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>
7 * Copyright (C) 2013,2014 Cloudwatt <libre.licensing@cloudwatt.com>
9 * Author: Loic Dachary <loic@dachary.org>
11 * This is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public
13 * License version 2.1, as published by the Free Software
14 * Foundation. See file COPYING.
18 /* Object Store Device (OSD) Monitor
21 #ifndef CEPH_OSDMONITOR_H
22 #define CEPH_OSDMONITOR_H
27 #include "include/types.h"
28 #include "include/encoding.h"
29 #include "common/simple_cache.hpp"
30 #include "common/PriorityCache.h"
31 #include "msg/Messenger.h"
33 #include "osd/OSDMap.h"
34 #include "osd/OSDMapMapping.h"
36 #include "CreatingPGs.h"
37 #include "PaxosService.h"
44 #include "erasure-code/ErasureCodeInterface.h"
45 #include "mon/MonOpRequest.h"
46 #include <boost/functional/hash.hpp>
47 // re-include our assert to clobber the system one; fix dout:
48 #include "include/ceph_assert.h"
50 /// information about a particular peer's failure reports for one osd
51 struct failure_reporter_t
{
52 utime_t failed_since
; ///< when they think it failed
53 MonOpRequestRef op
; ///< failure op request
55 failure_reporter_t() {}
56 explicit failure_reporter_t(utime_t s
) : failed_since(s
) {}
57 ~failure_reporter_t() { }
60 /// information about all failure reports for one osd
61 struct failure_info_t
{
62 map
<int, failure_reporter_t
> reporters
; ///< reporter -> failed_since etc
63 utime_t max_failed_since
; ///< most recent failed_since
67 utime_t
get_failed_since() {
68 if (max_failed_since
== utime_t() && !reporters
.empty()) {
69 // the old max must have canceled; recalculate.
70 for (map
<int, failure_reporter_t
>::iterator p
= reporters
.begin();
73 if (p
->second
.failed_since
> max_failed_since
)
74 max_failed_since
= p
->second
.failed_since
;
76 return max_failed_since
;
79 // set the message for the latest report. return any old op request we had,
80 // if any, so we can discard it.
81 MonOpRequestRef
add_report(int who
, utime_t failed_since
,
83 map
<int, failure_reporter_t
>::iterator p
= reporters
.find(who
);
84 if (p
== reporters
.end()) {
85 if (max_failed_since
!= utime_t() && max_failed_since
< failed_since
)
86 max_failed_since
= failed_since
;
87 p
= reporters
.insert(map
<int, failure_reporter_t
>::value_type(who
, failure_reporter_t(failed_since
))).first
;
90 MonOpRequestRef ret
= p
->second
.op
;
95 void take_report_messages(list
<MonOpRequestRef
>& ls
) {
96 for (map
<int, failure_reporter_t
>::iterator p
= reporters
.begin();
100 ls
.push_back(p
->second
.op
);
101 p
->second
.op
.reset();
106 MonOpRequestRef
cancel_report(int who
) {
107 map
<int, failure_reporter_t
>::iterator p
= reporters
.find(who
);
108 if (p
== reporters
.end())
109 return MonOpRequestRef();
110 MonOpRequestRef ret
= p
->second
.op
;
112 max_failed_since
= utime_t();
118 class LastEpochClean
{
120 vector
<epoch_t
> epoch_by_pg
;
121 ps_t next_missing
= 0;
122 epoch_t floor
= std::numeric_limits
<epoch_t
>::max();
123 void report(ps_t pg
, epoch_t last_epoch_clean
);
125 std::map
<uint64_t, Lec
> report_by_pool
;
127 void report(const pg_t
& pg
, epoch_t last_epoch_clean
);
128 void remove_pool(uint64_t pool
);
129 epoch_t
get_lower_bound(const OSDMap
& latest
) const;
131 void dump(Formatter
*f
) const;
135 struct osdmap_manifest_t
{
136 // all the maps we have pinned -- i.e., won't be removed unless
137 // they are inside a trim interval.
