1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
2 // vim: ts=8 sw=2 smarttab
4 * Ceph - scalable distributed file system
6 * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
8 * This is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License version 2.1, as published by the Free Software
11 * Foundation. See file COPYING.
18 #include <boost/statechart/custom_reaction.hpp>
19 #include <boost/statechart/event.hpp>
20 #include <boost/statechart/simple_state.hpp>
21 #include <boost/statechart/state.hpp>
22 #include <boost/statechart/state_machine.hpp>
23 #include <boost/statechart/transition.hpp>
24 #include <boost/statechart/event_base.hpp>
25 #include <boost/scoped_ptr.hpp>
26 #include <boost/circular_buffer.hpp>
27 #include "include/memory.h"
28 #include "include/mempool.h"
30 // re-include our assert to clobber boost's
31 #include "include/assert.h"
33 #include "include/types.h"
34 #include "include/stringify.h"
35 #include "osd_types.h"
36 #include "include/xlist.h"
37 #include "SnapMapper.h"
39 #include "common/Timer.h"
43 #include "messages/MOSDPGLog.h"
44 #include "include/str_list.h"
45 #include "PGBackend.h"
55 // #include "include/unordered_map.h"
56 // #include "include/unordered_set.h"
58 //#define DEBUG_RECOVERY_OIDS // track set of recovering oids explicitly, to find counting bugs
69 typedef OpRequest::Ref OpRequestRef
;
77 void intrusive_ptr_add_ref(PG
*pg
);
78 void intrusive_ptr_release(PG
*pg
);
80 using state_history_entry
= std::tuple
<utime_t
, utime_t
, const char*>;
81 using embedded_state
= std::pair
<utime_t
, const char*>;
83 struct PGStateInstance
{
84 // Time spent in pg states
86 void setepoch(const epoch_t current_epoch
) {
87 this_epoch
= current_epoch
;
90 void enter_state(const utime_t entime
, const char* state
) {
91 embedded_states
.push(std::make_pair(entime
, state
));
94 void exit_state(const utime_t extime
) {
95 embedded_state this_state
= embedded_states
.top();
96 state_history
.push_back(state_history_entry
{
97 this_state
.first
, extime
, this_state
.second
});
98 embedded_states
.pop();
103 std::vector
<state_history_entry
> state_history
;
104 std::stack
<embedded_state
> embedded_states
;
107 class PGStateHistory
{
108 // Member access protected with the PG lock
110 PGStateHistory() : buffer(10) {}
112 void enter(PG
* pg
, const utime_t entime
, const char* state
);
114 void exit(const char* state
);
120 void set_pg_in_destructor() { pg_in_destructor
= true; }
122 void dump(Formatter
* f
) const;
125 bool pg_in_destructor
= false;
126 PG
* thispg
= nullptr;
127 std::unique_ptr
<PGStateInstance
> tmppi
;
128 PGStateInstance
* pi
= nullptr;
129 boost::circular_buffer
<std::unique_ptr
<PGStateInstance
>> buffer
;
134 #include "common/tracked_int_ptr.hpp"
135 uint64_t get_with_id(PG
*pg
);
136 void put_with_id(PG
*pg
, uint64_t id
);
137 typedef TrackedIntPtr
<PG
> PGRef
;
139 typedef boost::intrusive_ptr
<PG
> PGRef
;
142 class PGRecoveryStats
{
143 struct per_state_info
{
144 uint64_t enter
, exit
; // enter/exit counts
146 utime_t event_time
; // time spent processing events
147 utime_t total_time
; // total time in state
148 utime_t min_time
, max_time
;
150 // cppcheck-suppress unreachableCode
151 per_state_info() : enter(0), exit(0), events(0) {}
153 map
<const char *,per_state_info
> info
;
157 PGRecoveryStats() : lock("PGRecoverStats::lock") {}
160 Mutex::Locker
l(lock
);
163 void dump(ostream
& out
) {
164 Mutex::Locker
l(lock
);
165 for (map
<const char *,per_state_info
>::iterator p
= info
.begin(); p
!= info
.end(); ++p
) {
166 per_state_info
& i
= p
->second
;
167 out
<< i
.enter
<< "\t" << i
.exit
<< "\t"
168 << i
.events
<< "\t" << i
.event_time
<< "\t"
169 << i
.total_time
<< "\t"
170 << i
.min_time
<< "\t" << i
.max_time
<< "\t"
175 void dump_formatted(Formatter
*f
) {
176 Mutex::Locker
l(lock
);
177 f
->open_array_section("pg_recovery_stats");
178 for (map
<const char *,per_state_info
>::iterator p
= info
.begin();
179 p
!= info
.end(); ++p
) {
180 per_state_info
& i
= p
->second
;
181 f
->open_object_section("recovery_state");
182 f
->dump_int("enter", i
.enter
);
183 f
->dump_int("exit", i
.exit
);
184 f
->dump_int("events", i
.events
);
185 f
->dump_stream("event_time") << i
.event_time
;
186 f
->dump_stream("total_time") << i
.total_time
;
187 f
->dump_stream("min_time") << i
.min_time
;
188 f
->dump_stream("max_time") << i
.max_time
;
189 vector
<string
> states
;
190 get_str_vec(p
->first
, "/", states
);
191 f
->open_array_section("nested_states");
192 for (vector
<string
>::iterator st
= states
.begin();
193 st
!= states
.end(); ++st
) {
194 f
->dump_string("state", *st
);
202 void log_enter(const char *s
) {
203 Mutex::Locker
l(lock
);
206 void log_exit(const char *s
, utime_t dur
, uint64_t events
, utime_t event_dur
) {
207 Mutex::Locker
l(lock
);
208 per_state_info
&i
= info
[s
];
211 if (dur
> i
.max_time
)
213 if (dur
< i
.min_time
|| i
.min_time
== utime_t())
216 i
.event_time
+= event_dur
;
222 epoch_t cached_epoch
;
228 SnapContext snapc
; // the default pool snapc, ready to go.
230 interval_set
<snapid_t
> cached_removed_snaps
; // current removed_snaps set
231 interval_set
<snapid_t
> newly_removed_snaps
; // newly removed in the last epoch
233 PGPool(CephContext
* cct
, OSDMapRef map
, int64_t i
)
235 cached_epoch(map
->get_epoch()),
237 name(map
->get_pool_name(id
)),
238 auid(map
->get_pg_pool(id
)->auid
) {
239 const pg_pool_t
*pi
= map
->get_pg_pool(id
);
242 snapc
= pi
->get_snap_context();
243 pi
->build_removed_snaps(cached_removed_snaps
);
246 void update(OSDMapRef map
);
249 /** PG - Replica Placement Group
253 class PG
: public DoutPrefixProvider
{
258 SnapMapper snap_mapper
;
259 bool eio_errors_to_process
= false;
261 virtual PGBackend
*get_pgbackend() = 0;
263 std::string
gen_prefix() const override
;
264 CephContext
*get_cct() const override
{ return cct
; }
265 unsigned get_subsys() const override
{ return ceph_subsys_osd
; }
268 void update_snap_mapper_bits(uint32_t bits
) {
269 snap_mapper
.update_bits(bits
);
271 /// get_is_recoverable_predicate: caller owns returned pointer and must delete when done
272 IsPGRecoverablePredicate
*get_is_recoverable_predicate() {
273 return get_pgbackend()->get_is_recoverable_predicate();
276 OSDMapRef osdmap_ref
;
277 OSDMapRef last_persisted_osdmap_ref
;
280 void requeue_map_waiters();
282 void update_osdmap_ref(OSDMapRef newmap
) {
283 assert(_lock
.is_locked_by_me());
284 osdmap_ref
= std::move(newmap
);
288 OSDMapRef
get_osdmap() const {
295 /** locking and reference counting.
296 * I destroy myself when the reference count hits zero.
297 * lock() should be called before doing anything.
298 * get() should be called on pointer copy (to another thread, etc.).
299 * put() should be called on destruction of some previously copied pointer.
300 * unlock() when done with the current pointer (_most common_).
303 std::atomic_uint ref
{0};
307 map
<uint64_t, string
> _live_ids
;
308 map
<string
, uint64_t> _tag_counts
;
313 bool deleting
; // true while in removing or OSD is shutting down
315 ZTracer::Endpoint trace_endpoint
;
317 void lock_suspend_timeout(ThreadPool::TPHandle
&handle
);
318 void lock(bool no_lockdep
= false) const;
319 void unlock() const {
320 //generic_dout(0) << this << " " << info.pgid << " unlock" << dendl;
322 assert(!dirty_big_info
);
326 bool is_locked() const {
327 return _lock
.is_locked();
331 uint64_t get_with_id();
332 void put_with_id(uint64_t);
333 void dump_live_ids();
335 void get(const char* tag
);
336 void put(const char* tag
);
338 bool dirty_info
, dirty_big_info
;
341 bool is_ec_pg() const {
342 return pool
.info
.ec_pool();
345 pg_info_t info
; ///< current pg info
346 pg_info_t last_written_info
; ///< last written info
348 static const __u8 cur_struct_v
= 10;
349 // v10 is the new past_intervals encoding
350 // v9 was fastinfo_key addition
351 // v8 was the move to a per-pg pgmeta object
352 // v7 was SnapMapper addition in 86658392516d5175b2756659ef7ffaaf95b0f8ad
353 // (first appeared in cuttlefish).
