void ReplicationCoordinatorImpl::_startVoteRequester_inlock(long long newTerm) {
invariant(_voteRequester);
const auto lastOpTime = _getMyLastAppliedOpTime_inlock();
_voteRequester.reset(new VoteRequester);
StatusWith<executor::TaskExecutor::EventHandle> nextPhaseEvh = _voteRequester->start(
_replExecutor.get(), _rsConfig, _selfIndex, newTerm, false, lastOpTime, -1);
if (nextPhaseEvh.getStatus() == ErrorCodes::ShutdownInProgress) {
return;
}
fassert(28643, nextPhaseEvh.getStatus());
_replExecutor
->onEvent(
nextPhaseEvh.getValue(),
[=](const executor::TaskExecutor::CallbackArgs&) { _onVoteRequestComplete(newTerm); })
.status_with_transitional_ignore();
}
void ReplicationCoordinatorImpl::_startElectSelfV1_inlock(
TopologyCoordinator::StartElectionReason reason) {
invariant(!_voteRequester);
invariant(!_freshnessChecker);
switch (_rsConfigState) {
case kConfigSteady:
break;
case kConfigInitiating:
case kConfigReconfiguring:
case kConfigHBReconfiguring:
LOG(2) << "Not standing for election; processing a configuration change";
// Transition out of candidate role.
_topCoord->processLoseElection();
return;
default:
severe() << "Entered replica set election code while in illegal config state "
<< int(_rsConfigState);
fassertFailed(28641);
}
auto finishedEvent = _makeEvent();
if (!finishedEvent) {
return;
}
_electionFinishedEvent = finishedEvent;
auto dryRunFinishedEvent = _makeEvent();
if (!dryRunFinishedEvent) {
return;
}
_electionDryRunFinishedEvent = dryRunFinishedEvent;
LoseElectionDryRunGuardV1 lossGuard(this);
invariant(_rsConfig.getMemberAt(_selfIndex).isElectable());
const auto lastOpTime = _getMyLastAppliedOpTime_inlock();
if (lastOpTime == OpTime()) {
log() << "not trying to elect self, "
"do not yet have a complete set of data from any point in time";
return;
}
long long term = _topCoord->getTerm();
int primaryIndex = -1;
log() << "conducting a dry run election to see if we could be elected. current term: " << term;
_voteRequester.reset(new VoteRequester);
// Only set primaryIndex if the primary's vote is required during the dry run.
if (reason == TopologyCoordinator::StartElectionReason::kCatchupTakeover) {
primaryIndex = _topCoord->getCurrentPrimaryIndex();
}
StatusWith<executor::TaskExecutor::EventHandle> nextPhaseEvh =
_voteRequester->start(_replExecutor.get(),
_rsConfig,
_selfIndex,
term,
true, // dry run
lastOpTime,
primaryIndex);
if (nextPhaseEvh.getStatus() == ErrorCodes::ShutdownInProgress) {
return;
}
fassert(28685, nextPhaseEvh.getStatus());
_replExecutor
->onEvent(nextPhaseEvh.getValue(),
[=](const executor::TaskExecutor::CallbackArgs&) { _onDryRunComplete(term); })
.status_with_transitional_ignore();
lossGuard.dismiss();
}
void ReplicationCoordinatorImpl::_startElectSelf_inlock() {
invariant(!_freshnessChecker);
invariant(!_electCmdRunner);
switch (_rsConfigState) {
case kConfigSteady:
break;
case kConfigInitiating:
case kConfigReconfiguring:
case kConfigHBReconfiguring:
LOG(2) << "Not standing for election; processing a configuration change";
// Transition out of candidate role.
