_change(const requested_scrub_t& request_flags) | |
459 | { | |
460 | dout(10) << __func__ << (is_primary() ? " Primary " : " Replica ") | |
461 | << " flags: " << request_flags << dendl; | |
f67539c2 | 462 | |
20effc67 TL |
463 | if (!m_scrub_job) { |
464 | return; | |
465 | } | |
f67539c2 | 466 | |
20effc67 | 467 | dout(15) << __func__ << " scrub-job state: " << m_scrub_job->state_desc() << dendl; |
f67539c2 | 468 | |
20effc67 TL |
469 | if (is_primary()) { |
470 | auto suggested = determine_scrub_time(request_flags); | |
471 | m_osds->get_scrub_services().register_with_osd(m_scrub_job, suggested); | |
f67539c2 | 472 | } else { |
20effc67 | 473 | m_osds->get_scrub_services().remove_from_osd_queue(m_scrub_job); |
f67539c2 TL |
474 | } |
475 | ||
20effc67 TL |
476 | dout(15) << __func__ << " done " << registration_state() << dendl; |
477 | } | |
478 | ||
479 | void PgScrubber::on_maybe_registration_change(const requested_scrub_t& request_flags) | |
480 | { | |
481 | dout(10) << __func__ << (is_primary() ? " Primary " : " Replica/other ") | |
482 | << registration_state() << " flags: " << request_flags << dendl; | |
483 | ||
484 | on_primary_change(request_flags); | |
485 | dout(15) << __func__ << " done " << registration_state() << dendl; | |
486 | } | |
f67539c2 | 487 | |
20effc67 TL |
488 | void PgScrubber::update_scrub_job(const requested_scrub_t& request_flags) |
489 | { | |
490 | dout(10) << __func__ << " flags: " << request_flags << dendl; | |
491 | ||
492 | { | |
493 | // verify that the 'in_q' status matches our "Primariority" | |
494 | if (m_scrub_job && is_primary() && !m_scrub_job->in_queues) { | |
495 | dout(1) << __func__ << " !!! primary but not scheduled! " << dendl; | |
496 | } | |
497 | } | |
498 | ||
499 | if (is_primary() && m_scrub_job) { | |
500 | auto suggested = determine_scrub_time(request_flags); | |
501 | m_osds->get_scrub_services().update_job(m_scrub_job, suggested); | |
502 | } | |
f67539c2 | 503 | |
20effc67 | 504 | dout(15) << __func__ << " done " << registration_state() << dendl; |
f67539c2 TL |
505 | } |
506 | ||
20effc67 TL |
507 | ScrubQueue::sched_params_t |
508 | PgScrubber::determine_scrub_time(const requested_scrub_t& request_flags) const | |
f67539c2 | 509 | { |
20effc67 TL |
510 | ScrubQueue::sched_params_t res; |
511 | ||
512 | if (!is_primary()) { | |
513 | return res; // with ok_to_scrub set to 'false' | |
514 | } | |
515 | ||
516 | if (request_flags.must_scrub || request_flags.need_auto) { | |
517 | ||
518 | // Set the smallest time that isn't utime_t() | |
519 | res.proposed_time = PgScrubber::scrub_must_stamp(); | |
520 | res.is_must = ScrubQueue::must_scrub_t::mandatory; | |
521 | // we do not need the interval data in this case | |
522 | ||
523 | } else if (m_pg->info.stats.stats_invalid && | |
524 | m_pg->cct->_conf->osd_scrub_invalid_stats) { | |
525 | res.proposed_time = ceph_clock_now(); | |
526 | res.is_must = ScrubQueue::must_scrub_t::mandatory; | |
527 | ||
528 | } else { | |
529 | res.proposed_time = m_pg->info.history.last_scrub_stamp; | |
530 | res.min_interval = | |
531 | m_pg->get_pool().info.opts.value_or(pool_opts_t::SCRUB_MIN_INTERVAL, 0.0); | |
532 | res.max_interval = | |
533 | m_pg->get_pool().info.opts.value_or(pool_opts_t::SCRUB_MAX_INTERVAL, 0.0); | |
f67539c2 | 534 | } |
20effc67 TL |
535 | |
536 | dout(15) << __func__ << " suggested: " << res.proposed_time << " hist: " | |
537 | << m_pg->info.history.last_scrub_stamp << " v:" << m_pg->info.stats.stats_invalid | |
538 | << " / " << m_pg->cct->_conf->osd_scrub_invalid_stats << " must:" | |
539 | << (res.is_must==ScrubQueue::must_scrub_t::mandatory ? "y" : "n" ) | |
540 | << " pool min: " << res.min_interval | |
541 | << dendl; | |
542 | return res; | |
f67539c2 TL |
543 | } |
544 | ||
545 | void PgScrubber::scrub_requested(scrub_level_t scrub_level, | |
546 | scrub_type_t scrub_type, | |
547 | requested_scrub_t& req_flags) | |
548 | { | |
549 | dout(10) << __func__ << (scrub_level == scrub_level_t::deep ? " deep " : " shallow ") | |
550 | << (scrub_type == scrub_type_t::do_repair ? " repair-scrub " : " not-repair ") | |
20effc67 TL |
551 | << " prev stamp: " << m_scrub_job->get_sched_time() |
552 | << " registered? " << registration_state() | |
f67539c2 TL |
553 | << dendl; |
554 | ||
f67539c2 TL |
555 | req_flags.must_scrub = true; |
556 | req_flags.must_deep_scrub = | |
557 | (scrub_level == scrub_level_t::deep) || (scrub_type == scrub_type_t::do_repair); | |
558 | req_flags.must_repair = (scrub_type == scrub_type_t::do_repair); | |
559 | // User might intervene, so clear this | |
560 | req_flags.need_auto = false; | |
561 | req_flags.req_scrub = true; | |
562 | ||
563 | dout(20) << __func__ << " pg(" << m_pg_id << ") planned:" << req_flags << dendl; | |
564 | ||
20effc67 TL |
565 | update_scrub_job(req_flags); |
566 | m_pg->publish_stats_to_osd(); | |
f67539c2 TL |
567 | } |
568 | ||
20effc67 TL |
569 | |
570 | void PgScrubber::request_rescrubbing(requested_scrub_t& request_flags) | |
f67539c2 | 571 | { |
20effc67 | 572 | dout(10) << __func__ << " flags: " << request_flags << dendl; |
f67539c2 | 573 | |
20effc67 TL |
574 | request_flags.need_auto = true; |
575 | update_scrub_job(request_flags); | |
f67539c2 TL |
576 | } |
577 | ||
578 | bool PgScrubber::reserve_local() | |
579 | { | |
580 | // try to create the reservation object (which translates into asking the | |
581 | // OSD for the local scrub resource). If failing - undo it immediately | |
582 | ||
20effc67 TL |
583 | m_local_osd_resource.emplace(m_osds); |
584 | if (m_local_osd_resource->is_reserved()) { | |
585 | dout(15) << __func__ << ": local resources reserved" << dendl; | |
586 | return true; | |
f67539c2 TL |
587 | } |
588 | ||
20effc67 TL |
589 | dout(10) << __func__ << ": failed to reserve local scrub resources" << dendl; |
590 | m_local_osd_resource.reset(); | |
591 | return false; | |
f67539c2 TL |
592 | } |
593 | ||
594 | // ---------------------------------------------------------------------------- | |
595 | ||
596 | bool PgScrubber::has_pg_marked_new_updates() const | |
597 | { | |
598 | auto last_applied = m_pg->recovery_state.get_last_update_applied(); | |
599 | dout(10) << __func__ << " recovery last: " << last_applied | |
600 | << " vs. scrub's: " << m_subset_last_update << dendl; | |
601 | ||
602 | return last_applied >= m_subset_last_update; | |
603 | } | |
604 | ||
605 | void PgScrubber::set_subset_last_update(eversion_t e) | |
606 | { | |
607 | m_subset_last_update = e; | |
608 | dout(15) << __func__ << " last-update: " << e << dendl; | |
609 | } | |
610 | ||
20effc67 TL |
611 | void PgScrubber::on_applied_when_primary(const eversion_t& applied_version) |
612 | { | |
613 | // we are only interested in updates if we are the Primary, and in state | |
614 | // WaitLastUpdate | |
615 | if (m_fsm->is_accepting_updates() && (applied_version >= m_subset_last_update)) { | |
616 | m_osds->queue_scrub_applied_update(m_pg, m_pg->is_scrub_blocking_ops()); | |
617 | dout(15) << __func__ << " update: " << applied_version | |
618 | << " vs. required: " << m_subset_last_update << dendl; | |
619 | } | |
620 | } | |
621 | ||
f67539c2 | 622 | /* |
20effc67 | 623 | * The selected range is set directly into 'm_start' and 'm_end' |
f67539c2 TL |
624 | * setting: |
625 | * - m_subset_last_update | |
626 | * - m_max_end | |
627 | * - end | |
628 | * - start | |
f67539c2 TL |
629 | */ |
630 | bool PgScrubber::select_range() | |
631 | { | |
632 | m_primary_scrubmap = ScrubMap{}; | |
633 | m_received_maps.clear(); | |
634 | ||
635 | /* get the start and end of our scrub chunk | |
636 | * | |
637 | * Our scrub chunk has an important restriction we're going to need to | |
638 | * respect. We can't let head be start or end. | |
639 | * Using a half-open interval means that if end == head, | |
640 | * we'd scrub/lock head and the clone right next to head in different | |
641 | * chunks which would allow us to miss clones created between | |
642 | * scrubbing that chunk and scrubbing the chunk including head. | |
643 | * This isn't true for any of the other clones since clones can | |
644 | * only be created "just to the left of" head. There is one exception | |
645 | * to this: promotion of clones which always happens to the left of the | |
646 | * left-most clone, but promote_object checks the scrubber in that | |
647 | * case, so it should be ok. Also, it's ok to "miss" clones at the | |
648 | * left end of the range if we are a tier because they may legitimately | |
649 | * not exist (see _scrub). | |
650 | */ | |
20effc67 TL |
651 | int min_idx = static_cast<int>(std::max<int64_t>( |
652 | 3, m_pg->get_cct()->_conf->osd_scrub_chunk_min / (int)preemption_data.chunk_divisor())); | |
f67539c2 | 653 | |
20effc67 TL |
654 | int max_idx = static_cast<int>(std::max<int64_t>(min_idx, m_pg->get_cct()->_conf->osd_scrub_chunk_max / |
655 | (int)preemption_data.chunk_divisor())); | |
f67539c2 TL |
656 | |
657 | dout(10) << __func__ << " Min: " << min_idx << " Max: " << max_idx | |
658 | << " Div: " << preemption_data.chunk_divisor() << dendl; | |
659 | ||
660 | hobject_t start = m_start; | |
661 | hobject_t candidate_end; | |
662 | std::vector<hobject_t> objects; | |
663 | int ret = m_pg->get_pgbackend()->objects_list_partial(start, min_idx, max_idx, &objects, | |
664 | &candidate_end); | |
665 | ceph_assert(ret >= 0); | |
666 | ||
667 | if (!objects.empty()) { | |
668 | ||
669 | hobject_t back = objects.back(); | |
670 | while (candidate_end.is_head() && candidate_end == back.get_head()) { | |
671 | candidate_end = back; | |
672 | objects.pop_back(); | |
673 | if (objects.empty()) { | |
674 | ceph_assert(0 == | |
675 | "Somehow we got more than 2 objects which" | |
676 | "have the same head but are not clones"); | |
677 | } | |
678 | back = objects.back(); | |
679 | } | |
680 | ||
681 | if (candidate_end.is_head()) { | |
682 | ceph_assert(candidate_end != back.get_head()); | |
683 | candidate_end = candidate_end.get_object_boundary(); | |
684 | } | |
685 | ||
686 | } else { | |
687 | ceph_assert(candidate_end.is_max()); | |
688 | } | |
689 | ||
690 | // is that range free for us? if not - we will be rescheduled later by whoever | |
691 | // triggered us this time | |
692 | ||
693 | if (!m_pg->_range_available_for_scrub(m_start, candidate_end)) { | |
694 | // we'll be requeued by whatever made us unavailable for scrub | |
695 | dout(10) << __func__ << ": scrub blocked somewhere in range " | |
696 | << "[" << m_start << ", " << candidate_end << ")" << dendl; | |
697 | return false; | |
698 | } | |
699 | ||
700 | m_end = candidate_end; | |
701 | if (m_end > m_max_end) | |
702 | m_max_end = m_end; | |
703 | ||
704 | dout(15) << __func__ << " range selected: " << m_start << " //// " << m_end << " //// " | |
705 | << m_max_end << dendl; | |
20effc67 TL |
706 | |
707 | // debug: be 'blocked' if told so by the 'pg scrub_debug block' asok command | |
708 | if (m_debug_blockrange > 0) { | |
709 | m_debug_blockrange--; | |
710 | return false; | |
711 | } | |
f67539c2 TL |
712 | return true; |
713 | } | |
714 | ||
20effc67 TL |
715 | void PgScrubber::select_range_n_notify() |
716 | { | |
717 | if (select_range()) { | |
718 | // the next chunk to handle is not blocked | |
