1 // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
2 // vim: ts=8 sw=2 smarttab
4 * Ceph - scalable distributed file system
6 * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
8 * This is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License version 2.1, as published by the Free Software
11 * Foundation. See file COPYING.
16 * This is the top level monitor. It runs on each machine in the Monitor
17 * Cluster. The election of a leader for the paxos algorithm only happens
18 * once per machine via the elector. There is a separate paxos instance (state)
19 * kept for each of the system components: Object Store Device (OSD) Monitor,
20 * Placement Group (PG) Monitor, Metadata Server (MDS) Monitor, and Client Monitor.
23 #ifndef CEPH_MONITOR_H
24 #define CEPH_MONITOR_H
31 #include "include/types.h"
32 #include "include/health.h"
33 #include "msg/Messenger.h"
35 #include "common/Timer.h"
37 #include "health_check.h"
42 #include "MonCommand.h"
45 #include "common/config_obs.h"
46 #include "common/LogClient.h"
47 #include "auth/AuthClient.h"
48 #include "auth/AuthServer.h"
49 #include "auth/cephx/CephxKeyServer.h"
50 #include "auth/AuthMethodList.h"
51 #include "auth/KeyRing.h"
52 #include "include/common_fwd.h"
53 #include "messages/MMonCommand.h"
54 #include "mon/MonitorDBStore.h"
55 #include "mgr/MgrClient.h"
57 #include "mon/MonOpRequest.h"
58 #include "common/WorkQueue.h"
60 using namespace TOPNSPC::common
;
62 #define CEPH_MON_PROTOCOL 13 /* cluster internal */
66 l_cluster_first
= 555000,
68 l_cluster_num_mon_quorum
,
74 l_cluster_osd_bytes_used
,
75 l_cluster_osd_bytes_avail
,
78 l_cluster_num_pg_active_clean
,
79 l_cluster_num_pg_active
,
80 l_cluster_num_pg_peering
,
82 l_cluster_num_object_degraded
,
83 l_cluster_num_object_misplaced
,
84 l_cluster_num_object_unfound
,
104 class AdminSocketHook
;
106 #define COMPAT_SET_LOC "feature_set"
108 class Monitor
: public Dispatcher
,
111 public md_config_obs_t
{
114 const char **orig_argv
= nullptr;
119 Messenger
*messenger
;
120 ConnectionRef con_self
;
121 ceph::mutex lock
= ceph::make_mutex("Monitor::lock");
124 ThreadPool cpu_tp
; ///< threadpool for CPU intensive work
126 ceph::mutex auth_lock
= ceph::make_mutex("Monitor::auth_lock");
128 /// true if we have ever joined a quorum. if false, we are either a
129 /// new cluster, a newly joining monitor, or a just-upgraded
131 bool has_ever_joined
;
133 PerfCounters
*logger
, *cluster_logger
;
134 bool cluster_logger_registered
;
136 void register_cluster_logger();
137 void unregister_cluster_logger();
142 std::set
<entity_addrvec_t
> extra_probe_peers
;
144 LogClient log_client
;
146 LogChannelRef audit_clog
;
148 KeyServer key_server
;
150 AuthMethodList auth_cluster_required
;
151 AuthMethodList auth_service_required
;
155 std::vector
<MonCommand
> leader_mon_commands
; // quorum leader's commands
156 std::vector
<MonCommand
> local_mon_commands
; // commands i support
157 ceph::buffer::list local_mon_commands_bl
; // encoded version of above
159 std::vector
<MonCommand
> prenautilus_local_mon_commands
;
160 ceph::buffer::list prenautilus_local_mon_commands_bl
;
162 Messenger
*mgr_messenger
;
163 MgrClient mgr_client
;
164 uint64_t mgr_proxy_bytes
= 0; // in-flight proxied mgr command message bytes
165 std::string gss_ktfile_client
{};
170 // -- local storage --
172 MonitorDBStore
*store
;
173 static const std::string MONITOR_NAME
;
174 static const std::string MONITOR_STORE_PREFIX
;
176 // -- monitor state --
187 int state
= STATE_INIT
;
190 static const char *get_state_name(int s
) {
192 case STATE_PROBING
: return "probing";
193 case STATE_SYNCHRONIZING
: return "synchronizing";
194 case STATE_ELECTING
: return "electing";
195 case STATE_LEADER
: return "leader";
196 case STATE_PEON
: return "peon";
197 case STATE_SHUTDOWN
: return "shutdown";
198 default: return "???";
201 const char *get_state_name() const {
202 return get_state_name(state
);
205 bool is_init() const { return state
== STATE_INIT
; }
206 bool is_shutdown() const { return state
== STATE_SHUTDOWN
; }
207 bool is_probing() const { return state
== STATE_PROBING
; }
208 bool is_synchronizing() const { return state
== STATE_SYNCHRONIZING
; }
209 bool is_electing() const { return state
== STATE_ELECTING
; }
210 bool is_leader() const { return state
== STATE_LEADER
; }
211 bool is_peon() const { return state
== STATE_PEON
; }