138 set
<version_t
> pinned
;
140 osdmap_manifest_t() {}
142 version_t
get_last_pinned() const
144 set
<version_t
>::const_reverse_iterator it
= pinned
.crbegin();
145 if (it
== pinned
.crend()) {
151 version_t
get_first_pinned() const
153 set
<version_t
>::const_iterator it
= pinned
.cbegin();
154 if (it
== pinned
.cend()) {
160 bool is_pinned(version_t v
) const
162 return pinned
.find(v
) != pinned
.end();
165 void pin(version_t v
)
170 version_t
get_lower_closest_pinned(version_t v
) const {
171 set
<version_t
>::const_iterator p
= pinned
.lower_bound(v
);
172 if (p
== pinned
.cend()) {
175 if (p
== pinned
.cbegin()) {
183 void encode(bufferlist
& bl
) const
185 ENCODE_START(1, 1, bl
);
190 void decode(bufferlist::const_iterator
& bl
)
197 void decode(bufferlist
& bl
) {
198 auto p
= bl
.cbegin();
202 void dump(Formatter
*f
) {
203 f
->dump_unsigned("first_pinned", get_first_pinned());
204 f
->dump_unsigned("last_pinned", get_last_pinned());
205 f
->open_array_section("pinned_maps");
206 for (auto& i
: pinned
) {
207 f
->dump_unsigned("epoch", i
);
212 WRITE_CLASS_ENCODER(osdmap_manifest_t
);
214 class OSDMonitor
: public PaxosService
,
215 public md_config_obs_t
{
222 const char** get_tracked_conf_keys() const override
;
223 void handle_conf_change(const ConfigProxy
& conf
,
224 const std::set
<std::string
> &changed
) override
;
226 OSDMap::Incremental pending_inc
;
227 map
<int, bufferlist
> pending_metadata
;
228 set
<int> pending_metadata_rm
;
229 map
<int, failure_info_t
> failure_info
;
230 map
<int,utime_t
> down_pending_out
; // osd down -> out
231 bool priority_convert
= false;
232 map
<int64_t,set
<snapid_t
>> pending_pseudo_purged_snaps
;
233 std::shared_ptr
<PriorityCache::PriCache
> rocksdb_binned_kv_cache
= nullptr;
234 std::shared_ptr
<PriorityCache::Manager
> pcm
= nullptr;
235 ceph::mutex balancer_lock
= ceph::make_mutex("OSDMonitor::balancer_lock");
237 map
<int,double> osd_weight
;
239 using osdmap_key_t
= std::pair
<version_t
, uint64_t>;
240 using osdmap_cache_t
= SimpleLRU
<osdmap_key_t
,
242 std::less
<osdmap_key_t
>,
243 boost::hash
<osdmap_key_t
>>;
244 osdmap_cache_t inc_osd_cache
;
245 osdmap_cache_t full_osd_cache
;
247 bool has_osdmap_manifest
;
248 osdmap_manifest_t osdmap_manifest
;
250 bool check_failures(utime_t now
);
251 bool check_failure(utime_t now
, int target_osd
, failure_info_t
& fi
);
252 void force_failure(int target_osd
, int by
);
254 bool _have_pending_crush();
255 CrushWrapper
&_get_stable_crush();
256 void _get_pending_crush(CrushWrapper
& newcrush
);
264 struct CleanUpmapJob
: public ParallelPGMapper::Job
{
266 const OSDMap
& osdmap
;
267 OSDMap::Incremental
& pending_inc
;
268 // lock to protect pending_inc form changing
269 // when checking is done
270 ceph::mutex pending_inc_lock
=
271 ceph::make_mutex("CleanUpmapJob::pending_inc_lock");
273 CleanUpmapJob(CephContext
*cct
, const OSDMap
& om
, OSDMap::Incremental
& pi
)
274 : ParallelPGMapper::Job(&om
),
279 void process(const vector
<pg_t
>& to_check
) override
{
280 vector
<pg_t
> to_cancel
;
281 map
<pg_t
, mempool::osdmap::vector
<pair
<int,int>>> to_remap
;
282 osdmap
.check_pg_upmaps(cct
, to_check
, &to_cancel
, &to_remap
);
