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
);
603 pending_fsmap
.modify_daemon(gid
, [fscid
, leaderinfo
](
604 MDSMap::mds_info_t
*info
) {
605 info
->standby_for_rank
= leaderinfo
->rank
;
606 info
->standby_for_fscid
= fscid
;
611 // initialize the beacon timer
612 last_beacon
[gid
].stamp
= ceph_clock_now();
613 last_beacon
[gid
].seq
= seq
;
616 if (!pending_fsmap
.compat
.writeable(m
->get_compat())) {
617 dout(10) << " fsmap " << pending_fsmap
.compat
618 << " can't write to new mds' " << m
->get_compat()
619 << ", updating fsmap and killing old mds's"
621 pending_fsmap
.update_compat(m
->get_compat());
624 update_metadata(m
->get_global_id(), m
->get_sys_info());
627 const MDSMap::mds_info_t
&info
= pending_fsmap
.get_info_gid(gid
);
628 // Old MDS daemons don't mention that they're standby replay until
629 // after they've sent their boot beacon, so update this field.
630 if (info
.standby_replay
!= m
->get_standby_replay()) {
631 pending_fsmap
.modify_daemon(info
.global_id
, [&m
](
632 MDSMap::mds_info_t
*i
)
634 i
->standby_replay
= m
->get_standby_replay();
638 if (info
.state
== MDSMap::STATE_STOPPING
&& state
!= MDSMap::STATE_STOPPED
) {
639 // we can't transition to any other states from STOPPING
640 dout(0) << "got beacon for MDS in STATE_STOPPING, ignoring requested state change"
647 dout(10) << "prepare_beacon clearing laggy flag on " << addr
<< dendl
;
648 pending_fsmap
.modify_daemon(info
.global_id
, [](MDSMap::mds_info_t
*info
)
655 dout(10) << "prepare_beacon mds." << info
.rank
656 << " " << ceph_mds_state_name(info
.state
)
657 << " -> " << ceph_mds_state_name(state
)
658 << " standby_for_rank=" << m
->get_standby_for_rank()
660 if (state
== MDSMap::STATE_STOPPED
) {
661 auto erased
= pending_fsmap
.stop(gid
);
662 erased
.push_back(gid
);
664 for (const auto &erased_gid
: erased
) {
665 last_beacon
.erase(erased_gid
);
666 if (pending_daemon_health
.count(erased_gid
)) {
667 pending_daemon_health
.erase(erased_gid
);
668 pending_daemon_health_rm
.insert(erased_gid
);
671 } else if (state
== MDSMap::STATE_DAMAGED
) {
672 if (!mon
->osdmon()->is_writeable()) {
673 dout(4) << __func__
<< ": DAMAGED from rank " << info
.rank
674 << " waiting for osdmon writeable to blacklist it" << dendl
;
675 mon
->osdmon()->wait_for_writeable(op
, new C_RetryMessage(this, op
));
679 // Record this MDS rank as damaged, so that other daemons
680 // won't try to run it.
681 dout(4) << __func__
<< ": marking rank "
682 << info
.rank
<< " damaged" << dendl
;
684 utime_t until
= ceph_clock_now();
685 until
+= g_conf
->mds_blacklist_interval
;
686 const auto blacklist_epoch
= mon
->osdmon()->blacklist(info
.addr
, until
);
687 request_proposal(mon
->osdmon());
688 pending_fsmap
.damaged(gid
, blacklist_epoch
);
689 last_beacon
.erase(gid
);
691 // Respond to MDS, so that it knows it can continue to shut down
694 mon
->monmap
->fsid
, m
->get_global_id(),
695 m
->get_name(), fsmap
.get_epoch(), state
, seq
,
696 CEPH_FEATURES_SUPPORTED_DEFAULT
));
697 } else if (state
== MDSMap::STATE_DNE
) {
698 if (!mon
->osdmon()->is_writeable()) {
699 dout(4) << __func__
<< ": DNE from rank " << info
.rank
700 << " waiting for osdmon writeable to blacklist it" << dendl
;
701 mon
->osdmon()->wait_for_writeable(op
, new C_RetryMessage(this, op
));
706 assert(mon
->osdmon()->is_writeable());
707 request_proposal(mon
->osdmon());
709 // Respond to MDS, so that it knows it can continue to shut down
712 mon
->monmap
->fsid
, m
->get_global_id(),
713 m
->get_name(), fsmap
.get_epoch(), state
, seq
,
714 CEPH_FEATURES_SUPPORTED_DEFAULT
));
715 } else if (info
.state
== MDSMap::STATE_STANDBY
&& state
!= info
.state
) {
716 // Standby daemons should never modify their own
717 // state. Reject any attempts to do so.
718 derr
<< "standby " << gid
<< " attempted to change state to "
719 << ceph_mds_state_name(state
) << ", rejecting" << dendl
;
721 } else if (info
.state
!= MDSMap::STATE_STANDBY
&& state
!= info
.state
&&
722 !MDSMap::state_transition_valid(info
.state
, state
)) {
723 // Validate state transitions for daemons that hold a rank
724 derr
<< "daemon " << gid
<< " (rank " << info
.rank
<< ") "
725 << "reported invalid state transition "
726 << ceph_mds_state_name(info
.state
) << " -> "
727 << ceph_mds_state_name(state
) << dendl
;
730 // Made it through special cases and validations, record the
731 // daemon's reported state to the FSMap.