213 const utime_t
&get_leader_since() const;
215 void prepare_new_fingerprint(MonitorDBStore::TransactionRef t
);
217 std::vector
<DaemonHealthMetric
> get_health_metrics();
221 std::unique_ptr
<Paxos
> paxos
;
223 friend class Elector
;
225 /// features we require of peers (based on on-disk compatset)
226 uint64_t required_features
;
228 int leader
; // current leader (to best of knowledge)
229 std::set
<int> quorum
; // current active set of monitors (if !starting)
230 ceph::mono_clock::time_point quorum_since
; // when quorum formed
231 utime_t leader_since
; // when this monitor became the leader, if it is the leader
232 utime_t exited_quorum
; // time detected as not in quorum; 0 if in
234 // map of counts of connected clients, by type and features, for
236 std::map
<int,FeatureMap
> quorum_feature_map
;
239 * Intersection of quorum member's connection feature bits.
241 uint64_t quorum_con_features
;
243 * Intersection of quorum members mon-specific feature bits
245 mon_feature_t quorum_mon_features
;
247 ceph_release_t quorum_min_mon_release
{ceph_release_t::unknown
};
249 std::set
<std::string
> outside_quorum
;
251 bool stretch_mode_engaged
{false};
252 bool degraded_stretch_mode
{false};
253 bool recovering_stretch_mode
{false};
254 string stretch_bucket_divider
;
255 map
<string
, set
<string
>> dead_mon_buckets
; // bucket->mon ranks, locations with no live mons
256 set
<string
> up_mon_buckets
; // locations with a live mon
257 void do_stretch_mode_election_work();
259 bool session_stretch_allowed(MonSession
*s
, MonOpRequestRef
& op
);
260 void disconnect_disallowed_stretch_sessions();
261 void set_elector_disallowed_leaders(bool allow_election
);
263 map
<string
,string
> crush_loc
;
264 bool need_set_crush_loc
{false};
266 bool is_stretch_mode() { return stretch_mode_engaged
; }
267 bool is_degraded_stretch_mode() { return degraded_stretch_mode
; }
268 bool is_recovering_stretch_mode() { return recovering_stretch_mode
; }
271 * This set of functions maintains the in-memory stretch state
272 * and sets up transitions of the map states by calling in to
273 * MonmapMonitor and OSDMonitor.
275 * The [maybe_]go_* functions are called on the leader to
276 * decide if transitions should happen; the trigger_* functions
277 * set up the map transitions; and the set_* functions actually
278 * change the memory state -- but these are only called
279 * via OSDMonitor::update_from_paxos, to guarantee consistent
280 * updates across the entire cluster.
282 void try_engage_stretch_mode();
283 void maybe_go_degraded_stretch_mode();
284 void trigger_degraded_stretch_mode(const set
<string
>& dead_mons
,
285 const set
<int>& dead_buckets
);
286 void set_degraded_stretch_mode();
287 void go_recovery_stretch_mode();
288 void set_recovery_stretch_mode();
289 void trigger_healthy_stretch_mode();
290 void set_healthy_stretch_mode();
291 void enable_stretch_mode();
292 void set_mon_crush_location(const string
& loc
);
298 * @defgroup Monitor_h_scrub
301 version_t scrub_version
; ///< paxos version we are scrubbing
302 std::map
<int,ScrubResult
> scrub_result
; ///< results so far
305 * trigger a cross-mon scrub
307 * Verify all mons are storing identical content
311 void handle_scrub(MonOpRequestRef op
);
312 bool _scrub(ScrubResult
*r
,
313 std::pair
<std::string
,std::string
> *start
,
315 void scrub_check_results();
316 void scrub_timeout();
319 void scrub_update_interval(ceph::timespan interval
);
321 Context
*scrub_event
; ///< periodic event to trigger scrub (leader)
322 Context
*scrub_timeout_event
; ///< scrub round timeout (leader)
323 void scrub_event_start();
324 void scrub_event_cancel();
325 void scrub_reset_timeout();
326 void scrub_cancel_timeout();
329 std::pair
<std::string
,std::string
> last_key
; ///< last scrubbed key
332 ScrubState() : finished(false) { }
333 virtual ~ScrubState() { }
335 std::shared_ptr
<ScrubState
> scrub_state
; ///< keeps track of current scrub
338 * @defgroup Monitor_h_sync Synchronization
342 * @} // provider state
344 struct SyncProvider
{
345 entity_addrvec_t addrs
;
346 uint64_t cookie
; ///< unique cookie for this sync attempt
347 utime_t timeout
; ///< when we give up and expire this attempt
348 version_t last_committed
; ///< last paxos version on peer
349 std::pair
<std::string
,std::string
> last_key
; ///< last key sent to (or on) peer
350 bool full
; ///< full scan?