283 // don't bother taking lock if nothing changes
284 if (!to_cancel
.empty() || !to_remap
.empty()) {
285 std::lock_guard
l(pending_inc_lock
);
286 osdmap
.clean_pg_upmaps(cct
, &pending_inc
, to_cancel
, to_remap
);
290 void process(int64_t poolid
, unsigned ps_begin
, unsigned ps_end
) override
{}
291 void complete() override
{}
292 }; // public as this will need to be accessible from TestTestOSDMap.cc
296 void create_initial() override
;
297 void get_store_prefixes(std::set
<string
>& s
) const override
;
300 void update_from_paxos(bool *need_bootstrap
) override
;
301 void create_pending() override
; // prepare a new pending
302 void encode_pending(MonitorDBStore::TransactionRef t
) override
;
303 void on_active() override
;
304 void on_restart() override
;
305 void on_shutdown() override
;
307 /* osdmap full map prune */
308 void load_osdmap_manifest();
309 bool should_prune() const;
310 void _prune_update_trimmed(
311 MonitorDBStore::TransactionRef tx
,
313 void prune_init(osdmap_manifest_t
& manifest
);
314 bool _prune_sanitize_options() const;
315 bool is_prune_enabled() const;
316 bool is_prune_supported() const;
317 bool do_prune(MonitorDBStore::TransactionRef tx
);
319 // Priority cache control
320 uint32_t mon_osd_cache_size
= 0; ///< Number of cached OSDMaps
321 uint64_t rocksdb_cache_size
= 0; ///< Cache for kv Db
322 double cache_kv_ratio
= 0; ///< Cache ratio dedicated to kv
323 double cache_inc_ratio
= 0; ///< Cache ratio dedicated to inc
324 double cache_full_ratio
= 0; ///< Cache ratio dedicated to full
325 uint64_t mon_memory_base
= 0; ///< Mon base memory for cache autotuning
326 double mon_memory_fragmentation
= 0; ///< Expected memory fragmentation
327 uint64_t mon_memory_target
= 0; ///< Mon target memory for cache autotuning
328 uint64_t mon_memory_min
= 0; ///< Min memory to cache osdmaps
329 bool mon_memory_autotune
= false; ///< Cache auto tune setting
330 int register_cache_with_pcm();
331 int _set_cache_sizes();
332 int _set_cache_ratios();
333 void _set_new_cache_sizes();
334 void _set_cache_autotuning();
335 int _update_mon_cache_settings();
337 friend struct OSDMemCache
;
338 friend struct IncCache
;
339 friend struct FullCache
;
342 * we haven't delegated full version stashing to paxosservice for some time
343 * now, making this function useless in current context.
345 void encode_full(MonitorDBStore::TransactionRef t
) override
{ }
347 * do not let paxosservice periodically stash full osdmaps, or we will break our
348 * locally-managed full maps. (update_from_paxos loads the latest and writes them
349 * out going forward from there, but if we just synced that may mean we skip some.)
351 bool should_stash_full() override
{
356 * hook into trim to include the oldest full map in the trim transaction
358 * This ensures that anyone post-sync will have enough to rebuild their
361 void encode_trim_extra(MonitorDBStore::TransactionRef tx
, version_t first
) override
;
363 void update_msgr_features();
364 int check_cluster_features(uint64_t features
, stringstream
&ss
);
366 * check if the cluster supports the features required by the
367 * given crush map. Outputs the daemons which don't support it
368 * to the stringstream.
