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
28 #include "include/types.h"
29 #include "include/encoding.h"
30 #include "common/simple_cache.hpp"
31 #include "common/PriorityCache.h"
32 #include "msg/Messenger.h"
34 #include "osd/OSDMap.h"
35 #include "osd/OSDMapMapping.h"
37 #include "CreatingPGs.h"
38 #include "PaxosService.h"
40 #include "erasure-code/ErasureCodeInterface.h"
41 #include "mon/MonOpRequest.h"
42 #include <boost/functional/hash.hpp>
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 std::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 (auto p
= reporters
.begin(); p
!= reporters
.end(); ++p
)
71 if (p
->second
.failed_since
> max_failed_since
)
72 max_failed_since
= p
->second
.failed_since
;
74 return max_failed_since
;
77 // set the message for the latest report. return any old op request we had,
78 // if any, so we can discard it.
79 MonOpRequestRef
add_report(int who
, utime_t failed_since
,
81 auto p
= reporters
.find(who
);
82 if (p
== reporters
.end()) {
83 if (max_failed_since
!= utime_t() && max_failed_since
< failed_since
)
84 max_failed_since
= failed_since
;
85 p
= reporters
.insert(std::map
<int, failure_reporter_t
>::value_type(who
, failure_reporter_t(failed_since
))).first
;
88 MonOpRequestRef ret
= p
->second
.op
;
93 void take_report_messages(std::list
<MonOpRequestRef
>& ls
) {
94 for (auto p
= reporters
.begin(); p
!= reporters
.end(); ++p
) {
96 ls
.push_back(p
->second
.op
);
102 MonOpRequestRef
cancel_report(int who
) {
103 auto p
= reporters
.find(who
);
104 if (p
== reporters
.end())
105 return MonOpRequestRef();
106 MonOpRequestRef ret
= p
->second
.op
;
108 max_failed_since
= utime_t();
114 class LastEpochClean
{
116 std::vector
<epoch_t
> epoch_by_pg
;
117 ps_t next_missing
= 0;
118 epoch_t floor
= std::numeric_limits
<epoch_t
>::max();
119 void report(ps_t pg
, epoch_t last_epoch_clean
);
121 std::map
<uint64_t, Lec
> report_by_pool
;
123 void report(const pg_t
& pg
, epoch_t last_epoch_clean
);
124 void remove_pool(uint64_t pool
);
125 epoch_t
get_lower_bound(const OSDMap
& latest
) const;
127 void dump(Formatter
*f
) const;
131 struct osdmap_manifest_t
{
132 // all the maps we have pinned -- i.e., won't be removed unless
133 // they are inside a trim interval.
134 std::set
<version_t
> pinned
;
136 osdmap_manifest_t() {}
138 version_t
get_last_pinned() const
140 auto it
= pinned
.crbegin();
141 if (it
== pinned
.crend()) {
147 version_t
get_first_pinned() const
149 auto it
= pinned
.cbegin();
150 if (it
== pinned
.cend()) {
156 bool is_pinned(version_t v
) const
158 return pinned
.find(v
) != pinned
.end();
161 void pin(version_t v
)
166 version_t
get_lower_closest_pinned(version_t v
) const {
167 auto p
= pinned
.lower_bound(v
);
168 if (p
== pinned
.cend()) {
171 if (p
== pinned
.cbegin()) {
179 void encode(ceph::buffer::list
& bl
) const
181 ENCODE_START(1, 1, bl
);
186 void decode(ceph::buffer::list::const_iterator
& bl
)
193 void decode(ceph::buffer::list
& bl
) {
194 auto p
= bl
.cbegin();
198 void dump(ceph::Formatter
*f
) {
199 f
->dump_unsigned("first_pinned", get_first_pinned());
200 f
->dump_unsigned("last_pinned", get_last_pinned());
201 f
->open_array_section("pinned_maps");
202 for (auto& i
: pinned
) {
203 f
->dump_unsigned("epoch", i
);
