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 #include <boost/utility.hpp>
17 #include <boost/regex.hpp>
19 #include "MDSMonitor.h"
20 #include "FSCommands.h"
22 #include "MonitorDBStore.h"
23 #include "OSDMonitor.h"
24 #include "PGMonitor.h"
26 #include "common/strtol.h"
27 #include "common/perf_counters.h"
28 #include "common/config.h"
29 #include "common/cmdparse.h"
30 #include "messages/MMDSMap.h"
31 #include "messages/MFSMap.h"
32 #include "messages/MFSMapUser.h"
33 #include "messages/MMDSLoadTargets.h"
34 #include "messages/MMonCommand.h"
35 #include "messages/MGenericMessage.h"
37 #include "include/assert.h"
38 #include "include/str_list.h"
39 #include "include/stringify.h"
40 #include "mds/mdstypes.h"
43 #define dout_subsys ceph_subsys_mon
45 #define dout_prefix _prefix(_dout, mon, fsmap)
46 static ostream
& _prefix(std::ostream
*_dout
, Monitor
*mon
, FSMap
const& fsmap
) {
47 return *_dout
<< "mon." << mon
->name
<< "@" << mon
->rank
48 << "(" << mon
->get_state_name()
49 << ").mds e" << fsmap
.get_epoch() << " ";
53 * Specialized implementation of cmd_getval to allow us to parse
54 * out strongly-typedef'd types
56 template<> bool cmd_getval(CephContext
*cct
, const cmdmap_t
& cmdmap
,
57 const std::string
& k
, mds_gid_t
&val
)
59 return cmd_getval(cct
, cmdmap
, k
, (int64_t&)val
);
62 template<> bool cmd_getval(CephContext
*cct
, const cmdmap_t
& cmdmap
,
63 const std::string
& k
, mds_rank_t
&val
)
65 return cmd_getval(cct
, cmdmap
, k
, (int64_t&)val
);
68 template<> bool cmd_getval(CephContext
*cct
, const cmdmap_t
& cmdmap
,
69 const std::string
& k
, MDSMap::DaemonState
&val
)
71 return cmd_getval(cct
, cmdmap
, k
, (int64_t&)val
);
74 static const string
MDS_METADATA_PREFIX("mds_metadata");
79 void MDSMonitor::print_map(FSMap
&m
, int dbl
)
81 dout(dbl
) << "print_map\n";
87 void MDSMonitor::create_initial()
89 dout(10) << "create_initial" << dendl
;
93 void MDSMonitor::update_from_paxos(bool *need_bootstrap
)
95 version_t version
= get_last_committed();
96 if (version
== fsmap
.epoch
)
99 dout(10) << __func__
<< " version " << version
100 << ", my e " << fsmap
.epoch
<< dendl
;
101 assert(version
> fsmap
.epoch
);
108 int err
= get_version(version
, fsmap_bl
);
111 assert(fsmap_bl
.length() > 0);
112 dout(10) << __func__
<< " got " << version
<< dendl
;
113 fsmap
.decode(fsmap_bl
);
116 dout(4) << "new map" << dendl
;
118 if (!g_conf
->mon_mds_skip_sanity
) {
126 void MDSMonitor::init()
128 (void)load_metadata(pending_metadata
);
131 void MDSMonitor::create_pending()
133 pending_fsmap
= fsmap
;
134 pending_fsmap
.epoch
++;
136 dout(10) << "create_pending e" << pending_fsmap
.epoch
<< dendl
;
139 void MDSMonitor::encode_pending(MonitorDBStore::TransactionRef t
)
141 dout(10) << "encode_pending e" << pending_fsmap
.epoch
<< dendl
;
144 // print map iff 'debug mon = 30' or higher
145 print_map(pending_fsmap
, 30);
146 if (!g_conf
->mon_mds_skip_sanity
) {
147 pending_fsmap
.sanity();
150 // Set 'modified' on maps modified this epoch
151 for (auto &i
: fsmap
.filesystems
) {
152 if (i
.second
->mds_map
.epoch
== fsmap
.epoch
) {
153 i
.second
->mds_map
.modified
= ceph_clock_now();
158 assert(get_last_committed() + 1 == pending_fsmap
.epoch
);
160 pending_fsmap
.encode(fsmap_bl
, mon
->get_quorum_con_features());
162 /* put everything in the transaction */
163 put_version(t
, pending_fsmap
.epoch
, fsmap_bl
);
164 put_last_committed(t
, pending_fsmap
.epoch
);
166 // Encode MDSHealth data
167 for (std::map
<uint64_t, MDSHealth
>::iterator i
= pending_daemon_health
.begin();
168 i
!= pending_daemon_health
.end(); ++i
) {
170 i
->second
.encode(bl
);
171 t
->put(MDS_HEALTH_PREFIX
, stringify(i
->first
), bl
);
174 for (std::set
<uint64_t>::iterator i
= pending_daemon_health_rm
.begin();
175 i
!= pending_daemon_health_rm
.end(); ++i
) {
176 t
->erase(MDS_HEALTH_PREFIX
, stringify(*i
));
178 pending_daemon_health_rm
.clear();
179 remove_from_metadata(t
);
182 health_check_map_t new_checks
;
183 const auto info_map
= pending_fsmap
.get_mds_info();
184 for (const auto &i
: info_map
) {
185 const auto &gid
= i
.first
;
186 const auto &info
= i
.second
;
187 if (pending_daemon_health_rm
.count(gid
)) {
191 auto p
= pending_daemon_health
.find(gid
);
192 if (p
!= pending_daemon_health
.end()) {
196 mon
->store
->get(MDS_HEALTH_PREFIX
, stringify(gid
), bl
);
198 derr
<< "Missing health data for MDS " << gid
<< dendl
;
201 bufferlist::iterator bl_i
= bl
.begin();
204 for (const auto &metric
: health
.metrics
) {
205 int const rank
= info
.rank
;
206 health_check_t
*check
= &new_checks
.get_or_add(
207 mds_metric_name(metric
.type
),
209 mds_metric_summary(metric
.type
));
211 ss
<< "mds" << info
.name
<< "(mds." << rank
<< "): " << metric
.message
;
212 for (auto p
= metric
.metadata
.begin();
213 p
!= metric
.metadata
.end();
215 if (p
!= metric
.metadata
.begin()) {
218 ss
<< p
->first
<< ": " << p
->second
;
220 check
->detail
.push_back(ss
.str());
223 pending_fsmap
.get_health_checks(&new_checks
);
224 for (auto& p
: new_checks
.checks
) {
225 p
.second
.summary
= boost::regex_replace(
227 boost::regex("%num%"),
228 stringify(p
.second
.detail
.size()));
229 p
.second
.summary
= boost::regex_replace(
231 boost::regex("%plurals%"),
232 p
.second
.detail
.size() > 1 ? "s" : "");
233 p
.second
.summary
= boost::regex_replace(
235 boost::regex("%isorare%"),
236 p
.second
.detail
.size() > 1 ? "are" : "is");
238 encode_health(new_checks
, t
);
241 version_t
MDSMonitor::get_trim_to()
244 if (g_conf
->mon_mds_force_trim_to
> 0 &&
245 g_conf
->mon_mds_force_trim_to
< (int)get_last_committed()) {
246 floor
= g_conf
->mon_mds_force_trim_to
;
247 dout(10) << __func__
<< " explicit mon_mds_force_trim_to = "
251 unsigned max
= g_conf
->mon_max_mdsmap_epochs
;
252 version_t last
= get_last_committed();
254 if (last
- get_first_committed() > max
&& floor
< last
- max
)
259 void MDSMonitor::update_logger()
261 dout(10) << "update_logger" << dendl
;
266 for (const auto &i
: fsmap
.filesystems
) {
267 const MDSMap
&mds_map
= i
.second
->mds_map
;
269 up
+= mds_map
.get_num_up_mds();
270 in
+= mds_map
.get_num_in_mds();
271 failed
+= mds_map
.get_num_failed_mds();
273 mon
->cluster_logger
->set(l_cluster_num_mds_up
, up
);
274 mon
->cluster_logger
->set(l_cluster_num_mds_in
, in
);
275 mon
->cluster_logger
->set(l_cluster_num_mds_failed
, failed
);
276 mon
->cluster_logger
->set(l_cluster_mds_epoch
, fsmap
.get_epoch());
279 bool MDSMonitor::preprocess_query(MonOpRequestRef op
)
281 op
->mark_mdsmon_event(__func__
);
282 PaxosServiceMessage
*m
= static_cast<PaxosServiceMessage
*>(op
->get_req());
283 dout(10) << "preprocess_query " << *m
<< " from " << m
->get_orig_source_inst() << dendl
;
285 switch (m
->get_type()) {
288 return preprocess_beacon(op
);
290 case MSG_MON_COMMAND
:
291 return preprocess_command(op
);
293 case MSG_MDS_OFFLOAD_TARGETS
:
294 return preprocess_offload_targets(op
);
302 void MDSMonitor::_note_beacon(MMDSBeacon
*m
)
304 mds_gid_t gid
= mds_gid_t(m
->get_global_id());
305 version_t seq
= m
->get_seq();
307 dout(15) << "_note_beacon " << *m
<< " noting time" << dendl
;
308 last_beacon
[gid
].stamp
= ceph_clock_now();
309 last_beacon
[gid
].seq
= seq
;
312 bool MDSMonitor::preprocess_beacon(MonOpRequestRef op
)
314 op
->mark_mdsmon_event(__func__
);
315 MMDSBeacon
*m
= static_cast<MMDSBeacon
*>(op
->get_req());
316 MDSMap::DaemonState state
= m
->get_state();
317 mds_gid_t gid
= m
->get_global_id();
318 version_t seq
= m
->get_seq();
319 MDSMap::mds_info_t info
;
320 epoch_t effective_epoch
= 0;
322 // check privileges, ignore if fails
323 MonSession
*session
= m
->get_session();
325 if (!session
->is_capable("mds", MON_CAP_X
)) {
326 dout(0) << "preprocess_beacon got MMDSBeacon from entity with insufficient privileges "
327 << session
->caps
<< dendl
;
331 if (m
->get_fsid() != mon
->monmap
->fsid
) {
332 dout(0) << "preprocess_beacon on fsid " << m
->get_fsid() << " != " << mon
->monmap
->fsid
<< dendl
;
336 dout(12) << "preprocess_beacon " << *m
337 << " from " << m
->get_orig_source_inst()
338 << " " << m
->get_compat()
341 // make sure the address has a port
342 if (m
->get_orig_source_addr().get_port() == 0) {
343 dout(1) << " ignoring boot message without a port" << dendl
;
348 if (!m
->get_compat().writeable(fsmap
.compat
)) {
349 dout(1) << " mds " << m
->get_source_inst() << " can't write to fsmap " << fsmap
.compat
<< dendl
;
354 if (!mon
->is_leader())
357 // booted, but not in map?