732 pending_fsmap
.modify_daemon(gid
, [state
, seq
](MDSMap::mds_info_t
*info
) {
734 info
->state_seq
= seq
;
739 dout(7) << "prepare_beacon pending map now:" << dendl
;
740 print_map(pending_fsmap
);
742 wait_for_finished_proposal(op
, new FunctionContext([op
, this](int r
){
744 _updated(op
); // success
745 else if (r
== -ECANCELED
) {
748 dispatch(op
); // try again
755 bool MDSMonitor::prepare_offload_targets(MonOpRequestRef op
)
757 op
->mark_mdsmon_event(__func__
);
758 MMDSLoadTargets
*m
= static_cast<MMDSLoadTargets
*>(op
->get_req());
759 mds_gid_t gid
= m
->global_id
;
760 if (pending_fsmap
.gid_has_rank(gid
)) {
761 dout(10) << "prepare_offload_targets " << gid
<< " " << m
->targets
<< dendl
;
762 pending_fsmap
.update_export_targets(gid
, m
->targets
);
764 dout(10) << "prepare_offload_targets " << gid
<< " not in map" << dendl
;
769 bool MDSMonitor::should_propose(double& delay
)
771 // delegate to PaxosService to assess whether we should propose
772 return PaxosService::should_propose(delay
);
775 void MDSMonitor::_updated(MonOpRequestRef op
)
777 op
->mark_mdsmon_event(__func__
);
778 MMDSBeacon
*m
= static_cast<MMDSBeacon
*>(op
->get_req());
779 dout(10) << "_updated " << m
->get_orig_source() << " " << *m
<< dendl
;
780 mon
->clog
->info() << m
->get_orig_source_inst() << " "
781 << ceph_mds_state_name(m
->get_state());
783 if (m
->get_state() == MDSMap::STATE_STOPPED
) {
784 // send the map manually (they're out of the map, so they won't get it automatic)
786 null_map
.epoch
= fsmap
.epoch
;
787 null_map
.compat
= fsmap
.compat
;
788 mon
->send_reply(op
, new MMDSMap(mon
->monmap
->fsid
, &null_map
));
790 mon
->send_reply(op
, new MMDSBeacon(mon
->monmap
->fsid
,
796 CEPH_FEATURES_SUPPORTED_DEFAULT
));
800 void MDSMonitor::on_active()
805 if (mon
->is_leader()) {
806 mon
->clog
->debug() << "fsmap " << fsmap
;
810 void MDSMonitor::get_health(list
<pair
<health_status_t
, string
> >& summary
,
811 list
<pair
<health_status_t
, string
> > *detail
,
812 CephContext
* cct
) const
814 fsmap
.get_health(summary
, detail
);
816 // For each MDS GID...
817 const auto info_map
= fsmap
.get_mds_info();
818 for (const auto &i
: info_map
) {
819 const auto &gid
= i
.first
;
820 const auto &info
= i
.second
;
824 mon
->store
->get(MDS_HEALTH_PREFIX
, stringify(gid
), bl
);
826 derr
<< "Missing health data for MDS " << gid
<< dendl
;
830 bufferlist::iterator bl_i
= bl
.begin();
833 for (const auto &metric
: health
.metrics
) {
834 int const rank
= info
.rank
;
835 std::ostringstream message
;
836 message
<< "mds" << rank
<< ": " << metric
.message
;
837 summary
.push_back(std::make_pair(metric
.sev
, message
.str()));
840 // There is no way for us to clealy associate detail entries with summary entries (#7192), so
841 // we duplicate the summary message in the detail string and tag the metadata on.
842 std::ostringstream detail_message
;
843 detail_message
<< message
.str();
844 if (metric
.metadata
.size()) {
845 detail_message
<< "(";
846 auto k
= metric
.metadata
.begin();
847 while (k
!= metric
.metadata
.end()) {
848 detail_message
<< k
->first
<< ": " << k
->second
;
849 if (boost::next(k
) != metric
.metadata
.end()) {
850 detail_message
<< ", ";
854 detail_message
<< ")";
856 detail
->push_back(std::make_pair(metric
.sev
, detail_message
.str()));
862 void MDSMonitor::dump_info(Formatter
*f
)
864 f
->open_object_section("fsmap");
868 f
->dump_unsigned("mdsmap_first_committed", get_first_committed());
869 f
->dump_unsigned("mdsmap_last_committed", get_last_committed());
872 bool MDSMonitor::preprocess_command(MonOpRequestRef op
)
874 op
->mark_mdsmon_event(__func__
);
875 MMonCommand
*m
= static_cast<MMonCommand
*>(op
->get_req());
880 map
<string
, cmd_vartype
> cmdmap
;
881 if (!cmdmap_from_json(m
->cmd
, &cmdmap
, ss
)) {
882 // ss has reason for failure
883 string rs
= ss
.str();
884 mon
->reply_command(op
, -EINVAL
, rs
, rdata
, get_last_committed());
889 cmd_getval(g_ceph_context
, cmdmap
, "prefix", prefix
);
891 cmd_getval(g_ceph_context
, cmdmap
, "format", format
, string("plain"));
892 boost::scoped_ptr
<Formatter
> f(Formatter::create(format
));
894 MonSession
*session
= m
->get_session();
896 mon
->reply_command(op
, -EACCES
, "access denied", rdata
, get_last_committed());
900 if (prefix
== "mds stat") {
902 f
->open_object_section("mds_stat");
910 } else if (prefix
== "mds dump") {
915 if (cmd_getval(g_ceph_context
, cmdmap
, "epoch", epocharg
)) {
918 int err
= get_version(epoch
, b
);
919 if (err
== -ENOENT
) {
931 const MDSMap
*mdsmap
= nullptr;
933 blank
.epoch
= fsmap
.epoch
;
934 if (fsmap
.legacy_client_fscid
!= FS_CLUSTER_ID_NONE
) {
935 mdsmap
= &(fsmap
.filesystems
[fsmap
.legacy_client_fscid
]->mds_map
);
940 f
->open_object_section("mdsmap");
941 mdsmap
->dump(f
.get());
951 ss
<< "dumped fsmap epoch " << p
->get_epoch();
957 } else if (prefix
== "fs dump") {
962 if (cmd_getval(g_ceph_context
, cmdmap
, "epoch", epocharg
)) {
965 int err
= get_version(epoch
, b
);
966 if (err
== -ENOENT
) {
979 f
->open_object_section("fsmap");
990 ss
<< "dumped fsmap epoch " << p
->get_epoch();
995 } else if (prefix
== "mds metadata") {
997 f
.reset(Formatter::create("json-pretty"));
1000 bool all
= !cmd_getval(g_ceph_context
, cmdmap
, "who", who
);
1001 dout(1) << "all = " << all
<< dendl
;
1004 // Dump all MDSs' metadata
1005 const auto all_info
= fsmap
.get_mds_info();
1007 f
->open_array_section("mds_metadata");
1008 for(const auto &i
: all_info
) {
1009 const auto &info
= i
.second
;
1011 f
->open_object_section("mds");
1012 f
->dump_string("name", info
.name
);
1013 std::ostringstream get_err
;
1014 r
= dump_metadata(info
.name
, f
.get(), get_err
);
1015 if (r
== -EINVAL
|| r
== -ENOENT
) {
1016 // Drop error, list what metadata we do have
1017 dout(1) << get_err
.str() << dendl
;
1019 } else if (r
!= 0) {
1020 derr
<< "Unexpected error reading metadata: " << cpp_strerror(r
)
1022 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 bool batched_propose
= false;
1288 for (auto h
: handlers
) {
1289 if (h
->can_handle(prefix
)) {
1290 batched_propose
= h
->batched_propose();
1291 if (batched_propose
) {
1294 r
= h
->handle(mon
, pending_fsmap
, op
, cmdmap
, ss
);
1295 if (batched_propose
) {
1300 // message has been enqueued for retry; return.