719 | dout(20) << __func__ << ": selection OK" << dendl; | |
720 | m_osds->queue_scrub_chunk_free(m_pg, Scrub::scrub_prio_t::low_priority); | |
721 | ||
722 | } else { | |
723 | // we will wait for the objects range to become available for scrubbing | |
724 | dout(10) << __func__ << ": selected chunk is busy" << dendl; | |
725 | m_osds->queue_scrub_chunk_busy(m_pg, Scrub::scrub_prio_t::low_priority); | |
726 | } | |
727 | } | |
728 | ||
f67539c2 TL |
729 | bool PgScrubber::write_blocked_by_scrub(const hobject_t& soid) |
730 | { | |
731 | if (soid < m_start || soid >= m_end) { | |
732 | return false; | |
733 | } | |
734 | ||
735 | dout(20) << __func__ << " " << soid << " can preempt? " | |
736 | << preemption_data.is_preemptable() << " already preempted? " | |
737 | << preemption_data.was_preempted() << dendl; | |
738 | ||
739 | if (preemption_data.was_preempted()) { | |
740 | // otherwise - write requests arriving while 'already preempted' is set | |
741 | // but 'preemptable' is not - will not be allowed to continue, and will | |
742 | // not be requeued on time. | |
743 | return false; | |
744 | } | |
745 | ||
746 | if (preemption_data.is_preemptable()) { | |
747 | ||
748 | dout(10) << __func__ << " " << soid << " preempted" << dendl; | |
749 | ||
750 | // signal the preemption | |
751 | preemption_data.do_preempt(); | |
752 | m_end = m_start; // free the range we were scrubbing | |
753 | ||
754 | return false; | |
755 | } | |
756 | return true; | |
757 | } | |
758 | ||
759 | bool PgScrubber::range_intersects_scrub(const hobject_t& start, const hobject_t& end) | |
760 | { | |
761 | // does [start, end] intersect [scrubber.start, scrubber.m_max_end) | |
762 | return (start < m_max_end && end >= m_start); | |
763 | } | |
764 | ||
20effc67 TL |
765 | Scrub::BlockedRangeWarning PgScrubber::acquire_blocked_alarm() |
766 | { | |
767 | return std::make_unique<blocked_range_t>(m_osds, ceph::timespan{300s}, m_pg_id); | |
768 | } | |
769 | ||
f67539c2 TL |
770 | /** |
771 | * if we are required to sleep: | |
772 | * arrange a callback sometimes later. | |
773 | * be sure to be able to identify a stale callback. | |
774 | * Otherwise: perform a requeue (i.e. - rescheduling thru the OSD queue) | |
775 | * anyway. | |
776 | */ | |
777 | void PgScrubber::add_delayed_scheduling() | |
778 | { | |
20effc67 | 779 | m_end = m_start; // not blocking any range now |
f67539c2 TL |
780 | |
781 | milliseconds sleep_time{0ms}; | |
782 | if (m_needs_sleep) { | |
20effc67 TL |
783 | double scrub_sleep = |
784 | 1000.0 * m_osds->get_scrub_services().scrub_sleep_time(m_flags.required); | |
785 | sleep_time = milliseconds{int64_t(scrub_sleep)}; | |
f67539c2 TL |
786 | } |
787 | dout(15) << __func__ << " sleep: " << sleep_time.count() << "ms. needed? " | |
788 | << m_needs_sleep << dendl; | |
789 | ||
790 | if (sleep_time.count()) { | |
791 | // schedule a transition for some 'sleep_time' ms in the future | |
792 | ||
793 | m_needs_sleep = false; | |
794 | m_sleep_started_at = ceph_clock_now(); | |
795 | ||
796 | // the following log line is used by osd-scrub-test.sh | |
797 | dout(20) << __func__ << " scrub state is PendingTimer, sleeping" << dendl; | |
798 | ||
799 | // the 'delayer' for crimson is different. Will be factored out. | |
800 | ||
801 | spg_t pgid = m_pg->get_pgid(); | |
20effc67 TL |
802 | auto callbk = new LambdaContext([osds = m_osds, pgid, scrbr = this]( |
803 | [[maybe_unused]] int r) mutable { | |
f67539c2 TL |
804 | PGRef pg = osds->osd->lookup_lock_pg(pgid); |
805 | if (!pg) { | |
806 | lgeneric_subdout(g_ceph_context, osd, 10) | |
807 | << "scrub_requeue_callback: Could not find " | |
808 | << "PG " << pgid << " can't complete scrub requeue after sleep" << dendl; | |
809 | return; | |
810 | } | |
811 | scrbr->m_needs_sleep = true; | |
812 | lgeneric_dout(scrbr->get_pg_cct(), 7) | |
813 | << "scrub_requeue_callback: slept for " | |
814 | << ceph_clock_now() - scrbr->m_sleep_started_at << ", re-queuing scrub" << dendl; | |
815 | ||
816 | scrbr->m_sleep_started_at = utime_t{}; | |
817 | osds->queue_for_scrub_resched(&(*pg), Scrub::scrub_prio_t::low_priority); | |
818 | pg->unlock(); | |
819 | }); | |
820 | ||
821 | std::lock_guard l(m_osds->sleep_lock); | |
822 | m_osds->sleep_timer.add_event_after(sleep_time.count() / 1000.0f, callbk); | |
823 | ||
824 | } else { | |
825 | // just a requeue | |
826 | m_osds->queue_for_scrub_resched(m_pg, Scrub::scrub_prio_t::high_priority); | |
827 | } | |
828 | } | |
829 | ||
830 | eversion_t PgScrubber::search_log_for_updates() const | |
831 | { | |
832 | auto& projected = m_pg->projected_log.log; | |
833 | auto pi = find_if( | |
834 | projected.crbegin(), projected.crend(), | |
835 | [this](const auto& e) -> bool { return e.soid >= m_start && e.soid < m_end; }); | |
836 | ||
837 | if (pi != projected.crend()) | |
838 | return pi->version; | |
839 | ||
840 | // there was no relevant update entry in the log | |
841 | ||
842 | auto& log = m_pg->recovery_state.get_pg_log().get_log().log; | |
843 | auto p = find_if(log.crbegin(), log.crend(), [this](const auto& e) -> bool { | |
844 | return e.soid >= m_start && e.soid < m_end; | |
845 | }); | |
846 | ||
847 | if (p == log.crend()) | |
848 | return eversion_t{}; | |
849 | else | |
850 | return p->version; | |
851 | } | |
852 | ||
20effc67 | 853 | void PgScrubber::get_replicas_maps(bool replica_can_preempt) |
f67539c2 TL |
854 | { |
855 | dout(10) << __func__ << " started in epoch/interval: " << m_epoch_start << "/" | |
856 | << m_interval_start | |
857 | << " pg same_interval_since: " << m_pg->info.history.same_interval_since | |
858 | << dendl; | |
859 | ||
f67539c2 TL |
860 | m_primary_scrubmap_pos.reset(); |
861 | ||
862 | // ask replicas to scan and send maps | |
863 | for (const auto& i : m_pg->get_acting_recovery_backfill()) { | |
864 | ||
865 | if (i == m_pg_whoami) | |
866 | continue; | |
867 | ||
f67539c2 TL |
868 | m_maps_status.mark_replica_map_request(i); |
869 | _request_scrub_map(i, m_subset_last_update, m_start, m_end, m_is_deep, | |
870 | replica_can_preempt); | |
871 | } | |
872 | ||
873 | dout(10) << __func__ << " awaiting" << m_maps_status << dendl; | |
f67539c2 TL |
874 | } |
875 | ||
876 | bool PgScrubber::was_epoch_changed() const | |
877 | { | |
878 | // for crimson we have m_pg->get_info().history.same_interval_since | |
879 | dout(10) << __func__ << " epoch_start: " << m_interval_start | |
880 | << " from pg: " << m_pg->get_history().same_interval_since << dendl; | |
881 | ||
882 | return m_interval_start < m_pg->get_history().same_interval_since; | |
883 | } | |
884 | ||
885 | void PgScrubber::mark_local_map_ready() | |
886 | { | |
887 | m_maps_status.mark_local_map_ready(); | |
888 | } | |
889 | ||
890 | bool PgScrubber::are_all_maps_available() const | |
891 | { | |
892 | return m_maps_status.are_all_maps_available(); | |
893 | } | |
894 | ||
895 | std::string PgScrubber::dump_awaited_maps() const | |
896 | { | |
897 | return m_maps_status.dump(); | |
898 | } | |
899 | ||
522d829b TL |
900 | void PgScrubber::update_op_mode_text() |
901 | { | |
902 | auto visible_repair = state_test(PG_STATE_REPAIR); | |
20effc67 | 903 | m_mode_desc = (visible_repair ? "repair" : (m_is_deep ? "deep-scrub" : "scrub")); |
522d829b TL |
904 | |
905 | dout(10) << __func__ << ": repair: visible: " << (visible_repair ? "true" : "false") | |
906 | << ", internal: " << (m_is_repair ? "true" : "false") | |
907 | << ". Displayed: " << m_mode_desc << dendl; | |
908 | } | |
909 | ||
f67539c2 TL |
910 | void PgScrubber::_request_scrub_map(pg_shard_t replica, |
911 | eversion_t version, | |
912 | hobject_t start, | |
913 | hobject_t end, | |
914 | bool deep, | |
915 | bool allow_preemption) | |
916 | { | |
917 | ceph_assert(replica != m_pg_whoami); | |
918 | dout(10) << __func__ << " scrubmap from osd." << replica | |
919 | << (deep ? " deep" : " shallow") << dendl; | |
920 | ||
921 | auto repscrubop = | |
922 | new MOSDRepScrub(spg_t(m_pg->info.pgid.pgid, replica.shard), version, | |
923 | get_osdmap_epoch(), m_pg->get_last_peering_reset(), start, end, deep, | |
924 | allow_preemption, m_flags.priority, m_pg->ops_blocked_by_scrub()); | |
925 | ||
926 | // default priority. We want the replica-scrub processed prior to any recovery | |
927 | // or client io messages (we are holding a lock!) | |
928 | m_osds->send_message_osd_cluster(replica.osd, repscrubop, get_osdmap_epoch()); | |
929 | } | |
930 | ||
931 | void PgScrubber::cleanup_store(ObjectStore::Transaction* t) | |
932 | { | |
933 | if (!m_store) | |
934 | return; | |
935 | ||
936 | struct OnComplete : Context { | |
937 | std::unique_ptr<Scrub::Store> store; | |
938 | explicit OnComplete(std::unique_ptr<Scrub::Store>&& store) : store(std::move(store)) | |
939 | {} | |
940 | void finish(int) override {} | |
941 | }; | |
942 | m_store->cleanup(t); | |
943 | t->register_on_complete(new OnComplete(std::move(m_store))); | |
944 | ceph_assert(!m_store); | |
945 | } | |
946 | ||
947 | void PgScrubber::on_init() | |
948 | { | |
949 | // going upwards from 'inactive' | |
950 | ceph_assert(!is_scrub_active()); | |
20effc67 | 951 | m_pg->reset_objects_scrubbed(); |
f67539c2 TL |
952 | preemption_data.reset(); |
953 | m_pg->publish_stats_to_osd(); | |
954 | m_interval_start = m_pg->get_history().same_interval_since; | |
955 | ||
956 | dout(10) << __func__ << " start same_interval:" << m_interval_start << dendl; | |
957 | ||
958 | // create a new store | |
959 | { | |
960 | ObjectStore::Transaction t; | |
961 | cleanup_store(&t); | |
962 | m_store.reset( | |
963 | Scrub::Store::create(m_pg->osd->store, &t, m_pg->info.pgid, m_pg->coll)); | |
964 | m_pg->osd->store->queue_transaction(m_pg->ch, std::move(t), nullptr); | |
965 | } | |
966 | ||
967 | m_start = m_pg->info.pgid.pgid.get_hobj_start(); | |
968 | m_active = true; | |
20effc67 TL |
969 | ++m_sessions_counter; |
970 | m_pg->publish_stats_to_osd(); | |
f67539c2 TL |
971 | } |
972 | ||
973 | void PgScrubber::on_replica_init() | |
974 | { | |
975 | m_active = true; | |
20effc67 | 976 | ++m_sessions_counter; |
f67539c2 TL |
977 | } |
978 | ||
979 | void PgScrubber::_scan_snaps(ScrubMap& smap) | |
980 | { | |
981 | hobject_t head; | |
982 | SnapSet snapset; | |
983 | ||
984 | // Test qa/standalone/scrub/osd-scrub-snaps.sh greps for the strings | |
985 | // in this function | |
986 | dout(15) << "_scan_snaps starts" << dendl; | |
987 | ||
988 | for (auto i = smap.objects.rbegin(); i != smap.objects.rend(); ++i) { | |
989 | ||
990 | const hobject_t& hoid = i->first; | |
991 | ScrubMap::object& o = i->second; | |
992 | ||
993 | dout(20) << __func__ << " " << hoid << dendl; | |
994 | ||
995 | ceph_assert(!hoid.is_snapdir()); | |
996 | if (hoid.is_head()) { | |
997 | // parse the SnapSet | |
998 | bufferlist bl; | |
999 | if (o.attrs.find(SS_ATTR) == o.attrs.end()) { | |
1000 | continue; | |
1001 | } | |
1002 | bl.push_back(o.attrs[SS_ATTR]); | |
1003 | auto p = bl.cbegin(); | |
1004 | try { | |
1005 | decode(snapset, p); | |
1006 | } catch (...) { | |
1007 | continue; | |
1008 | } | |
1009 | head = hoid.get_head(); | |
1010 | continue; | |
1011 | } | |
1012 | ||
1013 | if (hoid.snap < CEPH_MAXSNAP) { | |
1014 | // check and if necessary fix snap_mapper | |