354 static const __u8 compat_struct_v
= 7;
355 bool must_upgrade() {
356 return info_struct_v
< cur_struct_v
;
359 return info_struct_v
>= compat_struct_v
;
361 void upgrade(ObjectStore
*store
);
364 ObjectStore::CollectionHandle ch
;
366 static string
get_info_key(spg_t pgid
) {
367 return stringify(pgid
) + "_info";
369 static string
get_biginfo_key(spg_t pgid
) {
370 return stringify(pgid
) + "_biginfo";
372 static string
get_epoch_key(spg_t pgid
) {
373 return stringify(pgid
) + "_epoch";
375 ghobject_t pgmeta_oid
;
378 map
<hobject_t
, pg_missing_item
> needs_recovery_map
;
379 map
<hobject_t
, set
<pg_shard_t
> > missing_loc
;
380 set
<pg_shard_t
> missing_loc_sources
;
382 set
<pg_shard_t
> empty_set
;
384 boost::scoped_ptr
<IsPGReadablePredicate
> is_readable
;
385 boost::scoped_ptr
<IsPGRecoverablePredicate
> is_recoverable
;
386 explicit MissingLoc(PG
*pg
)
388 void set_backend_predicates(
389 IsPGReadablePredicate
*_is_readable
,
390 IsPGRecoverablePredicate
*_is_recoverable
) {
391 is_readable
.reset(_is_readable
);
392 is_recoverable
.reset(_is_recoverable
);
394 string
gen_prefix() const { return pg
->gen_prefix(); }
396 const hobject_t
&hoid
,
397 eversion_t
*v
= 0) const {
398 map
<hobject_t
, pg_missing_item
>::const_iterator i
=
399 needs_recovery_map
.find(hoid
);
400 if (i
== needs_recovery_map
.end())
406 bool is_deleted(const hobject_t
&hoid
) const {
407 auto i
= needs_recovery_map
.find(hoid
);
408 if (i
== needs_recovery_map
.end())
410 return i
->second
.is_delete();
412 bool is_unfound(const hobject_t
&hoid
) const {
413 return needs_recovery(hoid
) && !is_deleted(hoid
) && (
414 !missing_loc
.count(hoid
) ||
415 !(*is_recoverable
)(missing_loc
.find(hoid
)->second
));
417 bool readable_with_acting(
418 const hobject_t
&hoid
,
419 const set
<pg_shard_t
> &acting
) const;
420 uint64_t num_unfound() const {
422 for (map
<hobject_t
, pg_missing_item
>::const_iterator i
=
423 needs_recovery_map
.begin();
424 i
!= needs_recovery_map
.end();
426 if (is_unfound(i
->first
))
432 bool have_unfound() const {
433 for (map
<hobject_t
, pg_missing_item
>::const_iterator i
=
434 needs_recovery_map
.begin();
435 i
!= needs_recovery_map
.end();
437 if (is_unfound(i
->first
))
443 needs_recovery_map
.clear();
445 missing_loc_sources
.clear();
448 void add_location(const hobject_t
&hoid
, pg_shard_t location
) {
449 missing_loc
[hoid
].insert(location
);
451 void remove_location(const hobject_t
&hoid
, pg_shard_t location
) {
452 missing_loc
[hoid
].erase(location
);
454 void add_active_missing(const pg_missing_t
&missing
) {
455 for (map
<hobject_t
, pg_missing_item
>::const_iterator i
=
456 missing
.get_items().begin();
457 i
!= missing
.get_items().end();
459 map
<hobject_t
, pg_missing_item
>::const_iterator j
=
460 needs_recovery_map
.find(i
->first
);
461 if (j
== needs_recovery_map
.end()) {
462 needs_recovery_map
.insert(*i
);
464 lgeneric_dout(pg
->cct
, 0) << this << " " << pg
->info
.pgid
<< " unexpected need for "
465 << i
->first
<< " have " << j
->second
466 << " tried to add " << i
->second
<< dendl
;
467 assert(i
->second
.need
== j
->second
.need
);
472 void add_missing(const hobject_t
&hoid
, eversion_t need
, eversion_t have
) {
473 needs_recovery_map
[hoid
] = pg_missing_item(need
, have
);
475 void revise_need(const hobject_t
&hoid
, eversion_t need
) {
476 assert(needs_recovery(hoid
));
477 needs_recovery_map
[hoid
].need
= need
;
480 /// Adds info about a possible recovery source
481 bool add_source_info(
482 pg_shard_t source
, ///< [in] source
483 const pg_info_t
&oinfo
, ///< [in] info
484 const pg_missing_t
&omissing
, ///< [in] (optional) missing
485 ThreadPool::TPHandle
* handle
///< [in] ThreadPool handle
486 ); ///< @return whether a new object location was discovered
488 /// Adds recovery sources in batch
489 void add_batch_sources_info(
490 const set
<pg_shard_t
> &sources
, ///< [in] a set of resources which can be used for all objects
491 ThreadPool::TPHandle
* handle
///< [in] ThreadPool handle
494 /// Uses osdmap to update structures for now down sources
495 void check_recovery_sources(const OSDMapRef
& osdmap
);
497 /// Call when hoid is no longer missing in acting set
498 void recovered(const hobject_t
&hoid
) {
499 needs_recovery_map
.erase(hoid
);
500 missing_loc
.erase(hoid
);
503 /// Call to update structures for hoid after a change
505 const hobject_t
&hoid
,
507 const set
<pg_shard_t
> to_recover
,
508 const pg_info_t
&info
,
509 const pg_missing_t
&missing
,
510 const map
<pg_shard_t
, pg_missing_t
> &pmissing
,
511 const map
<pg_shard_t
, pg_info_t
> &pinfo
) {
513 boost::optional
<pg_missing_item
> item
;
514 auto miter
= missing
.get_items().find(hoid
);
515 if (miter
!= missing
.get_items().end()) {
516 item
= miter
->second
;
518 for (auto &&i
: to_recover
) {
521 auto pmiter
= pmissing
.find(i
);
522 assert(pmiter
!= pmissing
.end());
523 miter
= pmiter
->second
.get_items().find(hoid
);
524 if (miter
!= pmiter
->second
.get_items().end()) {
525 item
= miter
->second
;
531 return; // recovered!
533 needs_recovery_map
[hoid
] = *item
;
534 if (item
->is_delete())
537 missing_loc
.insert(make_pair(hoid
, set
<pg_shard_t
>())).first
;
538 assert(info
.last_backfill
.is_max());
539 assert(info
.last_update
>= item
->need
);
540 if (!missing
.is_missing(hoid
))
541 mliter
->second
.insert(self
);
542 for (auto &&i
: pmissing
) {
543 auto pinfoiter
= pinfo
.find(i
.first
);
544 assert(pinfoiter
!= pinfo
.end());
545 if (item
->need
<= pinfoiter
->second
.last_update
&&
546 hoid
<= pinfoiter
->second
.last_backfill
&&
547 !i
.second
.is_missing(hoid
))
548 mliter
->second
.insert(i
.first
);
552 const set
<pg_shard_t
> &get_locations(const hobject_t
&hoid
) const {
553 return missing_loc
.count(hoid
) ?
554 missing_loc
.find(hoid
)->second
: empty_set
;
556 const map
<hobject_t
, set
<pg_shard_t
>> &get_missing_locs() const {
559 const map
<hobject_t
, pg_missing_item
> &get_needs_recovery() const {
560 return needs_recovery_map
;
564 PastIntervals past_intervals
;
566 interval_set
<snapid_t
> snap_trimq
;
568 /* You should not use these items without taking their respective queue locks
569 * (if they have one) */
570 xlist
<PG
*>::item stat_queue_item
;
572 bool recovery_queued
;
574 int recovery_ops_active
;
575 set
<pg_shard_t
> waiting_on_backfill
;
576 #ifdef DEBUG_RECOVERY_OIDS
577 set
<hobject_t
> recovering_oids
;
581 int role
; // 0 = primary, 1 = replica, -1=none.
582 unsigned state
; // PG_STATE_*
584 bool send_notify
; ///< true if we are non-primary and should notify the primary
587 eversion_t last_update_ondisk
; // last_update that has committed; ONLY DEFINED WHEN is_active()
588 eversion_t last_complete_ondisk
; // last_complete that has committed.
589 eversion_t last_update_applied
;
592 struct C_UpdateLastRollbackInfoTrimmedToApplied
: Context
{
596 C_UpdateLastRollbackInfoTrimmedToApplied(PG
*pg
, epoch_t e
, eversion_t v
)
597 : pg(pg
), e(e
), v(v
) {}
598 void finish(int) override
{
600 if (!pg
->pg_has_reset_since(e
)) {
601 pg
->last_rollback_info_trimmed_to_applied
= v
;
606 // entries <= last_rollback_info_trimmed_to_applied have been trimmed,
607 // and the transaction has applied
608 eversion_t last_rollback_info_trimmed_to_applied
;
613 pg_shard_t pg_whoami
;
614 pg_shard_t up_primary
;
615 vector
<int> up
, acting
, want_acting
;
616 set
<pg_shard_t
> actingbackfill
, actingset
, upset
;
617 map
<pg_shard_t
,eversion_t
> peer_last_complete_ondisk
;
618 eversion_t min_last_complete_ondisk
; // up: min over last_complete_ondisk, peer_last_complete_ondisk
619 eversion_t pg_trim_to
;
621 set
<int> blocked_by
; ///< osds we are blocked by (for pg stats)
623 // [primary only] content recovery state
626 struct BufferedRecoveryMessages
{
627 map
<int, map
<spg_t
, pg_query_t
> > query_map
;
628 map
<int, vector
<pair
<pg_notify_t
, PastIntervals
> > > info_map
;
629 map
<int, vector
<pair
<pg_notify_t
, PastIntervals
> > > notify_list
;
633 bool dne() { return info
.dne(); }
636 map
<int, map
<spg_t
, pg_query_t
> > *query_map
;
637 map
<int, vector
<pair
<pg_notify_t
, PastIntervals
> > > *info_map
;
638 map
<int, vector
<pair
<pg_notify_t
, PastIntervals
> > > *notify_list
;
639 set
<PGRef
> created_pgs
;
640 C_Contexts
*on_applied
;
642 ObjectStore::Transaction
*transaction
;
643 ThreadPool::TPHandle
* handle
;
644 RecoveryCtx(map
<int, map
<spg_t
, pg_query_t
> > *query_map
,
646 vector
<pair
<pg_notify_t
, PastIntervals
> > > *info_map
,
648 vector
<pair
<pg_notify_t
, PastIntervals
> > > *notify_list
,
649 C_Contexts
*on_applied
,
651 ObjectStore::Transaction
*transaction
)
652 : query_map(query_map
), info_map(info_map
),
653 notify_list(notify_list
),
654 on_applied(on_applied
),
656 transaction(transaction
),
659 RecoveryCtx(BufferedRecoveryMessages
&buf
, RecoveryCtx
&rctx
)
660 : query_map(&(buf
.query_map
)),
661 info_map(&(buf
.info_map
)),
662 notify_list(&(buf
.notify_list
)),
663 on_applied(rctx
.on_applied
),
664 on_safe(rctx
.on_safe
),
665 transaction(rctx
.transaction
),
666 handle(rctx
.handle
) {}
668 void accept_buffered_messages(BufferedRecoveryMessages
&m
) {
672 for (map
<int, map
<spg_t
, pg_query_t
> >::iterator i
= m
.query_map
.begin();
673 i
!= m
.query_map
.end();
675 map
<spg_t
, pg_query_t
> &omap
= (*query_map
)[i
->first
];
676 for (map
<spg_t
, pg_query_t
>::iterator j
= i
->second
.begin();
677 j
!= i
->second
.end();
679 omap
[j
->first
] = j
->second
;
682 for (map
<int, vector
<pair
<pg_notify_t
, PastIntervals
> > >::iterator i
683 = m
.info_map
.begin();
684 i
!= m
.info_map
.end();
686 vector
<pair
<pg_notify_t
, PastIntervals
> > &ovec
=
687 (*info_map
)[i
->first
];
688 ovec
.reserve(ovec
.size() + i
->second
.size());
689 ovec
.insert(ovec
.end(), i
->second
.begin(), i
->second
.end());
691 for (map
<int, vector
<pair
<pg_notify_t
, PastIntervals
> > >::iterator i
692 = m
.notify_list
.begin();
693 i
!= m
.notify_list
.end();
695 vector
<pair
<pg_notify_t
, PastIntervals
> > &ovec
=
696 (*notify_list
)[i
->first
];
697 ovec
.reserve(ovec
.size() + i
->second
.size());
698 ovec
.insert(ovec
.end(), i
->second
.begin(), i
->second
.end());
704 PGStateHistory pgstate_history
;
707 const char *state_name
;
710 const char *get_state_name() { return state_name
; }
711 NamedState(PG
*pg_
, const char *state_name_
)
712 : state_name(state_name_
), enter_time(ceph_clock_now()), pg(pg_
) {
713 pg
->pgstate_history
.enter(pg
, enter_time
, state_name
);
715 virtual ~NamedState() { pg
->pgstate_history
.exit(state_name
); }
723 * peer_info -- projected (updates _before_ replicas ack)
724 * peer_missing -- committed (updates _after_ replicas ack)
728 set
<pg_shard_t
> stray_set
; // non-acting osds that have PG data.