1301 dout(4) << __func__
<< " enqueue for retry by prepare_command" << dendl
;
1305 // On successful updates, print the updated map
1306 print_map(pending_fsmap
);
1308 // Successful or not, we're done: respond.
1314 r
= filesystem_command(op
, prefix
, cmdmap
, ss
);
1317 } else if (r
== -EAGAIN
) {
1318 // Do not reply, the message has been enqueued for retry
1319 dout(4) << __func__
<< " enqueue for retry by filesystem_command" << dendl
;
1321 } else if (r
!= -ENOSYS
) {
1325 // Only handle legacy commands if there is a filesystem configured
1326 if (pending_fsmap
.legacy_client_fscid
== FS_CLUSTER_ID_NONE
) {
1327 if (pending_fsmap
.filesystems
.size() == 0) {
1328 ss
<< "No filesystem configured: use `ceph fs new` to create a filesystem";
1330 ss
<< "No filesystem set for use with legacy commands";
1336 r
= legacy_filesystem_command(op
, prefix
, cmdmap
, ss
);
1338 if (r
== -ENOSYS
&& ss
.str().empty()) {
1339 ss
<< "unrecognized command";
1343 dout(4) << __func__
<< " done, r=" << r
<< dendl
;
1344 /* Compose response */
1349 // success.. delay reply
1350 wait_for_finished_proposal(op
, new Monitor::C_Command(mon
, op
, r
, rs
,
1351 get_last_committed() + 1));
1352 if (batched_propose
) {
1353 force_immediate_propose();
1357 // reply immediately
1358 mon
->reply_command(op
, r
, rs
, rdata
, get_last_committed());
1365 * Given one of the following forms:
1370 * Parse into a mds_role_t. The rank-only form is only valid
1371 * if legacy_client_ns is set.
1373 int MDSMonitor::parse_role(
1374 const std::string
&role_str
,
1378 const FSMap
*relevant_fsmap
= &fsmap
;
1379 if (mon
->is_leader()) {
1380 relevant_fsmap
= &pending_fsmap
;
1382 return relevant_fsmap
->parse_role(role_str
, role
, ss
);
1385 int MDSMonitor::filesystem_command(
1387 std::string
const &prefix
,
1388 map
<string
, cmd_vartype
> &cmdmap
,
1389 std::stringstream
&ss
)
1391 dout(4) << __func__
<< " prefix='" << prefix
<< "'" << dendl
;
1392 op
->mark_mdsmon_event(__func__
);
1395 cmd_getval(g_ceph_context
, cmdmap
, "who", whostr
);
1397 if (prefix
== "mds stop" ||
1398 prefix
== "mds deactivate") {
1401 r
= parse_role(whostr
, &role
, ss
);
1405 auto fs
= pending_fsmap
.get_filesystem(role
.fscid
);
1407 if (!fs
->mds_map
.is_active(role
.rank
)) {
1409 ss
<< "mds." << role
<< " not active ("
1410 << ceph_mds_state_name(fs
->mds_map
.get_state(role
.rank
)) << ")";
1411 } else if (fs
->mds_map
.get_root() == role
.rank
||
1412 fs
->mds_map
.get_tableserver() == role
.rank
) {
1414 ss
<< "can't tell the root (" << fs
->mds_map
.get_root()
1415 << ") or tableserver (" << fs
->mds_map
.get_tableserver()
1416 << ") to deactivate";
1417 } else if (role
.rank
!= fs
->mds_map
.get_last_in_mds()) {
1419 ss
<< "mds." << role
<< " doesn't have the max rank ("
1420 << fs
->mds_map
.get_last_in_mds() << ")";
1421 } else if (fs
->mds_map
.get_num_in_mds() <= size_t(fs
->mds_map
.get_max_mds())) {
1423 ss
<< "must decrease max_mds or else MDS will immediately reactivate";
1426 mds_gid_t gid
= fs
->mds_map
.up
.at(role
.rank
);
1427 ss
<< "telling mds." << role
<< " "
1428 << pending_fsmap
.get_info_gid(gid
).addr
<< " to deactivate";
1430 pending_fsmap
.modify_daemon(gid
, [](MDSMap::mds_info_t
*info
) {
1431 info
->state
= MDSMap::STATE_STOPPING
;
1434 } else if (prefix
== "mds set_state") {
1436 if (!cmd_getval(g_ceph_context
, cmdmap
, "gid", gid
)) {
1437 ss
<< "error parsing 'gid' value '"
1438 << cmd_vartype_stringify(cmdmap
["gid"]) << "'";
1441 MDSMap::DaemonState state
;
1442 if (!cmd_getval(g_ceph_context
, cmdmap
, "state", state
)) {
1443 ss
<< "error parsing 'state' string value '"
1444 << cmd_vartype_stringify(cmdmap
["state"]) << "'";
1447 if (pending_fsmap
.gid_exists(gid
)) {
1448 pending_fsmap
.modify_daemon(gid
, [state
](MDSMap::mds_info_t
*info
) {
1449 info
->state
= state
;
1451 ss
<< "set mds gid " << gid
<< " to state " << state
<< " "
1452 << ceph_mds_state_name(state
);
1455 } else if (prefix
== "mds fail") {
1457 cmd_getval(g_ceph_context
, cmdmap
, "who", who
);
1458 r
= fail_mds(ss
, who
);
1459 if (r
< 0 && r
== -EAGAIN
) {
1460 mon
->osdmon()->wait_for_writeable(op
, new C_RetryMessage(this, op
));
1461 return -EAGAIN
; // don't propose yet; wait for message to be retried
1463 } else if (prefix
== "mds rm") {
1465 if (!cmd_getval(g_ceph_context
, cmdmap
, "gid", gid
)) {
1466 ss
<< "error parsing 'gid' value '"
1467 << cmd_vartype_stringify(cmdmap
["gid"]) << "'";
1470 if (!pending_fsmap
.gid_exists(gid
)) {
1471 ss
<< "mds gid " << gid
<< " dne";
1474 MDSMap::DaemonState state
= pending_fsmap
.get_info_gid(gid
).state
;
1476 ss
<< "cannot remove active mds." << pending_fsmap
.get_info_gid(gid
).name
1477 << " rank " << pending_fsmap
