1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
3 * Ceph - scalable distributed file system
5 * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
6 * Copyright (C) 2013 Cloudwatt <libre.licensing@cloudwatt.com>
8 * Author: Loic Dachary <loic@dachary.org>
10 * This is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License version 2.1, as published by the Free Software
13 * Foundation. See file COPYING.
17 #ifndef CEPH_REPLICATEDPG_H
18 #define CEPH_REPLICATEDPG_H
20 #include <boost/tuple/tuple.hpp>
21 #include "include/assert.h"
24 #include "TierAgentState.h"
25 #include "messages/MOSDOpReply.h"
26 #include "common/Checksummer.h"
27 #include "common/sharedptr_registry.hpp"
28 #include "ReplicatedBackend.h"
29 #include "PGTransaction.h"
31 class CopyFromCallback
;
32 class PromoteCallback
;
37 struct TierAgentState
;
42 void intrusive_ptr_add_ref(PrimaryLogPG
*pg
);
43 void intrusive_ptr_release(PrimaryLogPG
*pg
);
44 uint64_t get_with_id(PrimaryLogPG
*pg
);
45 void put_with_id(PrimaryLogPG
*pg
, uint64_t id
);
48 typedef TrackedIntPtr
<PrimaryLogPG
> PrimaryLogPGRef
;
50 typedef boost::intrusive_ptr
<PrimaryLogPG
> PrimaryLogPGRef
;
53 struct inconsistent_snapset_wrapper
;
55 class PrimaryLogPG
: public PG
, public PGBackend::Listener
{
60 MEMPOOL_CLASS_HELPERS();
63 * state associated with a copy operation
69 * CopyResults stores the object metadata of interest to a copy initiator.
72 ceph::real_time mtime
; ///< the copy source's mtime
73 uint64_t object_size
; ///< the copied object's size
74 bool started_temp_obj
; ///< true if the callback needs to delete temp object
75 hobject_t temp_oid
; ///< temp object (if any)
78 * Function to fill in transaction; if non-empty the callback
79 * must execute it before any other accesses to the object
80 * (in order to complete the copy).
82 std::function
<void(PGTransaction
*)> fill_in_final_tx
;
84 version_t user_version
; ///< The copy source's user version
85 bool should_requeue
; ///< op should be requeued on cancel
86 vector
<snapid_t
> snaps
; ///< src's snaps (if clone)
87 snapid_t snap_seq
; ///< src's snap_seq (if head)
88 librados::snap_set_t snapset
; ///< src snapset (if head)
91 uint32_t flags
; // object_copy_data_t::FLAG_*
92 uint32_t source_data_digest
, source_omap_digest
;
93 uint32_t data_digest
, omap_digest
;
94 mempool::osd_pglog::vector
<pair
<osd_reqid_t
, version_t
> > reqids
; // [(reqid, user_version)]
95 map
<string
, bufferlist
> attrs
; // xattrs
96 uint64_t truncate_seq
;
97 uint64_t truncate_size
;
98 bool is_data_digest() {
99 return flags
& object_copy_data_t::FLAG_DATA_DIGEST
;
101 bool is_omap_digest() {
102 return flags
& object_copy_data_t::FLAG_OMAP_DIGEST
;
105 : object_size(0), started_temp_obj(false),
107 should_requeue(false), mirror_snapset(false),
110 source_data_digest(-1), source_omap_digest(-1),
111 data_digest(-1), omap_digest(-1),
112 truncate_seq(0), truncate_size(0)
118 ObjectContextRef obc
;
120 object_locator_t oloc
;
126 ceph_tid_t objecter_tid
;
127 ceph_tid_t objecter_tid2
;
129 object_copy_cursor_t cursor
;
130 map
<string
,bufferlist
> attrs
;
132 bufferlist omap_header
;
133 bufferlist omap_data
;
136 object_copy_cursor_t temp_cursor
;
139 * For CopyOp the process is:
140 * step1: read the data(attr/omap/data) from the source object
141 * step2: handle those data(w/ those data create a new object)
142 * src_obj_fadvise_flags used in step1;
143 * dest_obj_fadvise_flags used in step2
145 unsigned src_obj_fadvise_flags
;
146 unsigned dest_obj_fadvise_flags
;
148 CopyOp(CopyCallback
*cb_
, ObjectContextRef _obc
, hobject_t s
,
153 unsigned src_obj_fadvise_flags
,
154 unsigned dest_obj_fadvise_flags
)
155 : cb(cb_
), obc(_obc
), src(s
), oloc(l
), flags(f
),
160 src_obj_fadvise_flags(src_obj_fadvise_flags
),
161 dest_obj_fadvise_flags(dest_obj_fadvise_flags
)
163 results
.user_version
= v
;
164 results
.mirror_snapset
= mirror_snapset
;
167 typedef ceph::shared_ptr
<CopyOp
> CopyOpRef
;
170 * The CopyCallback class defines an interface for completions to the
171 * copy_start code. Users of the copy infrastructure must implement
172 * one and give an instance of the class to start_copy.
174 * The implementer is responsible for making sure that the CopyCallback
175 * can associate itself with the correct copy operation.
177 typedef boost::tuple
<int, CopyResults
*> CopyCallbackResults
;
179 friend class CopyFromCallback
;
180 friend class CopyFromFinisher
;
181 friend class PromoteCallback
;
186 ceph_tid_t objecter_tid
;
188 version_t user_version
;
190 bool canceled
; ///< true if canceled
192 ProxyReadOp(OpRequestRef _op
, hobject_t oid
, vector
<OSDOp
>& _ops
)
193 : op(_op
), soid(oid
),
194 objecter_tid(0), ops(_ops
),
195 user_version(0), data_offset(0),
198 typedef ceph::shared_ptr
<ProxyReadOp
> ProxyReadOpRef
;
200 struct ProxyWriteOp
{
204 ceph_tid_t objecter_tid
;
206 version_t user_version
;
212 ProxyWriteOp(OpRequestRef _op
, hobject_t oid
, vector
<OSDOp
>& _ops
, osd_reqid_t _reqid
)
213 : ctx(NULL
), op(_op
), soid(oid
),
214 objecter_tid(0), ops(_ops
),
215 user_version(0), sent_reply(false),
219 typedef ceph::shared_ptr
<ProxyWriteOp
> ProxyWriteOpRef
;
222 ObjectContextRef obc
; ///< obc we are flushing
223 OpRequestRef op
; ///< initiating op
224 list
<OpRequestRef
> dup_ops
; ///< bandwagon jumpers
225 version_t flushed_version
; ///< user version we are flushing
226 ceph_tid_t objecter_tid
; ///< copy-from request tid
227 int rval
; ///< copy-from result
228 bool blocking
; ///< whether we are blocking updates
229 bool removal
; ///< we are removing the backend object
230 boost::optional
<std::function
<void()>> on_flush
; ///< callback, may be null
233 : flushed_version(0), objecter_tid(0), rval(0),
234 blocking(false), removal(false) {}
235 ~FlushOp() { assert(!on_flush
); }
237 typedef ceph::shared_ptr
<FlushOp
> FlushOpRef
;
239 boost::scoped_ptr
<PGBackend
> pgbackend
;
240 PGBackend
*get_pgbackend() override
{
241 return pgbackend
.get();
245 DoutPrefixProvider
*get_dpp() override
{
249 void on_local_recover(
250 const hobject_t
&oid
,
251 const ObjectRecoveryInfo
&recovery_info
,
252 ObjectContextRef obc
,
254 ObjectStore::Transaction
*t
256 void on_peer_recover(
258 const hobject_t
&oid
,
259 const ObjectRecoveryInfo
