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.
18 #include "common/Timer.h"
19 #include "MonitorDBStore.h"
20 #include "messages/MMonElection.h"
22 #include "common/config.h"
23 #include "include/assert.h"
25 #define dout_subsys ceph_subsys_mon
27 #define dout_prefix _prefix(_dout, mon, epoch)
28 static ostream& _prefix(std::ostream *_dout, Monitor *mon, epoch_t epoch) {
29 return *_dout << "mon." << mon->name << "@" << mon->rank
30 << "(" << mon->get_state_name()
31 << ").elector(" << epoch << ") ";
37 epoch = mon->store->get(Monitor::MONITOR_NAME, "election_epoch");
39 dout(1) << "init, first boot, initializing epoch at 1 " << dendl;
41 } else if (epoch % 2) {
42 dout(1) << "init, last seen epoch " << epoch
43 << ", mid-election, bumping" << dendl;
45 auto t(std::make_shared<MonitorDBStore::Transaction>());
46 t->put(Monitor::MONITOR_NAME, "election_epoch", epoch);
47 mon->store->apply_transaction(t);
49 dout(1) << "init, last seen epoch " << epoch << dendl;
53 void Elector::shutdown()
58 void Elector::bump_epoch(epoch_t e)
60 dout(10) << "bump_epoch " << epoch << " to " << e << dendl;
63 auto t(std::make_shared<MonitorDBStore::Transaction>());
64 t->put(Monitor::MONITOR_NAME, "election_epoch", epoch);
65 mon->store->apply_transaction(t);
69 // clear up some state
78 dout(0) << "not starting new election -- not participating" << dendl;
81 dout(5) << "start -- can i be leader?" << dendl;
86 // start by trying to elect me
88 bump_epoch(epoch+1); // odd == election cycle
90 // do a trivial db write just to ensure it is writeable.
91 auto t(std::make_shared<MonitorDBStore::Transaction>());
92 t->put(Monitor::MONITOR_NAME, "election_writeable_test", rand());
93 int r = mon->store->apply_transaction(t);
96 start_stamp = ceph_clock_now();
98 acked_me[mon->rank].cluster_features = CEPH_FEATURES_ALL;
99 acked_me[mon->rank].mon_features = ceph::features::mon::get_supported();
100 mon->collect_metadata(&acked_me[mon->rank].metadata);
103 // bcast to everyone else
104 for (unsigned i=0; i<mon->monmap->size(); ++i) {
105 if ((int)i == mon->rank) continue;
107 new MMonElection(MMonElection::OP_PROPOSE, epoch, mon->monmap);
108 m->mon_features = ceph::features::mon::get_supported();
109 mon->messenger->send_message(m, mon->monmap->get_inst(i));
115 void Elector::defer(int who)
117 dout(5) << "defer to " << who << dendl;
127 ack_stamp = ceph_clock_now();
128 MMonElection *m = new MMonElection(MMonElection::OP_ACK, epoch, mon->monmap);
129 m->mon_features = ceph::features::mon::get_supported();
130 mon->collect_metadata(&m->metadata);
132 // This field is unused completely in luminous, but jewel uses it to
133 // determine whether we are a dumpling mon due to some crufty old
134 // code. It only needs to see this buffer non-empty, so put
135 // something useless there.
136 m->sharing_bl = mon->get_local_commands_bl(mon->get_required_mon_features());
138 mon->messenger->send_message(m, mon->monmap->get_inst(who));
141 reset_timer(1.0); // give the leader some extra time to declare victory
145 void Elector::reset_timer(double plus)
150 * This class is used as the callback when the expire_event timer fires up.
152 * If the expire_event is fired, then it means that we had an election going,
153 * either started by us or by some other participant, but it took too long,
156 * When the election expires, we will check if we were the ones who won, and
157 * if so we will declare victory. If that is not the case, then we assume
158 * that the one we defered to didn't declare victory quickly enough (in fact,
159 * as far as we know, we may even be dead); so, just propose ourselves as the
162 expire_event = mon->timer.add_event_after(
163 g_conf->mon_election_timeout + plus,
164 new C_MonContext(mon, [this](int) {
170 void Elector::cancel_timer()
173 mon->timer.cancel_event(expire_event);
178 void Elector::expire()
180 dout(5) << "election timer expired" << dendl;
184 acked_me.size() > (unsigned)(mon->monmap->size() / 2)) {
188 // whoever i deferred to didn't declare victory quickly enough.
189 if (mon->has_ever_joined)
197 void Elector::victory()
202 uint64_t cluster_features = CEPH_FEATURES_ALL;
203 mon_feature_t mon_features = ceph::features::mon::get_supported();
205 map<int,Metadata> metadata;
206 for (map<int, elector_info_t>::iterator p = acked_me.begin();
209 quorum.insert(p->first);
210 cluster_features &= p->second.cluster_features;
211 mon_features &= p->second.mon_features;
212 metadata[p->first] = p->second.metadata;
217 assert(epoch % 2 == 1); // election
218 bump_epoch(epoch+1); // is over!