208 WRITE_CLASS_ENCODER(osdmap_manifest_t
);
210 class OSDMonitor
: public PaxosService
,
211 public md_config_obs_t
{
218 const char** get_tracked_conf_keys() const override
;
219 void handle_conf_change(const ConfigProxy
& conf
,
220 const std::set
<std::string
> &changed
) override
;
222 OSDMap::Incremental pending_inc
;
223 std::map
<int, ceph::buffer::list
> pending_metadata
;
224 std::set
<int> pending_metadata_rm
;
225 std::map
<int, failure_info_t
> failure_info
;
226 std::map
<int,utime_t
> down_pending_out
; // osd down -> out
227 bool priority_convert
= false;
228 std::map
<int64_t,std::set
<snapid_t
>> pending_pseudo_purged_snaps
;
229 std::shared_ptr
<PriorityCache::PriCache
> rocksdb_binned_kv_cache
= nullptr;
230 std::shared_ptr
<PriorityCache::Manager
> pcm
= nullptr;
231 ceph::mutex balancer_lock
= ceph::make_mutex("OSDMonitor::balancer_lock");
233 std::map
<int,double> osd_weight
;
235 using osdmap_key_t
= std::pair
<version_t
, uint64_t>;
236 using osdmap_cache_t
= SimpleLRU
<osdmap_key_t
,
238 std::less
<osdmap_key_t
>,
239 boost::hash
<osdmap_key_t
>>;
240 osdmap_cache_t inc_osd_cache
;
241 osdmap_cache_t full_osd_cache
;
243 bool has_osdmap_manifest
;
244 osdmap_manifest_t osdmap_manifest
;
246 bool check_failures(utime_t now
);
247 bool check_failure(utime_t now
, int target_osd
, failure_info_t
& fi
);
248 void force_failure(int target_osd
, int by
);
250 bool _have_pending_crush();
251 CrushWrapper
&_get_stable_crush();
252 void _get_pending_crush(CrushWrapper
& newcrush
);
260 struct CleanUpmapJob
: public ParallelPGMapper::Job
{
262 const OSDMap
& osdmap
;
263 OSDMap::Incremental
& pending_inc
;
264 // lock to protect pending_inc form changing
265 // when checking is done
266 ceph::mutex pending_inc_lock
=
267 ceph::make_mutex("CleanUpmapJob::pending_inc_lock");
269 CleanUpmapJob(CephContext
*cct
, const OSDMap
& om
, OSDMap::Incremental
& pi
)
270 : ParallelPGMapper::Job(&om
),
275 void process(const std::vector
<pg_t
>& to_check
) override
{
276 std::vector
<pg_t
> to_cancel
;
277 std::map
<pg_t
, mempool::osdmap::vector
<std::pair
<int,int>>> to_remap
;
278 osdmap
.check_pg_upmaps(cct
, to_check
, &to_cancel
, &to_remap
);
279 // don't bother taking lock if nothing changes
280 if (!to_cancel
.empty() || !to_remap
.empty()) {
281 std::lock_guard
l(pending_inc_lock
);
282 osdmap
.clean_pg_upmaps(cct
, &pending_inc
, to_cancel
, to_remap
);
286 void process(int64_t poolid
, unsigned ps_begin
, unsigned ps_end
) override
{}
287 void complete() override
{}
288 }; // public as this will need to be accessible from TestTestOSDMap.cc
292 void create_initial() override
;
293 void get_store_prefixes(std::set
<std::string
>& s
) const override
;
296 void update_from_paxos(bool *need_bootstrap
) override
;
297 void create_pending() override
; // prepare a new pending
298 void encode_pending(MonitorDBStore::TransactionRef t
) override
;
299 void on_active() override
;
300 void on_restart() override
;
301 void on_shutdown() override
;
303 /* osdmap full map prune */
304 void load_osdmap_manifest();
305 bool should_prune() const;
306 void _prune_update_trimmed(
307 MonitorDBStore::TransactionRef tx
,
309 void prune_init(osdmap_manifest_t
& manifest
);
310 bool _prune_sanitize_options() const;