&recovery_info
261 void begin_peer_recover(
263 const hobject_t oid
) override
;
264 void on_global_recover(
265 const hobject_t
&oid
,
266 const object_stat_sum_t
&stat_diff
,
267 bool is_delete
) override
;
268 void failed_push(const list
<pg_shard_t
> &from
, const hobject_t
&soid
) override
;
269 void primary_failed(const hobject_t
&soid
) override
;
270 bool primary_error(const hobject_t
& soid
, eversion_t v
) override
;
271 void cancel_pull(const hobject_t
&soid
) override
;
273 const hobject_t
&soid
,
274 const object_stat_sum_t
&delta_stats
) override
;
275 void on_primary_error(const hobject_t
&oid
, eversion_t v
) override
;
276 void backfill_add_missing(const hobject_t
&oid
, eversion_t v
) override
;
277 void remove_missing_object(const hobject_t
&oid
,
279 Context
*on_complete
) override
;
281 template<class T
> class BlessedGenContext
;
282 class BlessedContext
;
283 Context
*bless_context(Context
*c
) override
;
285 GenContext
<ThreadPool::TPHandle
&> *bless_gencontext(
286 GenContext
<ThreadPool::TPHandle
&> *c
) override
;
288 void send_message(int to_osd
, Message
*m
) override
{
289 osd
->send_message_osd_cluster(to_osd
, m
, get_osdmap()->get_epoch());
291 void queue_transaction(ObjectStore::Transaction
&& t
,
292 OpRequestRef op
) override
{
293 osd
->store
->queue_transaction(osr
.get(), std::move(t
), 0, 0, 0, op
);
295 void queue_transactions(vector
<ObjectStore::Transaction
>& tls
,
296 OpRequestRef op
) override
{
297 osd
->store
->queue_transactions(osr
.get(), tls
, 0, 0, 0, op
, NULL
);
299 epoch_t
get_epoch() const override
{
300 return get_osdmap()->get_epoch();
302 epoch_t
get_interval_start_epoch() const override
{
303 return info
.history
.same_interval_since
;
305 epoch_t
get_last_peering_reset_epoch() const override
{
306 return get_last_peering_reset();
308 const set
<pg_shard_t
> &get_actingbackfill_shards() const override
{
309 return actingbackfill
;
311 const set
<pg_shard_t
> &get_acting_shards() const override
{
314 const set
<pg_shard_t
> &get_backfill_shards() const override
{
315 return backfill_targets
;
318 std::string
gen_dbg_prefix() const override
{ return gen_prefix(); }
320 const map
<hobject_t
, set
<pg_shard_t
>>
321 &get_missing_loc_shards() const override
{
322 return missing_loc
.get_missing_locs();
324 const map
<pg_shard_t
, pg_missing_t
> &get_shard_missing() const override
{
327 using PGBackend::Listener::get_shard_missing
;
328 const map
<pg_shard_t
, pg_info_t
> &get_shard_info() const override
{
331 using PGBackend::Listener::get_shard_info
;
332 const pg_missing_tracker_t
&get_local_missing() const override
{
333 return pg_log
.get_missing();
335 const PGLog
&get_log() const override
{
338 bool pgb_is_primary() const override
{
341 OSDMapRef
pgb_get_osdmap() const override
{
344 const pg_info_t
&get_info() const override
{
347 const pg_pool_t
&get_pool() const override
{
351 ObjectContextRef
get_obc(
352 const hobject_t
&hoid
,
353 const map
<string
, bufferlist
> &attrs
) override
{
354 return get_object_context(hoid
, true, &attrs
);
357 bool try_lock_for_read(
358 const hobject_t
&hoid
,
359 ObcLockManager
&manager
) override
{
360 if (is_missing_object(hoid
))
362 auto obc
= get_object_context(hoid
, false, nullptr);
365 return manager
.try_get_read_lock(hoid
, obc
);
368 void release_locks(ObcLockManager
&manager
) override
{
369 release_object_locks(manager
);
372 void pgb_set_object_snap_mapping(
373 const hobject_t
&soid
,
374 const set
<snapid_t
> &snaps
,
375 ObjectStore::Transaction
*t
) override
{
376 return update_object_snap_mapping(t
, soid
, snaps
);
378 void pgb_clear_object_snap_mapping(
379 const hobject_t
&soid
,
380 ObjectStore::Transaction
*t
) override
{
381 return clear_object_snap_mapping(t
, soid
);
385 const vector
<pg_log_entry_t
> &logv
,
386 const boost::optional
<pg_hit_set_history_t
> &hset_history
,
387 const eversion_t
&trim_to
,
388 const eversion_t
&roll_forward_to
,
389 bool transaction_applied
,
390 ObjectStore::Transaction
&t
) override
{
392 info
.hit_set
= *hset_history
;
394 append_log(logv
, trim_to
, roll_forward_to
, t
, transaction_applied
);
397 struct C_OSD_OnApplied
;
399 const eversion_t
&applied_version
) override
;
403 const hobject_t
&hoid
) override
{
404 if (peer
== get_primary())
406 assert(peer_info
.count(peer
));
408 hoid
.pool
!= (int64_t)info
.pgid
.pool() ||
409 hoid
<= last_backfill_started
||
410 hoid
<= peer_info
[peer
].last_backfill
;
412 assert(is_backfill_targets(peer
));
416 void update_peer_last_complete_ondisk(
418 eversion_t lcod
) override
{
419 peer_last_complete_ondisk
[fromosd
] = lcod
;
422 void update_last_complete_ondisk(
423 eversion_t lcod
) override
{
424 last_complete_ondisk
= lcod
;
428 const pg_stat_t
&stat
) override
{
432 void schedule_recovery_work(
433 GenContext
<ThreadPool::TPHandle
&> *c
) override
;
435 pg_shard_t
whoami_shard() const override
{
438 spg_t
primary_spg_t() const override
{
439 return spg_t(info
.pgid
.pgid
, primary
.shard
);
441 pg_shard_t
primary_shard() const override
{
444 uint64_t min_peer_features() const override
{
445 return get_min_peer_features();
448 void send_message_osd_cluster(
449 int peer
, Message
*m
, epoch_t from_epoch
) override
;
450 void send_message_osd_cluster(
451 Message
*m
, Connection
*con
) override
;
452 void send_message_osd_cluster(
453 Message
*m
, const ConnectionRef
& con
) override
;
454 ConnectionRef
get_con_osd_cluster(int peer
, epoch_t from_epoch
) override
;
455 entity_name_t
get_cluster_msgr_name() override
{
456 return osd
->get_cluster_msgr_name();
459 PerfCounters
*get_logger() override
;
461 ceph_tid_t
get_tid() override
{ return osd
->get_tid(); }
463 LogClientTemp
clog_error() override
{ return osd
->clog
->error(); }
464 LogClientTemp
clog_warn() override
{ return osd
->clog
->warn(); }
466 struct watch_disconnect_t
{
469 bool send_disconnect
;
470 watch_disconnect_t(uint64_t c
, entity_name_t n
, bool sd
)
471 : cookie(c
), name(n
), send_disconnect(sd
) {}
473 void complete_disconnect_watches(
474 ObjectContextRef obc
,
475 const list
<watch_disconnect_t
> &to_disconnect
);
478 virtual ~OpFinisher() {
481 virtual int execute() = 0;
485 * Capture all object state associated with an in-progress read or write.