.get_info_gid(gid
).rank
;
1480 pending_fsmap
.erase(gid
, {});
1481 ss
<< "removed mds gid " << gid
;
1485 } else if (prefix
== "mds rmfailed") {
1487 if (!cmd_getval(g_ceph_context
, cmdmap
, "confirm", confirm
) ||
1488 confirm
!= "--yes-i-really-mean-it") {
1489 ss
<< "WARNING: this can make your filesystem inaccessible! "
1490 "Add --yes-i-really-mean-it if you are sure you wish to continue.";
1494 std::string role_str
;
1495 cmd_getval(g_ceph_context
, cmdmap
, "who", role_str
);
1497 int r
= parse_role(role_str
, &role
, ss
);
1499 ss
<< "invalid role '" << role_str
<< "'";
1503 pending_fsmap
.modify_filesystem(
1505 [role
](std::shared_ptr
<Filesystem
> fs
)
1507 fs
->mds_map
.failed
.erase(role
.rank
);
1510 ss
<< "removed failed mds." << role
;
1512 } else if (prefix
== "mds compat rm_compat") {
1514 if (!cmd_getval(g_ceph_context
, cmdmap
, "feature", f
)) {
1515 ss
<< "error parsing feature value '"
1516 << cmd_vartype_stringify(cmdmap
["feature"]) << "'";
1519 if (pending_fsmap
.compat
.compat
.contains(f
)) {
1520 ss
<< "removing compat feature " << f
;
1521 CompatSet modified
= pending_fsmap
.compat
;
1522 modified
.compat
.remove(f
);
1523 pending_fsmap
.update_compat(modified
);
1525 ss
<< "compat feature " << f
<< " not present in " << pending_fsmap
.compat
;
1528 } else if (prefix
== "mds compat rm_incompat") {
1530 if (!cmd_getval(g_ceph_context
, cmdmap
, "feature", f
)) {
1531 ss
<< "error parsing feature value '"
1532 << cmd_vartype_stringify(cmdmap
["feature"]) << "'";
1535 if (pending_fsmap
.compat
.incompat
.contains(f
)) {
1536 ss
<< "removing incompat feature " << f
;
1537 CompatSet modified
= pending_fsmap
.compat
;
1538 modified
.incompat
.remove(f
);
1539 pending_fsmap
.update_compat(modified
);
1541 ss
<< "incompat feature " << f
<< " not present in " << pending_fsmap
.compat
;
1544 } else if (prefix
== "mds repaired") {
1545 std::string role_str
;
1546 cmd_getval(g_ceph_context
, cmdmap
, "rank", role_str
);
1548 r
= parse_role(role_str
, &role
, ss
);
1553 bool modified
= pending_fsmap
.undamaged(role
.fscid
, role
.rank
);
1555 dout(4) << "repaired: restoring rank " << role
<< dendl
;
1557 dout(4) << "repaired: no-op on rank " << role
<< dendl
;
1569 * Helper to legacy_filesystem_command
1571 void MDSMonitor::modify_legacy_filesystem(
1572 std::function
<void(std::shared_ptr
<Filesystem
> )> fn
)
1574 pending_fsmap
.modify_filesystem(
1575 pending_fsmap
.legacy_client_fscid
,
1583 * Handle a command that affects the filesystem (i.e. a filesystem
1584 * must exist for the command to act upon).
1586 * @retval 0 Command was successfully handled and has side effects
1587 * @retval -EAGAIN Messages has been requeued for retry
1588 * @retval -ENOSYS Unknown command
1589 * @retval < 0 An error has occurred; **ss** may have been set.
1591 int MDSMonitor::legacy_filesystem_command(
1593 std::string
const &prefix
,
1594 map
<string
, cmd_vartype
> &cmdmap
,
1595 std::stringstream
&ss
)
1597 dout(4) << __func__
<< " prefix='" << prefix
<< "'" << dendl
;
1598 op
->mark_mdsmon_event(__func__
);
1601 cmd_getval(g_ceph_context
, cmdmap
, "who", whostr
);
1603 assert (pending_fsmap
.legacy_client_fscid
!= FS_CLUSTER_ID_NONE
);
1605 if (prefix
== "mds set_max_mds") {
1606 // NOTE: deprecated by "fs set max_mds"
1608 if (!cmd_getval(g_ceph_context
, cmdmap
, "maxmds", maxmds
) || maxmds
<= 0) {
1612 const MDSMap
& mdsmap
=
1613 pending_fsmap
.filesystems
.at(pending_fsmap
.legacy_client_fscid
)->mds_map
;
1615 if (!mdsmap
.allows_multimds() &&
1616 maxmds
> mdsmap
.get_max_mds() &&
1618 ss
<< "multi-MDS clusters are not enabled; set 'allow_multimds' to enable";
1622 if (maxmds
> MAX_MDS
) {
1623 ss
<< "may not have more than " << MAX_MDS
<< " MDS ranks";
1627 modify_legacy_filesystem(
1628 [maxmds
](std::shared_ptr
<Filesystem
> fs
)
1630 fs
->mds_map
.set_max_mds(maxmds
);
1634 ss
<< "max_mds = " << maxmds
;
1635 } else if (prefix
== "mds cluster_down") {
1636 // NOTE: deprecated by "fs set cluster_down"
1637 modify_legacy_filesystem(
1638 [](std::shared_ptr
<Filesystem
> fs
)
1640 fs
->mds_map
.set_flag(CEPH_MDSMAP_DOWN
);
1642 ss
<< "marked fsmap DOWN";
1644 } else if (prefix
== "mds cluster_up") {
1645 // NOTE: deprecated by "fs set cluster_up"
1646 modify_legacy_filesystem(
1647 [](std::shared_ptr
<Filesystem
> fs
)
1649 fs
->mds_map
.clear_flag(CEPH_MDSMAP_DOWN
);
1651 ss
<< "unmarked fsmap DOWN";
1661 void MDSMonitor::check_subs()
1663 std::list
<std::string
> types
;
1665 // Subscriptions may be to "mdsmap" (MDS and legacy clients),
1666 // "mdsmap.<namespace>", or to "fsmap" for the full state of all
1667 // filesystems. Build a list of all the types we service
1668 // subscriptions for.