358 if (!pending_fsmap
.gid_exists(gid
)) {
359 if (state
!= MDSMap::STATE_BOOT
) {
360 dout(7) << "mds_beacon " << *m
<< " is not in fsmap (state "
361 << ceph_mds_state_name(state
) << ")" << dendl
;
364 null_map
.epoch
= fsmap
.epoch
;
365 null_map
.compat
= fsmap
.compat
;
366 mon
->send_reply(op
, new MMDSMap(mon
->monmap
->fsid
, &null_map
));
369 return false; // not booted yet.
372 dout(10) << __func__
<< ": GID exists in map: " << gid
<< dendl
;
373 info
= pending_fsmap
.get_info_gid(gid
);
376 if (info
.state_seq
> seq
) {
377 dout(7) << "mds_beacon " << *m
<< " has old seq, ignoring" << dendl
;
381 // Work out the latest epoch that this daemon should have seen
383 fs_cluster_id_t fscid
= pending_fsmap
.mds_roles
.at(gid
);
384 if (fscid
== FS_CLUSTER_ID_NONE
) {
385 effective_epoch
= pending_fsmap
.standby_epochs
.at(gid
);
387 effective_epoch
= pending_fsmap
.get_filesystem(fscid
)->mds_map
.epoch
;
389 if (effective_epoch
!= m
->get_last_epoch_seen()) {
390 dout(10) << "mds_beacon " << *m
391 << " ignoring requested state, because mds hasn't seen latest map" << dendl
;
398 return false; // no longer laggy, need to update map.
400 if (state
== MDSMap::STATE_BOOT
) {
401 // ignore, already booted.
404 // is there a state change here?
405 if (info
.state
!= state
) {
406 // legal state change?
407 if ((info
.state
== MDSMap::STATE_STANDBY
||
408 info
.state
== MDSMap::STATE_STANDBY_REPLAY
) && state
> 0) {
409 dout(10) << "mds_beacon mds can't activate itself (" << ceph_mds_state_name(info
.state
)
410 << " -> " << ceph_mds_state_name(state
) << ")" << dendl
;
414 if ((state
== MDSMap::STATE_STANDBY
|| state
== MDSMap::STATE_STANDBY_REPLAY
)
415 && info
.rank
!= MDS_RANK_NONE
)
417 dout(4) << "mds_beacon MDS can't go back into standby after taking rank: "
418 "held rank " << info
.rank
<< " while requesting state "
419 << ceph_mds_state_name(state
) << dendl
;
427 // Comparing known daemon health with m->get_health()
428 // and return false (i.e. require proposal) if they
429 // do not match, to update our stored
430 if (!(pending_daemon_health
[gid
] == m
->get_health())) {
431 dout(20) << __func__
<< " health metrics for gid " << gid
<< " were updated" << dendl
;
437 // note time and reply
438 assert(effective_epoch
> 0);
441 new MMDSBeacon(mon
->monmap
->fsid
, m
->get_global_id(), m
->get_name(),
442 effective_epoch
, state
, seq
,
443 CEPH_FEATURES_SUPPORTED_DEFAULT
));
447 // I won't reply this beacon, drop it.
452 bool MDSMonitor::preprocess_offload_targets(MonOpRequestRef op
)
454 op
->mark_mdsmon_event(__func__
);
455 MMDSLoadTargets
*m
= static_cast<MMDSLoadTargets
*>(op
->get_req());
456 dout(10) << "preprocess_offload_targets " << *m
<< " from " << m
->get_orig_source() << dendl
;
458 // check privileges, ignore message if fails
459 MonSession
*session
= m
->get_session();
462 if (!session
->is_capable("mds", MON_CAP_X
)) {
463 dout(0) << "preprocess_offload_targets got MMDSLoadTargets from entity with insufficient caps "
464 << session
->caps
<< dendl
;
468 if (fsmap
.gid_exists(m
->global_id
) &&
469 m
->targets
== fsmap
.get_info_gid(m
->global_id
).export_targets
)
479 bool MDSMonitor::prepare_update(MonOpRequestRef op
)
481 op
->mark_mdsmon_event(__func__
);
482 PaxosServiceMessage
*m
= static_cast<PaxosServiceMessage
*>(op
->get_req());
483 dout(7) << "prepare_update " << *m
<< dendl
;
485 switch (m
->get_type()) {
488 return prepare_beacon(op
);
490 case MSG_MON_COMMAND
:
491 return prepare_command(op
);
493 case MSG_MDS_OFFLOAD_TARGETS
:
494 return prepare_offload_targets(op
);
503 bool MDSMonitor::prepare_beacon(MonOpRequestRef op
)
505 op
->mark_mdsmon_event(__func__
);
506 MMDSBeacon
*m
= static_cast<MMDSBeacon
*>(op
->get_req());
507 // -- this is an update --
508 dout(12) << "prepare_beacon " << *m
<< " from " << m
->get_orig_source_inst() << dendl
;
509 entity_addr_t addr
= m
->get_orig_source_inst().addr
;
510 mds_gid_t gid
= m
->get_global_id();
511 MDSMap::DaemonState state
= m
->get_state();
512 version_t seq
= m
->get_seq();
514 dout(20) << __func__
<< " got health from gid " << gid
<< " with " << m
->get_health().metrics
.size() << " metrics." << dendl
;
516 // Calculate deltas of health metrics created and removed
517 // Do this by type rather than MDSHealthMetric equality, because messages can
518 // change a lot when they include e.g. a number of items.
519 const auto &old_health
= pending_daemon_health
[gid
].metrics
;
520 const auto &new_health
= m
->get_health().metrics
;
522 std::set
<mds_metric_t
> old_types
;
523 for (const auto &i
: old_health
) {
524 old_types
.insert(i
.type
);
527 std::set
<mds_metric_t
> new_types
;
528 for (const auto &i
: new_health
) {
529 new_types
.insert(i
.type
);
532 for (const auto &new_metric
: new_health
) {
533 if (old_types
.count(new_metric
.type
) == 0) {
534 std::stringstream msg
;
535 msg
<< "MDS health message (" << m
->get_orig_source_inst().name
<< "): "
536 << new_metric
.message
;
537 if (new_metric
.sev
== HEALTH_ERR
) {
538 mon
->clog
->error() << msg
.str();
539 } else if (new_metric
.sev
== HEALTH_WARN
) {
540 mon
->clog
->warn() << msg
.str();
542 mon
->clog
->info() << msg
.str();
547 // Log the disappearance of health messages at INFO
548 for (const auto &old_metric
: old_health
) {
549 if (new_types
.count(old_metric
.type
) == 0) {
550 mon
->clog
->info() << "MDS health message cleared ("
551 << m
->get_orig_source_inst().name
<< "): " << old_metric
.message
;
556 pending_daemon_health
[gid
] = m
->get_health();
559 if (state
== MDSMap::STATE_BOOT
) {
560 // zap previous instance of this name?