221 for (set<int>::iterator p = quorum.begin();
224 if (*p == mon->rank) continue;
225 MMonElection *m = new MMonElection(MMonElection::OP_VICTORY, epoch,
228 m->quorum_features = cluster_features;
229 m->mon_features = mon_features;
230 m->sharing_bl = mon->get_local_commands_bl(mon_features);
231 mon->messenger->send_message(m, mon->monmap->get_inst(*p));
235 mon->win_election(epoch, quorum,
236 cluster_features, mon_features, metadata);
240 void Elector::handle_propose(MonOpRequestRef op)
242 op->mark_event("elector:handle_propose");
243 MMonElection *m = static_cast<MMonElection*>(op->get_req());
244 dout(5) << "handle_propose from " << m->get_source() << dendl;
245 int from = m->get_source().num();
247 assert(m->epoch % 2 == 1); // election
248 uint64_t required_features = mon->get_required_features();
249 mon_feature_t required_mon_features = mon->get_required_mon_features();
251 dout(10) << __func__ << " required features " << required_features
252 << " " << required_mon_features
253 << ", peer features " << m->get_connection()->get_features()
254 << " " << m->mon_features
257 if ((required_features ^ m->get_connection()->get_features()) &
259 dout(5) << " ignoring propose from mon" << from
260 << " without required features" << dendl;
263 } else if (!m->mon_features.contains_all(required_mon_features)) {
264 // all the features in 'required_mon_features' not in 'm->mon_features'
265 mon_feature_t missing = required_mon_features.diff(m->mon_features);
266 dout(5) << " ignoring propose from mon." << from
267 << " without required mon_features " << missing
270 } else if (m->epoch > epoch) {
271 bump_epoch(m->epoch);
272 } else if (m->epoch < epoch) {
273 // got an "old" propose,
274 if (epoch % 2 == 0 && // in a non-election cycle
275 mon->quorum.count(from) == 0) { // from someone outside the quorum
276 // a mon just started up, call a new election so they can rejoin!
277 dout(5) << " got propose from old epoch, quorum is " << mon->quorum
278 << ", " << m->get_source() << " must have just started" << dendl;
279 // we may be active; make sure we reset things in the monitor appropriately.
280 mon->start_election();
282 dout(5) << " ignoring old propose" << dendl;
287 if (mon->rank < from) {
288 // i would win over them.
289 if (leader_acked >= 0) { // we already acked someone
290 assert(leader_acked < from); // and they still win, of course
291 dout(5) << "no, we already acked " << leader_acked << dendl;
293 // wait, i should win!
295 mon->start_election();
299 // they would win over me
300 if (leader_acked < 0 || // haven't acked anyone yet, or
301 leader_acked > from || // they would win over who you did ack, or
302 leader_acked == from) { // this is the guy we're already deferring to
306 dout(5) << "no, we already acked " << leader_acked << dendl;
311 void Elector::handle_ack(MonOpRequestRef op)
313 op->mark_event("elector:handle_ack");
314 MMonElection *m = static_cast<MMonElection*>(op->get_req());
315 dout(5) << "handle_ack from " << m->get_source() << dendl;
316 int from = m->get_source().num();
318 assert(m->epoch % 2 == 1); // election
319 if (m->epoch > epoch) {
320 dout(5) << "woah, that's a newer epoch, i must have rebooted. bumping and re-starting!" << dendl;
321 bump_epoch(m->epoch);
325 assert(m->epoch == epoch);
326 uint64_t required_features = mon->get_required_features();
327 if ((required_features ^ m->get_connection()->get_features()) &
329 dout(5) << " ignoring ack from mon" << from
330 << " without required features" << dendl;
334 mon_feature_t required_mon_features = mon->get_required_mon_features();
335 if (!m->mon_features.contains_all(required_mon_features)) {
336 mon_feature_t missing = required_mon_features.diff(m->mon_features);
337 dout(5) << " ignoring ack from mon." << from
338 << " without required mon_features " << missing
345 acked_me[from].cluster_features = m->get_connection()->get_features();
346 acked_me[from].mon_features = m->mon_features;
347 acked_me[from].metadata = m->metadata;
348 dout(5) << " so far i have {";
349 for (map<int, elector_info_t>::const_iterator p = acked_me.begin();
352 if (p != acked_me.begin())
354 *_dout << " mon." << p->first << ":"
355 << " features " << p->second.cluster_features
356 << " " << p->second.mon_features;
358 *_dout << " }" << dendl;
360 // is that _everyone_?
361 if (acked_me.size() == mon->monmap->size()) {
362 // if yes, shortcut to election finish
366 // ignore, i'm deferring already.