1669 types
.push_back("fsmap");
1670 types
.push_back("fsmap.user");
1671 types
.push_back("mdsmap");
1672 for (const auto &i
: fsmap
.filesystems
) {
1673 auto fscid
= i
.first
;
1674 std::ostringstream oss
;
1675 oss
<< "mdsmap." << fscid
;
1676 types
.push_back(oss
.str());
1679 for (const auto &type
: types
) {
1680 if (mon
->session_map
.subs
.count(type
) == 0)
1682 xlist
<Subscription
*>::iterator p
= mon
->session_map
.subs
[type
]->begin();
1684 Subscription
*sub
= *p
;
1692 void MDSMonitor::check_sub(Subscription
*sub
)
1694 dout(20) << __func__
<< ": " << sub
->type
<< dendl
;
1696 if (sub
->type
== "fsmap") {
1697 if (sub
->next
<= fsmap
.get_epoch()) {
1698 sub
->session
->con
->send_message(new MFSMap(mon
->monmap
->fsid
, fsmap
));
1700 mon
->session_map
.remove_sub(sub
);
1702 sub
->next
= fsmap
.get_epoch() + 1;
1705 } else if (sub
->type
== "fsmap.user") {
1706 if (sub
->next
<= fsmap
.get_epoch()) {
1708 fsmap_u
.epoch
= fsmap
.get_epoch();
1709 fsmap_u
.legacy_client_fscid
= fsmap
.legacy_client_fscid
;
1710 for (auto p
= fsmap
.filesystems
.begin();
1711 p
!= fsmap
.filesystems
.end();
1713 FSMapUser::fs_info_t
& fs_info
= fsmap_u
.filesystems
[p
->first
];
1714 fs_info
.cid
= p
->first
;
1715 fs_info
.name
= p
->second
->mds_map
.fs_name
;
1717 sub
->session
->con
->send_message(new MFSMapUser(mon
->monmap
->fsid
, fsmap_u
));
1719 mon
->session_map
.remove_sub(sub
);
1721 sub
->next
= fsmap
.get_epoch() + 1;
1724 } else if (sub
->type
.compare(0, 6, "mdsmap") == 0) {
1725 if (sub
->next
> fsmap
.get_epoch()) {
1729 const bool is_mds
= sub
->session
->inst
.name
.is_mds();
1730 mds_gid_t mds_gid
= MDS_GID_NONE
;
1731 fs_cluster_id_t fscid
= FS_CLUSTER_ID_NONE
;
1733 // What (if any) namespace are you assigned to?
1734 auto mds_info
= fsmap
.get_mds_info();
1735 for (const auto &i
: mds_info
) {
1736 if (i
.second
.addr
== sub
->session
->inst
.addr
) {
1738 fscid
= fsmap
.mds_roles
.at(mds_gid
);
1742 // You're a client. Did you request a particular
1744 if (sub
->type
.find("mdsmap.") == 0) {
1745 auto namespace_id_str
= sub
->type
.substr(std::string("mdsmap.").size());
1746 dout(10) << __func__
<< ": namespace_id " << namespace_id_str
<< dendl
;
1748 fscid
= strict_strtoll(namespace_id_str
.c_str(), 10, &err
);
1750 // Client asked for a non-existent namespace, send them nothing
1751 dout(1) << "Invalid client subscription '" << sub
->type
1755 if (fsmap
.filesystems
.count(fscid
) == 0) {
1756 // Client asked for a non-existent namespace, send them nothing
1757 // TODO: something more graceful for when a client has a filesystem
1758 // mounted, and the fileysstem is deleted. Add a "shut down you fool"
1760 dout(1) << "Client subscribed to non-existent namespace '" <<
1761 fscid
<< "'" << dendl
;
1765 // Unqualified request for "mdsmap": give it the one marked
1766 // for use by legacy clients.