492 const ObjectState
*obs
; // Old objectstate
493 const SnapSet
*snapset
; // Old snapset
495 ObjectState new_obs
; // resulting ObjectState
496 SnapSet new_snapset
; // resulting SnapSet (in case of a write)
497 //pg_stat_t new_stats; // resulting Stats
498 object_stat_sum_t delta_stats
;
500 bool modify
; // (force) modification (even if op_t is empty)
501 bool user_modify
; // user-visible modification
502 bool undirty
; // user explicitly un-dirtying this object
503 bool cache_evict
; ///< true if this is a cache eviction
504 bool ignore_cache
; ///< true if IGNORE_CACHE flag is set
505 bool ignore_log_op_stats
; // don't log op stats
506 bool update_log_only
; ///< this is a write that returned an error - just record in pg log for dup detection
509 list
<pair
<watch_info_t
,bool> > watch_connects
; ///< new watch + will_ping flag
510 list
<watch_disconnect_t
> watch_disconnects
; ///< old watch + send_discon
511 list
<notify_info_t
> notifies
;
513 boost::optional
<uint64_t> watch_cookie
;
516 explicit NotifyAck(uint64_t notify_id
) : notify_id(notify_id
) {}
517 NotifyAck(uint64_t notify_id
, uint64_t cookie
, bufferlist
& rbl
)
518 : watch_cookie(cookie
), notify_id(notify_id
) {
522 list
<NotifyAck
> notify_acks
;
524 uint64_t bytes_written
, bytes_read
;
527 SnapContext snapc
; // writer snap context
528 eversion_t at_version
; // pg's current version pointer
529 version_t user_at_version
; // pg's current user version pointer
531 /// index of the current subop - only valid inside of do_osd_ops()
532 int current_osd_subop_num
;
533 /// total number of subops processed in this context for cls_cxx_subop_version()
534 int processed_subop_count
= 0;
536 PGTransactionUPtr op_t
;
537 vector
<pg_log_entry_t
> log
;
538 boost::optional
<pg_hit_set_history_t
> updated_hset_history
;
540 interval_set
<uint64_t> modified_ranges
;
541 ObjectContextRef obc
;
542 ObjectContextRef clone_obc
; // if we created a clone
543 ObjectContextRef snapset_obc
; // if we created/deleted a snapdir
545 // FIXME: we may want to kill this msgr hint off at some point!
546 boost::optional
<int> data_off
= boost::none
;
552 int num_read
; ///< count read ops
553 int num_write
; ///< count update ops
555 mempool::osd_pglog::vector
<pair
<osd_reqid_t
, version_t
> > extra_reqids
;
557 hobject_t new_temp_oid
, discard_temp_oid
; ///< temp objects we should start/stop tracking
559 list
<std::function
<void()>> on_applied
;
560 list
<std::function
<void()>> on_committed
;
561 list
<std::function
<void()>> on_finish
;
562 list
<std::function
<void()>> on_success
;
563 template <typename F
>
564 void register_on_finish(F
&&f
) {
565 on_finish
.emplace_back(std::forward
<F
>(f
));
567 template <typename F
>
568 void register_on_success(F
&&f
) {
569 on_success
.emplace_back(std::forward
<F
>(f
));
571 template <typename F
>
572 void register_on_applied(F
&&f
) {
573 on_applied
.emplace_back(std::forward
<F
>(f
));
575 template <typename F
>
576 void register_on_commit(F
&&f
) {
577 on_committed
.emplace_back(std::forward
<F
>(f
));
582 // pending async reads <off, len, op_flags> -> <outbl, outr>
583 list
<pair
<boost::tuple
<uint64_t, uint64_t, unsigned>,
584 pair
<bufferlist
*, Context
*> > > pending_async_reads
;
586 friend struct OnReadComplete
;
587 void start_async_reads(PrimaryLogPG
*pg
);
588 void finish_read(PrimaryLogPG
*pg
);
589 bool async_reads_complete() {
590 return inflightreads
== 0;
593 ObjectContext::RWState::State lock_type
;
594 ObcLockManager lock_manager
;
596 std::map
<int, std::unique_ptr
<OpFinisher
>> op_finishers
;
598 OpContext(const OpContext
& other
);
599 const OpContext
& operator=(const OpContext
& other
);
601 OpContext(OpRequestRef _op
, osd_reqid_t _reqid
, vector
<OSDOp
>* _ops
,
602 ObjectContextRef
& obc
,
604 op(_op
), reqid(_reqid
), ops(_ops
),
607 new_obs(obs
->oi
, obs
->exists
),
608 modify(false), user_modify(false), undirty(false), cache_evict(false),
609 ignore_cache(false), ignore_log_op_stats(false), update_log_only(false),
610 bytes_written(0), bytes_read(0), user_at_version(0),
611 current_osd_subop_num(0),
613 reply(NULL
), pg(_pg
),
618 lock_type(ObjectContext::RWState::RWNONE
) {
620 new_snapset
= obc
->ssc
->snapset
;
621 snapset
= &obc
->ssc
->snapset
;
624 OpContext(OpRequestRef _op
, osd_reqid_t _reqid
,
625 vector
<OSDOp
>* _ops
, PrimaryLogPG
*_pg
) :
626 op(_op
), reqid(_reqid
), ops(_ops
), obs(NULL
), snapset(0),
627 modify(false), user_modify(false), undirty(false), cache_evict(false),
628 ignore_cache(false), ignore_log_op_stats(false), update_log_only(false),
629 bytes_written(0), bytes_read(0), user_at_version(0),
630 current_osd_subop_num(0),
631 reply(NULL
), pg(_pg
),
635 lock_type(ObjectContext::RWState::RWNONE
) {}
636 void reset_obs(ObjectContextRef obc
) {
637 new_obs
= ObjectState(obc
->obs
.oi
, obc
->obs
.exists
);
639 new_snapset
= obc
->ssc
->snapset
;
640 snapset
= &obc
->ssc
->snapset
;
647 for (list
<pair
<boost::tuple
<uint64_t, uint64_t, unsigned>,
648 pair
<bufferlist
*, Context
*> > >::iterator i
=
649 pending_async_reads
.begin();
650 i
!= pending_async_reads
.end();
651 pending_async_reads
.erase(i
++)) {
652 delete i
->second
.second
;
655 uint64_t get_features() {
656 if (op
&& op
->get_req()) {
657 return op
->get_req()->get_connection()->get_features();
662 using OpContextUPtr
= std::unique_ptr
<OpContext
>;
663 friend struct OpContext
;
666 * State on the PG primary associated with the replicated mutation
672 xlist
<RepGather
*>::item queue_item
;
680 bool rep_aborted
, rep_done
;
684 const bool applies_with_commit
;
688 eversion_t pg_local_last_complete
;
690 ObcLockManager lock_manager
;
692 list
<std::function
<void()>> on_applied
;
693 list
<std::function
<void()>> on_committed
;
694 list
<std::function
<void()>> on_success
;
695 list
<std::function
<void()>> on_finish
;
698 OpContext
*c
, ceph_tid_t rt
,
700 bool applies_with_commit
) :
701 hoid(c
->obc
->obs
.oi
.soid
),
706 rep_aborted(false), rep_done(false),
707 all_applied(false), all_committed(false),
708 applies_with_commit(applies_with_commit
),
709 pg_local_last_complete(lc
),
710 lock_manager(std::move(c
->lock_manager
)),
711 on_applied(std::move(c
->on_applied
)),
712 on_committed(std::move(c
->on_committed
)),
713 on_success(std::move(c
->on_success
)),
714 on_finish(std::move(c
->on_finish
)) {}
717 ObcLockManager
&&manager
,
719 boost::optional
<std::function
<void(void)> > &&on_complete
,
722 bool applies_with_commit
,
729 rep_aborted(false), rep_done(false),
730 all_applied(false), all_committed(false),
731 applies_with_commit(applies_with_commit
),
732 pg_local_last_complete(lc
),
733 lock_manager(std::move(manager
)) {
735 on_success
.push_back(std::move(*on_complete
));
746 assert(on_applied
.empty());
748 //generic_dout(0) << "deleting " << this << dendl;
757 * Grabs locks for OpContext, should be cleaned up in close_op_ctx
759 * @param ctx [in,out] ctx to get locks for
760 * @return true on success, false if we are queued
762 bool get_rw_locks(bool write_ordered
, OpContext
*ctx
) {
763 /* If snapset_obc, !obc->obs->exists and we will always take the
764 * snapdir lock *before* the head lock. Since all callers will do
765 * this (read or write) if we get the first we will be guaranteed
768 if (write_ordered
&& ctx
->op
->may_read()) {
769 ctx
->lock_type
= ObjectContext::RWState::RWEXCL
;
770 } else if (write_ordered
) {
771 ctx
->lock_type
= ObjectContext::RWState::RWWRITE
;
773 assert(ctx
->op
->may_read());
774 ctx
->lock_type
= ObjectContext::RWState::RWREAD
;
777 if (ctx
->snapset_obc
) {
778 assert(!ctx
->obc
->obs
.exists
);
779 if (!ctx
->lock_manager
.get_lock_type(
781 ctx
->snapset_obc
->obs
.oi
.soid
,
784 ctx
->lock_type
= ObjectContext::RWState::RWNONE
;
788 if (ctx
->lock_manager
.get_lock_type(
790 ctx
->obc
->obs
.oi
.soid
,
795 assert(!ctx
->snapset_obc
);
796 ctx
->lock_type
= ObjectContext::RWState::RWNONE
;
802 * Cleans up OpContext
804 * @param ctx [in] ctx to clean up
806 void close_op_ctx(OpContext
*ctx
);
811 * @param manager [in] manager with locks to release
813 void release_object_locks(
814 ObcLockManager
&lock_manager
) {
815 list
<pair
<hobject_t
, list
<OpRequestRef
> > > to_req
;
816 bool requeue_recovery
= false;
817 bool requeue_snaptrim
= false;
818 lock_manager
.put_locks(
822 if (requeue_recovery
)
824 if (requeue_snaptrim
)
825 snap_trimmer_machine
.process_event(TrimWriteUnblocked());
827 if (!to_req
.empty()) {
828 // requeue at front of scrub blocking queue if we are blocked by scrub
829 for (auto &&p
: to_req
) {
830 if (scrubber
.write_blocked_by_scrub(p
.first
.get_head())) {
831 waiting_for_scrub
.splice(
832 waiting_for_scrub
.begin(),
837 requeue_ops(p
.second
);
845 xlist
<RepGather
*> repop_queue
;
847 friend class C_OSD_RepopApplied
;
848 friend class C_OSD_RepopCommit
;
849 void repop_all_applied(RepGather
*repop
);
850 void repop_all_committed(RepGather
*repop
);
851 void eval_repop(RepGather
*);
852 void issue_repop(RepGather
*repop
, OpContext
*ctx
);
853 RepGather
*new_repop(
855 ObjectContextRef obc
,
857 boost::intrusive_ptr
<RepGather
> new_repop(
860 ObcLockManager
&&manager
,
862 boost::optional
<std::function
<void(void)> > &&on_complete
);
863 void remove_repop(RepGather
*repop
);
865 OpContextUPtr
simple_opc_create(ObjectContextRef obc
);
866 void simple_opc_submit(OpContextUPtr ctx
);
869 * Merge entries atomically into all actingbackfill osds
870 * adjusting missing and recovery state as necessary.