370 * @returns true if the map is passable, false otherwise
372 bool validate_crush_against_features(const CrushWrapper
*newcrush
,
374 void check_osdmap_subs();
375 void share_map_with_random_osd();
377 ceph::mutex prime_pg_temp_lock
=
378 ceph::make_mutex("OSDMonitor::prime_pg_temp_lock");
379 struct PrimeTempJob
: public ParallelPGMapper::Job
{
381 PrimeTempJob(const OSDMap
& om
, OSDMonitor
*m
)
382 : ParallelPGMapper::Job(&om
), osdmon(m
) {}
383 void process(int64_t pool
, unsigned ps_begin
, unsigned ps_end
) override
{
384 for (unsigned ps
= ps_begin
; ps
< ps_end
; ++ps
) {
386 osdmon
->prime_pg_temp(*osdmap
, pgid
);
389 void process(const vector
<pg_t
>& pgs
) override
{}
390 void complete() override
{}
392 void maybe_prime_pg_temp();
393 void prime_pg_temp(const OSDMap
& next
, pg_t pgid
);
395 ParallelPGMapper mapper
; ///< for background pg work
396 OSDMapMapping mapping
; ///< pg <-> osd mappings
397 unique_ptr
<ParallelPGMapper::Job
> mapping_job
; ///< background mapping job
398 void start_mapping();
400 void update_logger();
402 void handle_query(PaxosServiceMessage
*m
);
403 bool preprocess_query(MonOpRequestRef op
) override
; // true if processed.
404 bool prepare_update(MonOpRequestRef op
) override
;
405 bool should_propose(double &delay
) override
;
407 version_t
get_trim_to() const override
;
409 bool can_mark_down(int o
);
410 bool can_mark_up(int o
);
411 bool can_mark_out(int o
);
412 bool can_mark_in(int o
);
415 MOSDMap
*build_latest_full(uint64_t features
);
416 MOSDMap
*build_incremental(epoch_t first
, epoch_t last
, uint64_t features
);
417 void send_full(MonOpRequestRef op
);
418 void send_incremental(MonOpRequestRef op
, epoch_t first
);
420 // @param req an optional op request, if the osdmaps are replies to it. so
421 // @c Monitor::send_reply() can mark_event with it.
422 void send_incremental(epoch_t first
, MonSession
*session
, bool onetime
,
423 MonOpRequestRef req
= MonOpRequestRef());
426 void print_utilization(ostream
&out
, Formatter
*f
, bool tree
) const;
428 bool check_source(MonOpRequestRef op
, uuid_d fsid
);
430 bool preprocess_get_osdmap(MonOpRequestRef op
);
432 bool preprocess_mark_me_down(MonOpRequestRef op
);
434 friend class C_AckMarkedDown
;
435 bool preprocess_failure(MonOpRequestRef op
);
436 bool prepare_failure(MonOpRequestRef op
);
437 bool prepare_mark_me_down(MonOpRequestRef op
);
438 void process_failures();
439 void take_all_failures(list
<MonOpRequestRef
>& ls
);
441 bool preprocess_mark_me_dead(MonOpRequestRef op
);
442 bool prepare_mark_me_dead(MonOpRequestRef op
);
444 bool preprocess_full(MonOpRequestRef op
);
445 bool prepare_full(MonOpRequestRef op
);
447 bool preprocess_boot(MonOpRequestRef op
);
448 bool prepare_boot(MonOpRequestRef op
);
449 void _booted(MonOpRequestRef op
, bool logit
);
451 void update_up_thru(int from
, epoch_t up_thru
);
452 bool preprocess_alive(MonOpRequestRef op
);
453 bool prepare_alive(MonOpRequestRef op
);
454 void _reply_map(MonOpRequestRef op
, epoch_t e
);
456 bool preprocess_pgtemp(MonOpRequestRef op
);
457 bool prepare_pgtemp(MonOpRequestRef op
);
459 bool preprocess_pg_created(MonOpRequestRef op
);
460 bool prepare_pg_created(MonOpRequestRef op
);
462 bool preprocess_pg_ready_to_merge(MonOpRequestRef op
);