1015 | if (hoid.get_head() != head) { | |
1016 | derr << __func__ << " no head for " << hoid << " (have " << head << ")" << dendl; | |
1017 | continue; | |
1018 | } | |
1019 | set<snapid_t> obj_snaps; | |
1020 | auto p = snapset.clone_snaps.find(hoid.snap); | |
1021 | if (p == snapset.clone_snaps.end()) { | |
1022 | derr << __func__ << " no clone_snaps for " << hoid << " in " << snapset << dendl; | |
1023 | continue; | |
1024 | } | |
1025 | obj_snaps.insert(p->second.begin(), p->second.end()); | |
1026 | set<snapid_t> cur_snaps; | |
1027 | int r = m_pg->snap_mapper.get_snaps(hoid, &cur_snaps); | |
1028 | if (r != 0 && r != -ENOENT) { | |
1029 | derr << __func__ << ": get_snaps returned " << cpp_strerror(r) << dendl; | |
1030 | ceph_abort(); | |
1031 | } | |
1032 | if (r == -ENOENT || cur_snaps != obj_snaps) { | |
1033 | ObjectStore::Transaction t; | |
1034 | OSDriver::OSTransaction _t(m_pg->osdriver.get_transaction(&t)); | |
1035 | if (r == 0) { | |
1036 | r = m_pg->snap_mapper.remove_oid(hoid, &_t); | |
1037 | if (r != 0) { | |
1038 | derr << __func__ << ": remove_oid returned " << cpp_strerror(r) << dendl; | |
1039 | ceph_abort(); | |
1040 | } | |
1041 | m_pg->osd->clog->error() | |
1042 | << "osd." << m_pg->osd->whoami << " found snap mapper error on pg " | |
1043 | << m_pg->info.pgid << " oid " << hoid << " snaps in mapper: " << cur_snaps | |
1044 | << ", oi: " << obj_snaps << "...repaired"; | |
1045 | } else { | |
1046 | m_pg->osd->clog->error() | |
1047 | << "osd." << m_pg->osd->whoami << " found snap mapper error on pg " | |
1048 | << m_pg->info.pgid << " oid " << hoid << " snaps missing in mapper" | |
1049 | << ", should be: " << obj_snaps << " was " << cur_snaps << " r " << r | |
1050 | << "...repaired"; | |
1051 | } | |
1052 | m_pg->snap_mapper.add_oid(hoid, obj_snaps, &_t); | |
1053 | ||
1054 | // wait for repair to apply to avoid confusing other bits of the system. | |
1055 | { | |
1056 | dout(15) << __func__ << " wait on repair!" << dendl; | |
1057 | ||
1058 | ceph::condition_variable my_cond; | |
1059 | ceph::mutex my_lock = ceph::make_mutex("PG::_scan_snaps my_lock"); | |
1060 | int e = 0; | |
1061 | bool done; | |
1062 | ||
1063 | t.register_on_applied_sync(new C_SafeCond(my_lock, my_cond, &done, &e)); | |
1064 | ||
1065 | e = m_pg->osd->store->queue_transaction(m_pg->ch, std::move(t)); | |
1066 | if (e != 0) { | |
1067 | derr << __func__ << ": queue_transaction got " << cpp_strerror(e) << dendl; | |
1068 | } else { | |
1069 | std::unique_lock l{my_lock}; | |
1070 | my_cond.wait(l, [&done] { return done; }); | |
1071 | } | |
1072 | } | |
1073 | } | |
1074 | } | |
1075 | } | |
1076 | } | |
1077 | ||
1078 | int PgScrubber::build_primary_map_chunk() | |
1079 | { | |
20effc67 TL |
1080 | epoch_t map_building_since = m_pg->get_osdmap_epoch(); |
1081 | dout(20) << __func__ << ": initiated at epoch " << map_building_since << dendl; | |
1082 | ||
f67539c2 TL |
1083 | auto ret = build_scrub_map_chunk(m_primary_scrubmap, m_primary_scrubmap_pos, m_start, |
1084 | m_end, m_is_deep); | |
1085 | ||
20effc67 TL |
1086 | if (ret == -EINPROGRESS) { |
1087 | // reschedule another round of asking the backend to collect the scrub data | |
f67539c2 | 1088 | m_osds->queue_for_scrub_resched(m_pg, Scrub::scrub_prio_t::low_priority); |
20effc67 | 1089 | } |
f67539c2 TL |
1090 | return ret; |
1091 | } | |
1092 | ||
1093 | int PgScrubber::build_replica_map_chunk() | |
1094 | { | |
1095 | dout(10) << __func__ << " interval start: " << m_interval_start | |
20effc67 TL |
1096 | << " current token: " << m_current_token << " epoch: " << m_epoch_start |
1097 | << " deep: " << m_is_deep << dendl; | |
f67539c2 TL |
1098 | |
1099 | auto ret = build_scrub_map_chunk(replica_scrubmap, replica_scrubmap_pos, m_start, m_end, | |
1100 | m_is_deep); | |
1101 | ||
20effc67 | 1102 | switch (ret) { |
f67539c2 | 1103 | |
20effc67 TL |
1104 | case -EINPROGRESS: |
1105 | // must wait for the backend to finish. No external event source. | |
1106 | // (note: previous version used low priority here. Now switched to using the | |
1107 | // priority of the original message) | |
1108 | m_osds->queue_for_rep_scrub_resched(m_pg, m_replica_request_priority, | |
1109 | m_flags.priority, m_current_token); | |
1110 | break; | |
f67539c2 | 1111 | |
20effc67 TL |
1112 | case 0: { |
1113 | // finished! | |
1114 | m_cleaned_meta_map.clear_from(m_start); | |
1115 | m_cleaned_meta_map.insert(replica_scrubmap); | |
1116 | auto for_meta_scrub = clean_meta_map(); | |
1117 | _scan_snaps(for_meta_scrub); | |
1118 | ||
1119 | // the local map has been created. Send it to the primary. | |
1120 | // Note: once the message reaches the Primary, it may ask us for another | |
1121 | // chunk - and we better be done with the current scrub. Thus - the preparation of | |
1122 | // the reply message is separate, and we clear the scrub state before actually | |
1123 | // sending it. | |
1124 | ||
1125 | auto reply = prep_replica_map_msg(PreemptionNoted::no_preemption); | |
1126 | replica_handling_done(); | |
1127 | dout(15) << __func__ << " chunk map sent " << dendl; | |
1128 | send_replica_map(reply); | |
1129 | } break; | |
1130 | ||
1131 | default: | |
1132 | // negative retval: build_scrub_map_chunk() signalled an error | |
1133 | // Pre-Pacific code ignored this option, treating it as a success. | |
1134 | // \todo Add an error flag in the returning message. | |
1135 | dout(1) << "Error! Aborting. ActiveReplica::react(SchedReplica) Ret: " << ret | |
1136 | << dendl; | |
1137 | replica_handling_done(); | |
1138 | // only in debug mode for now: | |
1139 | assert(false && "backend error"); | |
1140 | break; | |
1141 | }; | |
f67539c2 TL |
1142 | |
1143 | return ret; | |
1144 | } | |
1145 | ||
1146 | int PgScrubber::build_scrub_map_chunk( | |
1147 | ScrubMap& map, ScrubMapBuilder& pos, hobject_t start, hobject_t end, bool deep) | |
1148 | { | |
1149 | dout(10) << __func__ << " [" << start << "," << end << ") " | |
1150 | << " pos " << pos << " Deep: " << deep << dendl; | |
1151 | ||
1152 | // start | |
1153 | while (pos.empty()) { | |
1154 | ||
1155 | pos.deep = deep; | |
1156 | map.valid_through = m_pg->info.last_update; | |
1157 | ||
1158 | // objects | |
1159 | vector<ghobject_t> rollback_obs; | |
1160 | pos.ret = | |
1161 | m_pg->get_pgbackend()->objects_list_range(start, end, &pos.ls, &rollback_obs); | |
1162 | dout(10) << __func__ << " while pos empty " << pos.ret << dendl; | |
1163 | if (pos.ret < 0) { | |
1164 | dout(5) << "objects_list_range error: " << pos.ret << dendl; | |
1165 | return pos.ret; | |
1166 | } | |
1167 | dout(10) << __func__ << " pos.ls.empty()? " << (pos.ls.empty() ? "+" : "-") << dendl; | |
1168 | if (pos.ls.empty()) { | |
1169 | break; | |
1170 | } | |
1171 | m_pg->_scan_rollback_obs(rollback_obs); | |
1172 | pos.pos = 0; | |
1173 | return -EINPROGRESS; | |
1174 | } | |
1175 | ||
1176 | // scan objects | |
1177 | while (!pos.done()) { | |
20effc67 | 1178 | |
f67539c2 | 1179 | int r = m_pg->get_pgbackend()->be_scan_list(map, pos); |
20effc67 | 1180 | dout(30) << __func__ << " BE returned " << r << dendl; |
f67539c2 TL |
1181 | if (r == -EINPROGRESS) { |
1182 | dout(20) << __func__ << " in progress" << dendl; | |
1183 | return r; | |
1184 | } | |
1185 | } | |
1186 | ||
1187 | // finish | |
1188 | dout(20) << __func__ << " finishing" << dendl; | |
1189 | ceph_assert(pos.done()); | |
1190 | m_pg->_repair_oinfo_oid(map); | |
1191 | ||
1192 | dout(20) << __func__ << " done, got " << map.objects.size() << " items" << dendl; | |
1193 | return 0; | |
1194 | } | |
1195 | ||
1196 | /* | |
1197 | * Process: | |
1198 | * Building a map of objects suitable for snapshot validation. | |
1199 | * The data in m_cleaned_meta_map is the left over partial items that need to | |
1200 | * be completed before they can be processed. | |
1201 | * | |
1202 | * Snapshots in maps precede the head object, which is why we are scanning backwards. | |
1203 | */ | |
1204 | ScrubMap PgScrubber::clean_meta_map() | |
1205 | { | |
1206 | ScrubMap for_meta_scrub; | |
1207 | ||
1208 | if (m_end.is_max() || m_cleaned_meta_map.objects.empty()) { | |
1209 | m_cleaned_meta_map.swap(for_meta_scrub); | |
1210 | } else { | |
1211 | auto iter = m_cleaned_meta_map.objects.end(); | |
1212 | --iter; // not empty, see 'if' clause | |
1213 | auto begin = m_cleaned_meta_map.objects.begin(); | |
1214 | if (iter->first.has_snapset()) { | |
1215 | ++iter; | |
1216 | } else { | |
1217 | while (iter != begin) { | |
1218 | auto next = iter--; | |
1219 | if (next->first.get_head() != iter->first.get_head()) { | |
1220 | ++iter; | |
1221 | break; | |
1222 | } | |
1223 | } | |
1224 | } | |
1225 | for_meta_scrub.objects.insert(begin, iter); | |
1226 | m_cleaned_meta_map.objects.erase(begin, iter); | |
1227 | } | |
1228 | ||
1229 | return for_meta_scrub; | |
1230 | } | |
1231 | ||
1232 | void PgScrubber::run_callbacks() | |
1233 | { | |
1234 | std::list<Context*> to_run; | |
1235 | to_run.swap(m_callbacks); | |
1236 | ||
1237 | for (auto& tr : to_run) { | |
1238 | tr->complete(0); | |
1239 | } | |
1240 | } | |
1241 | ||
1242 | void PgScrubber::maps_compare_n_cleanup() | |
1243 | { | |
1244 | scrub_compare_maps(); | |
1245 | m_start = m_end; | |
1246 | run_callbacks(); | |
1247 | requeue_waiting(); | |
20effc67 | 1248 | m_osds->queue_scrub_maps_compared(m_pg, Scrub::scrub_prio_t::low_priority); |
f67539c2 TL |
1249 | } |
1250 | ||
1251 | Scrub::preemption_t& PgScrubber::get_preemptor() | |
1252 | { | |
1253 | return preemption_data; | |
1254 | } | |
1255 | ||
f67539c2 | 1256 | /* |
20effc67 TL |
1257 | * Process note: called for the arriving "give me your map, replica!" request. |
1258 | * Unlike the original implementation, we do not requeue the Op waiting for | |
f67539c2 TL |
1259 | * updates. Instead - we trigger the FSM. |
1260 | */ | |
1261 | void PgScrubber::replica_scrub_op(OpRequestRef op) | |
1262 | { | |
20effc67 | 1263 | op->mark_started(); |
f67539c2 | 1264 | auto msg = op->get_req<MOSDRepScrub>(); |
20effc67 TL |
1265 | dout(10) << __func__ << " pg:" << m_pg->pg_id |
1266 | << " Msg: map_epoch:" << msg->map_epoch | |
1267 | << " min_epoch:" << msg->min_epoch << " deep?" << msg->deep << dendl; | |
f67539c2 | 1268 | |
20effc67 TL |
1269 | // are we still processing a previous scrub-map request without noticing that |
1270 | // the interval changed? won't see it here, but rather at the reservation | |
1271 | // stage. | |
f67539c2 TL |
1272 | |
1273 | if (msg->map_epoch < m_pg->info.history.same_interval_since) { | |
20effc67 TL |
1274 | dout(10) << "replica_scrub_op discarding old replica_scrub from " |
1275 | << msg->map_epoch << " < " | |
1276 | << m_pg->info.history.same_interval_since << dendl; | |
f67539c2 TL |
1277 | |
1278 | // is there a general sync issue? are we holding a stale reservation? | |
1279 | // not checking now - assuming we will actively react to interval change. | |
1280 | ||
1281 | return; | |
1282 | } | |
1283 | ||
20effc67 TL |
1284 | if (is_queued_or_active()) { |
1285 | // this is bug! | |
1286 | // Somehow, we have received a new scrub request from our Primary, before | |
1287 | // having finished with the previous one. Did we go through an interval | |