311 bool is_prune_enabled() const;
312 bool is_prune_supported() const;
313 bool do_prune(MonitorDBStore::TransactionRef tx
);
315 // Priority cache control
316 uint32_t mon_osd_cache_size
= 0; ///< Number of cached OSDMaps
317 uint64_t rocksdb_cache_size
= 0; ///< Cache for kv Db
318 double cache_kv_ratio
= 0; ///< Cache ratio dedicated to kv
319 double cache_inc_ratio
= 0; ///< Cache ratio dedicated to inc
320 double cache_full_ratio
= 0; ///< Cache ratio dedicated to full
321 uint64_t mon_memory_base
= 0; ///< Mon base memory for cache autotuning
322 double mon_memory_fragmentation
= 0; ///< Expected memory fragmentation
323 uint64_t mon_memory_target
= 0; ///< Mon target memory for cache autotuning
324 uint64_t mon_memory_min
= 0; ///< Min memory to cache osdmaps
325 bool mon_memory_autotune
= false; ///< Cache auto tune setting
326 int register_cache_with_pcm();
327 int _set_cache_sizes();
328 int _set_cache_ratios();
329 void _set_new_cache_sizes();
330 void _set_cache_autotuning();
331 int _update_mon_cache_settings();
333 friend struct OSDMemCache
;
334 friend struct IncCache
;
335 friend struct FullCache
;
338 * we haven't delegated full version stashing to paxosservice for some time
339 * now, making this function useless in current context.
341 void encode_full(MonitorDBStore::TransactionRef t
) override
{ }
343 * do not let paxosservice periodically stash full osdmaps, or we will break our
344 * locally-managed full maps. (update_from_paxos loads the latest and writes them
345 * out going forward from there, but if we just synced that may mean we skip some.)
347 bool should_stash_full() override
{
352 * hook into trim to include the oldest full map in the trim transaction
354 * This ensures that anyone post-sync will have enough to rebuild their
357 void encode_trim_extra(MonitorDBStore::TransactionRef tx
, version_t first
) override
;
359 void update_msgr_features();
361 * check if the cluster supports the features required by the
362 * given crush map. Outputs the daemons which don't support it
363 * to the stringstream.
365 * @returns true if the map is passable, false otherwise
367 bool validate_crush_against_features(const CrushWrapper
*newcrush
,
368 std::stringstream
&ss
);
369 void check_osdmap_subs();
370 void share_map_with_random_osd();
372 ceph::mutex prime_pg_temp_lock
=
373 ceph::make_mutex("OSDMonitor::prime_pg_temp_lock");
374 struct PrimeTempJob
: public ParallelPGMapper::Job
{
376 PrimeTempJob(const OSDMap
& om
, OSDMonitor
*m
)
377 : ParallelPGMapper::Job(&om
), osdmon(m
) {}
378 void process(int64_t pool
, unsigned ps_begin
, unsigned ps_end
) override
{
379 for (unsigned ps
= ps_begin
; ps
< ps_end
; ++ps
) {
381 osdmon
->prime_pg_temp(*osdmap
, pgid
);
384 void process(const std::vector
<pg_t
>& pgs
) override
{}
385 void complete() override
{}
387 void maybe_prime_pg_temp();
388 void prime_pg_temp(const OSDMap
& next
, pg_t pgid
);
390 ParallelPGMapper mapper
; ///< for background pg work
391 OSDMapMapping mapping
; ///< pg <-> osd mappings
392 std::unique_ptr
<ParallelPGMapper::Job
> mapping_job
; ///< background mapping job
393 void start_mapping();
395 void update_logger();
397 void handle_query(PaxosServiceMessage
*m
);
398 bool preprocess_query(MonOpRequestRef op
) override
; // true if processed.