463 bool prepare_pg_ready_to_merge(MonOpRequestRef op
);
465 int _check_remove_pool(int64_t pool_id
, const pg_pool_t
&pool
, ostream
*ss
);
466 bool _check_become_tier(
467 int64_t tier_pool_id
, const pg_pool_t
*tier_pool
,
468 int64_t base_pool_id
, const pg_pool_t
*base_pool
,
469 int *err
, ostream
*ss
) const;
470 bool _check_remove_tier(
471 int64_t base_pool_id
, const pg_pool_t
*base_pool
, const pg_pool_t
*tier_pool
,
472 int *err
, ostream
*ss
) const;
474 int _prepare_remove_pool(int64_t pool
, ostream
*ss
, bool no_fake
);
475 int _prepare_rename_pool(int64_t pool
, string newname
);
477 bool enforce_pool_op_caps(MonOpRequestRef op
);
478 bool preprocess_pool_op (MonOpRequestRef op
);
479 bool preprocess_pool_op_create (MonOpRequestRef op
);
480 bool prepare_pool_op (MonOpRequestRef op
);
481 bool prepare_pool_op_create (MonOpRequestRef op
);
482 bool prepare_pool_op_delete(MonOpRequestRef op
);
483 int crush_rename_bucket(const string
& srcname
,
484 const string
& dstname
,
486 void check_legacy_ec_plugin(const string
& plugin
,
487 const string
& profile
) const;
488 int normalize_profile(const string
& profilename
,
489 ErasureCodeProfile
&profile
,
492 int crush_rule_create_erasure(const string
&name
,
493 const string
&profile
,
496 int get_crush_rule(const string
&rule_name
,
499 int get_erasure_code(const string
&erasure_code_profile
,
500 ErasureCodeInterfaceRef
*erasure_code
,
502 int prepare_pool_crush_rule(const unsigned pool_type
,
503 const string
&erasure_code_profile
,
504 const string
&rule_name
,
507 bool erasure_code_profile_in_use(
508 const mempool::osdmap::map
<int64_t, pg_pool_t
> &pools
,
509 const string
&profile
,
511 int parse_erasure_code_profile(const vector
<string
> &erasure_code_profile
,
512 map
<string
,string
> *erasure_code_profile_map
,
514 int prepare_pool_size(const unsigned pool_type
,
515 const string
&erasure_code_profile
,
517 unsigned *size
, unsigned *min_size
,
519 int prepare_pool_stripe_width(const unsigned pool_type
,
520 const string
&erasure_code_profile
,
521 unsigned *stripe_width
,
523 int check_pg_num(int64_t pool
, int pg_num
, int size
, ostream
* ss
);
524 int prepare_new_pool(string
& name
,
526 const string
&crush_rule_name
,
527 unsigned pg_num
, unsigned pgp_num
,
530 const uint64_t target_size_bytes
,
531 const float target_size_ratio
,
532 const string
&erasure_code_profile
,
533 const unsigned pool_type
,
534 const uint64_t expected_num_objects
,
535 FastReadType fast_read
,
536 const string
& pg_autoscale_mode
,
538 int prepare_new_pool(MonOpRequestRef op
);
540 void set_pool_flags(int64_t pool_id
, uint64_t flags
);
541 void clear_pool_flags(int64_t pool_id
, uint64_t flags
);
542 bool update_pools_status();
544 bool _is_removed_snap(int64_t pool_id
, snapid_t snapid
);
545 bool _is_pending_removed_snap(int64_t pool_id
, snapid_t snapid
);
547 string
make_purged_snap_epoch_key(epoch_t epoch
);
548 string
make_purged_snap_key(int64_t pool
, snapid_t snap
);
549 string
make_purged_snap_key_value(int64_t pool
, snapid_t snap
, snapid_t num
,
550 epoch_t epoch
, bufferlist
*v
);
552 bool try_prune_purged_snaps();
553 int lookup_purged_snap(int64_t pool
, snapid_t snap
,
554 snapid_t
*begin
, snapid_t
*end
);
556 void insert_purged_snap_update(
558 snapid_t start