872 * Also used to store error log entries for dup detection.
874 void submit_log_entries(
875 const mempool::osd_pglog::list
<pg_log_entry_t
> &entries
,
876 ObcLockManager
&&manager
,
877 boost::optional
<std::function
<void(void)> > &&on_complete
,
878 OpRequestRef op
= OpRequestRef(),
880 struct LogUpdateCtx
{
881 boost::intrusive_ptr
<RepGather
> repop
;
882 set
<pg_shard_t
> waiting_on
;
884 void cancel_log_updates();
885 map
<ceph_tid_t
, LogUpdateCtx
> log_entry_update_waiting_on
;
889 HitSetRef hit_set
; ///< currently accumulating HitSet
890 utime_t hit_set_start_stamp
; ///< time the current HitSet started recording
893 void hit_set_clear(); ///< discard any HitSet state
894 void hit_set_setup(); ///< initialize HitSet state
895 void hit_set_create(); ///< create a new HitSet
896 void hit_set_persist(); ///< persist hit info
897 bool hit_set_apply_log(); ///< apply log entries to update in-memory HitSet
898 void hit_set_trim(OpContextUPtr
&ctx
, unsigned max
); ///< discard old HitSets
899 void hit_set_in_memory_trim(uint32_t max_in_memory
); ///< discard old in memory HitSets
900 void hit_set_remove_all();
902 hobject_t
get_hit_set_current_object(utime_t stamp
);
903 hobject_t
get_hit_set_archive_object(utime_t start
,
908 boost::scoped_ptr
<TierAgentState
> agent_state
;
910 void agent_setup(); ///< initialize agent state
911 bool agent_work(int max
) override
///< entry point to do some agent work
913 return agent_work(max
, max
);
915 bool agent_work(int max
, int agent_flush_quota
) override
;
916 bool agent_maybe_flush(ObjectContextRef
& obc
); ///< maybe flush
917 bool agent_maybe_evict(ObjectContextRef
& obc
, bool after_flush
); ///< maybe evict
919 void agent_load_hit_sets(); ///< load HitSets, if needed
921 /// estimate object atime and temperature
923 /// @param oid [in] object name
924 /// @param temperature [out] relative temperature (# consider both access time and frequency)
925 void agent_estimate_temp(const hobject_t
& oid
, int *temperature
);
928 void agent_stop() override
;
929 void agent_delay() override
;
931 /// clear agent state
932 void agent_clear() override
;
934 /// choose (new) agent mode(s), returns true if op is requeued
935 bool agent_choose_mode(bool restart
= false, OpRequestRef op
= OpRequestRef());
936 void agent_choose_mode_restart() override
;
938 /// true if we can send an ondisk/commit for v
939 bool already_complete(eversion_t v
);
940 /// true if we can send an ack for v
941 bool already_ack(eversion_t v
);
943 // projected object info
944 SharedLRU
<hobject_t
, ObjectContext
> object_contexts
;
945 // map from oid.snapdir() to SnapSetContext *
946 map
<hobject_t
, SnapSetContext
*> snapset_contexts
;
947 Mutex snapset_contexts_lock
;
949 // debug order that client ops are applied
950 map
<hobject_t
, map
<client_t
, ceph_tid_t
>> debug_op_order
;
952 void populate_obc_watchers(ObjectContextRef obc
);
953 void check_blacklisted_obc_watchers(ObjectContextRef obc
);
954 void check_blacklisted_watchers() override
;
955 void get_watchers(list
<obj_watch_item_t
> &pg_watchers
) override
;
956 void get_obc_watchers(ObjectContextRef obc
, list
<obj_watch_item_t
> &pg_watchers
);
958 void handle_watch_timeout(WatchRef watch
);
961 ObjectContextRef
create_object_context(const object_info_t
& oi
, SnapSetContext
*ssc
);
962 ObjectContextRef
get_object_context(
963 const hobject_t
& soid
,
965 const map
<string
, bufferlist
> *attrs
= 0
968 void context_registry_on_change();
969 void object_context_destructor_callback(ObjectContext
*obc
);
970 class C_PG_ObjectContext
;
972 int find_object_context(const hobject_t
& oid
,
973 ObjectContextRef
*pobc
,
975 bool map_snapid_to_clone
=false,
976 hobject_t
*missing_oid
=NULL
);
978 void add_object_context_to_pg_stat(ObjectContextRef obc
, pg_stat_t
*stat
);
980 void get_src_oloc(const object_t
& oid
, const object_locator_t
& oloc
, object_locator_t
& src_oloc
);
982 SnapSetContext
*get_snapset_context(
983 const hobject_t
& oid
,
985 const map
<string
, bufferlist
> *attrs
= 0,
986 bool oid_existed
= true //indicate this oid whether exsited in backend
988 void register_snapset_context(SnapSetContext
*ssc
) {
989 Mutex::Locker
l(snapset_contexts_lock
);
990 _register_snapset_context(ssc
);
992 void _register_snapset_context(SnapSetContext
*ssc
) {
993 assert(snapset_contexts_lock
.is_locked());
994 if (!ssc
->registered
) {
995 assert(snapset_contexts
.count(ssc
->oid
) == 0);
996 ssc
->registered
= true;
997 snapset_contexts
[ssc
->oid
] = ssc
;
1000 void put_snapset_context(SnapSetContext
*ssc
);
1002 map
<hobject_t
, ObjectContextRef
> recovering
;
1007 * peer_info[backfill_target].last_backfill == info.last_backfill on the peer.