399 bool prepare_update(MonOpRequestRef op
) override
;
400 bool should_propose(double &delay
) override
;
402 version_t
get_trim_to() const override
;
404 bool can_mark_down(int o
);
405 bool can_mark_up(int o
);
406 bool can_mark_out(int o
);
407 bool can_mark_in(int o
);
410 MOSDMap
*build_latest_full(uint64_t features
);
411 MOSDMap
*build_incremental(epoch_t first
, epoch_t last
, uint64_t features
);
412 void send_full(MonOpRequestRef op
);
413 void send_incremental(MonOpRequestRef op
, epoch_t first
);
416 * Make sure the existing (up) OSDs support the given features
417 * @return 0 on success, or an error code if any OSDs re missing features.
418 * @param ss Filled in with ane explanation of failure, if any
420 int check_cluster_features(uint64_t features
, std::stringstream
&ss
);
421 // @param req an optional op request, if the osdmaps are replies to it. so
422 // @c Monitor::send_reply() can mark_event with it.
423 void send_incremental(epoch_t first
, MonSession
*session
, bool onetime
,
424 MonOpRequestRef req
= MonOpRequestRef());
427 void print_utilization(std::ostream
&out
, ceph::Formatter
*f
, bool tree
) const;
429 bool check_source(MonOpRequestRef op
, uuid_d fsid
);
431 bool preprocess_get_osdmap(MonOpRequestRef op
);
433 bool preprocess_mark_me_down(MonOpRequestRef op
);
435 friend class C_AckMarkedDown
;
436 bool preprocess_failure(MonOpRequestRef op
);
437 bool prepare_failure(MonOpRequestRef op
);
438 bool prepare_mark_me_down(MonOpRequestRef op
);
439 void process_failures();
440 void take_all_failures(std::list
<MonOpRequestRef
>& ls
);
442 bool preprocess_mark_me_dead(MonOpRequestRef op
);
443 bool prepare_mark_me_dead(MonOpRequestRef op
);
445 bool preprocess_full(MonOpRequestRef op
);
446 bool prepare_full(MonOpRequestRef op
);
448 bool preprocess_boot(MonOpRequestRef op
);
449 bool prepare_boot(MonOpRequestRef op
);
450 void _booted(MonOpRequestRef op
, bool logit
);
452 void update_up_thru(int from
, epoch_t up_thru
);
453 bool preprocess_alive(MonOpRequestRef op
);
454 bool prepare_alive(MonOpRequestRef op
);
455 void _reply_map(MonOpRequestRef op
, epoch_t e
);
457 bool preprocess_pgtemp(MonOpRequestRef op
);
458 bool prepare_pgtemp(MonOpRequestRef op
);
460 bool preprocess_pg_created(MonOpRequestRef op
);
461 bool prepare_pg_created(MonOpRequestRef op
);
463 bool preprocess_pg_ready_to_merge(MonOpRequestRef op
);
464 bool prepare_pg_ready_to_merge(MonOpRequestRef op
);
466 int _check_remove_pool(int64_t pool_id
, const pg_pool_t
&pool
, std::ostream
*ss
);
467 bool _check_become_tier(
468 int64_t tier_pool_id
, const pg_pool_t
*tier_pool
,
469 int64_t base_pool_id
, const pg_pool_t
*base_pool
,
470 int *err
, std::ostream
*ss
) const;
471 bool _check_remove_tier(
472 int64_t base_pool_id
, const pg_pool_t
*base_pool