351 MonitorDBStore::Synchronizer synchronizer
; ///< iterator
353 SyncProvider() : cookie(0), last_committed(0), full(false) {}
355 void reset_timeout(CephContext
*cct
, int grace
) {
356 timeout
= ceph_clock_now();
361 std::map
<std::uint64_t, SyncProvider
> sync_providers
; ///< cookie -> SyncProvider for those syncing from us
362 uint64_t sync_provider_count
; ///< counter for issued cookies to keep them unique
365 * @} // requester state
367 entity_addrvec_t sync_provider
; ///< who we are syncing from
368 uint64_t sync_cookie
; ///< 0 if we are starting, non-zero otherwise
369 bool sync_full
; ///< true if we are a full sync, false for recent catch-up
370 version_t sync_start_version
; ///< last_committed at sync start
371 Context
*sync_timeout_event
; ///< timeout event
374 * floor for sync source
376 * When we sync we forget about our old last_committed value which
377 * can be dangerous. For example, if we have a cluster of:
383 * If something forces us to sync (say, corruption, or manual
384 * intervention, or bug), we forget last_committed, and might abort.
385 * If mon.a happens to be down when we come back, we will see:
390 * and sync from mon.b, at which point a+b will both have lc 80 and
391 * come online with a majority holding out of date commits.
393 * Avoid this by preserving our old last_committed value prior to
394 * sync and never going backwards.
396 version_t sync_last_committed_floor
;
399 * Obtain the synchronization target prefixes in set form.
401 * We consider a target prefix all those that are relevant when
402 * synchronizing two stores. That is, all those that hold paxos service's
403 * versions, as well as paxos versions, or any control keys such as the
404 * first or last committed version.
406 * Given the current design, this function should return the name of all and
407 * any available paxos service, plus the paxos name.
409 * @returns a set of strings referring to the prefixes being synchronized
411 std::set
<std::string
> get_sync_targets_names();
414 * Reset the monitor's sync-related data structures for syncing *from* a peer
416 void sync_reset_requester();
419 * Reset sync state related to allowing others to sync from us
421 void sync_reset_provider();
424 * Caled when a sync attempt times out (requester-side)
429 * Get the latest monmap for backup purposes during sync
431 void sync_obtain_latest_monmap(ceph::buffer::list
&bl
);
436 * Start pulling committed state from another monitor.
438 * @param entity where to pull committed state from
439 * @param full whether to do a full sync or just catch up on recent paxos
441 void sync_start(entity_addrvec_t
&addrs
, bool full
);
445 * force a sync on next mon restart
447 void sync_force(ceph::Formatter
*f
);
451 * store critical state for safekeeping during sync
453 * We store a few things on the side that we don't want to get clobbered by sync. This
454 * includes the latest monmap and a lower bound on last_committed.
456 void sync_stash_critical_state(MonitorDBStore::TransactionRef tx
);
459 * reset the sync timeout
461 * This is used on the client to restart if things aren't progressing
463 void sync_reset_timeout();
466 * trim stale sync provider state
468 * If someone is syncing from us and hasn't talked to us recently, expire their state.
470 void sync_trim_providers();
475 * Finish up a sync after we've gotten all of the chunks.