367 assert(leader_acked >= 0);
372 void Elector::handle_victory(MonOpRequestRef op)
374 op->mark_event("elector:handle_victory");
375 MMonElection *m = static_cast<MMonElection*>(op->get_req());
376 dout(5) << "handle_victory from " << m->get_source()
377 << " quorum_features " << m->quorum_features
378 << " " << m->mon_features
380 int from = m->get_source().num();
382 assert(from < mon->rank);
383 assert(m->epoch % 2 == 0);
387 // i should have seen this election if i'm getting the victory.
388 if (m->epoch != epoch + 1) {
389 dout(5) << "woah, that's a funny epoch, i must have rebooted. bumping and re-starting!" << dendl;
390 bump_epoch(m->epoch);
395 bump_epoch(m->epoch);
398 mon->lose_election(epoch, m->quorum, from,
399 m->quorum_features, m->mon_features);
404 // stash leader's commands
405 assert(m->sharing_bl.length());
406 vector<MonCommand> new_cmds;
407 bufferlist::iterator bi = m->sharing_bl.begin();
408 MonCommand::decode_vector(new_cmds, bi);
409 mon->set_leader_commands(new_cmds);
412 void Elector::nak_old_peer(MonOpRequestRef op)
414 op->mark_event("elector:nak_old_peer");
415 MMonElection *m = static_cast<MMonElection*>(op->get_req());
416 uint64_t supported_features = m->get_connection()->get_features();
417 uint64_t required_features = mon->get_required_features();
418 mon_feature_t required_mon_features = mon->get_required_mon_features();
419 dout(10) << "sending nak to peer " << m->get_source()
420 << " that only supports " << supported_features
421 << " " << m->mon_features
422 << " of the required " << required_features
423 << " " << required_mon_features
426 MMonElection *reply = new MMonElection(MMonElection::OP_NAK, m->epoch,
428 reply->quorum_features = required_features;
429 reply->mon_features = required_mon_features;
430 mon->features.encode(reply->sharing_bl);
431 m->get_connection()->send_message(reply);
434 void Elector::handle_nak(MonOpRequestRef op)
436 op->mark_event("elector:handle_nak");
437 MMonElection *m = static_cast<MMonElection*>(op->get_req());
438 dout(1) << "handle_nak from " << m->get_source()
439 << " quorum_features " << m->quorum_features
440 << " " << m->mon_features
444 bufferlist::iterator bi = m->sharing_bl.begin();
446 CompatSet diff = Monitor::get_supported_features().unsupported(other);
448 mon_feature_t mon_supported = ceph::features::mon::get_supported();
449 // all features in 'm->mon_features' not in 'mon_supported'
450 mon_feature_t mon_diff = m->mon_features.diff(mon_supported);
452 derr << "Shutting down because I do not support required monitor features: { "
453 << diff << " } " << mon_diff << dendl;
459 void Elector::dispatch(MonOpRequestRef op)
461 op->mark_event("elector:dispatch");
462 assert(op->is_type_election());
464 switch (op->get_req()->get_type()) {
466 case MSG_MON_ELECTION:
468 if (!participating) {
471 if (op->get_req()->get_source().num() >= mon->monmap->size()) {
472 dout(5) << " ignoring bogus election message with bad mon rank "
473 << op->get_req()->get_source() << dendl;
477 MMonElection *em = static_cast<MMonElection*>(op->get_req());
479 // assume an old message encoding would have matched
480 if (em->fsid != mon->monmap->fsid) {
481 dout(0) << " ignoring election msg fsid "
482 << em->fsid << " != " << mon->monmap->fsid << dendl;
486 if (!mon->monmap->contains(em->get_source_addr())) {
487 dout(1) << "discarding election message: " << em->get_source_addr()
488 << " not in my monmap " << *mon->monmap << dendl;
493 peermap.decode(em->monmap_bl);
494 if (peermap.epoch > mon->monmap->epoch) {
495 dout(0) << em->get_source_inst() << " has newer monmap epoch " << peermap.epoch
496 << " > my epoch " << mon->monmap->epoch
499 mon->monmap->decode(em->monmap_bl);
500 auto t(std::make_shared<MonitorDBStore::Transaction>());
501 t->put("monmap", mon->monmap->epoch, em->monmap_bl);
502 t->put("monmap", "last_committed", mon->monmap->epoch);
503 mon->store->apply_transaction(t);
504 //mon->monmon()->paxos->stash_latest(mon->monmap->epoch, em->monmap_bl);
509 if (peermap.epoch < mon->monmap->epoch) {
510 dout(0) << em->get_source_inst() << " has older monmap epoch " << peermap.epoch
511 << " < my epoch " << mon->monmap->epoch
516 case MMonElection::OP_PROPOSE:
521 if (em->epoch < epoch) {
522 dout(5) << "old epoch, dropping" << dendl;
527 case MMonElection::OP_ACK:
530 case MMonElection::OP_VICTORY:
533 case MMonElection::OP_NAK:
547 void Elector::start_participating()
549 if (!participating) {
550 participating = true;