, const pg_pool_t
*tier_pool
,
473 int *err
, std::ostream
*ss
) const;
475 int _prepare_remove_pool(int64_t pool
, std::ostream
*ss
, bool no_fake
);
476 int _prepare_rename_pool(int64_t pool
, std::string newname
);
478 bool enforce_pool_op_caps(MonOpRequestRef op
);
479 bool preprocess_pool_op (MonOpRequestRef op
);
480 bool preprocess_pool_op_create (MonOpRequestRef op
);
481 bool prepare_pool_op (MonOpRequestRef op
);
482 bool prepare_pool_op_create (MonOpRequestRef op
);
483 bool prepare_pool_op_delete(MonOpRequestRef op
);
484 int crush_rename_bucket(const std::string
& srcname
,
485 const std::string
& dstname
,
487 void check_legacy_ec_plugin(const std::string
& plugin
,
488 const std::string
& profile
) const;
489 int normalize_profile(const std::string
& profilename
,
490 ceph::ErasureCodeProfile
&profile
,
493 int crush_rule_create_erasure(const std::string
&name
,
494 const std::string
&profile
,
497 int get_crush_rule(const std::string
&rule_name
,
500 int get_erasure_code(const std::string
&erasure_code_profile
,
501 ceph::ErasureCodeInterfaceRef
*erasure_code
,
502 std::ostream
*ss
) const;
503 int prepare_pool_crush_rule(const unsigned pool_type
,
504 const std::string
&erasure_code_profile
,
505 const std::string
&rule_name
,
508 bool erasure_code_profile_in_use(
509 const mempool::osdmap::map
<int64_t, pg_pool_t
> &pools
,
510 const std::string
&profile
,
512 int parse_erasure_code_profile(const std::vector
<std::string
> &erasure_code_profile
,
513 std::map
<std::string
,std::string
> *erasure_code_profile_map
,
515 int prepare_pool_size(const unsigned pool_type
,
516 const std::string
&erasure_code_profile
,
518 unsigned *size
, unsigned *min_size
,
520 int prepare_pool_stripe_width(const unsigned pool_type
,
521 const std::string
&erasure_code_profile
,
522 unsigned *stripe_width
,
524 int check_pg_num(int64_t pool
, int pg_num
, int size
, std::ostream
* ss
);
525 int prepare_new_pool(std::string
& name
,
527 const std::string
&crush_rule_name
,
528 unsigned pg_num
, unsigned pgp_num
,
531 const uint64_t target_size_bytes
,
532 const float target_size_ratio
,
533 const std::string
&erasure_code_profile
,
534 const unsigned pool_type
,
535 const uint64_t expected_num_objects
,
536 FastReadType fast_read
,
537 const std::string
& pg_autoscale_mode
,
539 int prepare_new_pool(MonOpRequestRef op
);
541 void set_pool_flags(int64_t pool_id
, uint64_t flags
);
542 void clear_pool_flags(int64_t pool_id
, uint64_t flags
);
543 bool update_pools_status();
545 bool _is_removed_snap(int64_t pool_id
, snapid_t snapid
);
546 bool _is_pending_removed_snap(int64_t pool_id
, snapid_t snapid
);
548 std::string
make_purged_snap_epoch_key(epoch_t epoch
);
549 std::string
make_purged_snap_key(int64_t pool
, snapid_t snap
);
550 std::string
make_purged_snap_key_value(int64_t pool