, snapid_t end
,
560 MonitorDBStore::TransactionRef t
);
562 bool prepare_set_flag(MonOpRequestRef op
, int flag
);
563 bool prepare_unset_flag(MonOpRequestRef op
, int flag
);
565 void _pool_op_reply(MonOpRequestRef op
,
566 int ret
, epoch_t epoch
, bufferlist
*blp
=NULL
);
568 struct C_Booted
: public C_MonOp
{
571 C_Booted(OSDMonitor
*cm
, MonOpRequestRef op_
, bool l
=true) :
572 C_MonOp(op_
), cmon(cm
), logit(l
) {}
573 void _finish(int r
) override
{
575 cmon
->_booted(op
, logit
);
576 else if (r
== -ECANCELED
)
578 else if (r
== -EAGAIN
)
581 ceph_abort_msg("bad C_Booted return value");
585 struct C_ReplyMap
: public C_MonOp
{
588 C_ReplyMap(OSDMonitor
*o
, MonOpRequestRef op_
, epoch_t ee
)
589 : C_MonOp(op_
), osdmon(o
), e(ee
) {}
590 void _finish(int r
) override
{
592 osdmon
->_reply_map(op
, e
);
593 else if (r
== -ECANCELED
)
595 else if (r
== -EAGAIN
)
596 osdmon
->dispatch(op
);
598 ceph_abort_msg("bad C_ReplyMap return value");
601 struct C_PoolOp
: public C_MonOp
{
605 bufferlist reply_data
;
606 C_PoolOp(OSDMonitor
* osd
, MonOpRequestRef op_
, int rc
, int e
, bufferlist
*rd
=NULL
) :
607 C_MonOp(op_
), osdmon(osd
), replyCode(rc
), epoch(e
) {
611 void _finish(int r
) override
{
613 osdmon
->_pool_op_reply(op
, replyCode
, epoch
, &reply_data
);
614 else if (r
== -ECANCELED
)
616 else if (r
== -EAGAIN
)
617 osdmon
->dispatch(op
);
619 ceph_abort_msg("bad C_PoolOp return value");
623 bool preprocess_remove_snaps(MonOpRequestRef op
);
624 bool prepare_remove_snaps(MonOpRequestRef op
);
626 bool preprocess_get_purged_snaps(MonOpRequestRef op
);
628 int load_metadata(int osd
, map
<string
, string
>& m
, ostream
*err
);
629 void count_metadata(const string
& field
, Formatter
*f
);
631 void reencode_incremental_map(bufferlist
& bl
, uint64_t features
);
632 void reencode_full_map(bufferlist
& bl
, uint64_t features
);
634 void count_metadata(const string
& field
, map
<string
,int> *out
);
636 int get_osd_objectstore_type(int osd
, std::string
*type
);
637 bool is_pool_currently_all_bluestore(int64_t pool_id
, const pg_pool_t
&pool
,
640 // when we last received PG stats from each osd
641 map
<int,utime_t
> last_osd_report
;
642 // TODO: use last_osd_report to store the osd report epochs, once we don't
643 // need to upgrade from pre-luminous releases.
644 map
<int,epoch_t
> osd_epochs
;
645 LastEpochClean last_epoch_clean
;
646 bool preprocess_beacon(MonOpRequestRef op
);
647 bool prepare_beacon(MonOpRequestRef op
);
648 epoch_t
get_min_last_epoch_clean() const;
650 friend class C_UpdateCreatingPGs
;
651 std::map
<int, std::map
<epoch_t
, std::set
<spg_t
>>> creating_pgs_by_osd_epoch
;
652 std::vector
<pg_t
> pending_created_pgs
;
653 // the epoch when the pg mapping was calculated
654 epoch_t creating_pgs_epoch
= 0;
655 creating_pgs_t creating_pgs
;
656 mutable std::mutex creating_pgs_lock
;
658 creating_pgs_t
update_pending_pgs(const OSDMap::Incremental
& inc
,
659 const OSDMap
& nextmap
);
660 unsigned scan_for_creating_pgs(
661 const mempool::osdmap::map
<int64_t,pg_pool_t
>& pools
,
662 const mempool::osdmap::set
<int64_t>& removed_pools
,
664 creating_pgs_t
* creating_pgs
) const;
665 pair
<int32_t, pg_t
> get_parent_pg(pg_t pgid
) const;
666 void update_creating_pgs();
667 void check_pg_creates_subs();