729 eversion_t oldest_update
; // acting: lowest (valid) last_update in active set
730 map
<pg_shard_t
, pg_info_t
> peer_info
; // info from peers (stray or prior)
731 set
<pg_shard_t
> peer_purged
; // peers purged
732 map
<pg_shard_t
, pg_missing_t
> peer_missing
;
733 set
<pg_shard_t
> peer_log_requested
; // logs i've requested (and start stamps)
734 set
<pg_shard_t
> peer_missing_requested
;
736 // i deleted these strays; ignore racing PGInfo from them
737 set
<pg_shard_t
> peer_activated
;
739 // primary-only, recovery-only state
740 set
<pg_shard_t
> might_have_unfound
; // These osds might have objects on them
741 // which are unfound on the primary
742 epoch_t last_peering_reset
;
745 /* heartbeat peers */
746 void set_probe_targets(const set
<pg_shard_t
> &probe_set
);
747 void clear_probe_targets();
749 Mutex heartbeat_peer_lock
;
750 set
<int> heartbeat_peers
;
751 set
<int> probe_targets
;
756 * Represents the objects in a range [begin, end)
759 * 1) begin == end == hobject_t() indicates the the interval is unpopulated
760 * 2) Else, objects contains all objects in [begin, end)
762 struct BackfillInterval
{
763 // info about a backfill interval on a peer
764 eversion_t version
; /// version at which the scan occurred
765 map
<hobject_t
,eversion_t
> objects
;
771 *this = BackfillInterval();
774 /// clear objects list only
775 void clear_objects() {
779 /// reinstantiate with a new start+end position and sort order
780 void reset(hobject_t start
) {
785 /// true if there are no objects in this interval
787 return objects
.empty();
790 /// true if interval extends to the end of the range
791 bool extends_to_end() const {
795 /// removes items <= soid and adjusts begin to the first object
796 void trim_to(const hobject_t
&soid
) {
798 while (!objects
.empty() &&
799 objects
.begin()->first
<= soid
) {
804 /// Adjusts begin to the first object
806 if (!objects
.empty())
807 begin
= objects
.begin()->first
;
812 /// drop first entry, and adjust @begin accordingly
814 assert(!objects
.empty());
815 objects
.erase(objects
.begin());
820 void dump(Formatter
*f
) const {
821 f
->dump_stream("begin") << begin
;
822 f
->dump_stream("end") << end
;
823 f
->open_array_section("objects");
824 for (map
<hobject_t
, eversion_t
>::const_iterator i
=
828 f
->open_object_section("object");
829 f
->dump_stream("object") << i
->first
;
830 f
->dump_stream("version") << i
->second
;
838 BackfillInterval backfill_info
;
839 map
<pg_shard_t
, BackfillInterval
> peer_backfill_info
;
840 bool backfill_reserved
;
841 bool backfill_reserving
;
846 set
<pg_shard_t
> backfill_targets
;
848 bool is_backfill_targets(pg_shard_t osd
) {
849 return backfill_targets
.count(osd
);
855 * blocked request wait hierarchy
857 * In order to preserve request ordering we need to be careful about the
858 * order in which blocked requests get requeued. Generally speaking, we
859 * push the requests back up to the op_wq in reverse order (most recent
860 * request first) so that they come back out again in the original order.
861 * However, because there are multiple wait queues, we need to requeue
862 * waitlists in order. Generally speaking, we requeue the wait lists
863 * that are checked first.
865 * Here are the various wait lists, in the order they are used during
866 * request processing, with notes:
869 * - may start or stop blocking at any time (depending on client epoch)
870 * - waiting_for_peered
871 * - !is_peered() or flushes_in_progress
872 * - only starts blocking on interval change; never restarts
873 * - waiting_for_active
875 * - only starts blocking on interval change; never restarts
876 * - waiting_for_flush
877 * - is_active() and flushes_in_progress
878 * - waiting for final flush during activate
879 * - waiting_for_scrub
880 * - starts and stops blocking for varying intervals during scrub
881 * - waiting_for_unreadable_object
882 * - never restarts once object is readable (* except for EIO?)
883 * - waiting_for_degraded_object
884 * - never restarts once object is writeable (* except for EIO?)
885 * - waiting_for_blocked_object
886 * - starts and stops based on proxied op activity
888 * - starts and stops based on read/write activity
892 * 1. During and interval change, we requeue *everything* in the above order.
894 * 2. When an obc rwlock is released, we check for a scrub block and requeue
895 * the op there if it applies. We ignore the unreadable/degraded/blocked
896 * queues because we assume they cannot apply at that time (this is
897 * probably mostly true).
899 * 3. The requeue_ops helper will push ops onto the waiting_for_map list if
902 * These three behaviors are generally sufficient to maintain ordering, with
903 * the possible exception of cases where we make an object degraded or
904 * unreadable that was previously okay, e.g. when scrub or op processing
905 * encounter an unexpected error. FIXME.
909 unsigned flushes_in_progress
;
911 // ops with newer maps than our (or blocked behind them)
912 // track these by client, since inter-request ordering doesn't otherwise
914 unordered_map
<entity_name_t
,list
<OpRequestRef
>> waiting_for_map
;
916 // ops waiting on peered
917 list
<OpRequestRef
> waiting_for_peered
;
919 // ops waiting on active (require peered as well)
920 list
<OpRequestRef
> waiting_for_active
;
921 list
<OpRequestRef
> waiting_for_flush
;
922 list
<OpRequestRef
> waiting_for_scrub
;
924 list
<OpRequestRef
> waiting_for_cache_not_full
;
925 list
<OpRequestRef
> waiting_for_clean_to_primary_repair
;
926 map
<hobject_t
, list
<OpRequestRef
>> waiting_for_unreadable_object
,
927 waiting_for_degraded_object
,
928 waiting_for_blocked_object
;
930 set
<hobject_t
> objects_blocked_on_cache_full
;
931 map
<hobject_t
,snapid_t
> objects_blocked_on_degraded_snap
;
932 map
<hobject_t
,ObjectContextRef
> objects_blocked_on_snap_promotion
;
934 // Callbacks should assume pg (and nothing else) is locked
935 map
<hobject_t
, list
<Context
*>> callbacks_for_degraded_object
;
938 list
<pair
<OpRequestRef
, version_t
> > > waiting_for_ondisk
;
940 void requeue_object_waiters(map
<hobject_t
, list
<OpRequestRef
>>& m
);
941 void requeue_op(OpRequestRef op
);
942 void requeue_ops(list
<OpRequestRef
> &l
);
944 // stats that persist lazily
945 object_stat_collection_t unstable_stats
;
948 Mutex pg_stats_publish_lock
;
949 bool pg_stats_publish_valid
;
950 pg_stat_t pg_stats_publish
;
952 // for ordering writes
953 ceph::shared_ptr
<ObjectStore::Sequencer
> osr
;
955 void _update_calc_stats();
956 void _update_blocked_by();
957 void publish_stats_to_osd();
958 void clear_publish_stats();
961 void clear_primary_state();
963 bool is_actingbackfill(pg_shard_t osd
) const {
964 return actingbackfill
.count(osd
);
966 bool is_acting(pg_shard_t osd
) const {
967 return has_shard(pool
.info
.ec_pool(), acting
, osd
);
969 bool is_up(pg_shard_t osd
) const {
970 return has_shard(pool
.info
.ec_pool(), up
, osd
);
972 static bool has_shard(bool ec
, const vector
<int>& v
, pg_shard_t osd
) {
974 return v
.size() > (unsigned)osd
.shard
&& v
[osd
.shard
] == osd
.osd
;
976 return std::find(v
.begin(), v
.end(), osd
.osd
) != v
.end();
980 bool needs_recovery() const;
981 bool needs_backfill() const;
983 /// clip calculated priority to reasonable range
984 inline int clamp_recovery_priority(int priority
);
985 /// get log recovery reservation priority
986 unsigned get_recovery_priority();
987 /// get backfill reservation priority
988 unsigned get_backfill_priority();
990 void mark_clean(); ///< mark an active pg clean
991 void _change_recovery_force_mode(int new_mode
, bool clear
);
993 /// return [start,end) bounds for required past_intervals
994 static pair
<epoch_t
, epoch_t
> get_required_past_interval_bounds(
995 const pg_info_t
&info
,
996 epoch_t oldest_map
) {
998 info
.history
.last_epoch_clean
? info
.history
.last_epoch_clean
:
999 info
.history
.epoch_pool_created
,
1002 info
.history
.same_interval_since
,
1003 info
.history
.epoch_pool_created
);
1004 return make_pair(start
, end
);
1006 void check_past_interval_bounds() const;
1007 PastIntervals::PriorSet
build_prior();
1009 void remove_down_peer_info(const OSDMapRef osdmap
);
1011 bool adjust_need_up_thru(const OSDMapRef osdmap
);
1013 bool all_unfound_are_queried_or_lost(const OSDMapRef osdmap
) const;
1014 virtual void dump_recovery_info(Formatter
*f
) const = 0;
1016 bool calc_min_last_complete_ondisk() {
1017 eversion_t min
= last_complete_ondisk
;
1018 assert(!actingbackfill
.empty());
1019 for (set
<pg_shard_t
>::iterator i
= actingbackfill
.begin();
1020 i
!= actingbackfill
.end();
1022 if (*i
== get_primary()) continue;
1023 if (peer_last_complete_ondisk
.count(*i
) == 0)
1024 return false; // we don't have complete info
1025 eversion_t a
= peer_last_complete_ondisk
[*i
];
1029 if (min
== min_last_complete_ondisk
)
1031 min_last_complete_ondisk
= min
;
1035 virtual void calc_trim_to() = 0;
1037 void proc_replica_log(pg_info_t
&oinfo
, const pg_log_t
&olog
,
1038 pg_missing_t
& omissing
, pg_shard_t from
);
1039 void proc_master_log(ObjectStore::Transaction
& t
, pg_info_t
&oinfo
, pg_log_t
&olog
,
1040 pg_missing_t
& omissing
, pg_shard_t from
);
1041 bool proc_replica_info(
1042 pg_shard_t from
, const pg_info_t
&info
, epoch_t send_epoch
);
1044 struct PGLogEntryHandler
: public PGLog::LogEntryHandler
{
1046 ObjectStore::Transaction