, snapid_t snap
, snapid_t num
,
551 epoch_t epoch
, ceph::buffer::list
*v
);
553 bool try_prune_purged_snaps();
554 int lookup_purged_snap(int64_t pool
, snapid_t snap
,
555 snapid_t
*begin
, snapid_t
*end
);
557 void insert_purged_snap_update(
559 snapid_t start
, snapid_t end
,
561 MonitorDBStore::TransactionRef t
);
563 bool prepare_set_flag(MonOpRequestRef op
, int flag
);
564 bool prepare_unset_flag(MonOpRequestRef op
, int flag
);
566 void _pool_op_reply(MonOpRequestRef op
,
567 int ret
, epoch_t epoch
, ceph::buffer::list
*blp
=NULL
);
569 struct C_Booted
: public C_MonOp
{
572 C_Booted(OSDMonitor
*cm
, MonOpRequestRef op_
, bool l
=true) :
573 C_MonOp(op_
), cmon(cm
), logit(l
) {}
574 void _finish(int r
) override
{
576 cmon
->_booted(op
, logit
);
577 else if (r
== -ECANCELED
)
579 else if (r
== -EAGAIN
)
582 ceph_abort_msg("bad C_Booted return value");
586 struct C_ReplyMap
: public C_MonOp
{
589 C_ReplyMap(OSDMonitor
*o
, MonOpRequestRef op_
, epoch_t ee
)
590 : C_MonOp(op_
), osdmon(o
), e(ee
) {}
591 void _finish(int r
) override
{
593 osdmon
->_reply_map(op
, e
);
594 else if (r
== -ECANCELED
)
596 else if (r
== -EAGAIN
)
597 osdmon
->dispatch(op
);
599 ceph_abort_msg("bad C_ReplyMap return value");
602 struct C_PoolOp
: public C_MonOp
{
606 ceph::buffer::list reply_data
;
607 C_PoolOp(OSDMonitor
* osd
, MonOpRequestRef op_
, int rc
, int e
, ceph::buffer::list
*rd
=NULL
) :
608 C_MonOp(op_
), osdmon(osd
), replyCode(rc
), epoch(e
) {
612 void _finish(int r
) override
{
614 osdmon
->_pool_op_reply(op
, replyCode
, epoch
, &reply_data
);
615 else if (r
== -ECANCELED
)
617 else if (r
== -EAGAIN
)
618 osdmon
->dispatch(op
);
620 ceph_abort_msg("bad C_PoolOp return value");
624 bool preprocess_remove_snaps(MonOpRequestRef op
);
625 bool prepare_remove_snaps(MonOpRequestRef op
);
627 bool preprocess_get_purged_snaps(MonOpRequestRef op
);
629 int load_metadata(int osd
, std::map
<std::string
, std::string
>& m
,
631 void count_metadata(const std::string
& field
, ceph::Formatter
*f
);
633 void reencode_incremental_map(ceph::buffer::list
& bl
, uint64_t features
);
634 void reencode_full_map(ceph::buffer::list
& bl
, uint64_t features
);
636 void count_metadata(const std::string
& field
, std::map
<std::string
,int> *out
);
637 void get_versions(std::map
<std::string
, std::list
<std::string
>> &versions
);
639 int get_osd_objectstore_type(int osd
, std::string
*type
);
640 bool is_pool_currently_all_bluestore(int64_t pool_id
, const pg_pool_t
&pool
,
643 // when we last received PG stats from each osd and the osd's osd_beacon_report_interval
644 std::map
<int, std::pair
<utime_t
, int>> last_osd_report
;
645 // TODO: use last_osd_report to store the osd report epochs, once we don't
646 // need to upgrade from pre-luminous releases.