1767 if (fsmap
.legacy_client_fscid
!= FS_CLUSTER_ID_NONE
) {
1768 fscid
= fsmap
.legacy_client_fscid
;
1770 dout(1) << "Client subscribed for legacy filesystem but "
1771 "none is configured" << dendl
;
1776 dout(10) << __func__
<< ": is_mds=" << is_mds
<< ", fscid= " << fscid
<< dendl
;
1778 // Work out the effective latest epoch
1779 MDSMap
*mds_map
= nullptr;
1781 null_map
.compat
= fsmap
.compat
;
1782 if (fscid
== FS_CLUSTER_ID_NONE
) {
1783 // For a client, we should have already dropped out
1786 if (fsmap
.standby_daemons
.count(mds_gid
)) {
1787 // For an MDS, we need to feed it an MDSMap with its own state in
1788 null_map
.mds_info
[mds_gid
] = fsmap
.standby_daemons
[mds_gid
];
1789 null_map
.epoch
= fsmap
.standby_epochs
[mds_gid
];
1791 null_map
.epoch
= fsmap
.epoch
;
1793 mds_map
= &null_map
;
1795 // Check the effective epoch
1796 mds_map
= &(fsmap
.filesystems
.at(fscid
)->mds_map
);
1799 assert(mds_map
!= nullptr);
1800 dout(10) << __func__
<< " selected MDS map epoch " <<
1801 mds_map
->epoch
<< " for namespace " << fscid
<< " for subscriber "
1802 << sub
->session
->inst
.name
<< " who wants epoch " << sub
->next
<< dendl
;
1804 if (sub
->next
> mds_map
->epoch
) {
1807 auto msg
= new MMDSMap(mon
->monmap
->fsid
, mds_map
);
1809 sub
->session
->con
->send_message(msg
);
1811 mon
->session_map
.remove_sub(sub
);
1813 sub
->next
= mds_map
->get_epoch() + 1;
1819 void MDSMonitor::update_metadata(mds_gid_t gid
,
1820 const map
<string
, string
>& metadata
)
1822 if (metadata
.empty()) {
1825 pending_metadata
[gid
] = metadata
;
1827 MonitorDBStore::TransactionRef t
= paxos
->get_pending_transaction();
1829 ::encode(pending_metadata
, bl
);
1830 t
->put(MDS_METADATA_PREFIX
, "last_metadata", bl
);
1831 paxos
->trigger_propose();
1834 void MDSMonitor::remove_from_metadata(MonitorDBStore::TransactionRef t
)
1836 bool update
= false;
1837 for (map
<mds_gid_t
, Metadata
>::iterator i
= pending_metadata
.begin();
1838 i
!= pending_metadata
.end(); ) {
1839 if (!pending_fsmap
.gid_exists(i
->first
)) {
1840 pending_metadata
.erase(i
++);
1849 ::encode(pending_metadata
, bl
);
1850 t
->put(MDS_METADATA_PREFIX
, "last_metadata", bl
);
1853 int MDSMonitor::load_metadata(map
<mds_gid_t
, Metadata
>& m
)
1856 int r
= mon
->store
->get(MDS_METADATA_PREFIX
, "last_metadata", bl
);
1858 dout(1) << "Unable to load 'last_metadata'" << dendl
;
1862 bufferlist::iterator it
= bl
.begin();
1867 void MDSMonitor::count_metadata(const string
& field
, map
<string
,int> *out
)
1869 map
<mds_gid_t
,Metadata
> meta
;
1870 load_metadata(meta
);
1871 for (auto& p
: meta
) {
1872 auto q
= p
.second
.find(field
);
1873 if (q
== p
.second
.end()) {
1874 (*out
)["unknown"]++;
1876 (*out
)[q
->second
]++;
1881 void MDSMonitor::count_metadata(const string
& field
, Formatter
*f
)
1883 map
<string
,int> by_val
;
1884 count_metadata(field
, &by_val
);
1885 f
->open_object_section(field
.c_str());
1886 for (auto& p
: by_val
) {
1887 f
->dump_int(p
.first
.c_str(), p
.second
);
1892 int MDSMonitor::dump_metadata(const std::string
&who
, Formatter
*f
, ostream
& err
)
1896 mds_gid_t gid
= gid_from_arg(who
, err
);
1897 if (gid
== MDS_GID_NONE
) {
1901 map
<mds_gid_t
, Metadata
> metadata
;
1902 if (int r
= load_metadata(metadata
)) {
1903 err
<< "Unable to load 'last_metadata'";
1907 if (!metadata
.count(gid
)) {
1910 const Metadata
& m
= metadata
[gid
];
1911 for (Metadata::const_iterator p
= m
.begin(); p
!= m
.end(); ++p
) {
1912 f
->dump_string(p
->first
.c_str(), p
->second
);
1917 int MDSMonitor::print_nodes(Formatter
*f
)
1921 map
<mds_gid_t
, Metadata
> metadata
;
1922 if (int r
= load_metadata(metadata
)) {
1926 map
<string
, list
<int> > mdses
; // hostname => rank
1927 for (map
<mds_gid_t
, Metadata
>::iterator it
= metadata
.begin();
1928 it
!= metadata
.end(); ++it
) {
1929 const Metadata
& m
= it
->second
;
1930 Metadata::const_iterator hostname
= m
.find("hostname");
1931 if (hostname
== m
.end()) {
1932 // not likely though
1935 const mds_gid_t gid
= it
->first
;
1936 if (!fsmap
.gid_exists(gid
)) {
1937 dout(5) << __func__
<< ": GID " << gid
<< " not existent" << dendl
;
1940 const MDSMap::mds_info_t
& mds_info
= fsmap
.get_info_gid(gid
);
1941 // FIXME: include filesystem name with rank here
1942 mdses
[hostname
->second
].push_back(mds_info
.rank
);
1945 dump_services(f
, mdses
, "mds");
1950 * If a cluster is undersized (with respect to max_mds), then
1951 * attempt to find daemons to grow it.
1953 bool MDSMonitor::maybe_expand_cluster(std::shared_ptr
<Filesystem
> fs
)
1955 bool do_propose
= false;
1957 if (fs
->mds_map
.test_flag(CEPH_MDSMAP_DOWN
)) {
1961 while (fs
->mds_map
.get_num_in_mds() < size_t(fs
->mds_map
.get_max_mds()) &&
1962 !fs
->mds_map
.is_degraded()) {
1963 mds_rank_t mds
= mds_rank_t(0);
1965 while (fs
->mds_map
.is_in(mds
)) {
1968 mds_gid_t newgid
= pending_fsmap
.find_replacement_for({fs
->fscid
, mds
},
1969 name
, g_conf
->mon_force_standby_active
);
1970 if (newgid
== MDS_GID_NONE
) {
1974 dout(1) << "adding standby " << pending_fsmap
.get_info_gid(newgid
).addr
1975 << " as mds." << mds
<< dendl
;
1976 pending_fsmap
.promote(newgid
, fs
, mds
);
1985 * If a daemon is laggy, and a suitable replacement
1986 * is available, fail this daemon (remove from map) and pass its
1987 * role to another daemon.