*t
;
1047 PGLogEntryHandler(PG
*pg
, ObjectStore::Transaction
*t
) : pg(pg
), t(t
) {}
1050 void remove(const hobject_t
&hoid
) override
{
1051 pg
->get_pgbackend()->remove(hoid
, t
);
1053 void try_stash(const hobject_t
&hoid
, version_t v
) override
{
1054 pg
->get_pgbackend()->try_stash(hoid
, v
, t
);
1056 void rollback(const pg_log_entry_t
&entry
) override
{
1057 assert(entry
.can_rollback());
1058 pg
->get_pgbackend()->rollback(entry
, t
);
1060 void rollforward(const pg_log_entry_t
&entry
) override
{
1061 pg
->get_pgbackend()->rollforward(entry
, t
);
1063 void trim(const pg_log_entry_t
&entry
) override
{
1064 pg
->get_pgbackend()->trim(entry
, t
);
1068 void update_object_snap_mapping(
1069 ObjectStore::Transaction
*t
, const hobject_t
&soid
,
1070 const set
<snapid_t
> &snaps
);
1071 void clear_object_snap_mapping(
1072 ObjectStore::Transaction
*t
, const hobject_t
&soid
);
1073 void remove_snap_mapped_object(
1074 ObjectStore::Transaction
& t
, const hobject_t
& soid
);
1076 ObjectStore::Transaction
& t
, pg_info_t
&oinfo
,
1077 pg_log_t
&olog
, pg_shard_t from
);
1078 void rewind_divergent_log(ObjectStore::Transaction
& t
, eversion_t newhead
);
1079 bool search_for_missing(
1080 const pg_info_t
&oinfo
, const pg_missing_t
&omissing
,
1084 void check_for_lost_objects();
1085 void forget_lost_objects();
1087 void discover_all_missing(std::map
<int, map
<spg_t
,pg_query_t
> > &query_map
);
1089 void trim_write_ahead();
1091 map
<pg_shard_t
, pg_info_t
>::const_iterator
find_best_info(
1092 const map
<pg_shard_t
, pg_info_t
> &infos
,
1093 bool restrict_to_up_acting
,
1094 bool *history_les_bound
) const;
1095 static void calc_ec_acting(
1096 map
<pg_shard_t
, pg_info_t
>::const_iterator auth_log_shard
,
1098 const vector
<int> &acting
,
1099 pg_shard_t acting_primary
,
1100 const vector
<int> &up
,
1101 pg_shard_t up_primary
,
1102 const map
<pg_shard_t
, pg_info_t
> &all_info
,
1103 bool restrict_to_up_acting
,
1105 set
<pg_shard_t
> *backfill
,
1106 set
<pg_shard_t
> *acting_backfill
,
1107 pg_shard_t
*want_primary
,
1109 static void calc_replicated_acting(
1110 map
<pg_shard_t
, pg_info_t
>::const_iterator auth_log_shard
,
1112 const vector
<int> &acting
,
1113 pg_shard_t acting_primary
,
1114 const vector
<int> &up
,
1115 pg_shard_t up_primary
,
1116 const map
<pg_shard_t
, pg_info_t
> &all_info
,
1117 bool restrict_to_up_acting
,
1119 set
<pg_shard_t
> *backfill
,
1120 set
<pg_shard_t
> *acting_backfill
,
1121 pg_shard_t
*want_primary
,
1123 bool choose_acting(pg_shard_t
&auth_log_shard
,
1124 bool restrict_to_up_acting
,
1125 bool *history_les_bound
);
1126 void build_might_have_unfound();
1128 ObjectStore::Transaction
& t
,
1129 epoch_t activation_epoch
,
1130 list
<Context
*>& tfin
,
1131 map
<int, map
<spg_t
,pg_query_t
> >& query_map
,
1133 vector
<pair
<pg_notify_t
, PastIntervals
> > > *activator_map
,
1135 void _activate_committed(epoch_t epoch
, epoch_t activation_epoch
);
1136 void all_activated_and_committed();
1138 void proc_primary_info(ObjectStore::Transaction
&t
, const pg_info_t
&info
);
1140 bool have_unfound() const {
1141 return missing_loc
.have_unfound();
1143 uint64_t get_num_unfound() const {
1144 return missing_loc
.num_unfound();
1147 virtual void check_local() = 0;
1150 * @param ops_begun returns how many recovery ops the function started
1151 * @returns true if any useful work was accomplished; false otherwise
1153 virtual bool start_recovery_ops(
1155 ThreadPool::TPHandle
&handle
,
1156 uint64_t *ops_begun
) = 0;
1158 void purge_strays();
1160 void update_heartbeat_peers();
1162 Context
*finish_sync_event
;
1164 void finish_recovery(list
<Context
*>& tfin
);
1165 void _finish_recovery(Context
*c
);
1166 void cancel_recovery();
1167 void clear_recovery_state();
1168 virtual void _clear_recovery_state() = 0;
1169 virtual void check_recovery_sources(const OSDMapRef
& newmap
) = 0;
1170 void start_recovery_op(const hobject_t
& soid
);
1171 void finish_recovery_op(const hobject_t
& soid
, bool dequeue
=false);
1173 void split_into(pg_t child_pgid
, PG
*child
, unsigned split_bits
);
1174 virtual void _split_into(pg_t child_pgid
, PG
*child
, unsigned split_bits
) = 0;
1176 friend class C_OSD_RepModify_Commit
;
1179 Mutex backoff_lock
; // orders inside Backoff::lock
1180 map
<hobject_t
,set
<BackoffRef
>> backoffs
;
1182 void add_backoff(SessionRef s
, const hobject_t
& begin
, const hobject_t
& end
);
1183 void release_backoffs(const hobject_t
& begin
, const hobject_t
& end
);
1184 void release_backoffs(const hobject_t
& o
) {
1185 release_backoffs(o
, o
);
1187 void clear_backoffs();
1189 void add_pg_backoff(SessionRef s
) {
1190 hobject_t begin
= info
.pgid
.pgid
.get_hobj_start();
1191 hobject_t end
= info
.pgid
.pgid
.get_hobj_end(pool
.info
.get_pg_num());
1192 add_backoff(s
, begin
, end
);
1194 void release_pg_backoffs() {
1195 hobject_t begin
= info
.pgid
.pgid
.get_hobj_start();
1196 hobject_t end
= info
.pgid
.pgid
.get_hobj_end(pool
.info
.get_pg_num());
1197 release_backoffs(begin
, end
);
1200 void rm_backoff(BackoffRef b
);
1208 set
<pg_shard_t
> reserved_peers
;
1209 bool reserved
, reserve_failed
;
1210 epoch_t epoch_start
;
1212 // common to both scrubs
1214 set
<pg_shard_t
> waiting_on_whom
;
1217 int large_omap_objects
= 0;
1219 ScrubMap primary_scrubmap
;
1220 ScrubMapBuilder primary_scrubmap_pos
;
1221 epoch_t replica_scrub_start
= 0;
1222 ScrubMap replica_scrubmap
;
1223 ScrubMapBuilder replica_scrubmap_pos
;
1224 map
<pg_shard_t
, ScrubMap
> received_maps
;
1225 OpRequestRef active_rep_scrub
;
1226 utime_t scrub_reg_stamp
; // stamp we registered for
1229 bool sleeping
= false;
1230 bool needs_sleep
= true;
1231 utime_t sleep_start
;
1233 // flags to indicate explicitly requested scrubs (by admin)
1234 bool must_scrub
, must_deep_scrub
, must_repair
;
1236 // Priority to use for scrub scheduling
1239 // this flag indicates whether we would like to do auto-repair of the PG or not
1242 // Maps from objects with errors to missing/inconsistent peers
1243 map
<hobject_t
, set
<pg_shard_t
>> missing
;
1244 map
<hobject_t
, set
<pg_shard_t
>> inconsistent
;
1246 // Map from object with errors to good peers
1247 map
<hobject_t
, list
<pair
<ScrubMap::object
, pg_shard_t
> >> authoritative
;
1249 // Cleaned map pending snap metadata scrub
1250 ScrubMap cleaned_meta_map
;
1252 void clean_meta_map(ScrubMap
&for_meta_scrub
) {
1254 cleaned_meta_map
.objects
.empty()) {
1255 cleaned_meta_map
.swap(for_meta_scrub
);
1257 auto iter
= cleaned_meta_map
.objects
.end();
1258 --iter
; // not empty, see if clause
1259 auto begin
= cleaned_meta_map
.objects
.begin();
1260 if (iter
->first
.has_snapset()) {
1263 while (iter
!= begin
) {
1265 if (next
->first
.get_head() != iter
->first
.get_head()) {
1271 for_meta_scrub
.objects
.insert(begin
, iter
);
1272 cleaned_meta_map
.objects
.erase(begin
, iter
);
1276 // digest updates which we are waiting on
1277 int num_digest_updates_pending
;
1280 hobject_t start
, end
; // [start,end)
1281 hobject_t max_end
; // Largest end that may have been sent to replicas
1282 eversion_t subset_last_update
;
1284 // chunky scrub state
1294 WAIT_DIGEST_UPDATES
,
1299 std::unique_ptr
<Scrub::Store
> store
;
1303 int preempt_divisor
;
1305 list
<Context
*> callbacks
;
1306 void add_callback(Context
*context
) {
1307 callbacks
.push_back(context
);
1309 void run_callbacks() {
1310 list
<Context
*> to_run
;
1311 to_run
.swap(callbacks
);
1312 for (list
<Context
*>::iterator i
= to_run
.begin();
1319 static const char *state_string(const PG::Scrubber::State
& state
) {
1320 const char *ret
= NULL
;
1323 case INACTIVE
: ret
= "INACTIVE"; break;
1324 case NEW_CHUNK
: ret
= "NEW_CHUNK"; break;
1325 case WAIT_PUSHES
: ret
= "WAIT_PUSHES"; break;
1326 case WAIT_LAST_UPDATE
: ret
= "WAIT_LAST_UPDATE"; break;
1327 case BUILD_MAP
: ret
= "BUILD_MAP"; break;
1328 case BUILD_MAP_DONE
: ret
= "BUILD_MAP_DONE"; break;
1329 case WAIT_REPLICAS
: ret
= "WAIT_REPLICAS"; break;
1330 case COMPARE_MAPS
: ret
= "COMPARE_MAPS"; break;
1331 case WAIT_DIGEST_UPDATES
: ret
= "WAIT_DIGEST_UPDATES"; break;
1332 case FINISH
: ret
= "FINISH"; break;
1333 case BUILD_MAP_REPLICA
: ret
= "BUILD_MAP_REPLICA"; break;
1338 bool is_chunky_scrub_active() const { return state
!= INACTIVE
; }
1343 waiting_on_whom
.clear();
1344 if (active_rep_scrub
) {
1345 active_rep_scrub
= OpRequestRef();
1347 received_maps
.clear();
1350 must_deep_scrub
= false;
1351 must_repair
= false;
1352 auto_repair
= false;
1354 state
= PG::Scrubber::INACTIVE
;
1355 start
= hobject_t();
1357 max_end
= hobject_t();
1358 subset_last_update
= eversion_t();
1361 large_omap_objects
= 0;
1365 inconsistent
.clear();
1367 authoritative
.clear();
1368 num_digest_updates_pending
= 0;
1369 primary_scrubmap
= ScrubMap();
1370 primary_scrubmap_pos
.reset();
1371 replica_scrubmap
= ScrubMap();
1372 replica_scrubmap_pos
.reset();
1373 cleaned_meta_map
= ScrubMap();
1376 sleep_start
= utime_t();
1379 void create_results(const hobject_t
& obj
);
1380 void cleanup_store(ObjectStore::Transaction
*t
);
1383 bool scrub_after_recovery
;
1387 bool scrub_can_preempt
= false;
1388 bool scrub_preempted
= false;
1390 // we allow some number of preemptions of the scrub, which mean we do
1391 // not block. then we start to block. once we start blocking, we do
1392 // not stop until the scrub range is completed.