561 if (g_conf
->mds_enforce_unique_name
) {
562 bool failed_mds
= false;
563 while (mds_gid_t existing
= pending_fsmap
.find_mds_gid_by_name(m
->get_name())) {
564 if (!mon
->osdmon()->is_writeable()) {
565 mon
->osdmon()->wait_for_writeable(op
, new C_RetryMessage(this, op
));
568 mon
->clog
->info() << "MDS daemon '" << m
->get_name() << "' restarted";
569 fail_mds_gid(existing
);
573 assert(mon
->osdmon()->is_writeable());
574 request_proposal(mon
->osdmon());
578 // Add this daemon to the map
579 if (pending_fsmap
.mds_roles
.count(gid
) == 0) {
580 MDSMap::mds_info_t new_info
;
581 new_info
.global_id
= gid
;
582 new_info
.name
= m
->get_name();
583 new_info
.addr
= addr
;
584 new_info
.mds_features
= m
->get_mds_features();
585 new_info
.state
= MDSMap::STATE_STANDBY
;
586 new_info
.state_seq
= seq
;
587 new_info
.standby_for_rank
= m
->get_standby_for_rank();
588 new_info
.standby_for_name
= m
->get_standby_for_name();
589 new_info
.standby_for_fscid
= m
->get_standby_for_fscid();
590 new_info
.standby_replay
= m
->get_standby_replay();
591 pending_fsmap
.insert(new_info
);
594 // Resolve standby_for_name to a rank
595 const MDSMap::mds_info_t
&info
= pending_fsmap
.get_info_gid(gid
);
596 if (!info
.standby_for_name
.empty()) {
597 const MDSMap::mds_info_t
*leaderinfo
= fsmap
.find_by_name(
598 info
.standby_for_name
);
599 if (leaderinfo
&& (leaderinfo
->rank
>= 0)) {
600 auto fscid
= pending_fsmap
.mds_roles
.at(leaderinfo
->global_id
);
601 auto fs
= pending_fsmap
.get_filesystem(fscid
);
602 bool followable
= fs
->mds_map
.is_followable(leaderinfo
->rank
);
604 pending_fsmap
.modify_daemon(gid
, [fscid
, leaderinfo
, followable
](
605 MDSMap::mds_info_t
*info
) {
606 info
->standby_for_rank
= leaderinfo
->rank
;
607 info
->standby_for_fscid
= fscid
;
612 // initialize the beacon timer
613 last_beacon
[gid
].stamp
= ceph_clock_now();
614 last_beacon
[gid
].seq
= seq
;
617 if (!pending_fsmap
.compat
.writeable(m
->get_compat())) {
618 dout(10) << " fsmap " << pending_fsmap
.compat
619 << " can't write to new mds' " << m
->get_compat()
620 << ", updating fsmap and killing old mds's"
622 pending_fsmap
.update_compat(m
->get_compat());
625 update_metadata(m
->get_global_id(), m
->get_sys_info());
628 const MDSMap::mds_info_t
&info
= pending_fsmap
.get_info_gid(gid
);
629 // Old MDS daemons don't mention that they're standby replay until
630 // after they've sent their boot beacon, so update this field.
631 if (info
.standby_replay
!= m
->get_standby_replay()) {
632 pending_fsmap
.modify_daemon(info
.global_id
, [&m
](
633 MDSMap::mds_info_t
*i
)
635 i
->standby_replay
= m
->get_standby_replay();
639 if (info
.state
== MDSMap::STATE_STOPPING
&& state
!= MDSMap::STATE_STOPPED
) {
640 // we can't transition to any other states from STOPPING
641 dout(0) << "got beacon for MDS in STATE_STOPPING, ignoring requested state change"
648 dout(10) << "prepare_beacon clearing laggy flag on " << addr
<< dendl
;
649 pending_fsmap
.modify_daemon(info
.global_id
, [](MDSMap::mds_info_t
*info
)
656 dout(10) << "prepare_beacon mds." << info
.rank
657 << " " << ceph_mds_state_name(info
.state
)
658 << " -> " << ceph_mds_state_name(state
)
659 << " standby_for_rank=" << m
->get_standby_for_rank()
661 if (state
== MDSMap::STATE_STOPPED
) {
662 auto erased
= pending_fsmap
.stop(gid
);
663 erased
.push_back(gid
);
665 for (const auto &erased_gid
: erased
) {
666 last_beacon
.erase(erased_gid
);
667 if (pending_daemon_health
.count(erased_gid
)) {
668 pending_daemon_health
.erase(erased_gid
);
669 pending_daemon_health_rm
.insert(erased_gid
);
672 } else if (state
== MDSMap::STATE_DAMAGED
) {
673 if (!mon
->osdmon()->is_writeable()) {
674 dout(4) << __func__
<< ": DAMAGED from rank " << info
.rank
675 << " waiting for osdmon writeable to blacklist it" << dendl
;
676 mon
->osdmon()->wait_for_writeable(op
, new C_RetryMessage(this, op
));
680 // Record this MDS rank as damaged, so that other daemons
681 // won't try to run it.
682 dout(4) << __func__
<< ": marking rank "
683 << info
.rank
<< " damaged" << dendl
;
685 utime_t until
= ceph_clock_now();
686 until
+= g_conf
->mds_blacklist_interval
;
687 const auto blacklist_epoch
= mon
->osdmon()->blacklist(info
.addr
, until
);
688 request_proposal(mon
->osdmon());
689 pending_fsmap
.damaged(gid
, blacklist_epoch
);
690 last_beacon
.erase(gid
);
692 // Respond to MDS, so that it knows it can continue to shut down
695 mon
->monmap
->fsid
, m
->get_global_id(),
696 m
->get_name(), fsmap
.get_epoch(), state
, seq
,
697 CEPH_FEATURES_SUPPORTED_DEFAULT
));
698 } else if (state
== MDSMap::STATE_DNE
) {
699 if (!mon
->osdmon()->is_writeable()) {
700 dout(4) << __func__
<< ": DNE from rank " << info
.rank
701 << " waiting for osdmon writeable to blacklist it" << dendl
;
702 mon
->osdmon()->wait_for_writeable(op
, new C_RetryMessage(this, op
));
707 assert(mon
->osdmon()->is_writeable());
708 request_proposal(mon
->osdmon());
710 // Respond to MDS, so that it knows it can continue to shut down
713 mon
->monmap
->fsid
, m
->get_global_id(),
714 m
->get_name(), fsmap
.get_epoch(), state
, seq
,
715 CEPH_FEATURES_SUPPORTED_DEFAULT
));
716 } else if (info
.state
== MDSMap::STATE_STANDBY
&& state
!= info
.state
) {
717 // Standby daemons should never modify their own
718 // state. Reject any attempts to do so.
719 derr
<< "standby " << gid
<< " attempted to change state to "
720 << ceph_mds_state_name(state
) << ", rejecting" << dendl
;
722 } else if (info
.state
!= MDSMap::STATE_STANDBY
&& state
!= info
.state
&&
723 !MDSMap::state_transition_valid(info
.state
, state
)) {
724 // Validate state transitions for daemons that hold a rank
725 derr
<< "daemon " << gid
<< " (rank " << info
.rank
<< ") "
726 << "reported invalid state transition "
727 << ceph_mds_state_name(info
.state
) << " -> "
728 << ceph_mds_state_name(state
) << dendl
;
731 // Made it through special cases and validations, record the
732 // daemon's reported state to the FSMap.
733 pending_fsmap
.modify_daemon(gid
, [state
, seq
](MDSMap::mds_info_t
*info
) {
735 info
->state_seq
= seq
;
740 dout(7) << "prepare_beacon pending map now:" << dendl
;
741 print_map(pending_fsmap
);
743 wait_for_finished_proposal(op
, new FunctionContext([op
, this](int r
){
745 _updated(op
); // success
746 else if (r
== -ECANCELED
) {
749 dispatch(op
); // try again
756 bool MDSMonitor::prepare_offload_targets(MonOpRequestRef op
)
758 op
->mark_mdsmon_event(__func__
);
759 MMDSLoadTargets
*m
= static_cast<MMDSLoadTargets
*>(op
->get_req());
760 mds_gid_t gid
= m
->global_id
;
761 if (pending_fsmap
.gid_has_rank(gid
)) {
762 dout(10) << "prepare_offload_targets " << gid
<< " " << m
->targets
<< dendl
;
763 pending_fsmap
.update_export_targets(gid
, m
->targets
);
765 dout(10) << "prepare_offload_targets " << gid
<< " not in map" << dendl
;
770 bool MDSMonitor::should_propose(double& delay
)
772 // delegate to PaxosService to assess whether we should propose
773 return PaxosService::should_propose(delay
);
776 void MDSMonitor::_updated(MonOpRequestRef op
)
778 op
->mark_mdsmon_event(__func__
);
779 MMDSBeacon
*m
= static_cast<MMDSBeacon
*>(op
->get_req());
780 dout(10) << "_updated " << m
->get_orig_source() << " " << *m
<< dendl
;
781 mon
->clog
->info() << m
->get_orig_source_inst() << " "
782 << ceph_mds_state_name(m
->get_state());
784 if (m
->get_state() == MDSMap::STATE_STOPPED
) {
785 // send the map manually (they're out of the map, so they won't get it automatic)
787 null_map
.epoch
= fsmap
.epoch
;
788 null_map
.compat
= fsmap
.compat
;
789 mon
->send_reply(op
, new MMDSMap(mon
->monmap
->fsid
, &null_map
));
791 mon
->send_reply(op
, new MMDSBeacon(mon
->monmap
->fsid
,
797 CEPH_FEATURES_SUPPORTED_DEFAULT
));
801 void MDSMonitor::on_active()
806 if (mon
->is_leader()) {
807 mon
->clog
->debug() << "fsmap " << fsmap
;
811 void MDSMonitor::get_health(list
<pair
<health_status_t
, string
> >& summary
,
812 list
<pair
<health_status_t
, string
> > *detail
,
813 CephContext
* cct
) const
815 fsmap
.get_health(summary
, detail
);
817 // For each MDS GID...
818 const auto info_map
= fsmap
.get_mds_info();
819 for (const auto &i
: info_map
) {
820 const auto &gid
= i
.first
;
821 const auto &info
= i
.second
;
825 mon
->store
->get(MDS_HEALTH_PREFIX
, stringify(gid
), bl
);
827 derr
<< "Missing health data for MDS " << gid
<< dendl
;
831 bufferlist::iterator bl_i
= bl
.begin();
834 for (const auto &metric
: health
.metrics
) {
835 int const rank
= info
.rank
;
836 std::ostringstream message
;
837 message
<< "mds" << rank
<< ": " << metric
.message
;
838 summary
.push_back(std::make_pair(metric
.sev
, message
.str()));
841 // There is no way for us to clealy associate detail entries with summary entries (#7192), so
842 // we duplicate the summary message in the detail string and tag the metadata on.