1288 | // change without reseting the FSM? Possible responses: | |
1289 | // - crashing (the original assert_not_active() implemented that one), or | |
1290 | // - trying to recover: | |
1291 | // - (logging enough information to debug this scenario) | |
1292 | // - reset the FSM. | |
1293 | m_osds->clog->warn() << fmt::format( | |
1294 | "{}: error: a second scrub-op received while handling the previous one", | |
1295 | __func__); | |
1296 | ||
1297 | scrub_clear_state(); | |
1298 | m_osds->clog->warn() << fmt::format( | |
1299 | "{}: after a reset. Now handling the new OP", __func__); | |
1300 | } | |
1301 | // make sure the FSM is at NotActive | |
1302 | m_fsm->assert_not_active(); | |
1303 | ||
f67539c2 TL |
1304 | replica_scrubmap = ScrubMap{}; |
1305 | replica_scrubmap_pos = ScrubMapBuilder{}; | |
1306 | ||
1307 | m_replica_min_epoch = msg->min_epoch; | |
1308 | m_start = msg->start; | |
1309 | m_end = msg->end; | |
1310 | m_max_end = msg->end; | |
1311 | m_is_deep = msg->deep; | |
1312 | m_interval_start = m_pg->info.history.same_interval_since; | |
20effc67 TL |
1313 | m_replica_request_priority = msg->high_priority |
1314 | ? Scrub::scrub_prio_t::high_priority | |
1315 | : Scrub::scrub_prio_t::low_priority; | |
f67539c2 TL |
1316 | m_flags.priority = msg->priority ? msg->priority : m_pg->get_scrub_priority(); |
1317 | ||
1318 | preemption_data.reset(); | |
1319 | preemption_data.force_preemptability(msg->allow_preemption); | |
1320 | ||
1321 | replica_scrubmap_pos.reset(); | |
1322 | ||
20effc67 TL |
1323 | set_queued_or_active(); |
1324 | m_osds->queue_for_rep_scrub(m_pg, m_replica_request_priority, | |
1325 | m_flags.priority, m_current_token); | |
f67539c2 TL |
1326 | } |
1327 | ||
1328 | void PgScrubber::set_op_parameters(requested_scrub_t& request) | |
1329 | { | |
1330 | dout(10) << __func__ << " input: " << request << dendl; | |
1331 | ||
33c7a0ef TL |
1332 | set_queued_or_active(); // we are fully committed now. |
1333 | ||
f67539c2 TL |
1334 | // write down the epoch of starting a new scrub. Will be used |
1335 | // to discard stale messages from previous aborted scrubs. | |
1336 | m_epoch_start = m_pg->get_osdmap_epoch(); | |
1337 | ||
1338 | m_flags.check_repair = request.check_repair; | |
1339 | m_flags.auto_repair = request.auto_repair || request.need_auto; | |
1340 | m_flags.required = request.req_scrub || request.must_scrub; | |
1341 | ||
1342 | m_flags.priority = (request.must_scrub || request.need_auto) | |
1343 | ? get_pg_cct()->_conf->osd_requested_scrub_priority | |
1344 | : m_pg->get_scrub_priority(); | |
1345 | ||
1346 | state_set(PG_STATE_SCRUBBING); | |
1347 | ||
1348 | // will we be deep-scrubbing? | |
1349 | if (request.must_deep_scrub || request.need_auto || request.time_for_deep) { | |
1350 | state_set(PG_STATE_DEEP_SCRUB); | |
1351 | } | |
1352 | ||
522d829b TL |
1353 | // m_is_repair is set for either 'must_repair' or 'repair-on-the-go' (i.e. |
1354 | // deep-scrub with the auto_repair configuration flag set). m_is_repair value | |
1355 | // determines the scrubber behavior. | |
1356 | // PG_STATE_REPAIR, on the other hand, is only used for status reports (inc. the | |
1357 | // PG status as appearing in the logs). | |
1358 | m_is_repair = request.must_repair || m_flags.auto_repair; | |
1359 | if (request.must_repair) { | |
f67539c2 | 1360 | state_set(PG_STATE_REPAIR); |
522d829b | 1361 | // not calling update_op_mode_text() yet, as m_is_deep not set yet |
f67539c2 TL |
1362 | } |
1363 | ||
20effc67 | 1364 | // the publishing here is required for tests synchronization |
f67539c2 TL |
1365 | m_pg->publish_stats_to_osd(); |
1366 | m_flags.deep_scrub_on_error = request.deep_scrub_on_error; | |
1367 | } | |
1368 | ||
1369 | void PgScrubber::scrub_compare_maps() | |
1370 | { | |
1371 | dout(10) << __func__ << " has maps, analyzing" << dendl; | |
1372 | ||
1373 | // construct authoritative scrub map for type-specific scrubbing | |
1374 | m_cleaned_meta_map.insert(m_primary_scrubmap); | |
1375 | map<hobject_t, pair<std::optional<uint32_t>, std::optional<uint32_t>>> missing_digest; | |
1376 | ||
1377 | map<pg_shard_t, ScrubMap*> maps; | |
1378 | maps[m_pg_whoami] = &m_primary_scrubmap; | |
1379 | ||
1380 | for (const auto& i : m_pg->get_acting_recovery_backfill()) { | |
1381 | if (i == m_pg_whoami) | |
1382 | continue; | |
1383 | dout(2) << __func__ << " replica " << i << " has " | |
1384 | << m_received_maps[i].objects.size() << " items" << dendl; | |
1385 | maps[i] = &m_received_maps[i]; | |
1386 | } | |
1387 | ||
1388 | set<hobject_t> master_set; | |
1389 | ||
1390 | // Construct master set | |
1391 | for (const auto& map : maps) { | |
1392 | for (const auto& i : map.second->objects) { | |
1393 | master_set.insert(i.first); | |
1394 | } | |
1395 | } | |
1396 | ||
1397 | stringstream ss; | |
1398 | m_pg->get_pgbackend()->be_omap_checks(maps, master_set, m_omap_stats, ss); | |
1399 | ||
1400 | if (!ss.str().empty()) { | |
1401 | m_osds->clog->warn(ss); | |
1402 | } | |
1403 | ||
1404 | if (m_pg->recovery_state.get_acting_recovery_backfill().size() > 1) { | |
1405 | ||
1406 | dout(10) << __func__ << " comparing replica scrub maps" << dendl; | |
1407 | ||
1408 | // Map from object with errors to good peer | |
1409 | map<hobject_t, list<pg_shard_t>> authoritative; | |
1410 | ||
20effc67 | 1411 | dout(2) << __func__ << ": primary (" << m_pg->get_primary() << ") has " |
f67539c2 | 1412 | << m_primary_scrubmap.objects.size() << " items" << dendl; |
20effc67 | 1413 | m_pg->add_objects_scrubbed_count(m_primary_scrubmap.objects.size()); |
f67539c2 TL |
1414 | |
1415 | ss.str(""); | |
1416 | ss.clear(); | |
1417 | ||
1418 | m_pg->get_pgbackend()->be_compare_scrubmaps( | |
522d829b | 1419 | maps, master_set, m_is_repair, m_missing, m_inconsistent, |
f67539c2 TL |
1420 | authoritative, missing_digest, m_shallow_errors, m_deep_errors, m_store.get(), |
1421 | m_pg->info.pgid, m_pg->recovery_state.get_acting(), ss); | |
f67539c2 TL |
1422 | |
1423 | if (!ss.str().empty()) { | |
1424 | m_osds->clog->error(ss); | |
1425 | } | |
1426 | ||
1427 | for (auto& i : authoritative) { | |
1428 | list<pair<ScrubMap::object, pg_shard_t>> good_peers; | |
1429 | for (list<pg_shard_t>::const_iterator j = i.second.begin(); j != i.second.end(); | |
1430 | ++j) { | |
1431 | good_peers.emplace_back(maps[*j]->objects[i.first], *j); | |
1432 | } | |
1433 | m_authoritative.emplace(i.first, good_peers); | |
1434 | } | |
1435 | ||
1436 | for (auto i = authoritative.begin(); i != authoritative.end(); ++i) { | |
1437 | m_cleaned_meta_map.objects.erase(i->first); | |
1438 | m_cleaned_meta_map.objects.insert( | |
1439 | *(maps[i->second.back()]->objects.find(i->first))); | |
1440 | } | |
1441 | } | |
1442 | ||
1443 | auto for_meta_scrub = clean_meta_map(); | |
1444 | ||
1445 | // ok, do the pg-type specific scrubbing | |
1446 | ||
1447 | // (Validates consistency of the object info and snap sets) | |
1448 | scrub_snapshot_metadata(for_meta_scrub, missing_digest); | |
1449 | ||
1450 | // Called here on the primary can use an authoritative map if it isn't the primary | |
1451 | _scan_snaps(for_meta_scrub); | |
1452 | ||
1453 | if (!m_store->empty()) { | |
1454 | ||
522d829b | 1455 | if (m_is_repair) { |
f67539c2 TL |
1456 | dout(10) << __func__ << ": discarding scrub results" << dendl; |
1457 | m_store->flush(nullptr); | |
1458 | } else { | |
1459 | dout(10) << __func__ << ": updating scrub object" << dendl; | |
1460 | ObjectStore::Transaction t; | |
1461 | m_store->flush(&t); | |
1462 | m_pg->osd->store->queue_transaction(m_pg->ch, std::move(t), nullptr); | |
1463 | } | |
1464 | } | |
1465 | } | |
1466 | ||
20effc67 TL |
1467 | ScrubMachineListener::MsgAndEpoch PgScrubber::prep_replica_map_msg( |
1468 | PreemptionNoted was_preempted) | |
f67539c2 TL |
1469 | { |
1470 | dout(10) << __func__ << " min epoch:" << m_replica_min_epoch << dendl; | |
1471 | ||
20effc67 TL |
1472 | auto reply = |
1473 | make_message<MOSDRepScrubMap>(spg_t(m_pg->info.pgid.pgid, m_pg->get_primary().shard), | |
1474 | m_replica_min_epoch, m_pg_whoami); | |
f67539c2 TL |
1475 | |
1476 | reply->preempted = (was_preempted == PreemptionNoted::preempted); | |
1477 | ::encode(replica_scrubmap, reply->get_data()); | |
1478 | ||
20effc67 TL |
1479 | return ScrubMachineListener::MsgAndEpoch{reply, m_replica_min_epoch}; |
1480 | } | |
1481 | ||
1482 | void PgScrubber::send_replica_map(const MsgAndEpoch& preprepared) | |
1483 | { | |
1484 | m_pg->send_cluster_message(m_pg->get_primary().osd, preprepared.m_msg, | |
1485 | preprepared.m_epoch, false); | |
1486 | } | |
1487 | ||
1488 | void PgScrubber::send_preempted_replica() | |
1489 | { | |
1490 | auto reply = | |
1491 | make_message<MOSDRepScrubMap>(spg_t{m_pg->info.pgid.pgid, m_pg->get_primary().shard}, | |
1492 | m_replica_min_epoch, m_pg_whoami); | |
1493 | ||
1494 | reply->preempted = true; | |
1495 | ::encode(replica_scrubmap, reply->get_data()); // must not skip this | |
1496 | m_pg->send_cluster_message(m_pg->get_primary().osd, reply, m_replica_min_epoch, false); | |
f67539c2 TL |
1497 | } |
1498 | ||
1499 | /* | |
1500 | * - if the replica lets us know it was interrupted, we mark the chunk as interrupted. | |
1501 | * The state-machine will react to that when all replica maps are received. | |
1502 | * - when all maps are received, we signal the FSM with the GotReplicas event (see | |
1503 | * scrub_send_replmaps_ready()). Note that due to the no-reentrancy limitations of the | |
1504 | * FSM, we do not 'process' the event directly. Instead - it is queued for the OSD to | |
1505 | * handle. | |
1506 | */ | |
1507 | void PgScrubber::map_from_replica(OpRequestRef op) | |
1508 | { | |
1509 | auto m = op->get_req<MOSDRepScrubMap>(); | |
1510 | dout(15) << __func__ << " " << *m << dendl; | |
1511 | ||
1512 | if (m->map_epoch < m_pg->info.history.same_interval_since) { | |
1513 | dout(10) << __func__ << " discarding old from " << m->map_epoch << " < " | |
1514 | << m_pg->info.history.same_interval_since << dendl; | |
1515 | return; | |
1516 | } | |
1517 | ||
1518 | auto p = const_cast<bufferlist&>(m->get_data()).cbegin(); | |
1519 | ||
1520 | m_received_maps[m->from].decode(p, m_pg->info.pgid.pool()); | |
1521 | dout(15) << "map version is " << m_received_maps[m->from].valid_through << dendl; | |
1522 | ||
1523 | auto [is_ok, err_txt] = m_maps_status.mark_arriving_map(m->from); | |
1524 | if (!is_ok) { | |
1525 | // previously an unexpected map was triggering an assert. Now, as scrubs can be | |
1526 | // aborted at any time, the chances of this happening have increased, and aborting is | |
1527 | // not justified | |
1528 | dout(1) << __func__ << err_txt << " from OSD " << m->from << dendl; | |
1529 | return; | |
1530 | } | |
1531 | ||
1532 | if (m->preempted) { | |
1533 | dout(10) << __func__ << " replica was preempted, setting flag" << dendl; | |
1534 | preemption_data.do_preempt(); | |
1535 | } | |
1536 | ||
1537 | if (m_maps_status.are_all_maps_available()) { | |
1538 | dout(15) << __func__ << " all repl-maps available" << dendl; | |
1539 | m_osds->queue_scrub_got_repl_maps(m_pg, m_pg->is_scrub_blocking_ops()); | |
1540 | } | |
1541 | } | |
1542 | ||
1543 | void PgScrubber::handle_scrub_reserve_request(OpRequestRef op) | |