1393 bool write_blocked_by_scrub(const hobject_t
&soid
);
1395 /// true if the given range intersects the scrub interval in any way
1396 bool range_intersects_scrub(const hobject_t
&start
, const hobject_t
& end
);
1399 const hobject_t
& soid
, list
<pair
<ScrubMap::object
, pg_shard_t
> > *ok_peers
,
1400 pg_shard_t bad_peer
);
1402 void scrub(epoch_t queued
, ThreadPool::TPHandle
&handle
);
1403 void chunky_scrub(ThreadPool::TPHandle
&handle
);
1404 void scrub_compare_maps();
1406 * return true if any inconsistency/missing is repaired, false otherwise
1408 bool scrub_process_inconsistent();
1409 bool ops_blocked_by_scrub() const;
1410 void scrub_finish();
1411 void scrub_clear_state();
1412 void _scan_snaps(ScrubMap
&map
);
1413 void _repair_oinfo_oid(ScrubMap
&map
);
1414 void _scan_rollback_obs(
1415 const vector
<ghobject_t
> &rollback_obs
,
1416 ThreadPool::TPHandle
&handle
);
1417 void _request_scrub_map(pg_shard_t replica
, eversion_t version
,
1418 hobject_t start
, hobject_t end
, bool deep
,
1419 bool allow_preemption
);
1420 int build_scrub_map_chunk(
1422 ScrubMapBuilder
&pos
,
1423 hobject_t start
, hobject_t end
, bool deep
,
1424 ThreadPool::TPHandle
&handle
);
1426 * returns true if [begin, end) is good to scrub at this time
1427 * a false return value obliges the implementer to requeue scrub when the
1428 * condition preventing scrub clears
1430 virtual bool _range_available_for_scrub(
1431 const hobject_t
&begin
, const hobject_t
&end
) = 0;
1432 virtual void scrub_snapshot_metadata(
1434 const std::map
<hobject_t
,
1435 pair
<boost::optional
<uint32_t>,
1436 boost::optional
<uint32_t>>> &missing_digest
) { }
1437 virtual void _scrub_clear_state() { }
1438 virtual void _scrub_finish() { }
1439 virtual void split_colls(
1443 const pg_pool_t
*pool
,
1444 ObjectStore::Transaction
*t
) = 0;
1445 void clear_scrub_reserved();
1446 void scrub_reserve_replicas();
1447 void scrub_unreserve_replicas();
1448 bool scrub_all_replicas_reserved() const;
1450 void reg_next_scrub();
1451 void unreg_next_scrub();
1455 ThreadPool::TPHandle
&handle
);
1456 void do_replica_scrub_map(OpRequestRef op
);
1457 void sub_op_scrub_map(OpRequestRef op
);
1459 void handle_scrub_reserve_request(OpRequestRef op
);
1460 void handle_scrub_reserve_grant(OpRequestRef op
, pg_shard_t from
);
1461 void handle_scrub_reserve_reject(OpRequestRef op
, pg_shard_t from
);
1462 void handle_scrub_reserve_release(OpRequestRef op
);
1464 void reject_reservation();
1465 void schedule_backfill_retry(float retry
);
1466 void schedule_recovery_retry(float retry
);
1468 // -- recovery state --
1470 template <class EVT
>
1471 struct QueuePeeringEvt
: Context
{
1475 QueuePeeringEvt(PG
*pg
, epoch_t epoch
, EVT evt
) :
1476 pg(pg
), epoch(epoch
), evt(evt
) {}
1477 void finish(int r
) override
{
1479 pg
->queue_peering_event(PG::CephPeeringEvtRef(
1480 new PG::CephPeeringEvt(
1488 class CephPeeringEvt
{
1490 epoch_t epoch_requested
;
1491 boost::intrusive_ptr
< const boost::statechart::event_base
> evt
;
1494 MEMPOOL_CLASS_HELPERS();
1496 CephPeeringEvt(epoch_t epoch_sent
,
1497 epoch_t epoch_requested
,
1499 epoch_sent(epoch_sent
), epoch_requested(epoch_requested
),
1500 evt(evt_
.intrusive_from_this()) {
1502 out
<< "epoch_sent: " << epoch_sent
1503 << " epoch_requested: " << epoch_requested
<< " ";
1507 epoch_t
get_epoch_sent() { return epoch_sent
; }
1508 epoch_t
get_epoch_requested() { return epoch_requested
; }
1509 const boost::statechart::event_base
&get_event() { return *evt
; }
1510 string
get_desc() { return desc
; }
1512 typedef ceph::shared_ptr
<CephPeeringEvt
> CephPeeringEvtRef
;
1513 list
<CephPeeringEvtRef
> peering_queue
; // op queue
1514 list
<CephPeeringEvtRef
> peering_waiters
;
1516 struct QueryState
: boost::statechart::event
< QueryState
> {
1518 explicit QueryState(Formatter
*f
) : f(f
) {}
1519 void print(std::ostream
*out
) const {
1524 struct MInfoRec
: boost::statechart::event
< MInfoRec
> {
1528 MInfoRec(pg_shard_t from
, const pg_info_t
&info
, epoch_t msg_epoch
) :
1529 from(from
), info(info
), msg_epoch(msg_epoch
) {}
1530 void print(std::ostream
*out
) const {
1531 *out
<< "MInfoRec from " << from
<< " info: " << info
;
1535 struct MLogRec
: boost::statechart::event
< MLogRec
> {
1537 boost::intrusive_ptr
<MOSDPGLog
> msg
;
1538 MLogRec(pg_shard_t from
, MOSDPGLog
*msg
) :
1539 from(from
), msg(msg
) {}
1540 void print(std::ostream
*out
) const {
1541 *out
<< "MLogRec from " << from
;
1545 struct MNotifyRec
: boost::statechart::event
< MNotifyRec
> {
1549 MNotifyRec(pg_shard_t from
, const pg_notify_t
¬ify
, uint64_t f
) :
1550 from(from
), notify(notify
), features(f
) {}
1551 void print(std::ostream
*out
) const {
1552 *out
<< "MNotifyRec from " << from
<< " notify: " << notify
1553 << " features: 0x" << hex
<< features
<< dec
;
1557 struct MQuery
: boost::statechart::event
< MQuery
> {
1560 epoch_t query_epoch
;
1561 MQuery(pg_shard_t from
, const pg_query_t
&query
, epoch_t query_epoch
):
1562 from(from
), query(query
), query_epoch(query_epoch
) {}
1563 void print(std::ostream
*out
) const {
1564 *out
<< "MQuery from " << from
1565 << " query_epoch " << query_epoch
1566 << " query: " << query
;
1570 struct AdvMap
: boost::statechart::event
< AdvMap
> {
1573 vector
<int> newup
, newacting
;
1574 int up_primary
, acting_primary
;
1576 OSDMapRef osdmap
, OSDMapRef lastmap
,
1577 vector
<int>& newup
, int up_primary
,
1578 vector
<int>& newacting
, int acting_primary
):
1579 osdmap(osdmap
), lastmap(lastmap
),
1581 newacting(newacting
),
1582 up_primary(up_primary
),
1583 acting_primary(acting_primary
) {}
1584 void print(std::ostream
*out
) const {
1589 struct ActMap
: boost::statechart::event
< ActMap
> {
1590 ActMap() : boost::statechart::event
< ActMap
>() {}
1591 void print(std::ostream
*out
) const {
1595 struct Activate
: boost::statechart::event
< Activate
> {
1596 epoch_t activation_epoch
;
1597 explicit Activate(epoch_t q
) : boost::statechart::event
< Activate
>(),
1598 activation_epoch(q
) {}
1599 void print(std::ostream
*out
) const {
1600 *out
<< "Activate from " << activation_epoch
;
1603 struct RequestBackfillPrio
: boost::statechart::event
< RequestBackfillPrio
> {
1605 explicit RequestBackfillPrio(unsigned prio
) :
1606 boost::statechart::event
< RequestBackfillPrio
>(),
1608 void print(std::ostream
*out
) const {
1609 *out
<< "RequestBackfillPrio: priority " << priority
;
1612 #define TrivialEvent(T) struct T : boost::statechart::event< T > { \
1613 T() : boost::statechart::event< T >() {} \
1614 void print(std::ostream *out) const { \
1618 struct DeferBackfill
: boost::statechart::event
<DeferBackfill
> {
1620 explicit DeferBackfill(float delay
) : delay(delay
) {}
1621 void print(std::ostream
*out
) const {
1622 *out
<< "DeferBackfill: delay " << delay
;
1625 struct DeferRecovery
: boost::statechart::event
<DeferRecovery
> {
1627 explicit DeferRecovery(float delay
) : delay(delay
) {}
1628 void print(std::ostream
*out
) const {
1629 *out
<< "DeferRecovery: delay " << delay
;
1632 struct UnfoundBackfill
: boost::statechart::event
<UnfoundBackfill
> {
1633 explicit UnfoundBackfill() {}
1634 void print(std::ostream
*out
) const {
1635 *out
<< "UnfoundBackfill";
1638 struct UnfoundRecovery
: boost::statechart::event
<UnfoundRecovery
> {
1639 explicit UnfoundRecovery() {}
1640 void print(std::ostream
*out
) const {
1641 *out
<< "UnfoundRecovery";
1645 TrivialEvent(Initialize
)
1647 TrivialEvent(GotInfo
)
1648 TrivialEvent(NeedUpThru
)
1649 TrivialEvent(NullEvt
)
1650 TrivialEvent(FlushedEvt
)
1651 TrivialEvent(Backfilled
)
1652 TrivialEvent(LocalBackfillReserved
)
1653 TrivialEvent(RemoteBackfillReserved
)
1654 TrivialEvent(RejectRemoteReservation
)
1655 TrivialEvent(RemoteReservationRejected
)
1656 TrivialEvent(RemoteReservationCanceled
)
1657 TrivialEvent(RequestBackfill
)
1658 TrivialEvent(RequestRecovery
)
1659 TrivialEvent(RecoveryDone
)
1660 TrivialEvent(BackfillTooFull
)
1661 TrivialEvent(RecoveryTooFull
)
1663 TrivialEvent(MakePrimary
)
1664 TrivialEvent(MakeStray
)
1665 TrivialEvent(NeedActingChange
)
1666 TrivialEvent(IsIncomplete
)
1667 TrivialEvent(IsDown
)
1669 TrivialEvent(AllReplicasRecovered
)
1670 TrivialEvent(DoRecovery
)
1671 TrivialEvent(LocalRecoveryReserved
)
1672 TrivialEvent(RemoteRecoveryReserved
)
1673 TrivialEvent(AllRemotesReserved
)
1674 TrivialEvent(AllBackfillsReserved
)
1675 TrivialEvent(GoClean
)
1677 TrivialEvent(AllReplicasActivated
)
1679 TrivialEvent(IntervalFlush
)
1681 /* Encapsulates PG recovery process */
1682 class RecoveryState
{
1683 void start_handle(RecoveryCtx
*new_ctx
);
1686 void begin_block_outgoing();
1687 void end_block_outgoing();
1688 void clear_blocked_outgoing();
1693 class RecoveryMachine
: public boost::statechart::state_machine
< RecoveryMachine
, Initial
> {
1694 RecoveryState
*state
;
1699 uint64_t event_count
;
1701 void clear_event_counters() {
1702 event_time
= utime_t();
1706 void log_enter(const char *state_name
);
1707 void log_exit(const char *state_name
, utime_t duration
);
1709 RecoveryMachine(RecoveryState
*state
, PG
*pg
) : state(state
), pg(pg
), event_count(0) {}
1711 /* Accessor functions for state methods */
1712 ObjectStore::Transaction
* get_cur_transaction() {
1713 assert(state
->rctx
);
1714 assert(state
->rctx
->transaction
);
1715 return state
->rctx
->transaction
;
1718 void send_query(pg_shard_t to
, const pg_query_t
&query
) {
1719 assert(state
->rctx
);
1720 assert(state
->rctx
->query_map
);
1721 (*state
->rctx
->query_map
)[to
.osd
][spg_t(pg
->info
.pgid
.pgid
, to
.shard
)] =
1725 map
<int, map
<spg_t
, pg_query_t
> > *get_query_map() {
1726 assert(state
->rctx
);
1727 assert(state
->rctx
->query_map
);
1728 return state
->rctx
->query_map
;
1731 map
<int, vector
<pair
<pg_notify_t
, PastIntervals
> > > *get_info_map() {
1732 assert(state
->rctx
);
1733 assert(state
->rctx
->info_map
);
1734 return state
->rctx
->info_map
;
1737 list
< Context
* > *get_on_safe_context_list() {
1738 assert(state
->rctx
);
1739 assert(state
->rctx
->on_safe
);
1740 return &(state
->rctx
->on_safe
->contexts
);
1743 list
< Context
* > *get_on_applied_context_list() {
1744 assert(state
->rctx
);
1745 assert(state
->rctx
->on_applied
);
1746 return &(state
->rctx
->on_applied
->contexts
);
1749 RecoveryCtx
*get_recovery_ctx() { return &*(state
->rctx
); }
1751 void send_notify(pg_shard_t to
,
1752 const pg_notify_t
&info
, const PastIntervals
&pi
) {
1753 assert(state
->rctx
);
1754 assert(state
->rctx
->notify_list
);
1755 (*state
->rctx
->notify_list
)[to
.osd
].push_back(make_pair(info
, pi
));
1758 friend class RecoveryMachine
;
1778 // WaitRemoteBackfillReserved
1779 // WaitLocalBackfillReserved
1783 // WaitRemoteRecoveryReserved
1784 // WaitLocalRecoveryReserved
1788 // RepWaitBackfillReserved