1009 * objects prior to peer_info[backfill_target].last_backfill
1011 * - are included in the peer stats
1013 * objects \in (last_backfill, last_backfill_started]
1014 * - are on the peer or are in backfills_in_flight
1015 * - are not included in pg stats (yet)
1016 * - have their stats in pending_backfill_updates on the primary
1018 set
<hobject_t
> backfills_in_flight
;
1019 map
<hobject_t
, pg_stat_t
> pending_backfill_updates
;
1021 void dump_recovery_info(Formatter
*f
) const override
{
1022 f
->open_array_section("backfill_targets");
1023 for (set
<pg_shard_t
>::const_iterator p
= backfill_targets
.begin();
1024 p
!= backfill_targets
.end(); ++p
)
1025 f
->dump_stream("replica") << *p
;
1027 f
->open_array_section("waiting_on_backfill");
1028 for (set
<pg_shard_t
>::const_iterator p
= waiting_on_backfill
.begin();
1029 p
!= waiting_on_backfill
.end(); ++p
)
1030 f
->dump_stream("osd") << *p
;
1032 f
->dump_stream("last_backfill_started") << last_backfill_started
;
1034 f
->open_object_section("backfill_info");
1035 backfill_info
.dump(f
);
1039 f
->open_array_section("peer_backfill_info");
1040 for (map
<pg_shard_t
, BackfillInterval
>::const_iterator pbi
=
1041 peer_backfill_info
.begin();
1042 pbi
!= peer_backfill_info
.end(); ++pbi
) {
1043 f
->dump_stream("osd") << pbi
->first
;
1044 f
->open_object_section("BackfillInterval");
1045 pbi
->second
.dump(f
);
1051 f
->open_array_section("backfills_in_flight");
1052 for (set
<hobject_t
>::const_iterator i
= backfills_in_flight
.begin();
1053 i
!= backfills_in_flight
.end();
1055 f
->dump_stream("object") << *i
;
1060 f
->open_array_section("recovering");
1061 for (map
<hobject_t
, ObjectContextRef
>::const_iterator i
= recovering
.begin();
1062 i
!= recovering
.end();
1064 f
->dump_stream("object") << i
->first
;
1069 f
->open_object_section("pg_backend");
1070 pgbackend
->dump_recovery_info(f
);
1075 /// last backfill operation started
1076 hobject_t last_backfill_started
;
1079 int prep_object_replica_pushes(const hobject_t
& soid
, eversion_t v
,
1080 PGBackend::RecoveryHandle
*h
);
1081 int prep_object_replica_deletes(const hobject_t
& soid
, eversion_t v
,
1082 PGBackend::RecoveryHandle
*h
);
1084 void finish_degraded_object(const hobject_t
& oid
);
1086 // Cancels/resets pulls from peer
1087 void check_recovery_sources(const OSDMapRef
& map
) override
;
1089 int recover_missing(
1090 const hobject_t
& oid
,
1093 PGBackend::RecoveryHandle
*h
);
1100 ObjectContextRef obc
,
1101 const hobject_t
& head
, const hobject_t
& coid
,
1102 object_info_t
*poi
);
1103 void execute_ctx(OpContext
*ctx
);
1104 void finish_ctx(OpContext
*ctx
, int log_op_type
, bool maintain_ssc
=true);
1105 void reply_ctx(OpContext
*ctx
, int err
);
1106 void reply_ctx(OpContext
*ctx
, int err
, eversion_t v
, version_t uv
);
1107 void make_writeable(OpContext
*ctx
);
1108 void log_op_stats(OpContext
*ctx
);
1110 void write_update_size_and_usage(object_stat_sum_t
& stats
, object_info_t
& oi
,
1111 interval_set
<uint64_t>& modified
, uint64_t offset
,
1112 uint64_t length
, bool write_full
=false);
1113 void add_interval_usage(interval_set
<uint64_t>& s
, object_stat_sum_t
& st
);
1116 enum class cache_result_t
{
1122 REPLIED_WITH_EAGAIN
,
1124 cache_result_t
maybe_handle_cache_detail(OpRequestRef op
,
1126 ObjectContextRef obc
, int r
,
1127 hobject_t missing_oid
,
1130 ObjectContextRef
*promote_obc
);
1131 cache_result_t
maybe_handle_manifest_detail(OpRequestRef op
,
1133 ObjectContextRef obc
);
1134 bool maybe_handle_manifest(OpRequestRef op
,
1136 ObjectContextRef obc
) {
1137 return cache_result_t::NOOP
!= maybe_handle_manifest_detail(
1144 * This helper function is called from do_op if the ObjectContext lookup fails.
1145 * @returns true if the caching code is handling the Op, false otherwise.
1147 bool maybe_handle_cache(OpRequestRef op
,
1149 ObjectContextRef obc
, int r
,
1150 const hobject_t
& missing_oid
,
1152 bool in_hit_set
= false) {
1153 return cache_result_t::NOOP
!= maybe_handle_cache_detail(
1165 * This helper function checks if a promotion is needed.
1167 bool maybe_promote(ObjectContextRef obc
,
1168 const hobject_t
& missing_oid
,
1169 const object_locator_t
& oloc
,
1172 OpRequestRef promote_op
,
1173 ObjectContextRef
*promote_obc
= nullptr);
1175 * This helper function tells the client to redirect their request elsewhere.
1177 void do_cache_redirect(OpRequestRef op
);
1179 * This function attempts to start a promote. Either it succeeds,
1180 * or places op on a wait list. If op is null, failure means that
1181 * this is a noop. If a future user wants to be able to distinguish
1182 * these cases, a return value should be added.
1184 void promote_object(
1185 ObjectContextRef obc
, ///< [optional] obc
1186 const hobject_t
& missing_object
, ///< oid (if !obc)
1187 const object_locator_t
& oloc
, ///< locator for obc|oid
1188 OpRequestRef op
, ///< [optional] client op
1189 ObjectContextRef
*promote_obc
= nullptr ///< [optional] new obc for object
1192 int prepare_transaction(OpContext
*ctx
);
1193 list
<pair
<OpRequestRef
, OpContext
*> > in_progress_async_reads
;
1194 void complete_read_ctx(int result
, OpContext
*ctx
);
1196 // pg on-disk content
1197 void check_local() override
;
1199 void _clear_recovery_state() override
;
1201 bool start_recovery_ops(
1203 ThreadPool::TPHandle
&handle
, uint64_t *started
) override
;
1205 uint64_t recover_primary(uint64_t max
, ThreadPool::TPHandle
&handle
);
1206 uint64_t recover_replicas(uint64_t max
, ThreadPool::TPHandle
&handle
);
1207 hobject_t
earliest_peer_backfill() const;
1208 bool all_peer_done() const;
1210 * @param work_started will be set to true if recover_backfill got anywhere
1211 * @returns the number of operations started
1213 uint64_t recover_backfill(uint64_t max
, ThreadPool::TPHandle
&handle
,
1214 bool *work_started
);
1217 * scan a (hash) range of objects in the current pg
1219 * @begin first item should be >= this value
1220 * @min return at least this many items, unless we are done
1221 * @max return no more than this many items
1222 * @bi [out] resulting map of objects to eversion_t's
1225 int min
, int max
, BackfillInterval
*bi
,
1226 ThreadPool::TPHandle
&handle
1229 /// Update a hash range to reflect changes since the last scan
1231 BackfillInterval
*bi
, ///< [in,out] interval to update
1232 ThreadPool::TPHandle
&handle
///< [in] tp handle
1235 int prep_backfill_object_push(
1236 hobject_t oid
, eversion_t v
, ObjectContextRef obc
,
1237 vector
<pg_shard_t
> peers
,
1238 PGBackend::RecoveryHandle
*h
);
1239 void send_remove_op(const hobject_t
& oid
, eversion_t v
, pg_shard_t peer
);
1242 class C_OSD_OndiskWriteUnlock
;
1243 class C_OSD_AppliedRecoveredObject
;
1244 class C_OSD_CommittedPushedObject
;
1245 class C_OSD_AppliedRecoveredObjectReplica
;
1246 void sub_op_remove(OpRequestRef op
);
1248 void _applied_recovered_object(ObjectContextRef obc
);
1249 void _applied_recovered_object_replica();
1250 void _committed_pushed_object(epoch_t epoch
, eversion_t lc
);
1251 void recover_got(hobject_t oid
, eversion_t v
);
1254 map
<hobject_t
, CopyOpRef
> copy_ops
;
1256 int do_copy_get(OpContext
*ctx
, bufferlist::iterator
& bp
, OSDOp
& op
,
1257 ObjectContextRef
& obc
);
1258 int finish_copy_get();
1260 void fill_in_copy_get_noent(OpRequestRef
& op
, hobject_t oid
,
1264 * To copy an object, call start_copy.