1544 | { | |
1545 | dout(10) << __func__ << " " << *op->get_req() << dendl; | |
1546 | op->mark_started(); | |
1547 | auto request_ep = op->get_req<MOSDScrubReserve>()->get_map_epoch(); | |
1548 | ||
1549 | /* | |
1550 | * if we are currently holding a reservation, then: | |
1551 | * either (1) we, the scrubber, did not yet notice an interval change. The remembered | |
1552 | * reservation epoch is from before our interval, and we can silently discard the | |
1553 | * reservation (no message is required). | |
1554 | * or: | |
1555 | * (2) the interval hasn't changed, but the same Primary that (we think) holds the | |
1556 | * lock just sent us a new request. Note that we know it's the same Primary, as | |
1557 | * otherwise the interval would have changed. | |
1558 | * Ostensibly we can discard & redo the reservation. But then we | |
1559 | * will be temporarily releasing the OSD resource - and might not be able to grab it | |
20effc67 TL |
1560 | * again. Thus, we simply treat this as a successful new request |
1561 | * (but mark the fact that if there is a previous request from the primary to | |
1562 | * scrub a specific chunk - that request is now defunct). | |
f67539c2 TL |
1563 | */ |
1564 | ||
1565 | if (m_remote_osd_resource.has_value() && m_remote_osd_resource->is_stale()) { | |
1566 | // we are holding a stale reservation from a past epoch | |
1567 | m_remote_osd_resource.reset(); | |
20effc67 | 1568 | dout(10) << __func__ << " cleared existing stale reservation" << dendl; |
f67539c2 TL |
1569 | } |
1570 | ||
1571 | if (request_ep < m_pg->get_same_interval_since()) { | |
1572 | // will not ack stale requests | |
1573 | return; | |
1574 | } | |
1575 | ||
1576 | bool granted{false}; | |
1577 | if (m_remote_osd_resource.has_value()) { | |
1578 | ||
1579 | dout(10) << __func__ << " already reserved." << dendl; | |
20effc67 TL |
1580 | |
1581 | /* | |
1582 | * it might well be that we did not yet finish handling the latest scrub-op from | |
1583 | * our primary. This happens, for example, if 'noscrub' was set via a command, then | |
1584 | * reset. The primary in this scenario will remain in the same interval, but we do need | |
1585 | * to reset our internal state (otherwise - the first renewed 'give me your scrub map' | |
1586 | * from the primary will see us in active state, crashing the OSD). | |
1587 | */ | |
1588 | advance_token(); | |
f67539c2 TL |
1589 | granted = true; |
1590 | ||
1591 | } else if (m_pg->cct->_conf->osd_scrub_during_recovery || | |
1592 | !m_osds->is_recovery_active()) { | |
20effc67 | 1593 | m_remote_osd_resource.emplace(this, m_pg, m_osds, request_ep); |
f67539c2 TL |
1594 | // OSD resources allocated? |
1595 | granted = m_remote_osd_resource->is_reserved(); | |
1596 | if (!granted) { | |
1597 | // just forget it | |
1598 | m_remote_osd_resource.reset(); | |
1599 | dout(20) << __func__ << ": failed to reserve remotely" << dendl; | |
1600 | } | |
1601 | } | |
1602 | ||
1603 | dout(10) << __func__ << " reserved? " << (granted ? "yes" : "no") << dendl; | |
1604 | ||
1605 | Message* reply = new MOSDScrubReserve( | |
1606 | spg_t(m_pg->info.pgid.pgid, m_pg->get_primary().shard), request_ep, | |
1607 | granted ? MOSDScrubReserve::GRANT : MOSDScrubReserve::REJECT, m_pg_whoami); | |
1608 | ||
1609 | m_osds->send_message_osd_cluster(reply, op->get_req()->get_connection()); | |
1610 | } | |
1611 | ||
1612 | void PgScrubber::handle_scrub_reserve_grant(OpRequestRef op, pg_shard_t from) | |
1613 | { | |
1614 | dout(10) << __func__ << " " << *op->get_req() << dendl; | |
1615 | op->mark_started(); | |
1616 | ||
1617 | if (m_reservations.has_value()) { | |
1618 | m_reservations->handle_reserve_grant(op, from); | |
1619 | } else { | |
2a845540 TL |
1620 | dout(20) << __func__ << ": late/unsolicited reservation grant from osd " |
1621 | << from << " (" << op << ")" << dendl; | |
f67539c2 TL |
1622 | } |
1623 | } | |
1624 | ||
1625 | void PgScrubber::handle_scrub_reserve_reject(OpRequestRef op, pg_shard_t from) | |
1626 | { | |
1627 | dout(10) << __func__ << " " << *op->get_req() << dendl; | |
1628 | op->mark_started(); | |
1629 | ||
1630 | if (m_reservations.has_value()) { | |
1631 | // there is an active reservation process. No action is required otherwise. | |
1632 | m_reservations->handle_reserve_reject(op, from); | |
1633 | } | |
1634 | } | |
1635 | ||
1636 | void PgScrubber::handle_scrub_reserve_release(OpRequestRef op) | |
1637 | { | |
1638 | dout(10) << __func__ << " " << *op->get_req() << dendl; | |
1639 | op->mark_started(); | |
20effc67 TL |
1640 | |
1641 | /* | |
1642 | * this specific scrub session has terminated. All incoming events carrying the old | |
1643 | * tag will be discarded. | |
1644 | */ | |
1645 | advance_token(); | |
f67539c2 TL |
1646 | m_remote_osd_resource.reset(); |
1647 | } | |
1648 | ||
1649 | void PgScrubber::discard_replica_reservations() | |
1650 | { | |
1651 | dout(10) << __func__ << dendl; | |
1652 | if (m_reservations.has_value()) { | |
1653 | m_reservations->discard_all(); | |
1654 | } | |
1655 | } | |
1656 | ||
1657 | void PgScrubber::clear_scrub_reservations() | |
1658 | { | |
1659 | dout(10) << __func__ << dendl; | |
1660 | m_reservations.reset(); // the remote reservations | |
1661 | m_local_osd_resource.reset(); // the local reservation | |
1662 | m_remote_osd_resource.reset(); // we as replica reserved for a Primary | |
1663 | } | |
1664 | ||
1665 | void PgScrubber::message_all_replicas(int32_t opcode, std::string_view op_text) | |
1666 | { | |
1667 | ceph_assert(m_pg->recovery_state.get_backfill_targets().empty()); | |
1668 | ||
20effc67 | 1669 | std::vector<pair<int, Message*>> messages; |
f67539c2 TL |
1670 | messages.reserve(m_pg->get_actingset().size()); |
1671 | ||
1672 | epoch_t epch = get_osdmap_epoch(); | |
1673 | ||
1674 | for (auto& p : m_pg->get_actingset()) { | |
1675 | ||
1676 | if (p == m_pg_whoami) | |
1677 | continue; | |
1678 | ||
1679 | dout(10) << "scrub requesting " << op_text << " from osd." << p << " Epoch: " << epch | |
1680 | << dendl; | |
1681 | Message* m = new MOSDScrubReserve(spg_t(m_pg->info.pgid.pgid, p.shard), epch, opcode, | |
1682 | m_pg_whoami); | |
1683 | messages.push_back(std::make_pair(p.osd, m)); | |
1684 | } | |
1685 | ||
1686 | if (!messages.empty()) { | |
1687 | m_osds->send_message_osd_cluster(messages, epch); | |
1688 | } | |
1689 | } | |
1690 | ||
1691 | void PgScrubber::unreserve_replicas() | |
1692 | { | |
1693 | dout(10) << __func__ << dendl; | |
1694 | m_reservations.reset(); | |
1695 | } | |
1696 | ||
20effc67 TL |
1697 | void PgScrubber::set_reserving_now() |
1698 | { | |
1699 | m_osds->get_scrub_services().set_reserving_now(); | |
1700 | } | |
1701 | ||
1702 | void PgScrubber::clear_reserving_now() | |
1703 | { | |
1704 | m_osds->get_scrub_services().clear_reserving_now(); | |
1705 | } | |
1706 | ||
1707 | void PgScrubber::set_queued_or_active() | |
1708 | { | |
1709 | m_queued_or_active = true; | |
1710 | } | |
1711 | ||
1712 | void PgScrubber::clear_queued_or_active() | |
1713 | { | |
33c7a0ef TL |
1714 | if (m_queued_or_active) { |
1715 | m_queued_or_active = false; | |
1716 | // and just in case snap trimming was blocked by the aborted scrub | |
1717 | m_pg->snap_trimmer_scrub_complete(); | |
1718 | } | |
20effc67 TL |
1719 | } |
1720 | ||
1721 | bool PgScrubber::is_queued_or_active() const | |
1722 | { | |
1723 | return m_queued_or_active; | |
1724 | } | |
1725 | ||
f67539c2 TL |
1726 | [[nodiscard]] bool PgScrubber::scrub_process_inconsistent() |
1727 | { | |
522d829b TL |
1728 | dout(10) << __func__ << ": checking authoritative (mode=" |
1729 | << m_mode_desc << ", auth remaining #: " << m_authoritative.size() | |
1730 | << ")" << dendl; | |
f67539c2 TL |
1731 | |
1732 | // authoritative only store objects which are missing or inconsistent. | |
1733 | if (!m_authoritative.empty()) { | |
1734 | ||
1735 | stringstream ss; | |
522d829b | 1736 | ss << m_pg->info.pgid << " " << m_mode_desc << " " << m_missing.size() << " missing, " |
f67539c2 TL |
1737 | << m_inconsistent.size() << " inconsistent objects"; |
1738 | dout(2) << ss.str() << dendl; | |
1739 | m_osds->clog->error(ss); | |
1740 | ||
522d829b | 1741 | if (m_is_repair) { |
f67539c2 | 1742 | state_clear(PG_STATE_CLEAN); |
522d829b TL |
1743 | // we know we have a problem, so it's OK to set the user-visible flag |
1744 | // even if we only reached here via auto-repair | |
1745 | state_set(PG_STATE_REPAIR); | |
1746 | update_op_mode_text(); | |
f67539c2 TL |
1747 | |
1748 | for (const auto& [hobj, shrd_list] : m_authoritative) { | |
1749 | ||
1750 | auto missing_entry = m_missing.find(hobj); | |
1751 | ||
1752 | if (missing_entry != m_missing.end()) { | |
1753 | m_pg->repair_object(hobj, shrd_list, missing_entry->second); | |
1754 | m_fixed_count += missing_entry->second.size(); | |
1755 | } | |
1756 | ||
1757 | if (m_inconsistent.count(hobj)) { | |
1758 | m_pg->repair_object(hobj, shrd_list, m_inconsistent[hobj]); | |
1759 | m_fixed_count += m_inconsistent[hobj].size(); | |
1760 | } | |
1761 | } | |
1762 | } | |
1763 | } | |
522d829b | 1764 | return (!m_authoritative.empty() && m_is_repair); |
f67539c2 TL |
1765 | } |
1766 | ||
1767 | /* | |
1768 | * note: only called for the Primary. | |
1769 | */ | |
1770 | void PgScrubber::scrub_finish() | |
1771 | { | |
1772 | dout(10) << __func__ << " before flags: " << m_flags | |
20effc67 TL |
1773 | << ". repair state: " << (state_test(PG_STATE_REPAIR) ? "repair" : "no-repair") |
1774 | << ". deep_scrub_on_error: " << m_flags.deep_scrub_on_error << dendl; | |
f67539c2 TL |
1775 | |
1776 | ceph_assert(m_pg->is_locked()); | |
20effc67 | 1777 | ceph_assert(is_queued_or_active()); |
f67539c2 TL |
1778 | |
1779 | m_pg->m_planned_scrub = requested_scrub_t{}; | |
1780 | ||
1781 | // if the repair request comes from auto-repair and large number of errors, | |
1782 | // we would like to cancel auto-repair | |
522d829b | 1783 | if (m_is_repair && m_flags.auto_repair && |
f67539c2 TL |
1784 | m_authoritative.size() > m_pg->cct->_conf->osd_scrub_auto_repair_num_errors) { |
1785 | ||
1786 | dout(10) << __func__ << " undoing the repair" << dendl; | |
522d829b TL |
1787 | state_clear(PG_STATE_REPAIR); // not expected to be set, anyway |
1788 | m_is_repair = false; | |
1789 | update_op_mode_text(); | |
f67539c2 TL |
1790 | } |
1791 | ||
f67539c2 TL |
1792 | bool do_auto_scrub = false; |
1793 | ||
1794 | // if a regular scrub had errors within the limit, do a deep scrub to auto repair | |
20effc67 | 1795 | if (m_flags.deep_scrub_on_error && !m_authoritative.empty() && |
f67539c2 | 1796 | m_authoritative.size() <= m_pg->cct->_conf->osd_scrub_auto_repair_num_errors) { |
522d829b | 1797 | ceph_assert(!m_is_deep); |
f67539c2 TL |
1798 | do_auto_scrub = true; |
1799 | dout(15) << __func__ << " Try to auto repair after scrub errors" << dendl; | |
1800 | } | |
1801 | ||
1802 | m_flags.deep_scrub_on_error = false; | |
1803 | ||
1804 | // type-specific finish (can tally more errors) | |
1805 | _scrub_finish(); | |
1806 | ||
1807 | bool has_error = scrub_process_inconsistent(); | |
1808 | ||
1809 | { | |
1810 | stringstream oss; | |
522d829b | 1811 | oss << m_pg->info.pgid.pgid << " " << m_mode_desc << " "; |
f67539c2 TL |