1789 // RepWaitRecoveryReserved
1792 struct Crashed
: boost::statechart::state
< Crashed
, RecoveryMachine
>, NamedState
{
1793 explicit Crashed(my_context ctx
);
1798 struct Initial
: boost::statechart::state
< Initial
, RecoveryMachine
>, NamedState
{
1799 explicit Initial(my_context ctx
);
1802 typedef boost::mpl::list
<
1803 boost::statechart::transition
< Initialize
, Reset
>,
1804 boost::statechart::custom_reaction
< Load
>,
1805 boost::statechart::custom_reaction
< NullEvt
>,
1806 boost::statechart::transition
< boost::statechart::event_base
, Crashed
>
1809 boost::statechart::result
react(const Load
&);
1810 boost::statechart::result
react(const MNotifyRec
&);
1811 boost::statechart::result
react(const MInfoRec
&);
1812 boost::statechart::result
react(const MLogRec
&);
1813 boost::statechart::result
react(const boost::statechart::event_base
&) {
1814 return discard_event();
1818 struct Reset
: boost::statechart::state
< Reset
, RecoveryMachine
>, NamedState
{
1819 explicit Reset(my_context ctx
);
1822 typedef boost::mpl::list
<
1823 boost::statechart::custom_reaction
< QueryState
>,
1824 boost::statechart::custom_reaction
< AdvMap
>,
1825 boost::statechart::custom_reaction
< ActMap
>,
1826 boost::statechart::custom_reaction
< NullEvt
>,
1827 boost::statechart::custom_reaction
< FlushedEvt
>,
1828 boost::statechart::custom_reaction
< IntervalFlush
>,
1829 boost::statechart::transition
< boost::statechart::event_base
, Crashed
>
1831 boost::statechart::result
react(const QueryState
& q
);
1832 boost::statechart::result
react(const AdvMap
&);
1833 boost::statechart::result
react(const ActMap
&);
1834 boost::statechart::result
react(const FlushedEvt
&);
1835 boost::statechart::result
react(const IntervalFlush
&);
1836 boost::statechart::result
react(const boost::statechart::event_base
&) {
1837 return discard_event();
1843 struct Started
: boost::statechart::state
< Started
, RecoveryMachine
, Start
>, NamedState
{
1844 explicit Started(my_context ctx
);
1847 typedef boost::mpl::list
<
1848 boost::statechart::custom_reaction
< QueryState
>,
1849 boost::statechart::custom_reaction
< AdvMap
>,
1850 boost::statechart::custom_reaction
< NullEvt
>,
1851 boost::statechart::custom_reaction
< FlushedEvt
>,
1852 boost::statechart::custom_reaction
< IntervalFlush
>,
1853 boost::statechart::transition
< boost::statechart::event_base
, Crashed
>
1855 boost::statechart::result
react(const QueryState
& q
);
1856 boost::statechart::result
react(const AdvMap
&);
1857 boost::statechart::result
react(const FlushedEvt
&);
1858 boost::statechart::result
react(const IntervalFlush
&);
1859 boost::statechart::result
react(const boost::statechart::event_base
&) {
1860 return discard_event();
1867 struct Start
: boost::statechart::state
< Start
, Started
>, NamedState
{
1868 explicit Start(my_context ctx
);
1871 typedef boost::mpl::list
<
1872 boost::statechart::transition
< MakePrimary
, Primary
>,
1873 boost::statechart::transition
< MakeStray
, Stray
>
1878 struct WaitActingChange
;
1882 struct Primary
: boost::statechart::state
< Primary
, Started
, Peering
>, NamedState
{
1883 explicit Primary(my_context ctx
);
1886 typedef boost::mpl::list
<
1887 boost::statechart::custom_reaction
< ActMap
>,
1888 boost::statechart::custom_reaction
< MNotifyRec
>,
1889 boost::statechart::transition
< NeedActingChange
, WaitActingChange
>
1891 boost::statechart::result
react(const ActMap
&);
1892 boost::statechart::result
react(const MNotifyRec
&);
1895 struct WaitActingChange
: boost::statechart::state
< WaitActingChange
, Primary
>,
1897 typedef boost::mpl::list
<
1898 boost::statechart::custom_reaction
< QueryState
>,
1899 boost::statechart::custom_reaction
< AdvMap
>,
1900 boost::statechart::custom_reaction
< MLogRec
>,
1901 boost::statechart::custom_reaction
< MInfoRec
>,
1902 boost::statechart::custom_reaction
< MNotifyRec
>
1904 explicit WaitActingChange(my_context ctx
);
1905 boost::statechart::result
react(const QueryState
& q
);
1906 boost::statechart::result
react(const AdvMap
&);
1907 boost::statechart::result
react(const MLogRec
&);
1908 boost::statechart::result
react(const MInfoRec
&);
1909 boost::statechart::result
react(const MNotifyRec
&);
1916 struct Peering
: boost::statechart::state
< Peering
, Primary
, GetInfo
>, NamedState
{
1917 PastIntervals::PriorSet prior_set
;
1918 bool history_les_bound
; //< need osd_find_best_info_ignore_history_les
1920 explicit Peering(my_context ctx
);
1923 typedef boost::mpl::list
<
1924 boost::statechart::custom_reaction
< QueryState
>,
1925 boost::statechart::transition
< Activate
, Active
>,
1926 boost::statechart::custom_reaction
< AdvMap
>
1928 boost::statechart::result
react(const QueryState
& q
);
1929 boost::statechart::result
react(const AdvMap
&advmap
);
1932 struct WaitLocalRecoveryReserved
;
1934 struct Active
: boost::statechart::state
< Active
, Primary
, Activating
>, NamedState
{
1935 explicit Active(my_context ctx
);
1938 const set
<pg_shard_t
> remote_shards_to_reserve_recovery
;
1939 const set
<pg_shard_t
> remote_shards_to_reserve_backfill
;
1940 bool all_replicas_activated
;
1942 typedef boost::mpl::list
<
1943 boost::statechart::custom_reaction
< QueryState
>,
1944 boost::statechart::custom_reaction
< ActMap
>,
1945 boost::statechart::custom_reaction
< AdvMap
>,
1946 boost::statechart::custom_reaction
< MInfoRec
>,
1947 boost::statechart::custom_reaction
< MNotifyRec
>,
1948 boost::statechart::custom_reaction
< MLogRec
>,
1949 boost::statechart::custom_reaction
< Backfilled
>,
1950 boost::statechart::custom_reaction
< AllReplicasActivated
>,
1951 boost::statechart::custom_reaction
< DeferRecovery
>,
1952 boost::statechart::custom_reaction
< DeferBackfill
>,
1953 boost::statechart::custom_reaction
< UnfoundRecovery
>,
1954 boost::statechart::custom_reaction
< UnfoundBackfill
>,
1955 boost::statechart::custom_reaction
< DoRecovery
>
1957 boost::statechart::result
react(const QueryState
& q
);
1958 boost::statechart::result
react(const ActMap
&);
1959 boost::statechart::result
react(const AdvMap
&);
1960 boost::statechart::result
react(const MInfoRec
& infoevt
);
1961 boost::statechart::result
react(const MNotifyRec
& notevt
);
1962 boost::statechart::result
react(const MLogRec
& logevt
);
1963 boost::statechart::result
react(const Backfilled
&) {
1964 return discard_event();
1966 boost::statechart::result
react(const AllReplicasActivated
&);
1967 boost::statechart::result
react(const DeferRecovery
& evt
) {
1968 return discard_event();
1970 boost::statechart::result
react(const DeferBackfill
& evt
) {
1971 return discard_event();
1973 boost::statechart::result
react(const UnfoundRecovery
& evt
) {
1974 return discard_event();
1976 boost::statechart::result
react(const UnfoundBackfill
& evt
) {
1977 return discard_event();
1979 boost::statechart::result
react(const DoRecovery
&) {
1980 return discard_event();
1984 struct Clean
: boost::statechart::state
< Clean
, Active
>, NamedState
{
1985 typedef boost::mpl::list
<
1986 boost::statechart::transition
< DoRecovery
, WaitLocalRecoveryReserved
>
1988 explicit Clean(my_context ctx
);
1992 struct Recovered
: boost::statechart::state
< Recovered
, Active
>, NamedState
{
1993 typedef boost::mpl::list
<
1994 boost::statechart::transition
< GoClean
, Clean
>,
1995 boost::statechart::transition
< DoRecovery
, WaitLocalRecoveryReserved
>,
1996 boost::statechart::custom_reaction
< AllReplicasActivated
>
1998 explicit Recovered(my_context ctx
);
2000 boost::statechart::result
react(const AllReplicasActivated
&) {
2001 post_event(GoClean());
2002 return forward_event();
2006 struct Backfilling
: boost::statechart::state
< Backfilling
, Active
>, NamedState
{
2007 typedef boost::mpl::list
<
2008 boost::statechart::transition
< Backfilled
, Recovered
>,
2009 boost::statechart::custom_reaction
< DeferBackfill
>,
2010 boost::statechart::custom_reaction
< UnfoundBackfill
>,
2011 boost::statechart::custom_reaction
< RemoteReservationRejected
>
2013 explicit Backfilling(my_context ctx
);
2014 boost::statechart::result
react(const RemoteReservationRejected
& evt
);
2015 boost::statechart::result
react(const DeferBackfill
& evt
);
2016 boost::statechart::result
react(const UnfoundBackfill
& evt
);
2020 struct WaitRemoteBackfillReserved
: boost::statechart::state
< WaitRemoteBackfillReserved
, Active
>, NamedState
{
2021 typedef boost::mpl::list
<
2022 boost::statechart::custom_reaction
< RemoteBackfillReserved
>,
2023 boost::statechart::custom_reaction
< RemoteReservationRejected
>,
2024 boost::statechart::transition
< AllBackfillsReserved
, Backfilling
>
2026 set
<pg_shard_t
>::const_iterator backfill_osd_it
;
2027 explicit WaitRemoteBackfillReserved(my_context ctx
);
2029 boost::statechart::result
react(const RemoteBackfillReserved
& evt
);
2030 boost::statechart::result
react(const RemoteReservationRejected
& evt
);
2033 struct WaitLocalBackfillReserved
: boost::statechart::state
< WaitLocalBackfillReserved
, Active
>, NamedState
{
2034 typedef boost::mpl::list
<
2035 boost::statechart::transition
< LocalBackfillReserved
, WaitRemoteBackfillReserved
>
2037 explicit WaitLocalBackfillReserved(my_context ctx
);
2041 struct NotBackfilling
: boost::statechart::state
< NotBackfilling
, Active
>, NamedState
{
2042 typedef boost::mpl::list
<
2043 boost::statechart::transition
< RequestBackfill
, WaitLocalBackfillReserved
>,
2044 boost::statechart::custom_reaction
< RemoteBackfillReserved
>,
2045 boost::statechart::custom_reaction
< RemoteReservationRejected
>
2047 explicit NotBackfilling(my_context ctx
);
2049 boost::statechart::result
react(const RemoteBackfillReserved
& evt
);
2050 boost::statechart::result
react(const RemoteReservationRejected
& evt
);
2053 struct NotRecovering
: boost::statechart::state
< NotRecovering
, Active
>, NamedState
{
2054 typedef boost::mpl::list
<
2055 boost::statechart::transition
< DoRecovery
, WaitLocalRecoveryReserved
>,
2056 boost::statechart::custom_reaction
< DeferRecovery
>,
2057 boost::statechart::custom_reaction
< UnfoundRecovery
>
2059 explicit NotRecovering(my_context ctx
);
2060 boost::statechart::result
react(const DeferRecovery
& evt
) {
2062 return discard_event();
2064 boost::statechart::result
react(const UnfoundRecovery
& evt
) {
2066 return discard_event();
2071 struct RepNotRecovering
;
2072 struct ReplicaActive
: boost::statechart::state
< ReplicaActive
, Started
, RepNotRecovering
>, NamedState
{
2073 explicit ReplicaActive(my_context ctx
);
2076 typedef boost::mpl::list
<
2077 boost::statechart::custom_reaction
< QueryState
>,
2078 boost::statechart::custom_reaction
< ActMap
>,
2079 boost::statechart::custom_reaction
< MQuery
>,
2080 boost::statechart::custom_reaction
< MInfoRec
>,
2081 boost::statechart::custom_reaction
< MLogRec
>,