843 std::ostringstream detail_message
;
844 detail_message
<< message
.str();
845 if (metric
.metadata
.size()) {
846 detail_message
<< "(";
847 auto k
= metric
.metadata
.begin();
848 while (k
!= metric
.metadata
.end()) {
849 detail_message
<< k
->first
<< ": " << k
->second
;
850 if (boost::next(k
) != metric
.metadata
.end()) {
851 detail_message
<< ", ";
855 detail_message
<< ")";
857 detail
->push_back(std::make_pair(metric
.sev
, detail_message
.str()));
863 void MDSMonitor::dump_info(Formatter
*f
)
865 f
->open_object_section("fsmap");
869 f
->dump_unsigned("mdsmap_first_committed", get_first_committed());
870 f
->dump_unsigned("mdsmap_last_committed", get_last_committed());
873 bool MDSMonitor::preprocess_command(MonOpRequestRef op
)
875 op
->mark_mdsmon_event(__func__
);
876 MMonCommand
*m
= static_cast<MMonCommand
*>(op
->get_req());
881 map
<string
, cmd_vartype
> cmdmap
;
882 if (!cmdmap_from_json(m
->cmd
, &cmdmap
, ss
)) {
883 // ss has reason for failure
884 string rs
= ss
.str();
885 mon
->reply_command(op
, -EINVAL
, rs
, rdata
, get_last_committed());
890 cmd_getval(g_ceph_context
, cmdmap
, "prefix", prefix
);
892 cmd_getval(g_ceph_context
, cmdmap
, "format", format
, string("plain"));
893 boost::scoped_ptr
<Formatter
> f(Formatter::create(format
));
895 MonSession
*session
= m
->get_session();
897 mon
->reply_command(op
, -EACCES
, "access denied", rdata
, get_last_committed());
901 if (prefix
== "mds stat") {
903 f
->open_object_section("mds_stat");
911 } else if (prefix
== "mds dump") {
916 if (cmd_getval(g_ceph_context
, cmdmap
, "epoch", epocharg
)) {
919 int err
= get_version(epoch
, b
);
920 if (err
== -ENOENT
) {
932 const MDSMap
*mdsmap
= nullptr;
934 blank
.epoch
= fsmap
.epoch
;
935 if (fsmap
.legacy_client_fscid
!= FS_CLUSTER_ID_NONE
) {
936 mdsmap
= &(fsmap
.filesystems
[fsmap
.legacy_client_fscid
]->mds_map
);
941 f
->open_object_section("mdsmap");
942 mdsmap
->dump(f
.get());
952 ss
<< "dumped fsmap epoch " << p
->get_epoch();
958 } else if (prefix
== "fs dump") {
963 if (cmd_getval(g_ceph_context
, cmdmap
, "epoch", epocharg
)) {
966 int err
= get_version(epoch
, b
);
967 if (err
== -ENOENT
) {
980 f
->open_object_section("fsmap");
991 ss
<< "dumped fsmap epoch " << p
->get_epoch();
996 } else if (prefix
== "mds metadata") {
998 f
.reset(Formatter::create("json-pretty"));
1001 bool all
= !cmd_getval(g_ceph_context
, cmdmap
, "who", who
);
1002 dout(1) << "all = " << all
<< dendl
;
1005 // Dump all MDSs' metadata
1006 const auto all_info
= fsmap
.get_mds_info();
1008 f
->open_array_section("mds_metadata");
1009 for(const auto &i
: all_info
) {
1010 const auto &info
= i
.second
;
1012 f
->open_object_section("mds");
1013 f
->dump_string("name", info
.name
);
1014 std::ostringstream get_err
;
1015 r
= dump_metadata(info
.name
, f
.get(), get_err
);
1016 if (r
== -EINVAL
|| r
== -ENOENT
) {
1017 // Drop error, list what metadata we do have
1018 dout(1) << get_err
.str() << dendl
;
1020 } else if (r
!= 0) {
1021 derr
<< "Unexpected error reading metadata: " << cpp_strerror(r
)
1023 ss
<< get_err
.str();
1030 // Dump a single daemon's metadata
1031 f
->open_object_section("mds_metadata");
1032 r
= dump_metadata(who
, f
.get(), ss
);
1036 } else if (prefix
== "mds versions") {
1038 f
.reset(Formatter::create("json-pretty"));
1039 count_metadata("ceph_version", f
.get());
1042 } else if (prefix
== "mds count-metadata") {
1044 f
.reset(Formatter::create("json-pretty"));
1046 cmd_getval(g_ceph_context
, cmdmap
, "property", field
);
1047 count_metadata(field
, f
.get());
1050 } else if (prefix
== "mds getmap") {
1054 if (cmd_getval(g_ceph_context
, cmdmap
, "epoch", epocharg
)) {
1056 int err
= get_version(e
, b
);
1057 if (err
== -ENOENT
) {
1064 mm
.encode(rdata
, m
->get_connection()->get_features());
1065 ss
<< "got fsmap epoch " << mm
.get_epoch();
1069 fsmap
.encode(rdata
, m
->get_connection()->get_features());
1070 ss
<< "got fsmap epoch " << fsmap
.get_epoch();
1073 } else if (prefix
== "mds compat show") {
1075 f
->open_object_section("mds_compat");
1076 fsmap
.compat
.dump(f
.get());
1083 } else if (prefix
== "fs get") {
1085 cmd_getval(g_ceph_context
, cmdmap
, "fs_name", fs_name
);
1086 auto fs
= fsmap
.get_filesystem(fs_name
);
1087 if (fs
== nullptr) {
1088 ss
<< "filesystem '" << fs_name
<< "' not found";
1092 f
->open_object_section("filesystem");
1102 } else if (prefix
== "fs ls") {
1104 f
->open_array_section("filesystems");
1106 for (const auto i
: fsmap
.filesystems
) {
1107 const auto fs
= i
.second
;
1108 f
->open_object_section("filesystem");
1110 const MDSMap
&mds_map
= fs
->mds_map
;
1111 f
->dump_string("name", mds_map
.fs_name
);
1112 /* Output both the names and IDs of pools, for use by
1113 * humans and machines respectively */
1114 f
->dump_string("metadata_pool", mon
->osdmon()->osdmap
.get_pool_name(
1115 mds_map
.metadata_pool
));
1116 f
->dump_int("metadata_pool_id", mds_map
.metadata_pool
);
1117 f
->open_array_section("data_pool_ids");
1119 for (auto dpi
= mds_map
.data_pools
.begin();
1120 dpi
!= mds_map
.data_pools
.end(); ++dpi
) {
1121 f
->dump_int("data_pool_id", *dpi
);
1126 f
->open_array_section("data_pools");
1128 for (auto dpi
= mds_map
.data_pools
.begin();
1129 dpi
!= mds_map
.data_pools
.end(); ++dpi
) {
1130 const auto &name
= mon
->osdmon()->osdmap
.get_pool_name(
1132 f
->dump_string("data_pool", name
);
1144 for (const auto i
: fsmap
.filesystems
) {
1145 const auto fs
= i
.second
;
1146 const MDSMap
&mds_map
= fs
->mds_map
;
1147 const string
&md_pool_name
= mon
->osdmon()->osdmap
.get_pool_name(
1148 mds_map
.metadata_pool
);
1150 ds
<< "name: " << mds_map
.fs_name
<< ", metadata pool: "
1151 << md_pool_name
<< ", data pools: [";
1152 for (auto dpi
: mds_map
.data_pools
) {
1153 const string
&pool_name
= mon
->osdmon()->osdmap
.get_pool_name(dpi
);
1154 ds
<< pool_name
<< " ";
1156 ds
<< "]" << std::endl
;
1159 if (fsmap
.filesystems
.empty()) {
1160 ds
<< "No filesystems enabled" << std::endl
;
1170 mon
->reply_command(op
, r
, rs
, rdata
, get_last_committed());
1176 bool MDSMonitor::fail_mds_gid(mds_gid_t gid
)
1178 const MDSMap::mds_info_t info
= pending_fsmap
.get_info_gid(gid
);
1179 dout(10) << "fail_mds_gid " << gid
<< " mds." << info
.name
<< " role " << info
.rank
<< dendl
;
1181 epoch_t blacklist_epoch
= 0;
1182 if (info
.rank
>= 0 && info
.state
!= MDSMap::STATE_STANDBY_REPLAY
) {
1183 utime_t until
= ceph_clock_now();
1184 until
+= g_conf
->mds_blacklist_interval
;
1185 blacklist_epoch
= mon
->osdmon()->blacklist(info
.addr
, until
);
1188 pending_fsmap
.erase(gid
, blacklist_epoch
);
1189 last_beacon
.erase(gid
);
1190 if (pending_daemon_health
.count(gid
)) {
1191 pending_daemon_health
.erase(gid
);
1192 pending_daemon_health_rm
.insert(gid
);
1195 return blacklist_epoch
!= 0;
1198 mds_gid_t
MDSMonitor::gid_from_arg(const std::string
& arg
, std::ostream
&ss
)
1200 const FSMap
*relevant_fsmap
= mon
->is_leader() ? &pending_fsmap
: &fsmap
;
1202 // Try parsing as a role
1204 std::ostringstream ignore_err
; // Don't spam 'ss' with parse_role errors
1205 int r
= parse_role(arg
, &role
, ignore_err
);
1207 // See if a GID is assigned to this role
1208 auto fs
= relevant_fsmap
->get_filesystem(role
.fscid
);
1209 assert(fs