1812 | int total_errors = m_shallow_errors + m_deep_errors; |
1813 | if (total_errors) | |
1814 | oss << total_errors << " errors"; | |
1815 | else | |
1816 | oss << "ok"; | |
522d829b | 1817 | if (!m_is_deep && m_pg->info.stats.stats.sum.num_deep_scrub_errors) |
f67539c2 TL |
1818 | oss << " ( " << m_pg->info.stats.stats.sum.num_deep_scrub_errors |
1819 | << " remaining deep scrub error details lost)"; | |
522d829b | 1820 | if (m_is_repair) |
f67539c2 TL |
1821 | oss << ", " << m_fixed_count << " fixed"; |
1822 | if (total_errors) | |
1823 | m_osds->clog->error(oss); | |
1824 | else | |
1825 | m_osds->clog->debug(oss); | |
1826 | } | |
1827 | ||
1828 | // Since we don't know which errors were fixed, we can only clear them | |
1829 | // when every one has been fixed. | |
522d829b | 1830 | if (m_is_repair) { |
f67539c2 TL |
1831 | if (m_fixed_count == m_shallow_errors + m_deep_errors) { |
1832 | ||
522d829b | 1833 | ceph_assert(m_is_deep); |
f67539c2 TL |
1834 | m_shallow_errors = 0; |
1835 | m_deep_errors = 0; | |
1836 | dout(20) << __func__ << " All may be fixed" << dendl; | |
1837 | ||
1838 | } else if (has_error) { | |
1839 | ||
1840 | // Deep scrub in order to get corrected error counts | |
1841 | m_pg->scrub_after_recovery = true; | |
1842 | m_pg->m_planned_scrub.req_scrub = | |
1843 | m_pg->m_planned_scrub.req_scrub || m_flags.required; | |
1844 | ||
1845 | dout(20) << __func__ << " Current 'required': " << m_flags.required | |
1846 | << " Planned 'req_scrub': " << m_pg->m_planned_scrub.req_scrub << dendl; | |
1847 | ||
1848 | } else if (m_shallow_errors || m_deep_errors) { | |
1849 | ||
1850 | // We have errors but nothing can be fixed, so there is no repair | |
1851 | // possible. | |
1852 | state_set(PG_STATE_FAILED_REPAIR); | |
1853 | dout(10) << __func__ << " " << (m_shallow_errors + m_deep_errors) | |
1854 | << " error(s) present with no repair possible" << dendl; | |
1855 | } | |
1856 | } | |
1857 | ||
1858 | { | |
1859 | // finish up | |
1860 | ObjectStore::Transaction t; | |
1861 | m_pg->recovery_state.update_stats( | |
522d829b | 1862 | [this](auto& history, auto& stats) { |
f67539c2 TL |
1863 | dout(10) << "m_pg->recovery_state.update_stats()" << dendl; |
1864 | utime_t now = ceph_clock_now(); | |
1865 | history.last_scrub = m_pg->recovery_state.get_info().last_update; | |
1866 | history.last_scrub_stamp = now; | |
1867 | if (m_is_deep) { | |
1868 | history.last_deep_scrub = m_pg->recovery_state.get_info().last_update; | |
1869 | history.last_deep_scrub_stamp = now; | |
1870 | } | |
1871 | ||
522d829b | 1872 | if (m_is_deep) { |
f67539c2 TL |
1873 | if ((m_shallow_errors == 0) && (m_deep_errors == 0)) |
1874 | history.last_clean_scrub_stamp = now; | |
1875 | stats.stats.sum.num_shallow_scrub_errors = m_shallow_errors; | |
1876 | stats.stats.sum.num_deep_scrub_errors = m_deep_errors; | |
1877 | stats.stats.sum.num_large_omap_objects = m_omap_stats.large_omap_objects; | |
1878 | stats.stats.sum.num_omap_bytes = m_omap_stats.omap_bytes; | |
1879 | stats.stats.sum.num_omap_keys = m_omap_stats.omap_keys; | |
1880 | dout(25) << "scrub_finish shard " << m_pg_whoami | |
1881 | << " num_omap_bytes = " << stats.stats.sum.num_omap_bytes | |
1882 | << " num_omap_keys = " << stats.stats.sum.num_omap_keys << dendl; | |
1883 | } else { | |
1884 | stats.stats.sum.num_shallow_scrub_errors = m_shallow_errors; | |
1885 | // XXX: last_clean_scrub_stamp doesn't mean the pg is not inconsistent | |
1886 | // because of deep-scrub errors | |
1887 | if (m_shallow_errors == 0) | |
1888 | history.last_clean_scrub_stamp = now; | |
1889 | } | |
1890 | stats.stats.sum.num_scrub_errors = stats.stats.sum.num_shallow_scrub_errors + | |
1891 | stats.stats.sum.num_deep_scrub_errors; | |
1892 | if (m_flags.check_repair) { | |
1893 | m_flags.check_repair = false; | |
1894 | if (m_pg->info.stats.stats.sum.num_scrub_errors) { | |
1895 | state_set(PG_STATE_FAILED_REPAIR); | |
1896 | dout(10) << "scrub_finish " << m_pg->info.stats.stats.sum.num_scrub_errors | |
1897 | << " error(s) still present after re-scrub" << dendl; | |
1898 | } | |
1899 | } | |
1900 | return true; | |
1901 | }, | |
1902 | &t); | |
1903 | int tr = m_osds->store->queue_transaction(m_pg->ch, std::move(t), nullptr); | |
1904 | ceph_assert(tr == 0); | |
f67539c2 TL |
1905 | } |
1906 | ||
1907 | if (has_error) { | |
1908 | m_pg->queue_peering_event(PGPeeringEventRef(std::make_shared<PGPeeringEvent>( | |
1909 | get_osdmap_epoch(), get_osdmap_epoch(), PeeringState::DoRecovery()))); | |
1910 | } else { | |
522d829b | 1911 | m_is_repair = false; |
f67539c2 | 1912 | state_clear(PG_STATE_REPAIR); |
522d829b | 1913 | update_op_mode_text(); |
f67539c2 TL |
1914 | } |
1915 | ||
1916 | cleanup_on_finish(); | |
1917 | if (do_auto_scrub) { | |
1918 | request_rescrubbing(m_pg->m_planned_scrub); | |
1919 | } | |
1920 | ||
1921 | if (m_pg->is_active() && m_pg->is_primary()) { | |
1922 | m_pg->recovery_state.share_pg_info(); | |
1923 | } | |
1924 | } | |
1925 | ||
20effc67 | 1926 | void PgScrubber::on_digest_updates() |
f67539c2 | 1927 | { |
20effc67 TL |
1928 | dout(10) << __func__ << " #pending: " << num_digest_updates_pending |
1929 | << (m_end.is_max() ? " <last chunk>" : " <mid chunk>") | |
1930 | << (is_queued_or_active() ? "" : " ** not marked as scrubbing **") | |
1931 | << dendl; | |
f67539c2 | 1932 | |
20effc67 TL |
1933 | if (num_digest_updates_pending > 0) { |
1934 | // do nothing for now. We will be called again when new updates arrive | |
1935 | return; | |
1936 | } | |
f67539c2 | 1937 | |
20effc67 TL |
1938 | // got all updates, and finished with this chunk. Any more? |
1939 | if (m_end.is_max()) { | |
1940 | m_osds->queue_scrub_is_finished(m_pg); | |
f67539c2 | 1941 | } else { |
20effc67 TL |
1942 | // go get a new chunk (via "requeue") |
1943 | preemption_data.reset(); | |
1944 | m_osds->queue_scrub_next_chunk(m_pg, m_pg->is_scrub_blocking_ops()); | |
f67539c2 TL |
1945 | } |
1946 | } | |
1947 | ||
1948 | /* | |
1949 | * note that the flags-set fetched from the PG (m_pg->m_planned_scrub) | |
1950 | * is cleared once scrubbing starts; Some of the values dumped here are | |
1951 | * thus transitory. | |
1952 | */ | |
20effc67 TL |
1953 | void PgScrubber::dump_scrubber(ceph::Formatter* f, |
1954 | const requested_scrub_t& request_flags) const | |
f67539c2 TL |
1955 | { |
1956 | f->open_object_section("scrubber"); | |
20effc67 TL |
1957 | |
1958 | if (m_active) { // TBD replace with PR#42780's test | |
1959 | f->dump_bool("active", true); | |
1960 | dump_active_scrubber(f, state_test(PG_STATE_DEEP_SCRUB)); | |
1961 | } else { | |
1962 | f->dump_bool("active", false); | |
1963 | f->dump_bool("must_scrub", | |
1964 | (m_pg->m_planned_scrub.must_scrub || m_flags.required)); | |
1965 | f->dump_bool("must_deep_scrub", request_flags.must_deep_scrub); | |
1966 | f->dump_bool("must_repair", request_flags.must_repair); | |
1967 | f->dump_bool("need_auto", request_flags.need_auto); | |
1968 | ||
1969 | f->dump_stream("scrub_reg_stamp") << m_scrub_job->get_sched_time(); | |
1970 | ||
1971 | // note that we are repeating logic that is coded elsewhere (currently PG.cc). | |
1972 | // This is not optimal. | |
1973 | bool deep_expected = (ceph_clock_now() >= m_pg->next_deepscrub_interval()) || | |
1974 | request_flags.must_deep_scrub || request_flags.need_auto; | |
1975 | auto sched_state = | |
1976 | m_scrub_job->scheduling_state(ceph_clock_now(), deep_expected); | |
1977 | f->dump_string("schedule", sched_state); | |
1978 | } | |
1979 | ||
1980 | if (m_publish_sessions) { | |
1981 | f->dump_int("test_sequence", | |
1982 | m_sessions_counter); // an ever-increasing number used by tests | |
1983 | } | |
1984 | ||
1985 | f->close_section(); | |
1986 | } | |
1987 | ||
1988 | void PgScrubber::dump_active_scrubber(ceph::Formatter* f, bool is_deep) const | |
1989 | { | |
f67539c2 | 1990 | f->dump_stream("epoch_start") << m_interval_start; |
20effc67 TL |
1991 | f->dump_stream("start") << m_start; |
1992 | f->dump_stream("end") << m_end; | |
1993 | f->dump_stream("max_end") << m_max_end; | |
1994 | f->dump_stream("subset_last_update") << m_subset_last_update; | |
1995 | // note that m_is_deep will be set some time after PG_STATE_DEEP_SCRUB is | |
1996 | // asserted. Thus, using the latter. | |
1997 | f->dump_bool("deep", is_deep); | |
1998 | ||
1999 | // dump the scrub-type flags | |
2000 | f->dump_bool("req_scrub", m_flags.required); | |
2001 | f->dump_bool("auto_repair", m_flags.auto_repair); | |
2002 | f->dump_bool("check_repair", m_flags.check_repair); | |
2003 | f->dump_bool("deep_scrub_on_error", m_flags.deep_scrub_on_error); | |
2004 | f->dump_unsigned("priority", m_flags.priority); | |
2005 | ||
2006 | f->dump_int("shallow_errors", m_shallow_errors); | |
2007 | f->dump_int("deep_errors", m_deep_errors); | |
2008 | f->dump_int("fixed", m_fixed_count); | |
2009 | { | |
2010 | f->open_array_section("waiting_on_whom"); | |
2011 | for (const auto& p : m_maps_status.get_awaited()) { | |
2012 | f->dump_stream("shard") << p; | |
f67539c2 | 2013 | } |
20effc67 | 2014 | f->close_section(); |
f67539c2 | 2015 | } |
20effc67 | 2016 | f->dump_string("schedule", "scrubbing"); |
f67539c2 TL |
2017 | } |
2018 | ||
20effc67 TL |
2019 | pg_scrubbing_status_t PgScrubber::get_schedule() const |
2020 | { | |
2021 | dout(25) << __func__ << dendl; | |
2022 | ||
2023 | if (!m_scrub_job) { | |
2024 | return pg_scrubbing_status_t{}; | |
2025 | } | |
2026 | ||
2027 | auto now_is = ceph_clock_now(); | |
2028 | ||
2029 | if (m_active) { | |
2030 | // report current scrub info, including updated duration | |
2031 | int32_t duration = (utime_t{now_is} - scrub_begin_stamp).sec(); | |
2032 | ||
2033 | return pg_scrubbing_status_t{ | |
2034 | utime_t{}, | |
2035 | duration, | |
2036 | pg_scrub_sched_status_t::active, | |
2037 | true, // active | |
2038 | (m_is_deep ? scrub_level_t::deep : scrub_level_t::shallow), | |
2039 | false /* is periodic? unknown, actually */}; | |
2040 | } | |
2041 | if (m_scrub_job->state != ScrubQueue::qu_state_t::registered) { | |
2042 | return pg_scrubbing_status_t{utime_t{}, | |
2043 | 0, | |
2044 | pg_scrub_sched_status_t::not_queued, | |
2045 | false, | |
2046 | scrub_level_t::shallow, | |
2047 | false}; | |
2048 | } | |
2049 | ||
2050 | // Will next scrub surely be a deep one? note that deep-scrub might be | |
2051 | // selected even if we report a regular scrub here. | |
2052 | bool deep_expected = (now_is >= m_pg->next_deepscrub_interval()) || | |
2053 | m_pg->m_planned_scrub.must_deep_scrub || | |
2054 | m_pg->m_planned_scrub.need_auto; | |
2055 | scrub_level_t expected_level = | |
2056 | deep_expected ? scrub_level_t::deep : scrub_level_t::shallow; | |
2057 | bool periodic = !m_pg->m_planned_scrub.must_scrub && | |
2058 | !m_pg->m_planned_scrub.need_auto && | |
2059 | !m_pg->m_planned_scrub.must_deep_scrub; | |
2060 | ||
2061 | // are we ripe for scrubbing? | |
2062 | if (now_is > m_scrub_job->schedule.scheduled_at) { | |
2063 | // we are waiting for our turn at the OSD. | |
2064 | return pg_scrubbing_status_t{m_scrub_job->schedule.scheduled_at, | |
2065 | 0, | |
2066 | pg_scrub_sched_status_t::queued, | |
2067 | false, | |
2068 | expected_level, | |
2069 | periodic}; | |
2070 | } | |
2071 | ||