_topCoord->processLoseElection();
return;
default:
severe() << "Entered replica set election code while in illegal config state "
<< int(_rsConfigState);
fassertFailed(18913);
}
log() << "Standing for election";
const StatusWith<executor::TaskExecutor::EventHandle> finishEvh = _replExecutor->makeEvent();
if (finishEvh.getStatus() == ErrorCodes::ShutdownInProgress) {
return;
}
fassert(18680, finishEvh.getStatus());
_electionFinishedEvent = finishEvh.getValue();
LoseElectionGuard lossGuard(_topCoord.get(),
_replExecutor.get(),
&_freshnessChecker,
&_electCmdRunner,
&_electionFinishedEvent);
invariant(_rsConfig.getMemberAt(_selfIndex).isElectable());
OpTime lastOpTimeApplied(_getMyLastAppliedOpTime_inlock());
if (lastOpTimeApplied.isNull()) {
log() << "not trying to elect self, "
"do not yet have a complete set of data from any point in time"
" -- lastAppliedOpTime is null";
return;
}
_freshnessChecker.reset(new FreshnessChecker);
StatusWith<executor::TaskExecutor::EventHandle> nextPhaseEvh =
_freshnessChecker->start(_replExecutor.get(),
lastOpTimeApplied.getTimestamp(),
_rsConfig,
_selfIndex,
_topCoord->getMaybeUpHostAndPorts());
if (nextPhaseEvh.getStatus() == ErrorCodes::ShutdownInProgress) {
return;
}
fassert(18681, nextPhaseEvh.getStatus());
_replExecutor
->onEvent(nextPhaseEvh.getValue(),
stdx::bind(&ReplicationCoordinatorImpl::_onFreshnessCheckComplete, this))
.status_with_transitional_ignore();
lossGuard.dismiss();
}
void ReplicationCoordinatorImpl::_startElectSelfV1() {
invariant(!_voteRequester);
invariant(!_freshnessChecker);
stdx::unique_lock<stdx::mutex> lk(_mutex);
switch (_rsConfigState) {
case kConfigSteady:
break;
case kConfigInitiating:
case kConfigReconfiguring:
case kConfigHBReconfiguring:
LOG(2) << "Not standing for election; processing a configuration change";
// Transition out of candidate role.
_topCoord->processLoseElection();
return;
default:
severe() << "Entered replica set election code while in illegal config state "
<< int(_rsConfigState);
fassertFailed(28641);
}
auto finishedEvent = _makeEvent();
if (!finishedEvent) {
return;
}
_electionFinishedEvent = finishedEvent;
auto dryRunFinishedEvent = _makeEvent();
if (!dryRunFinishedEvent) {
return;
}
_electionDryRunFinishedEvent = dryRunFinishedEvent;
LoseElectionDryRunGuardV1 lossGuard(this);
invariant(_rsConfig.getMemberAt(_selfIndex).isElectable());
// Note: If we aren't durable, send last applied.
const auto lastOpTime = _isDurableStorageEngine() ? _getMyLastDurableOpTime_inlock()
: _getMyLastAppliedOpTime_inlock();
if (lastOpTime == OpTime()) {
log() << "not trying to elect self, "
"do not yet have a complete set of data from any point in time";
return;
}
log() << "conducting a dry run election to see if we could be elected";
_voteRequester.reset(new VoteRequester);
// This is necessary because the voteRequester may call directly into winning an
// election, if there are no other MaybeUp nodes. Winning an election attempts to lock
// _mutex again.
lk.unlock();
long long term = _topCoord->getTerm();
StatusWith<ReplicationExecutor::EventHandle> nextPhaseEvh =
_voteRequester->start(&_replExecutor,
_rsConfig,
_selfIndex,
_topCoord->getTerm(),
true, // dry run
lastOpTime);
if (nextPhaseEvh.getStatus() == ErrorCodes::ShutdownInProgress) {
return;
}
fassert(28685, nextPhaseEvh.getStatus());
_replExecutor.onEvent(nextPhaseEvh.getValue(),
stdx::bind(&ReplicationCoordinatorImpl::_onDryRunComplete, this, term));
lossGuard.dismiss();
}