2082 boost::statechart::custom_reaction
< Activate
>,
2083 boost::statechart::custom_reaction
< DeferRecovery
>,
2084 boost::statechart::custom_reaction
< DeferBackfill
>,
2085 boost::statechart::custom_reaction
< UnfoundRecovery
>,
2086 boost::statechart::custom_reaction
< UnfoundBackfill
>
2088 boost::statechart::result
react(const QueryState
& q
);
2089 boost::statechart::result
react(const MInfoRec
& infoevt
);
2090 boost::statechart::result
react(const MLogRec
& logevt
);
2091 boost::statechart::result
react(const ActMap
&);
2092 boost::statechart::result
react(const MQuery
&);
2093 boost::statechart::result
react(const Activate
&);
2094 boost::statechart::result
react(const DeferRecovery
& evt
) {
2095 return discard_event();
2097 boost::statechart::result
react(const DeferBackfill
& evt
) {
2098 return discard_event();
2100 boost::statechart::result
react(const UnfoundRecovery
& evt
) {
2101 return discard_event();
2103 boost::statechart::result
react(const UnfoundBackfill
& evt
) {
2104 return discard_event();
2108 struct RepRecovering
: boost::statechart::state
< RepRecovering
, ReplicaActive
>, NamedState
{
2109 typedef boost::mpl::list
<
2110 boost::statechart::transition
< RecoveryDone
, RepNotRecovering
>,
2111 // for compat with old peers
2112 boost::statechart::transition
< RemoteReservationRejected
, RepNotRecovering
>,
2113 boost::statechart::transition
< RemoteReservationCanceled
, RepNotRecovering
>,
2114 boost::statechart::custom_reaction
< BackfillTooFull
>
2116 explicit RepRecovering(my_context ctx
);
2117 boost::statechart::result
react(const BackfillTooFull
&evt
);
2121 struct RepWaitBackfillReserved
: boost::statechart::state
< RepWaitBackfillReserved
, ReplicaActive
>, NamedState
{
2122 typedef boost::mpl::list
<
2123 boost::statechart::custom_reaction
< RemoteBackfillReserved
>,
2124 boost::statechart::custom_reaction
< RejectRemoteReservation
>,
2125 boost::statechart::custom_reaction
< RemoteReservationRejected
>,
2126 boost::statechart::custom_reaction
< RemoteReservationCanceled
>
2128 explicit RepWaitBackfillReserved(my_context ctx
);
2130 boost::statechart::result
react(const RemoteBackfillReserved
&evt
);
2131 boost::statechart::result
react(const RejectRemoteReservation
&evt
);
2132 boost::statechart::result
react(const RemoteReservationRejected
&evt
);
2133 boost::statechart::result
react(const RemoteReservationCanceled
&evt
);
2136 struct RepWaitRecoveryReserved
: boost::statechart::state
< RepWaitRecoveryReserved
, ReplicaActive
>, NamedState
{
2137 typedef boost::mpl::list
<
2138 boost::statechart::custom_reaction
< RemoteRecoveryReserved
>,
2139 // for compat with old peers
2140 boost::statechart::custom_reaction
< RemoteReservationRejected
>,
2141 boost::statechart::custom_reaction
< RemoteReservationCanceled
>
2143 explicit RepWaitRecoveryReserved(my_context ctx
);
2145 boost::statechart::result
react(const RemoteRecoveryReserved
&evt
);
2146 boost::statechart::result
react(const RemoteReservationRejected
&evt
) {
2147 // for compat with old peers
2148 post_event(RemoteReservationCanceled());
2149 return discard_event();
2151 boost::statechart::result
react(const RemoteReservationCanceled
&evt
);
2154 struct RepNotRecovering
: boost::statechart::state
< RepNotRecovering
, ReplicaActive
>, NamedState
{
2155 typedef boost::mpl::list
<
2156 boost::statechart::custom_reaction
< RequestBackfillPrio
>,
2157 boost::statechart::transition
< RequestRecovery
, RepWaitRecoveryReserved
>,
2158 boost::statechart::custom_reaction
< RejectRemoteReservation
>,
2159 boost::statechart::transition
< RemoteReservationRejected
, RepNotRecovering
>,
2160 boost::statechart::transition
< RemoteReservationCanceled
, RepNotRecovering
>,
2161 boost::statechart::transition
< RecoveryDone
, RepNotRecovering
> // for compat with pre-reservation peers
2163 explicit RepNotRecovering(my_context ctx
);
2164 boost::statechart::result
react(const RequestBackfillPrio
&evt
);
2165 boost::statechart::result
react(const RejectRemoteReservation
&evt
);
2169 struct Recovering
: boost::statechart::state
< Recovering
, Active
>, NamedState
{
2170 typedef boost::mpl::list
<
2171 boost::statechart::custom_reaction
< AllReplicasRecovered
>,
2172 boost::statechart::custom_reaction
< DeferRecovery
>,
2173 boost::statechart::custom_reaction
< UnfoundRecovery
>,
2174 boost::statechart::custom_reaction
< RequestBackfill
>
2176 explicit Recovering(my_context ctx
);
2178 void release_reservations(bool cancel
= false);
2179 boost::statechart::result
react(const AllReplicasRecovered
&evt
);
2180 boost::statechart::result
react(const DeferRecovery
& evt
);
2181 boost::statechart::result
react(const UnfoundRecovery
& evt
);
2182 boost::statechart::result
react(const RequestBackfill
&evt
);
2185 struct WaitRemoteRecoveryReserved
: boost::statechart::state
< WaitRemoteRecoveryReserved
, Active
>, NamedState
{
2186 typedef boost::mpl::list
<
2187 boost::statechart::custom_reaction
< RemoteRecoveryReserved
>,
2188 boost::statechart::transition
< AllRemotesReserved
, Recovering
>
2190 set
<pg_shard_t
>::const_iterator remote_recovery_reservation_it
;
2191 explicit WaitRemoteRecoveryReserved(my_context ctx
);
2192 boost::statechart::result
react(const RemoteRecoveryReserved
&evt
);
2196 struct WaitLocalRecoveryReserved
: boost::statechart::state
< WaitLocalRecoveryReserved
, Active
>, NamedState
{
2197 typedef boost::mpl::list
<
2198 boost::statechart::transition
< LocalRecoveryReserved
, WaitRemoteRecoveryReserved
>,
2199 boost::statechart::custom_reaction
< RecoveryTooFull
>
2201 explicit WaitLocalRecoveryReserved(my_context ctx
);
2203 boost::statechart::result
react(const RecoveryTooFull
&evt
);
2206 struct Activating
: boost::statechart::state
< Activating
, Active
>, NamedState
{
2207 typedef boost::mpl::list
<
2208 boost::statechart::transition
< AllReplicasRecovered
, Recovered
>,
2209 boost::statechart::transition
< DoRecovery
, WaitLocalRecoveryReserved
>,
2210 boost::statechart::transition
< RequestBackfill
, WaitLocalBackfillReserved
>
2212 explicit Activating(my_context ctx
);
2216 struct Stray
: boost::statechart::state
< Stray
, Started
>, NamedState
{
2217 map
<int, pair
<pg_query_t
, epoch_t
> > pending_queries
;
2219 explicit Stray(my_context ctx
);
2222 typedef boost::mpl::list
<
2223 boost::statechart::custom_reaction
< MQuery
>,
2224 boost::statechart::custom_reaction
< MLogRec
>,
2225 boost::statechart::custom_reaction
< MInfoRec
>,
2226 boost::statechart::custom_reaction
< ActMap
>,
2227 boost::statechart::custom_reaction
< RecoveryDone
>
2229 boost::statechart::result
react(const MQuery
& query
);
2230 boost::statechart::result
react(const MLogRec
& logevt
);
2231 boost::statechart::result
react(const MInfoRec
& infoevt
);
2232 boost::statechart::result
react(const ActMap
&);
2233 boost::statechart::result
react(const RecoveryDone
&) {
2234 return discard_event();
2240 struct GetInfo
: boost::statechart::state
< GetInfo
, Peering
>, NamedState
{
2241 set
<pg_shard_t
> peer_info_requested
;
2243 explicit GetInfo(my_context ctx
);
2247 typedef boost::mpl::list
<
2248 boost::statechart::custom_reaction
< QueryState
>,
2249 boost::statechart::transition
< GotInfo
, GetLog
>,
2250 boost::statechart::custom_reaction
< MNotifyRec
>,
2251 boost::statechart::transition
< IsDown
, Down
>
2253 boost::statechart::result
react(const QueryState
& q
);
2254 boost::statechart::result
react(const MNotifyRec
& infoevt
);
2257 struct GotLog
: boost::statechart::event
< GotLog
> {
2258 GotLog() : boost::statechart::event
< GotLog
>() {}
2261 struct GetLog
: boost::statechart::state
< GetLog
, Peering
>, NamedState
{
2262 pg_shard_t auth_log_shard
;
2263 boost::intrusive_ptr
<MOSDPGLog
> msg
;
2265 explicit GetLog(my_context ctx
);
2268 typedef boost::mpl::list
<
2269 boost::statechart::custom_reaction
< QueryState
>,
2270 boost::statechart::custom_reaction
< MLogRec
>,
2271 boost::statechart::custom_reaction
< GotLog
>,
2272 boost::statechart::custom_reaction
< AdvMap
>,
2273 boost::statechart::transition
< IsIncomplete
, Incomplete
>
2275 boost::statechart::result
react(const AdvMap
&);
2276 boost::statechart::result
react(const QueryState
& q
);
2277 boost::statechart::result
react(const MLogRec
& logevt
);
2278 boost::statechart::result
react(const GotLog
&);
2283 struct GetMissing
: boost::statechart::state
< GetMissing
, Peering
>, NamedState
{
2284 set
<pg_shard_t
> peer_missing_requested
;
2286 explicit GetMissing(my_context ctx
);
2289 typedef boost::mpl::list
<
2290 boost::statechart::custom_reaction
< QueryState
>,
2291 boost::statechart::custom_reaction
< MLogRec
>,
2292 boost::statechart::transition
< NeedUpThru
, WaitUpThru
>
2294 boost::statechart::result
react(const QueryState
& q
);
2295 boost::statechart::result
react(const MLogRec
& logevt
);
2298 struct WaitUpThru
: boost::statechart::state
< WaitUpThru
, Peering
>, NamedState
{
2299 explicit WaitUpThru(my_context ctx
);
2302 typedef boost::mpl::list
<
2303 boost::statechart::custom_reaction
< QueryState
>,
2304 boost::statechart::custom_reaction
< ActMap
>,
2305 boost::statechart::custom_reaction
< MLogRec
>
2307 boost::statechart::result
react(const QueryState
& q
);
2308 boost::statechart::result
react(const ActMap
& am
);
2309 boost::statechart::result
react(const MLogRec
& logrec
);
2312 struct Down
: boost::statechart::state
< Down
, Peering
>, NamedState
{
2313 explicit Down(my_context ctx
);
2314 typedef boost::mpl::list
<
2315 boost::statechart::custom_reaction
< QueryState
>
2317 boost::statechart::result
react(const QueryState
& infoevt
);
2321 struct Incomplete
: boost::statechart::state
< Incomplete
, Peering
>, NamedState
{
2322 typedef boost::mpl::list
<
2323 boost::statechart::custom_reaction
< AdvMap
>,
2324 boost::statechart::custom_reaction
< MNotifyRec
>,
2325 boost::statechart::custom_reaction
< QueryState
>
2327 explicit Incomplete(my_context ctx
);
2328 boost::statechart::result
react(const AdvMap
&advmap
);
2329 boost::statechart::result
react(const MNotifyRec
& infoevt
);
2330 boost::statechart::result
react(const QueryState
& infoevt
);
2335 RecoveryMachine machine
;
2338 /// context passed in by state machine caller
2339 RecoveryCtx
*orig_ctx
;
2341 /// populated if we are buffering messages pending a flush
2342 boost::optional
<BufferedRecoveryMessages
> messages_pending_flush
;
2345 * populated between start_handle() and end_handle(), points into
2346 * the message lists for messages_pending_flush while blocking messages
2347 * or into orig_ctx otherwise
2349 boost::optional
<RecoveryCtx
> rctx
;
2352 explicit RecoveryState(PG
*pg
)
2353 : machine(this, pg
), pg(pg
), orig_ctx(0) {
2357 void handle_event(const boost::statechart::event_base
&evt
,
2358 RecoveryCtx
*rctx
) {
2360 machine
.process_event(evt
);