477 * @param last_committed final last_committed value from provider
479 void sync_finish(version_t last_committed
);
482 * request the next chunk from the provider
484 void sync_get_next_chunk();
487 * handle sync message
489 * @param m Sync message with operation type MMonSync::OP_START_CHUNKS
491 void handle_sync(MonOpRequestRef op
);
493 void _sync_reply_no_cookie(MonOpRequestRef op
);
495 void handle_sync_get_cookie(MonOpRequestRef op
);
496 void handle_sync_get_chunk(MonOpRequestRef op
);
497 void handle_sync_finish(MonOpRequestRef op
);
499 void handle_sync_cookie(MonOpRequestRef op
);
500 void handle_sync_forward(MonOpRequestRef op
);
501 void handle_sync_chunk(MonOpRequestRef op
);
502 void handle_sync_no_cookie(MonOpRequestRef op
);
505 * @} // Synchronization
508 std::list
<Context
*> waitfor_quorum
;
509 std::list
<Context
*> maybe_wait_for_quorum
;
512 * @defgroup Monitor_h_TimeCheck Monitor Clock Drift Early Warning System
515 * We use time checks to keep track of any clock drifting going on in the
516 * cluster. This is accomplished by periodically ping each monitor in the
517 * quorum and register its response time on a map, assessing how much its
518 * clock has drifted. We also take this opportunity to assess the latency
521 * This mechanism works as follows:
523 * - Leader sends out a 'PING' message to each other monitor in the quorum.
524 * The message is timestamped with the leader's current time. The leader's
525 * current time is recorded in a map, associated with each peon's
527 * - The peon replies to the leader with a timestamped 'PONG' message.
528 * - The leader calculates a delta between the peon's timestamp and its
529 * current time and stashes it.
530 * - The leader also calculates the time it took to receive the 'PONG'
531 * since the 'PING' was sent, and stashes an approximate latency estimate.
532 * - Once all the quorum members have pong'ed, the leader will share the
533 * clock skew and latency maps with all the monitors in the quorum.
535 std::map
<int, utime_t
> timecheck_waiting
;
536 std::map
<int, double> timecheck_skews
;
537 std::map
<int, double> timecheck_latencies
;
538 // odd value means we are mid-round; even value means the round has
540 version_t timecheck_round
;
541 unsigned int timecheck_acks
;
542 utime_t timecheck_round_start
;
543 friend class HealthMonitor
;
544 /* When we hit a skew we will start a new round based off of
545 * 'mon_timecheck_skew_interval'. Each new round will be backed off
546 * until we hit 'mon_timecheck_interval' -- which is the typical
547 * interval when not in the presence of a skew.
549 * This variable tracks the number of rounds with skews since last clean
550 * so that we can report to the user and properly adjust the backoff.
552 uint64_t timecheck_rounds_since_clean
;
556 Context
*timecheck_event
;
558 void timecheck_start();
559 void timecheck_finish();
560 void timecheck_start_round();
561 void timecheck_finish_round(bool success
= true);
562 void timecheck_cancel_round();
563 void timecheck_cleanup();
564 void timecheck_reset_event();
565 void timecheck_check_skews();
566 void timecheck_report();
568 health_status_t
timecheck_status(std::ostringstream
&ss
,
569 const double skew_bound
,
570 const double latency
);
571 void handle_timecheck_leader(MonOpRequestRef op
);
572 void handle_timecheck_peon(MonOpRequestRef op
);
573 void handle_timecheck(MonOpRequestRef op
);
576 * Returns 'true' if this is considered to be a skew; 'false' otherwise.
578 bool timecheck_has_skew(const double skew_bound
, double *abs
) const {
579 double abs_skew
= std::fabs(skew_bound
);
582 return (abs_skew
> g_conf()->mon_clock_drift_allowed
);
589 * Handle ping messages from others.
591 void handle_ping(MonOpRequestRef op
);
593 Context
*probe_timeout_event
= nullptr; // for probing
595 void reset_probe_timeout();
596 void cancel_probe_timeout();
597 void probe_timeout(int r
);
599 void _apply_compatset_features(CompatSet
&new_features
);
603 int get_leader() const { return leader
; }
604 std::string
get_leader_name() {
605 return quorum
.empty() ? std::string() : monmap
->get_name(leader
);
607 const std::set
<int>& get_quorum() const { return quorum
; }
608 std::list
<std::string
> get_quorum_names() {
609 std::list
<std::string
> q
;
610 for (auto p
= quorum
.begin(); p
!= quorum
.end(); ++p
)
611 q
.push_back(monmap
->get_name(*p
));
614 uint64_t get_quorum_con_features() const {
615 return quorum_con_features
;
617 mon_feature_t
get_quorum_mon_features() const {
618 return quorum_mon_features
;
620 uint64_t get_required_features() const {
621 return required_features
;
623 mon_feature_t
get_required_mon_features() const {
624 return monmap
->get_required_features();
626 void apply_quorum_to_compatset_features();
627 void apply_monmap_to_compatset_features();
628 void calc_quorum_requirements();
630 void get_combined_feature_map(FeatureMap
*fm
);
633 void _reset(); ///< called from bootstrap, start_, or join_election
634 void wait_for_paxos_write();
635 void _finish_svc_election(); ///< called by {win,lose}_election
639 void join_election();
640 void start_election();
641 void win_standalone_election();
642 // end election (called by Elector)
643 void win_election(epoch_t epoch
, const std::set
<int>& q
,
645 const mon_feature_t
& mon_features
,
646 ceph_release_t min_mon_release
,
647 const std::map
<int,Metadata
>& metadata
);
648 void lose_election(epoch_t epoch
, std::set
<int>& q
, int l
,
650 const mon_feature_t
& mon_features
,
651 ceph_release_t min_mon_release
);
652 // end election (called by Elector)
653 void finish_election();
655 void update_logger();
658 * Vector holding the Services serviced by this Monitor.