2072 | return pg_scrubbing_status_t{m_scrub_job->schedule.scheduled_at, | |
2073 | 0, | |
2074 | pg_scrub_sched_status_t::scheduled, | |
2075 | false, | |
2076 | expected_level, | |
2077 | periodic}; | |
2078 | } | |
f67539c2 TL |
2079 | |
2080 | void PgScrubber::handle_query_state(ceph::Formatter* f) | |
2081 | { | |
20effc67 | 2082 | dout(15) << __func__ << dendl; |
f67539c2 TL |
2083 | |
2084 | f->open_object_section("scrub"); | |
2085 | f->dump_stream("scrubber.epoch_start") << m_interval_start; | |
2086 | f->dump_bool("scrubber.active", m_active); | |
2087 | f->dump_stream("scrubber.start") << m_start; | |
2088 | f->dump_stream("scrubber.end") << m_end; | |
20effc67 TL |
2089 | f->dump_stream("scrubber.max_end") << m_max_end; |
2090 | f->dump_stream("scrubber.subset_last_update") << m_subset_last_update; | |
f67539c2 TL |
2091 | f->dump_bool("scrubber.deep", m_is_deep); |
2092 | { | |
2093 | f->open_array_section("scrubber.waiting_on_whom"); | |
2094 | for (const auto& p : m_maps_status.get_awaited()) { | |
2095 | f->dump_stream("shard") << p; | |
2096 | } | |
2097 | f->close_section(); | |
2098 | } | |
2099 | ||
2100 | f->dump_string("comment", "DEPRECATED - may be removed in the next release"); | |
2101 | ||
2102 | f->close_section(); | |
2103 | } | |
2104 | ||
20effc67 TL |
2105 | PgScrubber::~PgScrubber() |
2106 | { | |
2107 | if (m_scrub_job) { | |
2108 | // make sure the OSD won't try to scrub this one just now | |
2109 | rm_from_osd_scrubbing(); | |
2110 | m_scrub_job.reset(); | |
2111 | } | |
2112 | } | |
f67539c2 TL |
2113 | |
2114 | PgScrubber::PgScrubber(PG* pg) | |
2115 | : m_pg{pg} | |
2116 | , m_pg_id{pg->pg_id} | |
2117 | , m_osds{m_pg->osd} | |
2118 | , m_pg_whoami{pg->pg_whoami} | |
2119 | , preemption_data{pg} | |
2120 | { | |
f67539c2 TL |
2121 | m_fsm = std::make_unique<ScrubMachine>(m_pg, this); |
2122 | m_fsm->initiate(); | |
20effc67 TL |
2123 | |
2124 | m_scrub_job = ceph::make_ref<ScrubQueue::ScrubJob>(m_osds->cct, m_pg->pg_id, | |
2125 | m_osds->get_nodeid()); | |
2126 | } | |
2127 | ||
2128 | void PgScrubber::set_scrub_begin_time() | |
2129 | { | |
2130 | scrub_begin_stamp = ceph_clock_now(); | |
2a845540 TL |
2131 | m_osds->clog->debug() << fmt::format( |
2132 | "{} {} starts", | |
2133 | m_pg->info.pgid.pgid, | |
2134 | m_mode_desc); | |
20effc67 TL |
2135 | } |
2136 | ||
2137 | void PgScrubber::set_scrub_duration() | |
2138 | { | |
2139 | utime_t stamp = ceph_clock_now(); | |
2140 | utime_t duration = stamp - scrub_begin_stamp; | |
2141 | m_pg->recovery_state.update_stats([=](auto& history, auto& stats) { | |
2142 | stats.last_scrub_duration = ceill(duration.to_msec()/1000.0); | |
1d09f67e | 2143 | stats.scrub_duration = double(duration); |
20effc67 TL |
2144 | return true; |
2145 | }); | |
f67539c2 TL |
2146 | } |
2147 | ||
2148 | void PgScrubber::reserve_replicas() | |
2149 | { | |
2150 | dout(10) << __func__ << dendl; | |
20effc67 | 2151 | m_reservations.emplace(m_pg, m_pg_whoami, m_scrub_job); |
f67539c2 TL |
2152 | } |
2153 | ||
2154 | void PgScrubber::cleanup_on_finish() | |
2155 | { | |
2156 | dout(10) << __func__ << dendl; | |
2157 | ceph_assert(m_pg->is_locked()); | |
2158 | ||
2159 | state_clear(PG_STATE_SCRUBBING); | |
2160 | state_clear(PG_STATE_DEEP_SCRUB); | |
2161 | m_pg->publish_stats_to_osd(); | |
2162 | ||
2163 | clear_scrub_reservations(); | |
2164 | m_pg->publish_stats_to_osd(); | |
2165 | ||
2166 | requeue_waiting(); | |
2167 | ||
2168 | reset_internal_state(); | |
1d09f67e | 2169 | m_pg->publish_stats_to_osd(); |
f67539c2 TL |
2170 | m_flags = scrub_flags_t{}; |
2171 | ||
2172 | // type-specific state clear | |
2173 | _scrub_clear_state(); | |
2174 | } | |
2175 | ||
2176 | // uses process_event(), so must be invoked externally | |
2177 | void PgScrubber::scrub_clear_state() | |
2178 | { | |
2179 | dout(10) << __func__ << dendl; | |
2180 | ||
2181 | clear_pgscrub_state(); | |
2182 | m_fsm->process_event(FullReset{}); | |
2183 | } | |
2184 | ||
2185 | /* | |
2186 | * note: does not access the state-machine | |
2187 | */ | |
2188 | void PgScrubber::clear_pgscrub_state() | |
2189 | { | |
2190 | dout(10) << __func__ << dendl; | |
2191 | ceph_assert(m_pg->is_locked()); | |
2192 | ||
2193 | state_clear(PG_STATE_SCRUBBING); | |
2194 | state_clear(PG_STATE_DEEP_SCRUB); | |
2195 | ||
2196 | state_clear(PG_STATE_REPAIR); | |
2197 | ||
2198 | clear_scrub_reservations(); | |
2199 | m_pg->publish_stats_to_osd(); | |
2200 | ||
2201 | requeue_waiting(); | |
2202 | ||
2203 | reset_internal_state(); | |
2204 | m_flags = scrub_flags_t{}; | |
2205 | ||
2206 | // type-specific state clear | |
2207 | _scrub_clear_state(); | |
1d09f67e | 2208 | m_pg->publish_stats_to_osd(); |
f67539c2 TL |
2209 | } |
2210 | ||
2211 | void PgScrubber::replica_handling_done() | |
2212 | { | |
2213 | dout(10) << __func__ << dendl; | |
2214 | ||
2215 | state_clear(PG_STATE_SCRUBBING); | |
2216 | state_clear(PG_STATE_DEEP_SCRUB); | |
2217 | ||
2218 | reset_internal_state(); | |
f67539c2 TL |
2219 | } |
2220 | ||
2221 | /* | |
2222 | * note: performs run_callbacks() | |
2223 | * note: reservations-related variables are not reset here | |
2224 | */ | |
2225 | void PgScrubber::reset_internal_state() | |
2226 | { | |
2227 | dout(10) << __func__ << dendl; | |
2228 | ||
2229 | preemption_data.reset(); | |
2230 | m_maps_status.reset(); | |
2231 | m_received_maps.clear(); | |
2232 | ||
2233 | m_start = hobject_t{}; | |
2234 | m_end = hobject_t{}; | |
2235 | m_max_end = hobject_t{}; | |
2236 | m_subset_last_update = eversion_t{}; | |
2237 | m_shallow_errors = 0; | |
2238 | m_deep_errors = 0; | |
2239 | m_fixed_count = 0; | |
2240 | m_omap_stats = (const struct omap_stat_t){0}; | |
2241 | ||
2242 | run_callbacks(); | |
2243 | ||
2244 | m_inconsistent.clear(); | |
2245 | m_missing.clear(); | |
2246 | m_authoritative.clear(); | |
2247 | num_digest_updates_pending = 0; | |
2248 | m_primary_scrubmap = ScrubMap{}; | |
2249 | m_primary_scrubmap_pos.reset(); | |
2250 | replica_scrubmap = ScrubMap{}; | |
2251 | replica_scrubmap_pos.reset(); | |
2252 | m_cleaned_meta_map = ScrubMap{}; | |
2253 | m_needs_sleep = true; | |
2254 | m_sleep_started_at = utime_t{}; | |
2255 | ||
2256 | m_active = false; | |
20effc67 TL |
2257 | clear_queued_or_active(); |
2258 | ++m_sessions_counter; | |
2259 | } | |
2260 | ||
2261 | // note that only applicable to the Replica: | |
2262 | void PgScrubber::advance_token() | |
2263 | { | |
2264 | dout(10) << __func__ << " was: " << m_current_token << dendl; | |
2265 | m_current_token++; | |
2266 | ||
2267 | // when advance_token() is called, it is assumed that no scrubbing takes place. | |
2268 | // We will, though, verify that. And if we are actually still handling a stale request - | |
2269 | // both our internal state and the FSM state will be cleared. | |
2270 | replica_handling_done(); | |
2271 | m_fsm->process_event(FullReset{}); | |
2272 | } | |
2273 | ||
2274 | bool PgScrubber::is_token_current(Scrub::act_token_t received_token) | |
2275 | { | |
2276 | if (received_token == 0 || received_token == m_current_token) { | |
2277 | return true; | |
2278 | } | |
2279 | dout(5) << __func__ << " obsolete token (" << received_token | |
2280 | << " vs current " << m_current_token << dendl; | |
2281 | ||
2282 | return false; | |
f67539c2 TL |
2283 | } |
2284 | ||
2285 | const OSDMapRef& PgScrubber::get_osdmap() const | |
2286 | { | |
2287 | return m_pg->get_osdmap(); | |
2288 | } | |
2289 | ||
2290 | ostream& operator<<(ostream& out, const PgScrubber& scrubber) | |
2291 | { | |
2292 | return out << scrubber.m_flags; | |
2293 | } | |
2294 | ||
20effc67 TL |
2295 | std::ostream& PgScrubber::gen_prefix(std::ostream& out) const |
2296 | { | |
2297 | const auto fsm_state = m_fsm ? m_fsm->current_states_desc() : "- :"; | |
2298 | if (m_pg) { | |
2299 | return m_pg->gen_prefix(out) << "scrubber " << fsm_state << ": "; | |
2300 | } else { | |
2301 | return out << " scrubber [~] " << fsm_state << ": "; | |
2302 | } | |
2303 | } | |
2304 | ||
33c7a0ef TL |
2305 | void PgScrubber::log_cluster_warning(const std::string& warning) const |
2306 | { | |
2307 | m_osds->clog->do_log(CLOG_WARN, warning); | |
2308 | } | |
2309 | ||
f67539c2 TL |
2310 | ostream& PgScrubber::show(ostream& out) const |
2311 | { | |
2312 | return out << " [ " << m_pg_id << ": " << m_flags << " ] "; | |
2313 | } | |
2314 | ||
20effc67 TL |
2315 | int PgScrubber::asok_debug(std::string_view cmd, |
2316 | std::string param, | |
2317 | Formatter* f, | |
2318 | stringstream& ss) | |
2319 | { | |
2320 | dout(10) << __func__ << " cmd: " << cmd << " param: " << param << dendl; | |
2321 | ||
2322 | if (cmd == "block") { | |
2323 | // set a flag that will cause the next 'select_range' to report a blocked object | |
2324 | m_debug_blockrange = 1; | |
2325 | ||
2326 | } else if (cmd == "unblock") { | |
2327 | // send an 'unblock' event, as if a blocked range was freed | |
2328 | m_debug_blockrange = 0; | |
2329 | m_fsm->process_event(Unblocked{}); | |
2330 | ||
2331 | } else if ((cmd == "set") || (cmd == "unset")) { | |
2332 | ||
2333 | if (param == "sessions") { | |
2334 | // set/reset the inclusion of the scrub sessions counter in 'query' output | |
2335 | m_publish_sessions = (cmd == "set"); | |
2336 | ||
2337 | } else if (param == "block") { | |
2338 | if (cmd == "set") { | |
2339 | // set a flag that will cause the next 'select_range' to report a blocked object | |
2340 | m_debug_blockrange = 1; | |
2341 | } else { | |
2342 | // send an 'unblock' event, as if a blocked range was freed | |
2343 | m_debug_blockrange = 0; | |
2344 | m_fsm->process_event(Unblocked{}); | |
2345 | } | |
2346 | } | |
2347 | } | |
2348 | ||
2349 | return 0; | |
2350 | } | |
f67539c2 TL |
2351 | // ///////////////////// preemption_data_t ////////////////////////////////// |
2352 | ||
2353 | PgScrubber::preemption_data_t::preemption_data_t(PG* pg) : m_pg{pg} | |
2354 | { | |
2355 | m_left = static_cast<int>( | |
2356 | m_pg->get_cct()->_conf.get_val<uint64_t>("osd_scrub_max_preemptions")); | |
2357 | } | |
2358 | ||
2359 | void PgScrubber::preemption_data_t::reset() | |
2360 | { | |
2361 | std::lock_guard<std::mutex> lk{m_preemption_lock}; | |
2362 | ||
2363 | m_preemptable = false; | |
2364 | m_preempted = false; | |
2365 | m_left = | |
2366 | static_cast<int>(m_pg->cct->_conf.get_val<uint64_t>("osd_scrub_max_preemptions")); | |
2367 | m_size_divisor = 1; | |
2368 | } | |
2369 | ||
2370 | ||
2371 | // ///////////////////// ReplicaReservations ////////////////////////////////// | |
2372 | namespace Scrub { | |
2373 | ||
2374 | void ReplicaReservations::release_replica(pg_shard_t peer, epoch_t epoch) | |
2375 | { | |
20effc67 | 2376 | auto m = new MOSDScrubReserve(spg_t(m_pg_info.pgid.pgid, peer.shard), epoch, |
f67539c2 TL |
2377 | MOSDScrubReserve::RELEASE, m_pg->pg_whoami); |
2378 | m_osds->send_message_osd_cluster(peer.osd, m, epoch); | |
2379 | } | |
2380 | ||
20effc67 | 2381 | ReplicaReservations::ReplicaReservations(PG* pg, pg_shard_t whoami, ScrubQueue::ScrubJobRef scrubjob) |
f67539c2 TL |
2382 | : m_pg{pg} |
2383 | , m_acting_set{pg->get_actingset()} | |
20effc67 | 2384 | , m_osds{m_pg->get_pg_osd(ScrubberPasskey())} |
f67539c2 | 2385 | , m_pending{static_cast<int>(m_acting_set.size()) - 1} |