2364 void handle_event(CephPeeringEvtRef evt
,
2365 RecoveryCtx
*rctx
) {
2367 machine
.process_event(evt
->get_event());
2375 PG(OSDService
*o
, OSDMapRef curmap
,
2376 const PGPool
&pool
, spg_t p
);
2381 explicit PG(const PG
& rhs
);
2382 PG
& operator=(const PG
& rhs
);
2384 uint64_t peer_features
;
2385 uint64_t acting_features
;
2386 uint64_t upacting_features
;
2391 const spg_t
& get_pgid() const { return pg_id
; }
2393 void reset_min_peer_features() {
2394 peer_features
= CEPH_FEATURES_SUPPORTED_DEFAULT
;
2396 uint64_t get_min_peer_features() const { return peer_features
; }
2397 void apply_peer_features(uint64_t f
) { peer_features
&= f
; }
2399 uint64_t get_min_acting_features() const { return acting_features
; }
2400 uint64_t get_min_upacting_features() const { return upacting_features
; }
2401 bool perform_deletes_during_peering() const {
2402 return !(get_osdmap()->test_flag(CEPH_OSDMAP_RECOVERY_DELETES
));
2405 void init_primary_up_acting(
2406 const vector
<int> &newup
,
2407 const vector
<int> &newacting
,
2409 int new_acting_primary
) {
2412 for (uint8_t i
= 0; i
< acting
.size(); ++i
) {
2413 if (acting
[i
] != CRUSH_ITEM_NONE
)
2417 pool
.info
.ec_pool() ? shard_id_t(i
) : shard_id_t::NO_SHARD
));
2421 for (uint8_t i
= 0; i
< up
.size(); ++i
) {
2422 if (up
[i
] != CRUSH_ITEM_NONE
)
2426 pool
.info
.ec_pool() ? shard_id_t(i
) : shard_id_t::NO_SHARD
));
2428 if (!pool
.info
.ec_pool()) {
2429 up_primary
= pg_shard_t(new_up_primary
, shard_id_t::NO_SHARD
);
2430 primary
= pg_shard_t(new_acting_primary
, shard_id_t::NO_SHARD
);
2433 up_primary
= pg_shard_t();
2434 primary
= pg_shard_t();
2435 for (uint8_t i
= 0; i
< up
.size(); ++i
) {
2436 if (up
[i
] == new_up_primary
) {
2437 up_primary
= pg_shard_t(up
[i
], shard_id_t(i
));
2441 for (uint8_t i
= 0; i
< acting
.size(); ++i
) {
2442 if (acting
[i
] == new_acting_primary
) {
2443 primary
= pg_shard_t(acting
[i
], shard_id_t(i
));
2447 assert(up_primary
.osd
== new_up_primary
);
2448 assert(primary
.osd
== new_acting_primary
);
2450 pg_shard_t
get_primary() const { return primary
; }
2452 int get_role() const { return role
; }
2453 void set_role(int r
) { role
= r
; }
2455 bool is_primary() const { return pg_whoami
== primary
; }
2456 bool is_replica() const { return role
> 0; }
2458 epoch_t
get_last_peering_reset() const { return last_peering_reset
; }
2460 //int get_state() const { return state; }
2461 bool state_test(int m
) const { return (state
& m
) != 0; }
2462 void state_set(int m
) { state
|= m
; }
2463 void state_clear(int m
) { state
&= ~m
; }
2465 bool is_complete() const { return info
.last_complete
== info
.last_update
; }
2466 bool should_send_notify() const { return send_notify
; }
2468 int get_state() const { return state
; }
2469 bool is_active() const { return state_test(PG_STATE_ACTIVE
); }
2470 bool is_activating() const { return state_test(PG_STATE_ACTIVATING
); }
2471 bool is_peering() const { return state_test(PG_STATE_PEERING
); }
2472 bool is_down() const { return state_test(PG_STATE_DOWN
); }
2473 bool is_recovery_unfound() const { return state_test(PG_STATE_RECOVERY_UNFOUND
); }
2474 bool is_backfill_unfound() const { return state_test(PG_STATE_BACKFILL_UNFOUND
); }
2475 bool is_incomplete() const { return state_test(PG_STATE_INCOMPLETE
); }
2476 bool is_clean() const { return state_test(PG_STATE_CLEAN
); }
2477 bool is_degraded() const { return state_test(PG_STATE_DEGRADED
); }
2478 bool is_undersized() const { return state_test(PG_STATE_UNDERSIZED
); }
2480 bool is_scrubbing() const { return state_test(PG_STATE_SCRUBBING
); }
2481 bool is_remapped() const { return state_test(PG_STATE_REMAPPED
); }
2482 bool is_peered() const {
2483 return state_test(PG_STATE_ACTIVE
) || state_test(PG_STATE_PEERED
);
2486 bool is_empty() const { return info
.last_update
== eversion_t(0,0); }
2490 const vector
<int>& up
,
2492 const vector
<int>& acting
,
2494 const pg_history_t
& history
,
2495 const PastIntervals
& pim
,
2497 ObjectStore::Transaction
*t
);
2500 void do_pending_flush();
2502 static void _create(ObjectStore::Transaction
& t
, spg_t pgid
, int bits
);
2503 static void _init(ObjectStore::Transaction
& t
,
2504 spg_t pgid
, const pg_pool_t
*pool
);
2507 void prepare_write_info(map
<string
,bufferlist
> *km
);
2509 void update_store_with_options();
2510 void update_store_on_load();
2513 static int _prepare_write_info(
2515 map
<string
,bufferlist
> *km
,
2518 pg_info_t
&last_written_info
,
2519 PastIntervals
&past_intervals
,
2520 bool dirty_big_info
,
2523 PerfCounters
*logger
= nullptr);
2524 void write_if_dirty(ObjectStore::Transaction
& t
);
2526 PGLog::IndexedLog projected_log
;
2527 bool check_in_progress_op(
2528 const osd_reqid_t
&r
,
2529 eversion_t
*version
,
2530 version_t
*user_version
,
2531 int *return_code
) const;
2532 eversion_t projected_last_update
;
2533 eversion_t
get_next_version() const {
2534 eversion_t
at_version(
2535 get_osdmap()->get_epoch(),
2536 projected_last_update
.version
+1);
2537 assert(at_version
> info
.last_update
);
2538 assert(at_version
> pg_log
.get_head());
2539 assert(at_version
> projected_last_update
);
2543 void add_log_entry(const pg_log_entry_t
& e
, bool applied
);
2545 const vector
<pg_log_entry_t
>& logv
,
2547 eversion_t roll_forward_to
,
2548 ObjectStore::Transaction
&t
,
2549 bool transaction_applied
= true);
2550 bool check_log_for_corruption(ObjectStore
*store
);
2553 std::string
get_corrupt_pg_log_name() const;
2554 static int read_info(
2555 ObjectStore
*store
, spg_t pgid
, const coll_t
&coll
,
2556 bufferlist
&bl
, pg_info_t
&info
, PastIntervals
&past_intervals
,
2558 void read_state(ObjectStore
*store
, bufferlist
&bl
);
2559 static bool _has_removal_flag(ObjectStore
*store
, spg_t pgid
);
2560 static int peek_map_epoch(ObjectStore
*store
, spg_t pgid
,
2561 epoch_t
*pepoch
, bufferlist
*bl
);
2562 void update_snap_map(
2563 const vector
<pg_log_entry_t
> &log_entries
,
2564 ObjectStore::Transaction
& t
);
2566 void filter_snapc(vector
<snapid_t
> &snaps
);
2568 void log_weirdness();
2570 virtual void kick_snap_trim() = 0;
2571 virtual void snap_trimmer_scrub_complete() = 0;
2572 bool requeue_scrub(bool high_priority
= false);
2573 void queue_recovery();
2575 unsigned get_scrub_priority();
2577 /// share pg info after a pg is active
2578 void share_pg_info();
2581 bool append_log_entries_update_missing(
2582 const mempool::osd_pglog::list
<pg_log_entry_t
> &entries
,
2583 ObjectStore::Transaction
&t
,
2584 boost::optional
<eversion_t
> trim_to
,
2585 boost::optional
<eversion_t
> roll_forward_to
);
2588 * Merge entries updating missing as necessary on all
2589 * actingbackfill logs and missings (also missing_loc)
2591 void merge_new_log_entries(
2592 const mempool::osd_pglog::list
<pg_log_entry_t
> &entries
,
2593 ObjectStore::Transaction
&t
,
2594 boost::optional
<eversion_t
> trim_to
,
2595 boost::optional
<eversion_t
> roll_forward_to
);
2597 void reset_interval_flush();
2598 void start_peering_interval(
2599 const OSDMapRef lastmap
,
2600 const vector
<int>& newup
, int up_primary
,
2601 const vector
<int>& newacting
, int acting_primary
,
2602 ObjectStore::Transaction
*t
);
2603 void on_new_interval();
2604 virtual void _on_new_interval() = 0;
2605 void start_flush(ObjectStore::Transaction
*t
,
2606 list
<Context
*> *on_applied
,
2607 list
<Context
*> *on_safe
);
2608 void set_last_peering_reset();
2609 bool pg_has_reset_since(epoch_t e
) {
2610 assert(is_locked());
2611 return deleting
|| e
< get_last_peering_reset();
2614 void update_history(const pg_history_t
& history
);
2615 void fulfill_info(pg_shard_t from
, const pg_query_t
&query
,
2616 pair
<pg_shard_t
, pg_info_t
> ¬ify_info
);
2617 void fulfill_log(pg_shard_t from
, const pg_query_t
&query
, epoch_t query_epoch
);
2619 void check_full_transition(OSDMapRef lastmap
, OSDMapRef osdmap
);
2621 bool should_restart_peering(
2623 int newactingprimary
,
2624 const vector
<int>& newup
,
2625 const vector
<int>& newacting
,
2629 // OpRequest queueing
2630 bool can_discard_op(OpRequestRef
& op
);
2631 bool can_discard_scan(OpRequestRef op
);
2632 bool can_discard_backfill(OpRequestRef op
);
2633 bool can_discard_request(OpRequestRef
& op
);
2635 template<typename T
, int MSGTYPE
>
2636 bool can_discard_replica_op(OpRequestRef
& op
);
2638 bool old_peering_msg(epoch_t reply_epoch
, epoch_t query_epoch
);
2639 bool old_peering_evt(CephPeeringEvtRef evt
) {
2640 return old_peering_msg(evt
->get_epoch_sent(), evt
->get_epoch_requested());
2642 static bool have_same_or_newer_map(epoch_t cur_epoch
, epoch_t e
) {
2643 return e
<= cur_epoch
;
2645 bool have_same_or_newer_map(epoch_t e
) {
2646 return e
<= get_osdmap()->get_epoch();
2649 bool op_has_sufficient_caps(OpRequestRef
& op
);
2653 void take_waiters();
2654 void queue_peering_event(CephPeeringEvtRef evt
);
2655 void handle_peering_event(CephPeeringEvtRef evt
, RecoveryCtx
*rctx
);
2656 void queue_query(epoch_t msg_epoch
, epoch_t query_epoch
,
2657 pg_shard_t from
, const pg_query_t
& q
);
2658 void queue_null(epoch_t msg_epoch
, epoch_t query_epoch
);
2659 void queue_flushed(epoch_t started_at
);
2660 void handle_advance_map(
2661 OSDMapRef osdmap
, OSDMapRef lastmap
,
2662 vector
<int>& newup
, int up_primary
,
2663 vector
<int>& newacting
, int acting_primary
,
2665 void handle_activate_map(RecoveryCtx
*rctx
);
2666 void handle_create(RecoveryCtx
*rctx
);
2667 void handle_loaded(RecoveryCtx
*rctx
);
2668 void handle_query_state(Formatter
*f
);
2670 virtual void on_removal(ObjectStore::Transaction
*t
) = 0;
2674 virtual void do_request(
2676 ThreadPool::TPHandle
&handle
2679 virtual void do_op(OpRequestRef
& op
) = 0;
2680 virtual void do_sub_op(OpRequestRef op
) = 0;
2681 virtual void do_sub_op_reply(OpRequestRef op
) = 0;
2682 virtual void do_scan(
2684 ThreadPool::TPHandle
&handle
2686 virtual void do_backfill(OpRequestRef op
) = 0;
2687 virtual void snap_trimmer(epoch_t epoch_queued
) = 0;
2689 virtual int do_command(
2695 ceph_tid_t tid
) = 0;
2697 virtual void on_role_change() = 0;
2698 virtual void on_pool_change() = 0;
2699 virtual void on_change(ObjectStore::Transaction
*t
) = 0;
2700 virtual void on_activate() = 0;
2701 virtual void on_flushed() = 0;
2702 virtual void on_shutdown() = 0;
2703 virtual void check_blacklisted_watchers() = 0;
2704 virtual void get_watchers(std::list
<obj_watch_item_t
>&) = 0;
2706 virtual bool agent_work(int max
) = 0;
2707 virtual bool agent_work(int max
, int agent_flush_quota
) = 0;
2708 virtual void agent_stop() = 0;
2709 virtual void agent_delay() = 0;
2710 virtual void agent_clear() = 0;
2711 virtual void agent_choose_mode_restart() = 0;
2714 ostream
& operator<<(ostream
& out
, const PG
& pg
);
2716 ostream
& operator<<(ostream
& out
, const PG::BackfillInterval
& bi
);