!= nullptr); // parse_role ensures it exists
1210 if (fs
->mds_map
.is_up(role
.rank
)) {
1211 dout(10) << __func__
<< ": validated rank/GID " << role
1212 << " as a rank" << dendl
;
1213 return fs
->mds_map
.get_mds_info(role
.rank
).global_id
;
1217 // Try parsing as a gid
1219 unsigned long long maybe_gid
= strict_strtoll(arg
.c_str(), 10, &err
);
1221 // Not a role or a GID, try as a daemon name
1222 const MDSMap::mds_info_t
*mds_info
= relevant_fsmap
->find_by_name(arg
);
1224 ss
<< "MDS named '" << arg
1225 << "' does not exist, or is not up";
1226 return MDS_GID_NONE
;
1228 dout(10) << __func__
<< ": resolved MDS name '" << arg
1229 << "' to GID " << mds_info
->global_id
<< dendl
;
1230 return mds_info
->global_id
;
1232 // Not a role, but parses as a an integer, might be a GID
1233 dout(10) << __func__
<< ": treating MDS reference '" << arg
1234 << "' as an integer " << maybe_gid
<< dendl
;
1236 if (relevant_fsmap
->gid_exists(mds_gid_t(maybe_gid
))) {
1237 return mds_gid_t(maybe_gid
);
1241 dout(1) << __func__
<< ": rank/GID " << arg
1242 << " not a existent rank or GID" << dendl
;
1243 return MDS_GID_NONE
;
1246 int MDSMonitor::fail_mds(std::ostream
&ss
, const std::string
&arg
)
1248 mds_gid_t gid
= gid_from_arg(arg
, ss
);
1249 if (gid
== MDS_GID_NONE
) {
1252 if (!mon
->osdmon()->is_writeable()) {
1256 ss
<< "failed mds gid " << gid
;
1257 assert(mon
->osdmon()->is_writeable());
1258 request_proposal(mon
->osdmon());
1262 bool MDSMonitor::prepare_command(MonOpRequestRef op
)
1264 op
->mark_mdsmon_event(__func__
);
1265 MMonCommand
*m
= static_cast<MMonCommand
*>(op
->get_req());
1270 map
<string
, cmd_vartype
> cmdmap
;
1271 if (!cmdmap_from_json(m
->cmd
, &cmdmap
, ss
)) {
1272 string rs
= ss
.str();
1273 mon
->reply_command(op
, -EINVAL
, rs
, rdata
, get_last_committed());
1278 cmd_getval(g_ceph_context
, cmdmap
, "prefix", prefix
);
1280 /* Refuse access if message not associated with a valid session */
1281 MonSession
*session
= m
->get_session();
1283 mon
->reply_command(op
, -EACCES
, "access denied", rdata
, get_last_committed());
1287 for (auto h
: handlers
) {
1288 if (h
->can_handle(prefix
)) {
1289 r
= h
->handle(mon
, pending_fsmap
, op
, cmdmap
, ss
);
1291 // message has been enqueued for retry; return.
1292 dout(4) << __func__
<< " enqueue for retry by prepare_command" << dendl
;
1296 // On successful updates, print the updated map
1297 print_map(pending_fsmap
);
1299 // Successful or not, we're done: respond.
1305 r
= filesystem_command(op
, prefix
, cmdmap
, ss
);
1308 } else if (r
== -EAGAIN
) {
1309 // Do not reply, the message has been enqueued for retry
1310 dout(4) << __func__
<< " enqueue for retry by filesystem_command" << dendl
;
1312 } else if (r
!= -ENOSYS
) {
1316 // Only handle legacy commands if there is a filesystem configured
1317 if (pending_fsmap
.legacy_client_fscid
== FS_CLUSTER_ID_NONE
) {
1318 if (pending_fsmap
.filesystems
.size() == 0) {
1319 ss
<< "No filesystem configured: use `ceph fs new` to create a filesystem";
1321 ss
<< "No filesystem set for use with legacy commands";
1327 r
= legacy_filesystem_command(op
, prefix
, cmdmap
, ss
);
1329 if (r
== -ENOSYS
&& ss
.str().empty()) {
1330 ss
<< "unrecognized command";
1334 dout(4) << __func__
<< " done, r=" << r
<< dendl
;
1335 /* Compose response */
1340 // success.. delay reply
1341 wait_for_finished_proposal(op
, new Monitor::C_Command(mon
, op
, r
, rs
,
1342 get_last_committed() + 1));
1345 // reply immediately
1346 mon
->reply_command(op
, r
, rs
, rdata
, get_last_committed());
1353 * Given one of the following forms:
1358 * Parse into a mds_role_t. The rank-only form is only valid
1359 * if legacy_client_ns is set.
1361 int MDSMonitor::parse_role(
1362 const std::string
&role_str
,
1366 const FSMap
*relevant_fsmap
= &fsmap
;
1367 if (mon
->is_leader()) {
1368 relevant_fsmap
= &pending_fsmap
;
1370 return relevant_fsmap
->parse_role(role_str
, role
, ss
);
1373 int MDSMonitor::filesystem_command(
1375 std::string
const &prefix
,
1376 map
<string
, cmd_vartype
> &cmdmap
,
1377 std::stringstream
&ss
)
1379 dout(4) << __func__
<< " prefix='" << prefix
<< "'" << dendl
;
1380 op
->mark_mdsmon_event(__func__
);
1383 cmd_getval(g_ceph_context
, cmdmap
, "who", whostr
);
1385 if (prefix
== "mds stop" ||
1386 prefix
== "mds deactivate") {
1389 r
= parse_role(whostr
, &role
, ss
);
1393 auto fs
= pending_fsmap
.get_filesystem(role
.fscid
);
1395 if (!fs
->mds_map
.is_active(role
.rank
)) {
1397 ss
<< "mds." << role
<< " not active ("
1398 << ceph_mds_state_name(fs
->mds_map
.get_state(role
.rank
)) << ")";
1399 } else if (fs
->mds_map
.get_root() == role
.rank
||
1400 fs
->mds_map
.get_tableserver() == role
.rank
) {
1402 ss
<< "can't tell the root (" << fs
->mds_map
.get_root()
1403 << ") or tableserver (" << fs
->mds_map
.get_tableserver()
1404 << ") to deactivate";
1405 } else if (role
.rank
!= fs
->mds_map
.get_last_in_mds()) {
1407 ss
<< "mds." << role
<< " doesn't have the max rank ("
1408 << fs
->mds_map
.get_last_in_mds() << ")";
1409 } else if (fs
->mds_map
.get_num_in_mds() <= size_t(fs
->mds_map
.get_max_mds())) {
1411 ss
<< "must decrease max_mds or else MDS will immediately reactivate";
1414 mds_gid_t gid
= fs
->mds_map
.up
.at(role
.rank
);
1415 ss
<< "telling mds." << role
<< " "
1416 << pending_fsmap
.get_info_gid(gid
).addr
<< " to deactivate";
1418 pending_fsmap
.modify_daemon(gid
, [](MDSMap::mds_info_t
*info
) {
1419 info
->state
= MDSMap::STATE_STOPPING
;
1422 } else if (prefix
== "mds set_state") {
1424 if (!cmd_getval(g_ceph_context
, cmdmap
, "gid", gid
)) {
1425 ss
<< "error parsing 'gid' value '"
1426 << cmd_vartype_stringify(cmdmap
["gid"]) << "'";
1429 MDSMap::DaemonState state
;
1430 if (!cmd_getval(g_ceph_context
, cmdmap
, "state", state
)) {
1431 ss
<< "error parsing 'state' string value '"
1432 << cmd_vartype_stringify(cmdmap
["state"]) << "'";
1435 if (pending_fsmap
.gid_exists(gid
)) {
1436 pending_fsmap
.modify_daemon(gid
, [state
](MDSMap::mds_info_t
*info
) {
1437 info
->state
= state
;
1439 ss
<< "set mds gid " << gid
<< " to state " << state
<< " "
1440 << ceph_mds_state_name(state
);
1443 } else if (prefix
== "mds fail") {
1445 cmd_getval(g_ceph_context
, cmdmap
, "who", who
);
1446 r
= fail_mds(ss
, who
);
1447 if (r
< 0 && r
== -EAGAIN
) {
1448 mon
->osdmon()->wait_for_writeable(op
, new C_RetryMessage(this, op
));
1449 return -EAGAIN
; // don't propose yet; wait for message to be retried
1451 } else if (prefix
== "mds rm") {
1453 if (!cmd_getval(g_ceph_context
, cmdmap
, "gid", gid
)) {
1454 ss
<< "error parsing 'gid' value '"
1455 << cmd_vartype_stringify(cmdmap
["gid"]) << "'";
1458 if (!pending_fsmap
.gid_exists(gid
)) {
1459 ss
<< "mds gid " << gid
<< " dne";
1462 MDSMap::DaemonState state
= pending_fsmap
.get_info_gid(gid
).state
;
1464 ss
<< "cannot remove active mds." << pending_fsmap
.get_info_gid(gid
).name
1465 << " rank " << pending_fsmap
.get_info_gid(gid
).rank
;
1468 pending_fsmap
.erase(gid
, {});
1469 ss
<< "removed mds gid " << gid
;
1473 } else if (prefix
== "mds rmfailed") {
1475 if (!cmd_getval(g_ceph_context
, cmdmap
, "confirm", confirm
) ||
1476 confirm
!= "--yes-i-really-mean-it") {
1477 ss
<< "WARNING: this can make your filesystem inaccessible! "