1266 * @param cb: The CopyCallback to be activated when the copy is complete
1267 * @param obc: The ObjectContext we are copying into
1268 * @param src: The source object
1269 * @param oloc: the source object locator
1270 * @param version: the version of the source object to copy (0 for any)
1272 void start_copy(CopyCallback
*cb
, ObjectContextRef obc
, hobject_t src
,
1273 object_locator_t oloc
, version_t version
, unsigned flags
,
1274 bool mirror_snapset
, unsigned src_obj_fadvise_flags
,
1275 unsigned dest_obj_fadvise_flags
);
1276 void process_copy_chunk(hobject_t oid
, ceph_tid_t tid
, int r
);
1277 void _write_copy_chunk(CopyOpRef cop
, PGTransaction
*t
);
1278 uint64_t get_copy_chunk_size() const {
1279 uint64_t size
= cct
->_conf
->osd_copyfrom_max_chunk
;
1280 if (pool
.info
.requires_aligned_append()) {
1281 uint64_t alignment
= pool
.info
.required_alignment();
1282 if (size
% alignment
) {
1283 size
+= alignment
- (size
% alignment
);
1288 void _copy_some(ObjectContextRef obc
, CopyOpRef cop
);
1289 void finish_copyfrom(CopyFromCallback
*cb
);
1290 void finish_promote(int r
, CopyResults
*results
, ObjectContextRef obc
);
1291 void cancel_copy(CopyOpRef cop
, bool requeue
);
1292 void cancel_copy_ops(bool requeue
);
1294 friend struct C_Copyfrom
;
1297 map
<hobject_t
, FlushOpRef
> flush_ops
;
1299 /// start_flush takes ownership of on_flush iff ret == -EINPROGRESS
1301 OpRequestRef op
, ObjectContextRef obc
,
1302 bool blocking
, hobject_t
*pmissing
,
1303 boost::optional
<std::function
<void()>> &&on_flush
);
1304 void finish_flush(hobject_t oid
, ceph_tid_t tid
, int r
);
1305 int try_flush_mark_clean(FlushOpRef fop
);
1306 void cancel_flush(FlushOpRef fop
, bool requeue
);
1307 void cancel_flush_ops(bool requeue
);
1309 /// @return false if clone is has been evicted
1310 bool is_present_clone(hobject_t coid
);
1312 friend struct C_Flush
;
1315 bool _range_available_for_scrub(
1316 const hobject_t
&begin
, const hobject_t
&end
) override
;
1317 void scrub_snapshot_metadata(
1319 const std::map
<hobject_t
, pair
<uint32_t, uint32_t>> &missing_digest
) override
;
1320 void _scrub_clear_state() override
;
1321 void _scrub_finish() override
;
1322 object_stat_collection_t scrub_cstat
;
1324 void _split_into(pg_t child_pgid
, PG
*child
,
1325 unsigned split_bits
) override
;
1326 void apply_and_flush_repops(bool requeue
);
1328 void calc_trim_to() override
;
1329 int do_xattr_cmp_u64(int op
, __u64 v1
, bufferlist
& xattr
);
1330 int do_xattr_cmp_str(int op
, string
& v1s
, bufferlist
& xattr
);
1333 int do_checksum(OpContext
*ctx
, OSDOp
& osd_op
, bufferlist::iterator
*bl_it
);
1334 int finish_checksum(OSDOp
& osd_op
, Checksummer::CSumType csum_type
,
1335 bufferlist::iterator
*init_value_bl_it
,
1336 const bufferlist
&read_bl
);
1338 friend class C_ChecksumRead
;
1340 int do_extent_cmp(OpContext
*ctx
, OSDOp
& osd_op
);
1341 int finish_extent_cmp(OSDOp
& osd_op
, const bufferlist
&read_bl
);
1343 friend class C_ExtentCmpRead
;
1345 int do_read(OpContext
*ctx
, OSDOp
& osd_op
);
1346 int do_sparse_read(OpContext
*ctx
, OSDOp
& osd_op
);
1347 int do_writesame(OpContext
*ctx
, OSDOp
& osd_op
);
1349 bool pgls_filter(PGLSFilter
*filter
, hobject_t
& sobj
, bufferlist
& outdata
);
1350 int get_pgls_filter(bufferlist::iterator
& iter
, PGLSFilter
**pfilter
);
1352 map
<hobject_t
, list
<OpRequestRef
>> in_progress_proxy_ops
;
1353 void kick_proxy_ops_blocked(hobject_t
& soid
);
1354 void cancel_proxy_ops(bool requeue
);
1357 map
<ceph_tid_t
, ProxyReadOpRef
> proxyread_ops
;
1359 void do_proxy_read(OpRequestRef op
, ObjectContextRef obc
= NULL
);
1360 void finish_proxy_read(hobject_t oid
, ceph_tid_t tid
, int r
);
1361 void cancel_proxy_read(ProxyReadOpRef prdop
);
1363 friend struct C_ProxyRead
;
1366 map
<ceph_tid_t
, ProxyWriteOpRef
> proxywrite_ops
;
1368 void do_proxy_write(OpRequestRef op
, const hobject_t
& missing_oid
, ObjectContextRef obc
= NULL
);
1369 void finish_proxy_write(hobject_t oid
, ceph_tid_t tid
, int r
);
1370 void cancel_proxy_write(ProxyWriteOpRef pwop
);
1372 friend struct C_ProxyWrite_Commit
;
1375 PrimaryLogPG(OSDService
*o
, OSDMapRef curmap
,
1376 const PGPool
&_pool
, spg_t p
);
1377 ~PrimaryLogPG() override
{}
1385 ceph_tid_t tid
) override
;
1389 ThreadPool::TPHandle
&handle
) override
;
1390 void do_op(OpRequestRef
& op
) override
;
1391 void record_write_error(OpRequestRef op
, const hobject_t
&soid
,
1392 MOSDOpReply
*orig_reply
, int r
);
1393 void do_pg_op(OpRequestRef op
);
1394 void do_sub_op(OpRequestRef op
) override
;
1395 void do_sub_op_reply(OpRequestRef op
) override
;
1398 ThreadPool::TPHandle
&handle
) override
;
1399 void do_backfill(OpRequestRef op
) override
;
1400 void do_backfill_remove(OpRequestRef op
);
1402 void handle_backoff(OpRequestRef
& op
);
1404 int trim_object(bool first
, const hobject_t
&coid
, OpContextUPtr
*ctxp
);
1405 void snap_trimmer(epoch_t e
) override
;
1406 void kick_snap_trim() override
;
1407 void snap_trimmer_scrub_complete() override
;
1408 int do_osd_ops(OpContext
*ctx
, vector
<OSDOp
>& ops
);
1410 int _get_tmap(OpContext
*ctx
, bufferlist
*header
, bufferlist
*vals
);
1411 int do_tmap2omap(OpContext
*ctx
, unsigned flags
);
1412 int do_tmapup(OpContext
*ctx
, bufferlist::iterator
& bp
, OSDOp
& osd_op
);
1413 int do_tmapup_slow(OpContext
*ctx
, bufferlist::iterator
& bp
, OSDOp
& osd_op
, bufferlist
& bl
);
1415 void do_osd_op_effects(OpContext
*ctx
, const ConnectionRef
& conn
);
1417 int do_scrub_ls(MOSDOp
*op
, OSDOp
*osd_op
);
1418 hobject_t
earliest_backfill() const;
1419 bool check_src_targ(const hobject_t
& soid
, const hobject_t
& toid
) const;
1421 uint64_t temp_seq
; ///< last id for naming temp objects
1422 /// generate a new temp object name
1423 hobject_t
generate_temp_object(const hobject_t
& target
);
1424 /// generate a new temp object name (for recovery)
1425 hobject_t
get_temp_recovery_object(const hobject_t
& target
,
1426 eversion_t version
) override
;
1427 int get_recovery_op_priority() const {
1429 pool
.info
.opts
.get(pool_opts_t::RECOVERY_OP_PRIORITY
, &pri
);
1430 return pri
> 0 ? pri