647 std::map
<int,epoch_t
> osd_epochs
;
648 LastEpochClean last_epoch_clean
;
649 bool preprocess_beacon(MonOpRequestRef op
);
650 bool prepare_beacon(MonOpRequestRef op
);
651 epoch_t
get_min_last_epoch_clean() const;
653 friend class C_UpdateCreatingPGs
;
654 std::map
<int, std::map
<epoch_t
, std::set
<spg_t
>>> creating_pgs_by_osd_epoch
;
655 std::vector
<pg_t
> pending_created_pgs
;
656 // the epoch when the pg mapping was calculated
657 epoch_t creating_pgs_epoch
= 0;
658 creating_pgs_t creating_pgs
;
659 mutable std::mutex creating_pgs_lock
;
661 creating_pgs_t
update_pending_pgs(const OSDMap::Incremental
& inc
,
662 const OSDMap
& nextmap
);
663 unsigned scan_for_creating_pgs(
664 const mempool::osdmap::map
<int64_t,pg_pool_t
>& pools
,
665 const mempool::osdmap::set
<int64_t>& removed_pools
,
667 creating_pgs_t
* creating_pgs
) const;
668 std::pair
<int32_t, pg_t
> get_parent_pg(pg_t pgid
) const;
669 void update_creating_pgs();
670 void check_pg_creates_subs();
671 epoch_t
send_pg_creates(int osd
, Connection
*con
, epoch_t next
) const;
673 int32_t _allocate_osd_id(int32_t* existing_id
);
675 int get_grace_interval_threshold();
676 bool grace_interval_threshold_exceeded(int last_failed
);
677 void set_default_laggy_params(int target_osd
);
680 OSDMonitor(CephContext
*cct
, Monitor
&mn
, Paxos
&p
, const std::string
& service_name
);
682 void tick() override
; // check state, take actions
684 bool preprocess_command(MonOpRequestRef op
);
685 bool prepare_command(MonOpRequestRef op
);
686 bool prepare_command_impl(MonOpRequestRef op
, const cmdmap_t
& cmdmap
);
688 int validate_osd_create(
691 const bool check_osd_exists
,
692 int32_t* existing_id
,
693 std::stringstream
& ss
);
694 int prepare_command_osd_create(
697 int32_t* existing_id
,
698 std::stringstream
& ss
);
699 void do_osd_create(const int32_t id
, const uuid_d
& uuid
,
700 const std::string
& device_class
,
702 int prepare_command_osd_purge(int32_t id
, std::stringstream
& ss
);
703 int prepare_command_osd_destroy(int32_t id
, std::stringstream
& ss
);
704 int _prepare_command_osd_crush_remove(
705 CrushWrapper
&newcrush
,
710 void do_osd_crush_remove(CrushWrapper
& newcrush
);
711 int prepare_command_osd_crush_remove(
712 CrushWrapper
&newcrush
,
717 int prepare_command_osd_remove(int32_t id
);
718 int prepare_command_osd_new(
720 const cmdmap_t
& cmdmap
,
721 const std::map
<std::string
,std::string
>& secrets
,
722 std::stringstream
&ss
,
725 int prepare_command_pool_set(const cmdmap_t
& cmdmap
,
726 std::stringstream
& ss
);
728 int prepare_command_pool_application(const std::string
&prefix
,
729 const cmdmap_t
& cmdmap
,
730 std::stringstream
& ss
);
731 int preprocess_command_pool_application(const std::string
&prefix
,
732 const cmdmap_t
& cmdmap
,
733 std::stringstream
& ss
,
735 int _command_pool_application(const std::string
&prefix
,
736 const cmdmap_t
& cmdmap
,
737 std::stringstream
& ss
,
741 bool handle_osd_timeouts(const utime_t
&now
,
742 std::map
<int, std::pair
<utime_t
, int>> &last_osd_report
);
744 void send_latest(MonOpRequestRef op
, epoch_t start
=0);
745 void send_latest_now_nodelete(MonOpRequestRef op
, epoch_t start
=0) {
746 op
->mark_osdmon_event(__func__
);
747 send_incremental(op
, start
);
750 int get_version(version_t ver
, ceph::buffer::list
& bl
) override
;
751 int get_version(version_t ver
, uint64_t feature
, ceph::buffer::list
& bl
);
753 int get_version_full(version_t ver
, uint64_t feature
, ceph::buffer::list
& bl
);
754 int get_version_full(version_t ver
, ceph::buffer::list
& bl
) override
;
755 int get_inc(version_t ver
, OSDMap::Incremental
& inc
);
756 int get_full_from_pinned_map(version_t ver
, ceph::buffer::list
& bl
);
758 epoch_t
blocklist(const entity_addrvec_t
& av
, utime_t until
);
759 epoch_t
blocklist(entity_addr_t a
, utime_t until
);
761 void dump_info(ceph::Formatter
*f
);
762 int dump_osd_metadata(int osd
, ceph::Formatter
*f
, std::ostream
*err
);
763 void print_nodes(ceph::Formatter
*f
);
765 void check_osdmap_sub(Subscription
*sub
);
766 void check_pg_creates_sub(Subscription
*sub
);
768 void do_application_enable(int64_t pool_id
, const std::string
&app_name
,
769 const std::string
&app_key
="",
770 const std::string
&app_value
="",
772 void do_set_pool_opt(int64_t pool_id
, pool_opts_t::key_t opt
,
773 pool_opts_t::value_t
);
775 void add_flag(int flag
) {
776 if (!(osdmap
.flags
& flag
)) {
777 if (pending_inc
.new_flags
< 0)
778 pending_inc
.new_flags
= osdmap
.flags
;
779 pending_inc
.new_flags
|= flag
;
783 void remove_flag(int flag
) {
784 if(osdmap
.flags
& flag
) {
785 if (pending_inc
.new_flags
< 0)
786 pending_inc
.new_flags
= osdmap
.flags
;
787 pending_inc
.new_flags
&= ~flag
;
790 void convert_pool_priorities(void);
792 * Find the pools which are requested to be put into stretch mode,
793 * validate that they are allowed to be in stretch mode (eg, are replicated)
794 * and place copies of them in the pools set.