660 std::array
<std::unique_ptr
<PaxosService
>, PAXOS_NUM
> paxos_service
;
662 class MDSMonitor
*mdsmon() {
663 return (class MDSMonitor
*)paxos_service
[PAXOS_MDSMAP
].get();
666 class MonmapMonitor
*monmon() {
667 return (class MonmapMonitor
*)paxos_service
[PAXOS_MONMAP
].get();
670 class OSDMonitor
*osdmon() {
671 return (class OSDMonitor
*)paxos_service
[PAXOS_OSDMAP
].get();
674 class AuthMonitor
*authmon() {
675 return (class AuthMonitor
*)paxos_service
[PAXOS_AUTH
].get();
678 class LogMonitor
*logmon() {
679 return (class LogMonitor
*) paxos_service
[PAXOS_LOG
].get();
682 class MgrMonitor
*mgrmon() {
683 return (class MgrMonitor
*) paxos_service
[PAXOS_MGR
].get();
686 class MgrStatMonitor
*mgrstatmon() {
687 return (class MgrStatMonitor
*) paxos_service
[PAXOS_MGRSTAT
].get();
690 class HealthMonitor
*healthmon() {
691 return (class HealthMonitor
*) paxos_service
[PAXOS_HEALTH
].get();
694 class ConfigMonitor
*configmon() {
695 return (class ConfigMonitor
*) paxos_service
[PAXOS_CONFIG
].get();
698 class KVMonitor
*kvmon() {
699 return (class KVMonitor
*) paxos_service
[PAXOS_KV
].get();
703 friend class OSDMonitor
;
704 friend class MDSMonitor
;
705 friend class MonmapMonitor
;
706 friend class LogMonitor
;
707 friend class KVMonitor
;
710 MonSessionMap session_map
;
711 ceph::mutex session_map_lock
= ceph::make_mutex("Monitor::session_map_lock");
712 AdminSocketHook
*admin_hook
;
714 template<typename Func
, typename
...Args
>
715 void with_session_map(Func
&& func
) {
716 std::lock_guard
l(session_map_lock
);
717 std::forward
<Func
>(func
)(session_map
);
719 void send_latest_monmap(Connection
*con
);
722 void handle_get_version(MonOpRequestRef op
);
723 void handle_subscribe(MonOpRequestRef op
);
724 void handle_mon_get_map(MonOpRequestRef op
);
726 static void _generate_command_map(cmdmap_t
& cmdmap
,
727 std::map
<std::string
,std::string
> ¶m_str_map
);
728 static const MonCommand
*_get_moncommand(
729 const std::string
&cmd_prefix
,
730 const std::vector
<MonCommand
>& cmds
);
731 bool _allowed_command(MonSession
*s
, const std::string
& module
,
732 const std::string
& prefix
,
733 const cmdmap_t
& cmdmap
,
734 const std::map
<std::string
,std::string
>& param_str_map
,
735 const MonCommand
*this_cmd
);
736 void get_mon_status(ceph::Formatter
*f
);
737 void _quorum_status(ceph::Formatter
*f
, std::ostream
& ss
);
738 bool _add_bootstrap_peer_hint(std::string_view cmd
, const cmdmap_t
& cmdmap
,
740 void handle_tell_command(MonOpRequestRef op
);
741 void handle_command(MonOpRequestRef op
);
742 void handle_route(MonOpRequestRef op
);
744 int get_mon_metadata(int mon
, ceph::Formatter
*f
, std::ostream
& err
);
745 int print_nodes(ceph::Formatter
*f
, std::ostream
& err
);
747 // track metadata reported by win_election()
748 std::map
<int, Metadata
> mon_metadata
;
749 std::map
<int, Metadata
> pending_metadata
;
754 struct health_cache_t
{
755 health_status_t overall
;
759 // health_status_t doesn't really have a NONE value and we're not
760 // okay with setting something else (say, HEALTH_ERR). so just
764 } health_status_cache
;
766 Context
*health_tick_event
= nullptr;
767 Context
*health_interval_event
= nullptr;
769 void health_tick_start();
770 void health_tick_stop();
771 ceph::real_clock::time_point
health_interval_calc_next_update();
772 void health_interval_start();
773 void health_interval_stop();
774 void health_events_cleanup();