1989 void MDSMonitor::maybe_replace_gid(mds_gid_t gid
, const MDSMap::mds_info_t
& info
,
1990 bool *mds_propose
, bool *osd_propose
)
1992 assert(mds_propose
!= nullptr);
1993 assert(osd_propose
!= nullptr);
1995 const auto fscid
= pending_fsmap
.mds_roles
.at(gid
);
1997 // We will only take decisive action (replacing/removing a daemon)
1998 // if we have some indicating that some other daemon(s) are successfully
1999 // getting beacons through recently.
2000 utime_t latest_beacon
;
2001 for (const auto & i
: last_beacon
) {
2002 latest_beacon
= MAX(i
.second
.stamp
, latest_beacon
);
2004 const bool may_replace
= latest_beacon
>
2006 MAX(g_conf
->mds_beacon_interval
, g_conf
->mds_beacon_grace
* 0.5));
2009 // and is there a non-laggy standby that can take over for us?
2011 if (info
.rank
>= 0 &&
2012 info
.state
!= MDSMap::STATE_STANDBY
&&
2013 info
.state
!= MDSMap::STATE_STANDBY_REPLAY
&&
2015 !pending_fsmap
.get_filesystem(fscid
)->mds_map
.test_flag(CEPH_MDSMAP_DOWN
) &&
2016 (sgid
= pending_fsmap
.find_replacement_for({fscid
, info
.rank
}, info
.name
,
2017 g_conf
->mon_force_standby_active
)) != MDS_GID_NONE
)
2020 MDSMap::mds_info_t si
= pending_fsmap
.get_info_gid(sgid
);
2021 dout(10) << " replacing " << gid
<< " " << info
.addr
<< " mds."
2022 << info
.rank
<< "." << info
.inc
2023 << " " << ceph_mds_state_name(info
.state
)
2024 << " with " << sgid
<< "/" << si
.name
<< " " << si
.addr
<< dendl
;
2026 mon
->clog
->warn() << "MDS daemon '" << info
.name
<< "'"
2027 << " is not responding, replacing it "
2028 << "as rank " << info
.rank
2029 << " with standby '" << si
.name
<< "'";
2031 // Remember what NS the old one was in
2032 const fs_cluster_id_t fscid
= pending_fsmap
.mds_roles
.at(gid
);
2034 // Remove the old one
2035 *osd_propose
|= fail_mds_gid(gid
);
2037 // Promote the replacement
2038 auto fs
= pending_fsmap
.filesystems
.at(fscid
);
2039 pending_fsmap
.promote(sgid
, fs
, info
.rank
);
2041 *mds_propose
= true;
2042 } else if ((info
.state
== MDSMap::STATE_STANDBY_REPLAY
||
2043 info
.state
== MDSMap::STATE_STANDBY
) && may_replace
) {
2044 dout(10) << " failing and removing " << gid
<< " " << info
.addr
<< " mds." << info
.rank
2045 << "." << info
.inc
<< " " << ceph_mds_state_name(info
.state
)
2047 mon
->clog
->info() << "MDS standby '" << info
.name
2048 << "' is not responding, removing it from the set of "
2051 *mds_propose
= true;
2052 } else if (!info
.laggy()) {
2053 dout(10) << " marking " << gid
<< " " << info
.addr
<< " mds." << info
.rank
<< "." << info
.inc
2054 << " " << ceph_mds_state_name(info
.state
)
2055 << " laggy" << dendl
;
2056 pending_fsmap
.modify_daemon(info
.global_id
, [](MDSMap::mds_info_t
*info
) {
2057 info
->laggy_since
= ceph_clock_now();
2059 *mds_propose
= true;
2063 bool MDSMonitor::maybe_promote_standby(std::shared_ptr
<Filesystem
> fs
)
2065 assert(!fs
->mds_map
.test_flag(CEPH_MDSMAP_DOWN
));
2067 bool do_propose
= false;
2069 // have a standby take over?
2070 set
<mds_rank_t
> failed
;
2071 fs
->mds_map
.get_failed_mds_set(failed
);
2072 if (!failed
.empty()) {
2073 set
<mds_rank_t
>::iterator p
= failed
.begin();
2074 while (p
!= failed
.end()) {
2075 mds_rank_t f
= *p
++;
2076 mds_gid_t sgid
= pending_fsmap
.find_replacement_for({fs
->fscid
, f
}, {},
2077 g_conf
->mon_force_standby_active
);
2079 const MDSMap::mds_info_t si
= pending_fsmap
.get_info_gid(sgid
);
2080 dout(0) << " taking over failed mds." << f
<< " with " << sgid
2081 << "/" << si
.name
<< " " << si
.addr
<< dendl
;
2082 pending_fsmap
.promote(sgid
, fs
, f
);
2087 // There were no failures to replace, so try using any available standbys
2088 // as standby-replay daemons.
2090 // Take a copy of the standby GIDs so that we can iterate over
2091 // them while perhaps-modifying standby_daemons during the loop
2092 // (if we promote anyone they are removed from standby_daemons)
2093 std::vector
<mds_gid_t
> standby_gids
;
2094 for (const auto &j
: pending_fsmap
.standby_daemons
) {
2095 standby_gids
.push_back(j
.first
);
2098 for (const auto &gid
: standby_gids
) {
2099 const auto &info
= pending_fsmap
.standby_daemons
.at(gid
);
2100 assert(info
.state
== MDSMap::STATE_STANDBY
);
2102 if (!info
.standby_replay
) {
2107 * This mds is standby but has no rank assigned.
2108 * See if we can find it somebody to shadow
2110 dout(20) << "gid " << gid
<< " is standby and following nobody" << dendl
;
2112 // standby for someone specific?
2113 if (info
.standby_for_rank
>= 0) {
2114 // The mds_info_t may or may not tell us exactly which filesystem
2115 // the standby_for_rank refers to: lookup via legacy_client_fscid
2116 mds_role_t target_role
= {
2117 info
.standby_for_fscid
== FS_CLUSTER_ID_NONE
?