: cct
->_conf
->osd_recovery_op_priority
;
1432 void log_missing(unsigned missing
,
1433 const boost::optional
<hobject_t
> &head
,
1438 bool allow_incomplete_clones
);
1439 unsigned process_clones_to(const boost::optional
<hobject_t
> &head
,
1440 const boost::optional
<SnapSet
> &snapset
,
1444 bool allow_incomplete_clones
,
1445 boost::optional
<snapid_t
> target
,
1446 vector
<snapid_t
>::reverse_iterator
*curclone
,
1447 inconsistent_snapset_wrapper
&snap_error
);
1457 const pg_pool_t
*pool
,
1458 ObjectStore::Transaction
*t
) override
{
1459 coll_t target
= coll_t(child
);
1460 PG::_create(*t
, child
, split_bits
);
1461 t
->split_collection(
1466 PG::_init(*t
, child
, pool
);
1470 struct DoSnapWork
: boost::statechart::event
< DoSnapWork
> {
1471 DoSnapWork() : boost::statechart::event
< DoSnapWork
>() {}
1473 struct KickTrim
: boost::statechart::event
< KickTrim
> {
1474 KickTrim() : boost::statechart::event
< KickTrim
>() {}
1476 struct RepopsComplete
: boost::statechart::event
< RepopsComplete
> {
1477 RepopsComplete() : boost::statechart::event
< RepopsComplete
>() {}
1479 struct ScrubComplete
: boost::statechart::event
< ScrubComplete
> {
1480 ScrubComplete() : boost::statechart::event
< ScrubComplete
>() {}
1482 struct TrimWriteUnblocked
: boost::statechart::event
< TrimWriteUnblocked
> {
1483 TrimWriteUnblocked() : boost::statechart::event
< TrimWriteUnblocked
>() {}
1485 struct Reset
: boost::statechart::event
< Reset
> {
1486 Reset() : boost::statechart::event
< Reset
>() {}
1488 struct SnapTrimReserved
: boost::statechart::event
< SnapTrimReserved
> {
1489 SnapTrimReserved() : boost::statechart::event
< SnapTrimReserved
>() {}
1491 struct SnapTrimTimerReady
: boost::statechart::event
< SnapTrimTimerReady
> {
1492 SnapTrimTimerReady() : boost::statechart::event
< SnapTrimTimerReady
>() {}
1496 struct SnapTrimmer
: public boost::statechart::state_machine
< SnapTrimmer
, NotTrimming
> {
1498 explicit SnapTrimmer(PrimaryLogPG
*pg
) : pg(pg
) {}
1499 void log_enter(const char *state_name
);
1500 void log_exit(const char *state_name
, utime_t duration
);
1502 return pg
->is_clean() && !pg
->scrubber
.active
&& !pg
->snap_trimq
.empty();
1504 } snap_trimmer_machine
;
1506 struct WaitReservation
;
1507 struct Trimming
: boost::statechart::state
< Trimming
, SnapTrimmer
, WaitReservation
>, NamedState
{
1508 typedef boost::mpl::list
<
1509 boost::statechart::custom_reaction
< KickTrim
>,
1510 boost::statechart::transition
< Reset
, NotTrimming
>
1513 set
<hobject_t
> in_flight
;
1514 snapid_t snap_to_trim
;
1516 explicit Trimming(my_context ctx
)
1518 NamedState(context
< SnapTrimmer
>().pg
, "Trimming") {
1519 context
< SnapTrimmer
>().log_enter(state_name
);
1520 assert(context
< SnapTrimmer
>().can_trim());
1521 assert(in_flight
.empty());
1524 context
< SnapTrimmer
>().log_exit(state_name
, enter_time
);
1525 auto *pg
= context
< SnapTrimmer
>().pg
;
1526 pg
->osd
->snap_reserver
.cancel_reservation(pg
->get_pgid());
1527 pg
->state_clear(PG_STATE_SNAPTRIM
);
1528 pg
->publish_stats_to_osd();
1530 boost::statechart::result
react(const KickTrim
&) {
1531 return discard_event();
1535 /* SnapTrimmerStates */
1536 struct WaitTrimTimer
: boost::statechart::state
< WaitTrimTimer
, Trimming
>, NamedState
{
1537 typedef boost::mpl::list
<
1538 boost::statechart::custom_reaction
< SnapTrimTimerReady
>
1540 Context
*wakeup
= nullptr;
1541 explicit WaitTrimTimer(my_context ctx
)
1543 NamedState(context
< SnapTrimmer
>().pg
, "Trimming/WaitTrimTimer") {
1544 context
< SnapTrimmer
>().log_enter(state_name
);
1545 assert(context
<Trimming
>().in_flight
.empty());
1546 struct OnTimer
: Context
{
1549 OnTimer(PrimaryLogPGRef pg
, epoch_t epoch
) : pg(pg
), epoch(epoch
) {}
1550 void finish(int) override
{
1552 if (!pg
->pg_has_reset_since(epoch
))
1553 pg
->snap_trimmer_machine
.process_event(SnapTrimTimerReady());
1557 auto *pg
= context
< SnapTrimmer
>().pg
;
1558 if (pg
->cct
->_conf
->osd_snap_trim_sleep
> 0) {
1559 Mutex::Locker
l(pg
->osd
->snap_sleep_lock
);
1560 wakeup
= pg
->osd
->snap_sleep_timer
.add_event_after(
1561 pg
->cct
->_conf
->osd_snap_trim_sleep
,
1562 new OnTimer
{pg
, pg
->get_osdmap()->get_epoch()});
1564 post_event(SnapTrimTimerReady());
1568 context
< SnapTrimmer
>().log_exit(state_name
, enter_time
);
1569 auto *pg
= context
< SnapTrimmer
>().pg
;
1571 Mutex::Locker
l(pg
->osd
->snap_sleep_lock
);
1572 pg
->osd
->snap_sleep_timer
.cancel_event(wakeup
);
1576 boost::statechart::result
react(const SnapTrimTimerReady
&) {
1578 if (!context
< SnapTrimmer
>().can_trim()) {
1579 post_event(KickTrim());
1580 return transit
< NotTrimming
>();
1582 return transit
< AwaitAsyncWork
>();
1587 struct WaitRWLock
: boost::statechart::state
< WaitRWLock
, Trimming
>, NamedState
{
1588 typedef boost::mpl::list
<
1589 boost::statechart::custom_reaction
< TrimWriteUnblocked
>
1591 explicit WaitRWLock(my_context ctx
)
1593 NamedState(context
< SnapTrimmer
>().pg
, "Trimming/WaitRWLock") {
1594 context
< SnapTrimmer
>().log_enter(state_name
);
1595 assert(context
<Trimming
>().in_flight
.empty());
1598 context
< SnapTrimmer
>().log_exit(state_name
, enter_time
);
1600 boost::statechart::result
react(const TrimWriteUnblocked
&) {
1601 if (!context
< SnapTrimmer
>().can_trim()) {
1602 post_event(KickTrim());
1603 return transit
< NotTrimming
>();
1605 return transit
< AwaitAsyncWork
>();
1610 struct WaitRepops
: boost::statechart::state
< WaitRepops
, Trimming
>, NamedState
{
1611 typedef boost::mpl::list
<
1612 boost::statechart::custom_reaction
< RepopsComplete
>
1614 explicit WaitRepops(my_context ctx
)
1616 NamedState(context
< SnapTrimmer
>().pg
, "Trimming/WaitRepops") {
1617 context
< SnapTrimmer
>().log_enter(state_name
);
1618 assert(!context
<Trimming
>().in_flight
.empty());
1621 context
< SnapTrimmer
>().log_exit(state_name
, enter_time
);
1623 boost::statechart::result
react(const RepopsComplete
&) {
1624 if (!context
< SnapTrimmer
>().can_trim()) {
1625 post_event(KickTrim());
1626 return transit
< NotTrimming
>();
1628 return transit
< WaitTrimTimer
>();
1633 struct AwaitAsyncWork