776 void health_to_clog_update_conf(const std::set
<std::string
> &changed
);
778 void do_health_to_clog_interval();
779 void do_health_to_clog(bool force
= false);
782 const health_check_map_t
& updated
,
783 const health_check_map_t
& previous
,
784 MonitorDBStore::TransactionRef t
);
788 class HealthCheckLogStatus
{
790 health_status_t severity
;
791 std::string last_message
;
792 utime_t updated_at
= 0;
793 HealthCheckLogStatus(health_status_t severity_
,
794 const std::string
&last_message_
,
796 : severity(severity_
),
797 last_message(last_message_
),
798 updated_at(updated_at_
)
801 std::map
<std::string
, HealthCheckLogStatus
> health_check_log_times
;
805 void get_cluster_status(std::stringstream
&ss
, ceph::Formatter
*f
,
806 MonSession
*session
);
808 void reply_command(MonOpRequestRef op
, int rc
, const std::string
&rs
, version_t version
);
809 void reply_command(MonOpRequestRef op
, int rc
, const std::string
&rs
, ceph::buffer::list
& rdata
, version_t version
);
811 void reply_tell_command(MonOpRequestRef op
, int rc
, const std::string
&rs
);
815 void handle_probe(MonOpRequestRef op
);
817 * Handle a Probe Operation, replying with our name, quorum and known versions.
819 * We use the MMonProbe message class for anything and everything related with
820 * Monitor probing. One of the operations relates directly with the probing
821 * itself, in which we receive a probe request and to which we reply with
822 * our name, our quorum and the known versions for each Paxos service. Thus the
823 * redundant function name. This reply will obviously be sent to the one
824 * probing/requesting these infos.
826 * @todo Add @pre and @post
828 * @param m A Probe message, with an operation of type Probe.
830 void handle_probe_probe(MonOpRequestRef op
);
831 void handle_probe_reply(MonOpRequestRef op
);
834 struct RoutedRequest
{
836 ceph::buffer::list request_bl
;
839 uint64_t con_features
;
842 RoutedRequest() : tid(0), session(NULL
), con_features(0) {}
848 uint64_t routed_request_tid
;
849 std::map
<uint64_t, RoutedRequest
*> routed_requests
;
851 void forward_request_leader(MonOpRequestRef op
);
852 void handle_forward(MonOpRequestRef op
);
853 void send_reply(MonOpRequestRef op
, Message
*reply
);
854 void no_reply(MonOpRequestRef op
);
855 void resend_routed_requests();
856 void remove_session(MonSession
*s
);
857 void remove_all_sessions();
858 void waitlist_or_zap_client(MonOpRequestRef op
);
860 void send_mon_message(Message
*m
, int rank
);
861 /** can_change_external_state if we can do things like
862 * call elections as a result of the new map.
864 void notify_new_monmap(bool can_change_external_state
=false);
867 struct C_Command
: public C_MonOp
{
871 ceph::buffer::list rdata
;
873 C_Command(Monitor
&_mm
, MonOpRequestRef _op
, int r
, std::string s
, version_t v
) :
874 C_MonOp(_op
), mon(_mm
), rc(r
), rs(s
), version(v
){}
875 C_Command(Monitor
&_mm
, MonOpRequestRef _op
, int r
, std::string s
, ceph::buffer::list rd
, version_t v
) :
876 C_MonOp(_op
), mon(_mm
), rc(r
), rs(s
), rdata(rd
), version(v
){}
878 void _finish(int r
) override
{
879 auto m
= op
->get_req
<MMonCommand
>();
881 std::ostringstream ss
;
882 if (!op
->get_req()->get_connection()) {
883 ss
<< "connection dropped for command ";
885 MonSession
*s
= op
->get_session();
887 // if client drops we may not have a session to draw information from.