2118 pending_fsmap
.legacy_client_fscid
: info
.standby_for_fscid
,
2119 info
.standby_for_rank
};
2121 // It is possible that the map contains a standby_for_fscid
2122 // that doesn't correspond to an existing filesystem, especially
2123 // if we loaded from a version with a bug (#17466)
2124 if (info
.standby_for_fscid
!= FS_CLUSTER_ID_NONE
2125 && !pending_fsmap
.filesystem_exists(info
.standby_for_fscid
)) {
2126 derr
<< "gid " << gid
<< " has invalid standby_for_fscid "
2127 << info
.standby_for_fscid
<< dendl
;
2131 // If we managed to resolve a full target role
2132 if (target_role
.fscid
!= FS_CLUSTER_ID_NONE
) {
2133 auto fs
= pending_fsmap
.get_filesystem(target_role
.fscid
);
2134 if (fs
->mds_map
.is_followable(target_role
.rank
)) {
2135 do_propose
|= try_standby_replay(
2138 fs
->mds_map
.get_info(target_role
.rank
));
2146 for (auto fs_i
: pending_fsmap
.filesystems
) {
2147 const MDSMap
&mds_map
= fs_i
.second
->mds_map
;
2148 for (auto mds_i
: mds_map
.mds_info
) {
2149 MDSMap::mds_info_t
&cand_info
= mds_i
.second
;
2150 if (cand_info
.rank
>= 0 && mds_map
.is_followable(cand_info
.rank
)) {
2151 if ((info
.standby_for_name
.length() && info
.standby_for_name
!= cand_info
.name
) ||
2152 info
.standby_for_rank
!= MDS_RANK_NONE
) {
2153 continue; // we're supposed to follow someone else
2156 if (try_standby_replay(info
, *(fs_i
.second
), cand_info
)) {
2170 void MDSMonitor::tick()
2172 // make sure mds's are still alive
2173 // ...if i am an active leader
2174 if (!is_active()) return;
2176 dout(10) << fsmap
<< dendl
;
2178 bool do_propose
= false;
2180 if (!mon
->is_leader()) return;
2182 do_propose
|= pending_fsmap
.check_health();
2184 // expand mds cluster (add new nodes to @in)?
2185 for (auto i
: pending_fsmap
.filesystems
) {
2186 do_propose
|= maybe_expand_cluster(i
.second
);
2189 const auto now
= ceph_clock_now();
2190 if (last_tick
.is_zero()) {
2194 if (now
- last_tick
> (g_conf
->mds_beacon_grace
- g_conf
->mds_beacon_interval
)) {
2195 // This case handles either local slowness (calls being delayed
2196 // for whatever reason) or cluster election slowness (a long gap
2197 // between calls while an election happened)
2198 dout(4) << __func__
<< ": resetting beacon timeouts due to mon delay "
2199 "(slow election?) of " << now
- last_tick
<< " seconds" << dendl
;
2200 for (auto &i
: last_beacon
) {
2201 i
.second
.stamp
= now
;
2207 // check beacon timestamps
2208 utime_t cutoff
= now
;
2209 cutoff
-= g_conf
->mds_beacon_grace
;
2211 // make sure last_beacon is fully populated
2212 for (const auto &p
: pending_fsmap
.mds_roles
) {
2213 auto &gid
= p
.first
;
2214 if (last_beacon
.count(gid
) == 0) {
2215 last_beacon
[gid
].stamp
= now
;
2216 last_beacon
[gid
].seq
= 0;
2220 bool propose_osdmap
= false;
2221 bool osdmap_writeable
= mon
->osdmon()->is_writeable();
2222 auto p
= last_beacon
.begin();
2223 while (p
!= last_beacon
.end()) {
2224 mds_gid_t gid
= p
->first
;
2225 auto beacon_info
= p
->second
;
2228 if (!pending_fsmap
.gid_exists(gid
)) {
2230 last_beacon
.erase(gid
);
2234 if (beacon_info
.stamp
< cutoff
) {
2235 auto &info
= pending_fsmap
.get_info_gid(gid
);
2236 dout(1) << "no beacon from mds." << info
.rank
<< "." << info
.inc
2237 << " (gid: " << gid
<< " addr: " << info
.addr
2238 << " state: " << ceph_mds_state_name(info
.state
) << ")"
2239 << " since " << beacon_info
.stamp
<< dendl
;
2240 // If the OSDMap is writeable, we can blacklist things, so we can
2241 // try failing any laggy MDS daemons. Consider each one for failure.
2242 if (osdmap_writeable
) {
2243 maybe_replace_gid(gid
, info
, &do_propose
, &propose_osdmap
);
2247 if (propose_osdmap
) {
2248 request_proposal(mon
->osdmon());
2251 for (auto i
: pending_fsmap
.filesystems
) {
2253 if (!fs
->mds_map
.test_flag(CEPH_MDSMAP_DOWN
)) {
2254 do_propose
|= maybe_promote_standby(fs
);
2264 * finfo: the would-be follower
2265 * leader_fs: the Filesystem containing the would-be leader
2266 * ainfo: the would-be leader
2268 bool MDSMonitor::try_standby_replay(
2269 const MDSMap::mds_info_t
& finfo
,
2270 const Filesystem
&leader_fs
,
2271 const MDSMap::mds_info_t
& ainfo
)
2273 // someone else already following?
2274 if (leader_fs
.has_standby_replay(ainfo
.global_id
)) {
2275 dout(20) << " mds." << ainfo
.rank
<< " already has a follower" << dendl
;
2278 // Assign the new role to the standby
2279 dout(10) << " setting to follow mds rank " << ainfo
.rank
<< dendl
;
2280 pending_fsmap
.assign_standby_replay(finfo
.global_id
, leader_fs
.fscid
, ainfo
.rank
);
2285 MDSMonitor::MDSMonitor(Monitor
*mn
, Paxos
*p
, string service_name
)
2286 : PaxosService(mn
, p
, service_name
)
2288 handlers
= FileSystemCommandHandler::load(p
);
2291 void MDSMonitor::on_restart()
2293 // Clear out the leader-specific state.
2294 last_tick
= utime_t();
2295 last_beacon
.clear();