795 * This does not make any changes to the pools or state; it's just
796 * a safety-check-and-collect function.
798 void try_enable_stretch_mode_pools(stringstream
& ss
, bool *okay
,
800 set
<pg_pool_t
*>* pools
, const string
& new_crush_rule
);
802 * Check validity of inputs and OSD/CRUSH state to
803 * engage stretch mode. Designed to be used with
804 * MonmapMonitor::try_enable_stretch_mode() where we call both twice,
805 * first with commit=false to validate.
806 * @param ss: a stringstream to write errors into
807 * @param okay: Filled to true if okay, false if validation fails
808 * @param errcode: filled with -errno if there's a problem
809 * @param commit: true if we should commit the change, false if just testing
810 * @param dividing_bucket: the bucket type (eg 'dc') that divides the cluster
811 * @param bucket_count: The number of buckets required in peering.
812 * Currently must be 2.
813 * @param pools: The pg_pool_ts which are being set to stretch mode (obtained
814 * from try_enable_stretch_mode_pools()).
815 * @param new_crush_rule: The crush rule to set the pools to.
817 void try_enable_stretch_mode(stringstream
& ss
, bool *okay
,
818 int *errcode
, bool commit
,
819 const string
& dividing_bucket
,
820 uint32_t bucket_count
,
821 const set
<pg_pool_t
*>& pools
,
822 const string
& new_crush_rule
);
824 * Check the input dead_buckets mapping (buckets->dead monitors) to see
825 * if the OSDs are also down. If so, fill in really_down_buckets and
826 * really_down_mons and return true; else return false.
828 bool check_for_dead_crush_zones(const map
<string
,set
<string
>>& dead_buckets
,
829 set
<int> *really_down_buckets
,
830 set
<string
> *really_down_mons
);
832 * Set degraded mode in the OSDMap, adding the given dead buckets to the dead set
833 * and using the live_zones (should presently be size 1)
835 void trigger_degraded_stretch_mode(const set
<int>& dead_buckets
,
836 const set
<string
>& live_zones
);
838 * Set recovery stretch mode in the OSDMap, resetting pool size back to normal
840 void trigger_recovery_stretch_mode();
842 * Tells the OSD there's a new pg digest, in case it's interested.
843 * (It's interested when in recovering stretch mode.)
845 void notify_new_pg_digest();
847 * Check if we can exit recovery stretch mode and go back to normal.
848 * @param force If true, we will force the exit through once it is legal,
849 * without regard to the reported PG status.
851 void try_end_recovery_stretch_mode(bool force
);
853 * Sets the osdmap and pg_pool_t values back to healthy stretch mode status.
855 void trigger_healthy_stretch_mode();
857 utime_t stretch_recovery_triggered
; // what time we committed a switch to recovery mode