889 ss
<< "from='" << s
->name
<< " " << s
->addrs
<< "' "
890 << "entity='" << s
->entity_name
<< "' ";
892 ss
<< "session dropped for command ";
896 std::ostringstream ds
;
898 cmdmap_from_json(m
->cmd
, &cmdmap
, ds
);
899 cmd_getval(cmdmap
, "prefix", prefix
);
900 if (prefix
!= "config set" && prefix
!= "config-key set")
901 ss
<< "cmd='" << m
->cmd
<< "': finished";
903 mon
.audit_clog
->info() << ss
.str();
904 mon
.reply_command(op
, rc
, rs
, rdata
, version
);
906 else if (r
== -ECANCELED
)
908 else if (r
== -EAGAIN
)
911 ceph_abort_msg("bad C_Command return value");
916 class C_RetryMessage
: public C_MonOp
{
919 C_RetryMessage(Monitor
*m
, MonOpRequestRef op
) :
920 C_MonOp(op
), mon(m
) { }
922 void _finish(int r
) override
{
923 if (r
== -EAGAIN
|| r
>= 0)
924 mon
->dispatch_op(op
);
925 else if (r
== -ECANCELED
)
928 ceph_abort_msg("bad C_RetryMessage return value");
932 //ms_dispatch handles a lot of logic and we want to reuse it
933 //on forwarded messages, so we create a non-locking version for this class
934 void _ms_dispatch(Message
*m
);
935 bool ms_dispatch(Message
*m
) override
{
936 std::lock_guard l
{lock
};
940 void dispatch_op(MonOpRequestRef op
);
941 //mon_caps is used for un-connected messages from monitors
943 bool get_authorizer(int dest_type
, AuthAuthorizer
**authorizer
);
944 public: // for AuthMonitor msgr1:
945 int ms_handle_authentication(Connection
*con
) override
;
947 void ms_handle_accept(Connection
*con
) override
;
948 bool ms_handle_reset(Connection
*con
) override
;
949 void ms_handle_remote_reset(Connection
*con
) override
{}
950 bool ms_handle_refused(Connection
*con
) override
;
953 int get_auth_request(
955 AuthConnectionMeta
*auth_meta
,
957 std::vector
<uint32_t> *preferred_modes
,
958 ceph::buffer::list
*out
) override
;
959 int handle_auth_reply_more(
961 AuthConnectionMeta
*auth_meta
,
962 const ceph::buffer::list
& bl
,
963 ceph::buffer::list
*reply
) override
;
964 int handle_auth_done(
966 AuthConnectionMeta
*auth_meta
,
969 const ceph::buffer::list
& bl
,
970 CryptoKey
*session_key
,
971 std::string
*connection_secret
) override
;
972 int handle_auth_bad_method(
974 AuthConnectionMeta
*auth_meta
,
975 uint32_t old_auth_method
,
977 const std::vector
<uint32_t>& allowed_methods
,
978 const std::vector
<uint32_t>& allowed_modes
) override
;
981 int handle_auth_request(
983 AuthConnectionMeta
*auth_meta
,
985 uint32_t auth_method
,
986 const ceph::buffer::list
& bl
,
987 ceph::buffer::list
*reply
) override
;
990 int write_default_keyring(ceph::buffer::list
& bl
);
991 void extract_save_mon_key(KeyRing
& keyring
);
993 void collect_metadata(Metadata
*m
);
995 void count_metadata(const std::string
& field
, ceph::Formatter
*f
);
996 void count_metadata(const std::string
& field
, std::map
<std::string
,int> *out
);
997 // get_all_versions() gathers version information from daemons for health check
998 void get_all_versions(std::map
<string
, std::list
<std::string
>> &versions
);
999 void get_versions(std::map
<string
, std::list
<std::string
>> &versions
);
1002 static CompatSet
get_initial_supported_features();
1003 static CompatSet
get_supported_features();
1004 static CompatSet
get_legacy_features();
1005 /// read the ondisk features into the CompatSet pointed to by read_features
1006 static void read_features_off_disk(MonitorDBStore
*store
, CompatSet
*read_features
);
1007 void read_features();
1008 void write_features(MonitorDBStore::TransactionRef t
);
1010 OpTracker op_tracker
;
1013 Monitor(CephContext
*cct_