1478 "Add --yes-i-really-mean-it if you are sure you wish to continue.";
1482 std::string role_str
;
1483 cmd_getval(g_ceph_context
, cmdmap
, "who", role_str
);
1485 int r
= parse_role(role_str
, &role
, ss
);
1487 ss
<< "invalid role '" << role_str
<< "'";
1491 pending_fsmap
.modify_filesystem(
1493 [role
](std::shared_ptr
<Filesystem
> fs
)
1495 fs
->mds_map
.failed
.erase(role
.rank
);
1498 ss
<< "removed failed mds." << role
;
1500 } else if (prefix
== "mds compat rm_compat") {
1502 if (!cmd_getval(g_ceph_context
, cmdmap
, "feature", f
)) {
1503 ss
<< "error parsing feature value '"
1504 << cmd_vartype_stringify(cmdmap
["feature"]) << "'";
1507 if (pending_fsmap
.compat
.compat
.contains(f
)) {
1508 ss
<< "removing compat feature " << f
;
1509 CompatSet modified
= pending_fsmap
.compat
;
1510 modified
.compat
.remove(f
);
1511 pending_fsmap
.update_compat(modified
);
1513 ss
<< "compat feature " << f
<< " not present in " << pending_fsmap
.compat
;
1516 } else if (prefix
== "mds compat rm_incompat") {
1518 if (!cmd_getval(g_ceph_context
, cmdmap
, "feature", f
)) {
1519 ss
<< "error parsing feature value '"
1520 << cmd_vartype_stringify(cmdmap
["feature"]) << "'";
1523 if (pending_fsmap
.compat
.incompat
.contains(f
)) {
1524 ss
<< "removing incompat feature " << f
;
1525 CompatSet modified
= pending_fsmap
.compat
;
1526 modified
.incompat
.remove(f
);
1527 pending_fsmap
.update_compat(modified
);
1529 ss
<< "incompat feature " << f
<< " not present in " << pending_fsmap
.compat
;
1532 } else if (prefix
== "mds repaired") {
1533 std::string role_str
;
1534 cmd_getval(g_ceph_context
, cmdmap
, "rank", role_str
);
1536 r
= parse_role(role_str
, &role
, ss
);
1541 bool modified
= pending_fsmap
.undamaged(role
.fscid
, role
.rank
);
1543 dout(4) << "repaired: restoring rank " << role
<< dendl
;
1545 dout(4) << "repaired: no-op on rank " << role
<< dendl
;
1557 * Helper to legacy_filesystem_command
1559 void MDSMonitor::modify_legacy_filesystem(
1560 std::function
<void(std::shared_ptr
<Filesystem
> )> fn
)
1562 pending_fsmap
.modify_filesystem(
1563 pending_fsmap
.legacy_client_fscid
,
1571 * Handle a command that affects the filesystem (i.e. a filesystem
1572 * must exist for the command to act upon).
1574 * @retval 0 Command was successfully handled and has side effects
1575 * @retval -EAGAIN Messages has been requeued for retry
1576 * @retval -ENOSYS Unknown command
1577 * @retval < 0 An error has occurred; **ss** may have been set.
1579 int MDSMonitor::legacy_filesystem_command(
1581 std::string
const &prefix
,
1582 map
<string
, cmd_vartype
> &cmdmap
,
1583 std::stringstream
&ss
)
1585 dout(4) << __func__
<< " prefix='" << prefix
<< "'" << dendl
;
1586 op
->mark_mdsmon_event(__func__
);
1589 cmd_getval(g_ceph_context
, cmdmap
, "who", whostr
);
1591 assert (pending_fsmap
.legacy_client_fscid
!= FS_CLUSTER_ID_NONE
);
1593 if (prefix
== "mds set_max_mds") {
1594 // NOTE: deprecated by "fs set max_mds"
1596 if (!cmd_getval(g_ceph_context
, cmdmap
, "maxmds", maxmds
) || maxmds
<= 0) {
1600 const MDSMap
& mdsmap
=
1601 pending_fsmap
.filesystems
.at(pending_fsmap
.legacy_client_fscid
)->mds_map
;
1603 if (!mdsmap
.allows_multimds() &&
1604 maxmds
> mdsmap
.get_max_mds() &&
1606 ss
<< "multi-MDS clusters are not enabled; set 'allow_multimds' to enable";
1610 if (maxmds
> MAX_MDS
) {
1611 ss
<< "may not have more than " << MAX_MDS
<< " MDS ranks";
1615 modify_legacy_filesystem(
1616 [maxmds
](std::shared_ptr
<Filesystem
> fs
)
1618 fs
->mds_map
.set_max_mds(maxmds
);
1622 ss
<< "max_mds = " << maxmds
;
1623 } else if (prefix
== "mds cluster_down") {
1624 // NOTE: deprecated by "fs set cluster_down"
1625 modify_legacy_filesystem(
1626 [](std::shared_ptr
<Filesystem
> fs
)
1628 fs
->mds_map
.set_flag(CEPH_MDSMAP_DOWN
);
1630 ss
<< "marked fsmap DOWN";
1632 } else if (prefix
== "mds cluster_up") {
1633 // NOTE: deprecated by "fs set cluster_up"
1634 modify_legacy_filesystem(
1635 [](std::shared_ptr
<Filesystem
> fs
)
1637 fs
->mds_map
.clear_flag(CEPH_MDSMAP_DOWN
);
1639 ss
<< "unmarked fsmap DOWN";
1649 void MDSMonitor::check_subs()
1651 std::list
<std::string
> types
;
1653 // Subscriptions may be to "mdsmap" (MDS and legacy clients),
1654 // "mdsmap.<namespace>", or to "fsmap" for the full state of all
1655 // filesystems. Build a list of all the types we service
1656 // subscriptions for.
1657 types
.push_back("fsmap");
1658 types
.push_back("fsmap.user");
1659 types
.push_back("mdsmap");
1660 for (const auto &i
: fsmap
.filesystems
) {
1661 auto fscid
= i
.first
;
1662 std::ostringstream oss
;
1663 oss
<< "mdsmap." << fscid
;
1664 types
.push_back(oss
.str());
1667 for (const auto &type
: types
) {
1668 if (mon
->session_map
.subs
.count(type
) == 0)
1670 xlist
<Subscription
*>::iterator p
= mon
->session_map
.subs
[type
]->begin();
1672 Subscription
*sub
= *p
;
1680 void MDSMonitor::check_sub(Subscription
*sub
)
1682 dout(20) << __func__
<< ": " << sub
->type
<< dendl
;
1684 if (sub
->type
== "fsmap") {
1685 if (sub
->next
<= fsmap
.get_epoch()) {
1686 sub
->session
->con
->send_message(new MFSMap(mon
->monmap
->fsid
, fsmap
));
1688 mon
->session_map
.remove_sub(sub
);
1690 sub
->next
= fsmap
.get_epoch() + 1;
1693 } else if (sub
->type
== "fsmap.user") {
1694 if (sub
->next
<= fsmap
.get_epoch()) {
1696 fsmap_u
.epoch
= fsmap
.get_epoch();
1697 fsmap_u
.legacy_client_fscid
= fsmap
.legacy_client_fscid
;
1698 for (auto p
= fsmap
.filesystems
.begin();
1699 p
!= fsmap
.filesystems
.end();
1701 FSMapUser::fs_info_t
& fs_info
= fsmap_u
.filesystems
[p
->first
];
1702 fs_info
.cid
= p
->first
;
1703 fs_info
.name
= p
->second
->mds_map
.fs_name
;
1705 sub
->session
->con
->send_message(new MFSMapUser(mon
->monmap
->fsid
, fsmap_u
));
1707 mon
->session_map
.remove_sub(sub
);
1709 sub
->next
= fsmap
.get_epoch() + 1;
1712 } else if (sub
->type
.compare(0, 6, "mdsmap") == 0) {
1713 if (sub
->next
> fsmap
.get_epoch()) {
1717 const bool is_mds
= sub
->session
->inst
.name
.is_mds();
1718 mds_gid_t mds_gid
= MDS_GID_NONE
;
1719 fs_cluster_id_t fscid
= FS_CLUSTER_ID_NONE
;
1721 // What (if any) namespace are you assigned to?
1722 auto mds_info
= fsmap
.get_mds_info();
1723 for (const auto &i
: mds_info
) {
1724 if (i
.second
.addr
== sub
->session
->inst
.addr
) {
1726 fscid
= fsmap
.mds_roles
.at(mds_gid
);
1730 // You're a client. Did you request a particular
1732 if (sub
->type
.find("mdsmap.") == 0) {
1733 auto namespace_id_str
= sub
->type
.substr(std::string("mdsmap.").size());
1734 dout(10) << __func__
<< ": namespace_id " << namespace_id_str
<< dendl
;
1736 fscid
= strict_strtoll(namespace_id_str
.c_str(), 10, &err
);
1738 // Client asked for a non-existent namespace, send them nothing
1739 dout(1) << "Invalid client subscription '" << sub
->type
1743 if (fsmap
.filesystems
.count(fscid
) == 0) {
1744 // Client asked for a non-existent namespace, send them nothing
1745 // TODO: something more graceful for when a client has a filesystem
1746 // mounted, and the fileysstem is deleted. Add a "shut down you fool"
1748 dout(1) << "Client subscribed to non-existent namespace '" <<
1749 fscid
<< "'" << dendl
;
1753 // Unqualified request for "mdsmap": give it the one marked
1754 // for use by legacy clients.