: boost::statechart::state
< AwaitAsyncWork
, Trimming
>, NamedState
{
1634 typedef boost::mpl::list
<
1635 boost::statechart::custom_reaction
< DoSnapWork
>
1637 explicit AwaitAsyncWork(my_context ctx
);
1639 context
< SnapTrimmer
>().log_exit(state_name
, enter_time
);
1641 boost::statechart::result
react(const DoSnapWork
&);
1644 struct WaitReservation
: boost::statechart::state
< WaitReservation
, Trimming
>, NamedState
{
1645 /* WaitReservation is a sub-state of trimming simply so that exiting Trimming
1646 * always cancels the reservation */
1647 typedef boost::mpl::list
<
1648 boost::statechart::custom_reaction
< SnapTrimReserved
>
1650 struct ReservationCB
: public Context
{
1653 ReservationCB(PrimaryLogPG
*pg
) : pg(pg
), canceled(false) {}
1654 void finish(int) override
{
1657 pg
->snap_trimmer_machine
.process_event(SnapTrimReserved());
1661 assert(pg
->is_locked());
1666 ReservationCB
*pending
= nullptr;
1668 explicit WaitReservation(my_context ctx
)
1670 NamedState(context
< SnapTrimmer
>().pg
, "Trimming/WaitReservation") {
1671 context
< SnapTrimmer
>().log_enter(state_name
);
1672 assert(context
<Trimming
>().in_flight
.empty());
1673 auto *pg
= context
< SnapTrimmer
>().pg
;
1674 pending
= new ReservationCB(pg
);
1675 pg
->osd
->snap_reserver
.request_reservation(
1679 pg
->state_set(PG_STATE_SNAPTRIM_WAIT
);
1680 pg
->publish_stats_to_osd();
1682 boost::statechart::result
react(const SnapTrimReserved
&);
1684 context
< SnapTrimmer
>().log_exit(state_name
, enter_time
);
1688 auto *pg
= context
< SnapTrimmer
>().pg
;
1689 pg
->state_clear(PG_STATE_SNAPTRIM_WAIT
);
1690 pg
->state_clear(PG_STATE_SNAPTRIM_ERROR
);
1691 pg
->publish_stats_to_osd();
1695 struct WaitScrub
: boost::statechart::state
< WaitScrub
, SnapTrimmer
>, NamedState
{
1696 typedef boost::mpl::list
<
1697 boost::statechart::custom_reaction
< ScrubComplete
>,
1698 boost::statechart::custom_reaction
< KickTrim
>,
1699 boost::statechart::transition
< Reset
, NotTrimming
>
1701 explicit WaitScrub(my_context ctx
)
1703 NamedState(context
< SnapTrimmer
>().pg
, "Trimming/WaitScrub") {
1704 context
< SnapTrimmer
>().log_enter(state_name
);
1707 context
< SnapTrimmer
>().log_exit(state_name
, enter_time
);
1709 boost::statechart::result
react(const ScrubComplete
&) {
1710 post_event(KickTrim());
1711 return transit
< NotTrimming
>();
1713 boost::statechart::result
react(const KickTrim
&) {
1714 return discard_event();
1718 struct NotTrimming
: boost::statechart::state
< NotTrimming
, SnapTrimmer
>, NamedState
{
1719 typedef boost::mpl::list
<
1720 boost::statechart::custom_reaction
< KickTrim
>,
1721 boost::statechart::transition
< Reset
, NotTrimming
>
1723 explicit NotTrimming(my_context ctx
);
1725 boost::statechart::result
react(const KickTrim
&);
1728 int _verify_no_head_clones(const hobject_t
& soid
,
1730 // return true if we're creating a local object, false for a
1731 // whiteout or no change.
1732 void maybe_create_new_object(OpContext
*ctx
, bool ignore_transaction
=false);
1733 int _delete_oid(OpContext
*ctx
, bool no_whiteout
, bool try_no_whiteout
);
1734 int _rollback_to(OpContext
*ctx
, ceph_osd_op
& op
);
1736 bool is_missing_object(const hobject_t
& oid
) const;
1737 bool is_unreadable_object(const hobject_t
&oid
) const {
1738 return is_missing_object(oid
) ||
1739 !missing_loc
.readable_with_acting(oid
, actingset
);
1741 void maybe_kick_recovery(const hobject_t
&soid
);
1742 void wait_for_unreadable_object(const hobject_t
& oid
, OpRequestRef op
);
1743 void wait_for_all_missing(OpRequestRef op
);
1745 bool is_degraded_or_backfilling_object(const hobject_t
& oid
);
1746 void wait_for_degraded_object(const hobject_t
& oid
, OpRequestRef op
);
1748 void block_write_on_full_cache(
1749 const hobject_t
& oid
, OpRequestRef op
);
1750 void block_for_clean(
1751 const hobject_t
& oid
, OpRequestRef op
);
1752 void block_write_on_snap_rollback(
1753 const hobject_t
& oid
, ObjectContextRef obc
, OpRequestRef op
);
1754 void block_write_on_degraded_snap(const hobject_t
& oid
, OpRequestRef op
);
1756 bool maybe_await_blocked_snapset(const hobject_t
&soid
, OpRequestRef op
);
1757 void wait_for_blocked_object(const hobject_t
& soid
, OpRequestRef op
);
1758 void kick_object_context_blocked(ObjectContextRef obc
);
1760 void maybe_force_recovery();
1762 void mark_all_unfound_lost(
1766 eversion_t
pick_newest_available(const hobject_t
& oid
);
1768 void do_update_log_missing(
1771 void do_update_log_missing_reply(
1774 void on_role_change() override
;
1775 void on_pool_change() override
;
1776 void _on_new_interval() override
;
1777 void clear_async_reads();
1778 void on_change(ObjectStore::Transaction
*t
) override
;
1779 void on_activate() override
;
1780 void on_flushed() override
;
1781 void on_removal(ObjectStore::Transaction
*t
) override
;
1782 void on_shutdown() override
;
1783 bool check_failsafe_full(ostream
&ss
) override
;
1784 bool check_osdmap_full(const set
<pg_shard_t
> &missing_on
) override
;
1785 int rep_repair_primary_object(const hobject_t
& soid
, OpRequestRef op
);
1787 // attr cache handling
1788 void setattr_maybe_cache(
1789 ObjectContextRef obc
,
1794 void setattrs_maybe_cache(
1795 ObjectContextRef obc
,
1798 map
<string
, bufferlist
> &attrs
);
1799 void rmattr_maybe_cache(
1800 ObjectContextRef obc
,
1804 int getattr_maybe_cache(
1805 ObjectContextRef obc
,
1808 int getattrs_maybe_cache(
1809 ObjectContextRef obc
,
1810 map
<string
, bufferlist
> *out
);
1813 inline ostream
& operator<<(ostream
& out
, const PrimaryLogPG::RepGather
& repop
)
1815 out
<< "repgather(" << &repop
1817 << " rep_tid=" << repop
.rep_tid
1818 << " committed?=" << repop
.all_committed
1819 << " applied?=" << repop
.all_applied
1825 inline ostream
& operator<<(ostream
& out
,
1826 const PrimaryLogPG::ProxyWriteOpRef
& pwop
)
1828 out
<< "proxywrite(" << &pwop
1829 << " " << pwop
->user_version
1830 << " pwop_tid=" << pwop
->objecter_tid
;
1832 out
<< " op=" << *(pwop
->ctx
->op
->get_req());
1837 void intrusive_ptr_add_ref(PrimaryLogPG::RepGather
*repop
);
1838 void intrusive_ptr_release(PrimaryLogPG::RepGather
*repop
);