, std::string nm
, MonitorDBStore
*s
,
1014 Messenger
*m
, Messenger
*mgr_m
, MonMap
*map
);
1015 ~Monitor() override
;
1017 static int check_features(MonitorDBStore
*store
);
1020 const char** get_tracked_conf_keys() const override
;
1021 void handle_conf_change(const ConfigProxy
& conf
,
1022 const std::set
<std::string
> &changed
) override
;
1024 void update_log_clients();
1025 int sanitize_options();
1029 void refresh_from_paxos(bool *need_bootstrap
);
1033 void handle_signal(int sig
);
1035 int mkfs(ceph::buffer::list
& osdmapbl
);
1038 * check cluster_fsid file
1040 * @return EEXIST if file exists and doesn't match, 0 on match, or negative error code
1045 * write cluster_fsid file
1047 * @return 0 on success, or negative error code
1050 int write_fsid(MonitorDBStore::TransactionRef t
);
1052 int do_admin_command(std::string_view command
, const cmdmap_t
& cmdmap
,
1058 // don't allow copying
1059 Monitor(const Monitor
& rhs
);
1060 Monitor
& operator=(const Monitor
&rhs
);
1063 static void format_command_descriptions(const std::vector
<MonCommand
> &commands
,
1066 ceph::buffer::list
*rdata
);
1068 const std::vector
<MonCommand
> &get_local_commands(mon_feature_t f
) {
1069 if (f
.contains_all(ceph::features::mon::FEATURE_NAUTILUS
)) {
1070 return local_mon_commands
;
1072 return prenautilus_local_mon_commands
;
1075 const ceph::buffer::list
& get_local_commands_bl(mon_feature_t f
) {
1076 if (f
.contains_all(ceph::features::mon::FEATURE_NAUTILUS
)) {
1077 return local_mon_commands_bl
;
1079 return prenautilus_local_mon_commands_bl
;
1082 void set_leader_commands(const std::vector
<MonCommand
>& cmds
) {
1083 leader_mon_commands
= cmds
;
1086 bool is_keyring_required();
1089 #define CEPH_MON_FEATURE_INCOMPAT_BASE CompatSet::Feature (1, "initial feature set (~v.18)")
1090 #define CEPH_MON_FEATURE_INCOMPAT_GV CompatSet::Feature (2, "global version sequencing (v0.52)")
1091 #define CEPH_MON_FEATURE_INCOMPAT_SINGLE_PAXOS CompatSet::Feature (3, "single paxos with k/v store (v0.\?)")
1092 #define CEPH_MON_FEATURE_INCOMPAT_OSD_ERASURE_CODES CompatSet::Feature(4, "support erasure code pools")
1093 #define CEPH_MON_FEATURE_INCOMPAT_OSDMAP_ENC CompatSet::Feature(5, "new-style osdmap encoding")
1094 #define CEPH_MON_FEATURE_INCOMPAT_ERASURE_CODE_PLUGINS_V2 CompatSet::Feature(6, "support isa/lrc erasure code")
1095 #define CEPH_MON_FEATURE_INCOMPAT_ERASURE_CODE_PLUGINS_V3 CompatSet::Feature(7, "support shec erasure code")
1096 #define CEPH_MON_FEATURE_INCOMPAT_KRAKEN CompatSet::Feature(8, "support monmap features")
1097 #define CEPH_MON_FEATURE_INCOMPAT_LUMINOUS CompatSet::Feature(9, "luminous ondisk layout")
1098 #define CEPH_MON_FEATURE_INCOMPAT_MIMIC CompatSet::Feature(10, "mimic ondisk layout")
1099 #define CEPH_MON_FEATURE_INCOMPAT_NAUTILUS CompatSet::Feature(11, "nautilus ondisk layout")
1100 #define CEPH_MON_FEATURE_INCOMPAT_OCTOPUS CompatSet::Feature(12, "octopus ondisk layout")
1101 #define CEPH_MON_FEATURE_INCOMPAT_PACIFIC CompatSet::Feature(13, "pacific ondisk layout")
1102 // make sure you add your feature to Monitor::get_supported_features
1107 * new C_MonContext{...}
1111 * new C_MonContext(...)
1113 * because of gcc bug [1].
1115 * [1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85883
1117 template<typename T
>
1118 class C_MonContext
: public LambdaContext
<T
> {
1120 C_MonContext(const Monitor
* m
, T
&& f
) :
1121 LambdaContext
<T
>(std::forward
<T
>(f
)),
1124 void finish(int r
) override
{
1125 if (mon
->is_shutdown())
1127 LambdaContext
<T
>::finish(r
);