668 epoch_t
send_pg_creates(int osd
, Connection
*con
, epoch_t next
) const;
670 int32_t _allocate_osd_id(int32_t* existing_id
);
673 OSDMonitor(CephContext
*cct
, Monitor
*mn
, Paxos
*p
, const string
& service_name
);
675 void tick() override
; // check state, take actions
677 bool preprocess_command(MonOpRequestRef op
);
678 bool prepare_command(MonOpRequestRef op
);
679 bool prepare_command_impl(MonOpRequestRef op
, const cmdmap_t
& cmdmap
);
681 int validate_osd_create(
684 const bool check_osd_exists
,
685 int32_t* existing_id
,
687 int prepare_command_osd_create(
690 int32_t* existing_id
,
692 void do_osd_create(const int32_t id
, const uuid_d
& uuid
,
693 const string
& device_class
,
695 int prepare_command_osd_purge(int32_t id
, stringstream
& ss
);
696 int prepare_command_osd_destroy(int32_t id
, stringstream
& ss
);
697 int _prepare_command_osd_crush_remove(
698 CrushWrapper
&newcrush
,
703 void do_osd_crush_remove(CrushWrapper
& newcrush
);
704 int prepare_command_osd_crush_remove(
705 CrushWrapper
&newcrush
,
710 int prepare_command_osd_remove(int32_t id
);
711 int prepare_command_osd_new(
713 const cmdmap_t
& cmdmap
,
714 const map
<string
,string
>& secrets
,
718 int prepare_command_pool_set(const cmdmap_t
& cmdmap
,
721 int prepare_command_pool_application(const string
&prefix
,
722 const cmdmap_t
& cmdmap
,
724 int preprocess_command_pool_application(const string
&prefix
,
725 const cmdmap_t
& cmdmap
,
728 int _command_pool_application(const string
&prefix
,
729 const cmdmap_t
& cmdmap
,
734 bool handle_osd_timeouts(const utime_t
&now
,
735 std::map
<int,utime_t
> &last_osd_report
);
737 void send_latest(MonOpRequestRef op
, epoch_t start
=0);
738 void send_latest_now_nodelete(MonOpRequestRef op
, epoch_t start
=0) {
739 op
->mark_osdmon_event(__func__
);
740 send_incremental(op
, start
);
743 int get_version(version_t ver
, bufferlist
& bl
) override
;
744 int get_version(version_t ver
, uint64_t feature
, bufferlist
& bl
);
746 int get_version_full(version_t ver
, uint64_t feature
, bufferlist
& bl
);
747 int get_version_full(version_t ver
, bufferlist
& bl
) override
;
748 int get_inc(version_t ver
, OSDMap::Incremental
& inc
);
749 int get_full_from_pinned_map(version_t ver
, bufferlist
& bl
);
751 epoch_t
blacklist(const entity_addrvec_t
& av
, utime_t until
);
752 epoch_t
blacklist(entity_addr_t a
, utime_t until
);
754 void dump_info(Formatter
*f
);
755 int dump_osd_metadata(int osd
, Formatter
*f
, ostream
*err
);
756 void print_nodes(Formatter
*f
);
758 void check_osdmap_sub(Subscription
*sub
);
759 void check_pg_creates_sub(Subscription
*sub
);
761 void do_application_enable(int64_t pool_id
, const std::string
&app_name
,
762 const std::string
&app_key
="",
763 const std::string
&app_value
="",
765 void do_set_pool_opt(int64_t pool_id
, pool_opts_t::key_t opt
,
766 pool_opts_t::value_t
);
768 void add_flag(int flag
) {
769 if (!(osdmap
.flags
& flag
)) {
770 if (pending_inc
.new_flags
< 0)
771 pending_inc
.new_flags
= osdmap
.flags
;
772 pending_inc
.new_flags
|= flag
;
776 void remove_flag(int flag
) {
777 if(osdmap
.flags
& flag
) {
778 if (pending_inc
.new_flags
< 0)
779 pending_inc
.new_flags
= osdmap
.flags
;
780 pending_inc
.new_flags
&= ~flag
;
783 void convert_pool_priorities(void);