1755 if (fsmap
.legacy_client_fscid
!= FS_CLUSTER_ID_NONE
) {
1756 fscid
= fsmap
.legacy_client_fscid
;
1758 dout(1) << "Client subscribed for legacy filesystem but "
1759 "none is configured" << dendl
;
1764 dout(10) << __func__
<< ": is_mds=" << is_mds
<< ", fscid= " << fscid
<< dendl
;
1766 // Work out the effective latest epoch
1767 MDSMap
*mds_map
= nullptr;
1769 null_map
.compat
= fsmap
.compat
;
1770 if (fscid
== FS_CLUSTER_ID_NONE
) {
1771 // For a client, we should have already dropped out
1774 if (fsmap
.standby_daemons
.count(mds_gid
)) {
1775 // For an MDS, we need to feed it an MDSMap with its own state in
1776 null_map
.mds_info
[mds_gid
] = fsmap
.standby_daemons
[mds_gid
];
1777 null_map
.epoch
= fsmap
.standby_epochs
[mds_gid
];
1779 null_map
.epoch
= fsmap
.epoch
;
1781 mds_map
= &null_map
;
1783 // Check the effective epoch
1784 mds_map
= &(fsmap
.filesystems
.at(fscid
)->mds_map
);
1787 assert(mds_map
!= nullptr);
1788 dout(10) << __func__
<< " selected MDS map epoch " <<
1789 mds_map
->epoch
<< " for namespace " << fscid
<< " for subscriber "
1790 << sub
->session
->inst
.name
<< " who wants epoch " << sub
->next
<< dendl
;
1792 if (sub
->next
> mds_map
->epoch
) {
1795 auto msg
= new MMDSMap(mon
->monmap
->fsid
, mds_map
);
1797 sub
->session
->con
->send_message(msg
);
1799 mon
->session_map
.remove_sub(sub
);
1801 sub
->next
= mds_map
->get_epoch() + 1;
1807 void MDSMonitor::update_metadata(mds_gid_t gid
,
1808 const map
<string
, string
>& metadata
)
1810 if (metadata
.empty()) {
1813 pending_metadata
[gid
] = metadata
;
1815 MonitorDBStore::TransactionRef t
= paxos
->get_pending_transaction();
1817 ::encode(pending_metadata
, bl
);
1818 t
->put(MDS_METADATA_PREFIX
, "last_metadata", bl
);
1819 paxos
->trigger_propose();
1822 void MDSMonitor::remove_from_metadata(MonitorDBStore::TransactionRef t
)
1824 bool update
= false;
1825 for (map
<mds_gid_t
, Metadata
>::iterator i
= pending_metadata
.begin();
1826 i
!= pending_metadata
.end(); ) {
1827 if (!pending_fsmap
.gid_exists(i
->first
)) {
1828 pending_metadata
.erase(i
++);
1837 ::encode(pending_metadata
, bl
);
1838 t
->put(MDS_METADATA_PREFIX
, "last_metadata", bl
);
1841 int MDSMonitor::load_metadata(map
<mds_gid_t
, Metadata
>& m
)
1844 int r
= mon
->store
->get(MDS_METADATA_PREFIX
, "last_metadata", bl
);
1846 dout(1) << "Unable to load 'last_metadata'" << dendl
;
1850 bufferlist::iterator it
= bl
.begin();
1855 void MDSMonitor::count_metadata(const string
& field
, Formatter
*f
)
1857 map
<string
,int> by_val
;
1858 map
<mds_gid_t
,Metadata
> meta
;
1859 load_metadata(meta
);
1860 for (auto& p
: meta
) {
1861 auto q
= p
.second
.find(field
);
1862 if (q
== p
.second
.end()) {
1863 by_val
["unknown"]++;
1865 by_val
[q
->second
]++;
1868 f
->open_object_section(field
.c_str());
1869 for (auto& p
: by_val
) {
1870 f
->dump_int(p
.first
.c_str(), p
.second
);
1875 int MDSMonitor::dump_metadata(const std::string
&who
, Formatter
*f
, ostream
& err
)
1879 mds_gid_t gid
= gid_from_arg(who
, err
);
1880 if (gid
== MDS_GID_NONE
) {
1884 map
<mds_gid_t
, Metadata
> metadata
;
1885 if (int r
= load_metadata(metadata
)) {
1886 err
<< "Unable to load 'last_metadata'";
1890 if (!metadata
.count(gid
)) {
1893 const Metadata
& m
= metadata
[gid
];
1894 for (Metadata::const_iterator p
= m
.begin(); p
!= m
.end(); ++p
) {
1895 f
->dump_string(p
->first
.c_str(), p
->second
);
1900 int MDSMonitor::print_nodes(Formatter
*f
)
1904 map
<mds_gid_t
, Metadata
> metadata
;
1905 if (int r
= load_metadata(metadata
)) {
1909 map
<string
, list
<int> > mdses
; // hostname => rank
1910 for (map
<mds_gid_t
, Metadata
>::iterator it
= metadata
.begin();
1911 it
!= metadata
.end(); ++it
) {
1912 const Metadata
& m
= it
->second
;
1913 Metadata::const_iterator hostname
= m
.find("hostname");
1914 if (hostname
== m
.end()) {
1915 // not likely though
1918 const mds_gid_t gid
= it
->first
;
1919 if (!fsmap
.gid_exists(gid
)) {
1920 dout(5) << __func__
<< ": GID " << gid
<< " not existent" << dendl
;
1923 const MDSMap::mds_info_t
& mds_info
= fsmap
.get_info_gid(gid
);
1924 // FIXME: include filesystem name with rank here
1925 mdses
[hostname
->second
].push_back(mds_info
.rank
);
1928 dump_services(f
, mdses
, "mds");
1933 * If a cluster is undersized (with respect to max_mds), then
1934 * attempt to find daemons to grow it.
1936 bool MDSMonitor::maybe_expand_cluster(std::shared_ptr
<Filesystem
> fs
)
1938 bool do_propose
= false;
1940 if (fs
->mds_map
.test_flag(CEPH_MDSMAP_DOWN
)) {
1944 while (fs
->mds_map
.get_num_in_mds() < size_t(fs
->mds_map
.get_max_mds()) &&
1945 !fs
->mds_map
.is_degraded()) {
1946 mds_rank_t mds
= mds_rank_t(0);
1948 while (fs
->mds_map
.is_in(mds
)) {
1951 mds_gid_t newgid
= pending_fsmap
.find_replacement_for({fs
->fscid
, mds
},
1952 name
, g_conf
->mon_force_standby_active
);
1953 if (newgid
== MDS_GID_NONE
) {
1957 dout(1) << "adding standby " << pending_fsmap
.get_info_gid(newgid
).addr
1958 << " as mds." << mds
<< dendl
;
1959 pending_fsmap
.promote(newgid
, fs
, mds
);
1968 * If a daemon is laggy, and a suitable replacement
1969 * is available, fail this daemon (remove from map) and pass its
1970 * role to another daemon.
1972 void MDSMonitor::maybe_replace_gid(mds_gid_t gid
,
1973 const beacon_info_t
&beacon
,
1974 bool *mds_propose
, bool *osd_propose
)
1976 assert(mds_propose
!= nullptr);
1977 assert(osd_propose
!= nullptr);
1979 const MDSMap::mds_info_t info
= pending_fsmap
.get_info_gid(gid
);
1980 const auto fscid
= pending_fsmap
.mds_roles
.at(gid
);
1982 dout(10) << "no beacon from " << gid
<< " " << info
.addr
<< " mds."
1983 << info
.rank
<< "." << info
.inc
1984 << " " << ceph_mds_state_name(info
.state
)
1985 << " since " << beacon
.stamp
<< dendl
;
1987 // We will only take decisive action (replacing/removing a daemon)
1988 // if we have some indicating that some other daemon(s) are successfully
1989 // getting beacons through recently.
1990 utime_t latest_beacon
;
1991 for (const auto & i
: last_beacon
) {
1992 latest_beacon
= MAX(i
.second
.stamp
, latest_beacon
);
1994 const bool may_replace
= latest_beacon
>
1996 MAX(g_conf
->mds_beacon_interval
, g_conf
->mds_beacon_grace
* 0.5));
1999 // and is there a non-laggy standby that can take over for us?
2001 if (info
.rank
>= 0 &&
2002 info
.state
!= MDSMap::STATE_STANDBY
&&
2003 info
.state
!= MDSMap::STATE_STANDBY_REPLAY
&&
2005 !pending_fsmap
.get_filesystem(fscid
)->mds_map
.test_flag(CEPH_MDSMAP_DOWN
) &&
2006 (sgid
= pending_fsmap
.find_replacement_for({fscid
, info
.rank
}, info
.name
,
2007 g_conf
->mon_force_standby_active
)) != MDS_GID_NONE
)
2010 MDSMap::mds_info_t si
= pending_fsmap
.get_info_gid(sgid
);
2011 dout(10) << " replacing " << gid
<< " " << info
.addr
<< " mds."