20effc67 TL |
2386 | , m_pg_info{m_pg->get_pg_info(ScrubberPasskey())} |
2387 | , m_scrub_job{scrubjob} | |
f67539c2 TL |
2388 | { |
2389 | epoch_t epoch = m_pg->get_osdmap_epoch(); | |
2390 | ||
20effc67 TL |
2391 | { |
2392 | std::stringstream prefix; | |
2393 | prefix << "osd." << m_osds->whoami << " ep: " << epoch | |
2394 | << " scrubber::ReplicaReservations pg[" << pg->pg_id << "]: "; | |
2395 | m_log_msg_prefix = prefix.str(); | |
2396 | } | |
2397 | ||
f67539c2 TL |
2398 | // handle the special case of no replicas |
2399 | if (m_pending <= 0) { | |
2400 | // just signal the scrub state-machine to continue | |
2401 | send_all_done(); | |
2402 | ||
2403 | } else { | |
2404 | ||
2405 | for (auto p : m_acting_set) { | |
2406 | if (p == whoami) | |
2407 | continue; | |
20effc67 | 2408 | auto m = new MOSDScrubReserve(spg_t(m_pg_info.pgid.pgid, p.shard), epoch, |
f67539c2 TL |
2409 | MOSDScrubReserve::REQUEST, m_pg->pg_whoami); |
2410 | m_osds->send_message_osd_cluster(p.osd, m, epoch); | |
2411 | m_waited_for_peers.push_back(p); | |
20effc67 | 2412 | dout(10) << __func__ << ": reserve " << p.osd << dendl; |
f67539c2 TL |
2413 | } |
2414 | } | |
2415 | } | |
2416 | ||
2417 | void ReplicaReservations::send_all_done() | |
2418 | { | |
2419 | m_osds->queue_for_scrub_granted(m_pg, scrub_prio_t::low_priority); | |
2420 | } | |
2421 | ||
2422 | void ReplicaReservations::send_reject() | |
2423 | { | |
20effc67 | 2424 | m_scrub_job->resources_failure = true; |
f67539c2 TL |
2425 | m_osds->queue_for_scrub_denied(m_pg, scrub_prio_t::low_priority); |
2426 | } | |
2427 | ||
2428 | void ReplicaReservations::discard_all() | |
2429 | { | |
20effc67 | 2430 | dout(10) << __func__ << ": " << m_reserved_peers << dendl; |
f67539c2 TL |
2431 | |
2432 | m_had_rejections = true; // preventing late-coming responses from triggering events | |
2433 | m_reserved_peers.clear(); | |
2434 | m_waited_for_peers.clear(); | |
2435 | } | |
2436 | ||
2437 | ReplicaReservations::~ReplicaReservations() | |
2438 | { | |
2439 | m_had_rejections = true; // preventing late-coming responses from triggering events | |
2440 | ||
2441 | // send un-reserve messages to all reserved replicas. We do not wait for answer (there | |
2442 | // wouldn't be one). Other incoming messages will be discarded on the way, by our | |
2443 | // owner. | |
2444 | epoch_t epoch = m_pg->get_osdmap_epoch(); | |
2445 | ||
2446 | for (auto& p : m_reserved_peers) { | |
2447 | release_replica(p, epoch); | |
2448 | } | |
2449 | m_reserved_peers.clear(); | |
2450 | ||
2451 | // note: the release will follow on the heels of the request. When tried otherwise, | |
2452 | // grants that followed a reject arrived after the whole scrub machine-state was | |
2453 | // reset, causing leaked reservations. | |
2454 | for (auto& p : m_waited_for_peers) { | |
2455 | release_replica(p, epoch); | |
2456 | } | |
2457 | m_waited_for_peers.clear(); | |
2458 | } | |
2459 | ||
2460 | /** | |
2461 | * @ATTN we would not reach here if the ReplicaReservation object managed by the | |
2462 | * scrubber was reset. | |
2463 | */ | |
2464 | void ReplicaReservations::handle_reserve_grant(OpRequestRef op, pg_shard_t from) | |
2465 | { | |
20effc67 | 2466 | dout(10) << __func__ << ": granted by " << from << dendl; |
f67539c2 TL |
2467 | op->mark_started(); |
2468 | ||
2469 | { | |
2470 | // reduce the amount of extra release messages. Not a must, but the log is cleaner | |
2471 | auto w = find(m_waited_for_peers.begin(), m_waited_for_peers.end(), from); | |
2472 | if (w != m_waited_for_peers.end()) | |
2473 | m_waited_for_peers.erase(w); | |
2474 | } | |
2475 | ||
2476 | // are we forced to reject the reservation? | |
2477 | if (m_had_rejections) { | |
2478 | ||
20effc67 TL |
2479 | dout(10) << __func__ << ": rejecting late-coming reservation from " |
2480 | << from << dendl; | |
f67539c2 TL |
2481 | release_replica(from, m_pg->get_osdmap_epoch()); |
2482 | ||
2483 | } else if (std::find(m_reserved_peers.begin(), m_reserved_peers.end(), from) != | |
2484 | m_reserved_peers.end()) { | |
2485 | ||
20effc67 | 2486 | dout(10) << __func__ << ": already had osd." << from << " reserved" << dendl; |
f67539c2 TL |
2487 | |
2488 | } else { | |
2489 | ||
20effc67 TL |
2490 | dout(10) << __func__ << ": osd." << from << " scrub reserve = success" |
2491 | << dendl; | |
f67539c2 TL |
2492 | m_reserved_peers.push_back(from); |
2493 | if (--m_pending == 0) { | |
2494 | send_all_done(); | |
2495 | } | |
2496 | } | |
2497 | } | |
2498 | ||
2499 | void ReplicaReservations::handle_reserve_reject(OpRequestRef op, pg_shard_t from) | |
2500 | { | |
20effc67 TL |
2501 | dout(10) << __func__ << ": rejected by " << from << dendl; |
2502 | dout(15) << __func__ << ": " << *op->get_req() << dendl; | |
f67539c2 TL |
2503 | op->mark_started(); |
2504 | ||
2505 | { | |
2506 | // reduce the amount of extra release messages. Not a must, but the log is cleaner | |
2507 | auto w = find(m_waited_for_peers.begin(), m_waited_for_peers.end(), from); | |
2508 | if (w != m_waited_for_peers.end()) | |
2509 | m_waited_for_peers.erase(w); | |
2510 | } | |
2511 | ||
2512 | if (m_had_rejections) { | |
2513 | ||
2514 | // our failure was already handled when the first rejection arrived | |
20effc67 TL |
2515 | dout(15) << __func__ << ": ignoring late-coming rejection from " |
2516 | << from << dendl; | |
f67539c2 TL |
2517 | |
2518 | } else if (std::find(m_reserved_peers.begin(), m_reserved_peers.end(), from) != | |
2519 | m_reserved_peers.end()) { | |
2520 | ||
20effc67 | 2521 | dout(10) << __func__ << ": already had osd." << from << " reserved" << dendl; |
f67539c2 TL |
2522 | |
2523 | } else { | |
2524 | ||
20effc67 | 2525 | dout(10) << __func__ << ": osd." << from << " scrub reserve = fail" << dendl; |
f67539c2 TL |
2526 | m_had_rejections = true; // preventing any additional notifications |
2527 | send_reject(); | |
2528 | } | |
2529 | } | |
2530 | ||
20effc67 TL |
2531 | std::ostream& ReplicaReservations::gen_prefix(std::ostream& out) const |
2532 | { | |
2533 | return out << m_log_msg_prefix; | |
2534 | } | |
f67539c2 TL |
2535 | |
2536 | // ///////////////////// LocalReservation ////////////////////////////////// | |
2537 | ||
20effc67 TL |
2538 | // note: no dout()s in LocalReservation functions. Client logs interactions. |
2539 | LocalReservation::LocalReservation(OSDService* osds) | |
2540 | : m_osds{osds} | |
f67539c2 | 2541 | { |
20effc67 | 2542 | if (m_osds->get_scrub_services().inc_scrubs_local()) { |
f67539c2 | 2543 | // the failure is signalled by not having m_holding_local_reservation set |
20effc67 | 2544 | m_holding_local_reservation = true; |
f67539c2 | 2545 | } |
f67539c2 TL |
2546 | } |
2547 | ||
2548 | LocalReservation::~LocalReservation() | |
2549 | { | |
2550 | if (m_holding_local_reservation) { | |
2551 | m_holding_local_reservation = false; | |
20effc67 | 2552 | m_osds->get_scrub_services().dec_scrubs_local(); |
f67539c2 TL |
2553 | } |
2554 | } | |
2555 | ||
f67539c2 TL |
2556 | // ///////////////////// ReservedByRemotePrimary /////////////////////////////// |
2557 | ||
20effc67 TL |
2558 | ReservedByRemotePrimary::ReservedByRemotePrimary(const PgScrubber* scrubber, |
2559 | PG* pg, | |
2560 | OSDService* osds, | |
2561 | epoch_t epoch) | |
2562 | : m_scrubber{scrubber} | |
2563 | , m_pg{pg} | |
2564 | , m_osds{osds} | |
2565 | , m_reserved_at{epoch} | |
f67539c2 | 2566 | { |
20effc67 | 2567 | if (!m_osds->get_scrub_services().inc_scrubs_remote()) { |
f67539c2 TL |
2568 | dout(10) << __func__ << ": failed to reserve at Primary request" << dendl; |
2569 | // the failure is signalled by not having m_reserved_by_remote_primary set | |
2570 | return; | |
2571 | } | |
2572 | ||
2573 | dout(20) << __func__ << ": scrub resources reserved at Primary request" << dendl; | |
2574 | m_reserved_by_remote_primary = true; | |
2575 | } | |
2576 | ||
2577 | bool ReservedByRemotePrimary::is_stale() const | |
2578 | { | |
2579 | return m_reserved_at < m_pg->get_same_interval_since(); | |
2580 | } | |
2581 | ||
2582 | ReservedByRemotePrimary::~ReservedByRemotePrimary() | |
2583 | { | |
2584 | if (m_reserved_by_remote_primary) { | |
2585 | m_reserved_by_remote_primary = false; | |
20effc67 | 2586 | m_osds->get_scrub_services().dec_scrubs_remote(); |
f67539c2 TL |
2587 | } |
2588 | } | |
2589 | ||
20effc67 TL |
2590 | std::ostream& ReservedByRemotePrimary::gen_prefix(std::ostream& out) const |
2591 | { | |
2592 | return m_scrubber->gen_prefix(out); | |
2593 | } | |
2594 | ||
f67539c2 TL |
2595 | // ///////////////////// MapsCollectionStatus //////////////////////////////// |
2596 | ||
2597 | auto MapsCollectionStatus::mark_arriving_map(pg_shard_t from) | |
2598 | -> std::tuple<bool, std::string_view> | |
2599 | { | |
2600 | auto fe = std::find(m_maps_awaited_for.begin(), m_maps_awaited_for.end(), from); | |
2601 | if (fe != m_maps_awaited_for.end()) { | |
2602 | // we are indeed waiting for a map from this replica | |
2603 | m_maps_awaited_for.erase(fe); | |
2604 | return std::tuple{true, ""sv}; | |
2605 | } else { | |
2606 | return std::tuple{false, " unsolicited scrub-map"sv}; | |
2607 | } | |
2608 | } | |
2609 | ||
2610 | void MapsCollectionStatus::reset() | |
2611 | { | |
2612 | *this = MapsCollectionStatus{}; | |
2613 | } | |
2614 | ||
2615 | std::string MapsCollectionStatus::dump() const | |
2616 | { | |
2617 | std::string all; | |
2618 | for (const auto& rp : m_maps_awaited_for) { | |
2619 | all.append(rp.get_osd() + " "s); | |
2620 | } | |
2621 | return all; | |
2622 | } | |
2623 | ||
2624 | ostream& operator<<(ostream& out, const MapsCollectionStatus& sf) | |
2625 | { | |
2626 | out << " [ "; | |
2627 | for (const auto& rp : sf.m_maps_awaited_for) { | |
2628 | out << rp.get_osd() << " "; | |
2629 | } | |
2630 | if (!sf.m_local_map_ready) { | |
2631 | out << " local "; | |
2632 | } | |
2633 | return out << " ] "; | |
2634 | } | |
2635 | ||
20effc67 TL |
2636 | // ///////////////////// blocked_range_t /////////////////////////////// |
2637 | ||
2638 | blocked_range_t::blocked_range_t(OSDService* osds, ceph::timespan waittime, spg_t pg_id) | |
2639 | : m_osds{osds} | |
2640 | { | |
2641 | auto now_is = std::chrono::system_clock::now(); | |
2642 | m_callbk = new LambdaContext([now_is, pg_id, osds]([[maybe_unused]] int r) { | |
2643 | std::time_t now_c = std::chrono::system_clock::to_time_t(now_is); | |
2644 | char buf[50]; | |
2645 | strftime(buf, sizeof(buf), "%Y-%m-%dT%H:%M:%S", std::localtime(&now_c)); | |
2646 | lgeneric_subdout(g_ceph_context, osd, 10) | |
2647 | << "PgScrubber: " << pg_id << " blocked on an object for too long (since " << buf | |
2648 | << ")" << dendl; | |
2649 | osds->clog->warn() << "osd." << osds->whoami << " PgScrubber: " << pg_id << " blocked on an object for too long (since " << buf << ")"; | |
2650 | return; | |
2651 | }); | |
2652 | ||
2653 | std::lock_guard l(m_osds->sleep_lock); | |
2654 | m_osds->sleep_timer.add_event_after(waittime, m_callbk); | |
2655 | } | |
2656 | ||
2657 | blocked_range_t::~blocked_range_t() | |
2658 | { | |
2659 | std::lock_guard l(m_osds->sleep_lock); | |
2660 | m_osds->sleep_timer.cancel_event(m_callbk); | |
2661 | } | |
2662 | ||
f67539c2 | 2663 | } // namespace Scrub |