2012 << info
.rank
<< "." << info
.inc
2013 << " " << ceph_mds_state_name(info
.state
)
2014 << " with " << sgid
<< "/" << si
.name
<< " " << si
.addr
<< dendl
;
2016 mon
->clog
->warn() << "MDS daemon '" << info
.name
<< "'"
2017 << " is not responding, replacing it "
2018 << "as rank " << info
.rank
2019 << " with standby '" << si
.name
<< "'";
2021 // Remember what NS the old one was in
2022 const fs_cluster_id_t fscid
= pending_fsmap
.mds_roles
.at(gid
);
2024 // Remove the old one
2025 *osd_propose
|= fail_mds_gid(gid
);
2027 // Promote the replacement
2028 auto fs
= pending_fsmap
.filesystems
.at(fscid
);
2029 pending_fsmap
.promote(sgid
, fs
, info
.rank
);
2031 *mds_propose
= true;
2032 } else if ((info
.state
== MDSMap::STATE_STANDBY_REPLAY
||
2033 info
.state
== MDSMap::STATE_STANDBY
) && may_replace
) {
2034 dout(10) << " failing and removing " << gid
<< " " << info
.addr
<< " mds." << info
.rank
2035 << "." << info
.inc
<< " " << ceph_mds_state_name(info
.state
)
2037 mon
->clog
->info() << "MDS standby '" << info
.name
2038 << "' is not responding, removing it from the set of "
2041 *mds_propose
= true;
2042 } else if (!info
.laggy()) {
2043 dout(10) << " marking " << gid
<< " " << info
.addr
<< " mds." << info
.rank
<< "." << info
.inc
2044 << " " << ceph_mds_state_name(info
.state
)
2045 << " laggy" << dendl
;
2046 pending_fsmap
.modify_daemon(info
.global_id
, [](MDSMap::mds_info_t
*info
) {
2047 info
->laggy_since
= ceph_clock_now();
2049 *mds_propose
= true;
2053 bool MDSMonitor::maybe_promote_standby(std::shared_ptr
<Filesystem
> fs
)
2055 assert(!fs
->mds_map
.test_flag(CEPH_MDSMAP_DOWN
));
2057 bool do_propose
= false;
2059 // have a standby take over?
2060 set
<mds_rank_t
> failed
;
2061 fs
->mds_map
.get_failed_mds_set(failed
);
2062 if (!failed
.empty()) {
2063 set
<mds_rank_t
>::iterator p
= failed
.begin();
2064 while (p
!= failed
.end()) {
2065 mds_rank_t f
= *p
++;
2066 mds_gid_t sgid
= pending_fsmap
.find_replacement_for({fs
->fscid
, f
}, {},
2067 g_conf
->mon_force_standby_active
);
2069 const MDSMap::mds_info_t si
= pending_fsmap
.get_info_gid(sgid
);
2070 dout(0) << " taking over failed mds." << f
<< " with " << sgid
2071 << "/" << si
.name
<< " " << si
.addr
<< dendl
;
2072 pending_fsmap
.promote(sgid
, fs
, f
);
2077 // There were no failures to replace, so try using any available standbys
2078 // as standby-replay daemons.
2080 // Take a copy of the standby GIDs so that we can iterate over
2081 // them while perhaps-modifying standby_daemons during the loop
2082 // (if we promote anyone they are removed from standby_daemons)
2083 std::vector
<mds_gid_t
> standby_gids
;
2084 for (const auto &j
: pending_fsmap
.standby_daemons
) {
2085 standby_gids
.push_back(j
.first
);
2088 for (const auto &gid
: standby_gids
) {
2089 const auto &info
= pending_fsmap
.standby_daemons
.at(gid
);
2090 assert(info
.state
== MDSMap::STATE_STANDBY
);
2092 if (!info
.standby_replay
) {
2097 * This mds is standby but has no rank assigned.
2098 * See if we can find it somebody to shadow
2100 dout(20) << "gid " << gid
<< " is standby and following nobody" << dendl
;
2102 // standby for someone specific?
2103 if (info
.standby_for_rank
>= 0) {
2104 // The mds_info_t may or may not tell us exactly which filesystem
2105 // the standby_for_rank refers to: lookup via legacy_client_fscid
2106 mds_role_t target_role
= {
2107 info
.standby_for_fscid
== FS_CLUSTER_ID_NONE
?
2108 pending_fsmap
.legacy_client_fscid
: info
.standby_for_fscid
,
2109 info
.standby_for_rank
};
2111 // It is possible that the map contains a standby_for_fscid
2112 // that doesn't correspond to an existing filesystem, especially
2113 // if we loaded from a version with a bug (#17466)
2114 if (info
.standby_for_fscid
!= FS_CLUSTER_ID_NONE
2115 && !pending_fsmap
.filesystem_exists(info
.standby_for_fscid
)) {
2116 derr
<< "gid " << gid
<< " has invalid standby_for_fscid "
2117 << info
.standby_for_fscid
<< dendl
;
2121 // If we managed to resolve a full target role
2122 if (target_role
.fscid
!= FS_CLUSTER_ID_NONE
) {
2123 auto fs
= pending_fsmap
.get_filesystem(target_role
.fscid
);
2124 if (fs
->mds_map
.is_followable(target_role
.rank
)) {
2125 do_propose
|= try_standby_replay(
2128 fs
->mds_map
.get_info(target_role
.rank
));
2136 for (auto fs_i
: pending_fsmap
.filesystems
) {
2137 const MDSMap
&mds_map
= fs_i
.second
->mds_map
;
2138 for (auto mds_i
: mds_map
.mds_info
) {
2139 MDSMap::mds_info_t
&cand_info
= mds_i
.second
;
2140 if (cand_info
.rank
>= 0 && mds_map
.is_followable(cand_info
.rank
)) {
2141 if ((info
.standby_for_name
.length() && info
.standby_for_name
!= cand_info
.name
) ||
2142 info
.standby_for_rank
!= MDS_RANK_NONE
) {
2143 continue; // we're supposed to follow someone else
2146 if (try_standby_replay(info
, *(fs_i
.second
), cand_info
)) {
2160 void MDSMonitor::tick()
2162 // make sure mds's are still alive
2163 // ...if i am an active leader
2164 if (!is_active()) return;
2166 dout(10) << fsmap
<< dendl
;
2168 bool do_propose
= false;
2170 if (!mon
->is_leader()) return;
2172 do_propose
|= pending_fsmap
.check_health();
2174 // expand mds cluster (add new nodes to @in)?
2175 for (auto i
: pending_fsmap
.filesystems
) {
2176 do_propose
|= maybe_expand_cluster(i
.second
);
2179 const auto now
= ceph_clock_now();
2180 if (last_tick
.is_zero()) {
2184 if (now
- last_tick
> (g_conf
->mds_beacon_grace
- g_conf
->mds_beacon_interval
)) {
2185 // This case handles either local slowness (calls being delayed
2186 // for whatever reason) or cluster election slowness (a long gap
2187 // between calls while an election happened)
2188 dout(4) << __func__
<< ": resetting beacon timeouts due to mon delay "
2189 "(slow election?) of " << now
- last_tick
<< " seconds" << dendl
;
2190 for (auto &i
: last_beacon
) {
2191 i
.second
.stamp
= now
;
2197 // check beacon timestamps
2198 utime_t cutoff
= now
;
2199 cutoff
-= g_conf
->mds_beacon_grace
;
2201 // make sure last_beacon is fully populated
2202 for (const auto &p
: pending_fsmap
.mds_roles
) {
2203 auto &gid
= p
.first
;
2204 if (last_beacon
.count(gid
) == 0) {
2205 last_beacon
[gid
].stamp
= now
;
2206 last_beacon
[gid
].seq
= 0;
2210 // If the OSDMap is writeable, we can blacklist things, so we can
2211 // try failing any laggy MDS daemons. Consider each one for failure.
2212 if (mon
->osdmon()->is_writeable()) {
2213 bool propose_osdmap
= false;
2215 map
<mds_gid_t
, beacon_info_t
>::iterator p
= last_beacon
.begin();
2216 while (p
!= last_beacon
.end()) {
2217 mds_gid_t gid
= p
->first
;
2218 auto beacon_info
= p
->second
;
2221 if (!pending_fsmap
.gid_exists(gid
)) {
2223 last_beacon
.erase(gid
);
2227 if (beacon_info
.stamp
< cutoff
) {
2228 maybe_replace_gid(gid
, beacon_info
, &do_propose
, &propose_osdmap
);
2232 if (propose_osdmap
) {
2233 request_proposal(mon
->osdmon());
2237 for (auto i
: pending_fsmap
.filesystems
) {
2239 if (!fs
->mds_map
.test_flag(CEPH_MDSMAP_DOWN
)) {
2240 do_propose
|= maybe_promote_standby(fs
);
2250 * finfo: the would-be follower
2251 * leader_fs: the Filesystem containing the would-be leader
2252 * ainfo: the would-be leader
2254 bool MDSMonitor::try_standby_replay(
2255 const MDSMap::mds_info_t
& finfo
,
2256 const Filesystem
&leader_fs
,
2257 const MDSMap::mds_info_t
& ainfo
)
2259 // someone else already following?
2260 if (leader_fs
.has_standby_replay(ainfo
.global_id
)) {
2261 dout(20) << " mds." << ainfo
.rank
<< " already has a follower" << dendl
;
2264 // Assign the new role to the standby
2265 dout(10) << " setting to follow mds rank " << ainfo
.rank
<< dendl
;
2266 pending_fsmap
.assign_standby_replay(finfo
.global_id
, leader_fs
.fscid
, ainfo
.rank
);
2271 MDSMonitor::MDSMonitor(Monitor
*mn
, Paxos
*p
, string service_name
)
2272 : PaxosService(mn
, p
, service_name
)
2274 handlers
= FileSystemCommandHandler::load();
2277 void MDSMonitor::on_restart()
2279 // Clear out the leader-specific state.
2280 last_tick
= utime_t();
2281 last_beacon
.clear();