Status waitForWriteConcern(OperationContext* txn,
const OpTime& replOpTime,
const WriteConcernOptions& writeConcern,
WriteConcernResult* result) {
// We assume all options have been validated earlier, if not, programming error
dassert(validateWriteConcern(writeConcern).isOK());
// Next handle blocking on disk
Timer syncTimer;
switch (writeConcern.syncMode) {
case WriteConcernOptions::NONE:
break;
case WriteConcernOptions::FSYNC: {
StorageEngine* storageEngine = getGlobalServiceContext()->getGlobalStorageEngine();
if (!storageEngine->isDurable()) {
result->fsyncFiles = storageEngine->flushAllFiles(true);
} else {
// We only need to commit the journal if we're durable
txn->recoveryUnit()->waitUntilDurable();
}
break;
}
case WriteConcernOptions::JOURNAL:
txn->recoveryUnit()->waitUntilDurable();
break;
}
result->syncMillis = syncTimer.millis();
// Now wait for replication
if (replOpTime.isNull()) {
// no write happened for this client yet
return Status::OK();
}
// needed to avoid incrementing gleWtimeStats SERVER-9005
if (writeConcern.wNumNodes <= 1 && writeConcern.wMode.empty()) {
// no desired replication check
return Status::OK();
}
// Now we wait for replication
// Note that replica set stepdowns and gle mode changes are thrown as errors
repl::ReplicationCoordinator::StatusAndDuration replStatus =
repl::getGlobalReplicationCoordinator()->awaitReplication(txn, replOpTime, writeConcern);
if (replStatus.status == ErrorCodes::WriteConcernFailed) {
gleWtimeouts.increment();
result->err = "timeout";
result->wTimedOut = true;
}
// Add stats
result->writtenTo = repl::getGlobalReplicationCoordinator()->getHostsWrittenTo(replOpTime);
gleWtimeStats.recordMillis(durationCount<Milliseconds>(replStatus.duration));
result->wTime = durationCount<Milliseconds>(replStatus.duration);
return replStatus.status;
}
Status waitForWriteConcern(OperationContext* txn,
const OpTime& replOpTime,
const WriteConcernOptions& writeConcern,
WriteConcernResult* result) {
LOG(2) << "Waiting for write concern. OpTime: " << replOpTime
<< ", write concern: " << writeConcern.toBSON();
auto replCoord = repl::ReplicationCoordinator::get(txn);
MONGO_FAIL_POINT_PAUSE_WHILE_SET(hangBeforeWaitingForWriteConcern);
// Next handle blocking on disk
Timer syncTimer;
WriteConcernOptions writeConcernWithPopulatedSyncMode =
replCoord->populateUnsetWriteConcernOptionsSyncMode(writeConcern);
switch (writeConcernWithPopulatedSyncMode.syncMode) {
case WriteConcernOptions::SyncMode::UNSET:
severe() << "Attempting to wait on a WriteConcern with an unset sync option";
fassertFailed(34410);
case WriteConcernOptions::SyncMode::NONE:
break;
case WriteConcernOptions::SyncMode::FSYNC: {
StorageEngine* storageEngine = getGlobalServiceContext()->getGlobalStorageEngine();
if (!storageEngine->isDurable()) {
result->fsyncFiles = storageEngine->flushAllFiles(true);
} else {
// We only need to commit the journal if we're durable
txn->recoveryUnit()->waitUntilDurable();
}
break;
}
case WriteConcernOptions::SyncMode::JOURNAL:
if (replCoord->getReplicationMode() != repl::ReplicationCoordinator::Mode::modeNone) {
// Wait for ops to become durable then update replication system's
// knowledge of this.
OpTime appliedOpTime = replCoord->getMyLastAppliedOpTime();
txn->recoveryUnit()->waitUntilDurable();
replCoord->setMyLastDurableOpTimeForward(appliedOpTime);
} else {
txn->recoveryUnit()->waitUntilDurable();
}
break;
}
result->syncMillis = syncTimer.millis();
// Now wait for replication
if (replOpTime.isNull()) {
// no write happened for this client yet
return Status::OK();
}
// needed to avoid incrementing gleWtimeStats SERVER-9005
if (writeConcernWithPopulatedSyncMode.wNumNodes <= 1 &&
writeConcernWithPopulatedSyncMode.wMode.empty()) {
// no desired replication check
return Status::OK();
}
// Replica set stepdowns and gle mode changes are thrown as errors
repl::ReplicationCoordinator::StatusAndDuration replStatus =
replCoord->awaitReplication(txn, replOpTime, writeConcernWithPopulatedSyncMode);
if (replStatus.status == ErrorCodes::WriteConcernFailed) {
gleWtimeouts.increment();
result->err = "timeout";
result->wTimedOut = true;
}
// Add stats
result->writtenTo = repl::getGlobalReplicationCoordinator()->getHostsWrittenTo(
replOpTime,
writeConcernWithPopulatedSyncMode.syncMode == WriteConcernOptions::SyncMode::JOURNAL);
gleWtimeStats.recordMillis(durationCount<Milliseconds>(replStatus.duration));
result->wTime = durationCount<Milliseconds>(replStatus.duration);
return replStatus.status;
}
Status waitForWriteConcern(OperationContext* txn,
const OpTime& replOpTime,
const WriteConcernOptions& writeConcern,
WriteConcernResult* result) {
// We assume all options have been validated earlier, if not, programming error.
// Passing localDB name is a hack to avoid more rigorous check that performed for non local DB.
dassert(validateWriteConcern(txn, writeConcern, kLocalDB).isOK());
// We should never be waiting for write concern while holding any sort of lock, because this may
// lead to situations where the replication heartbeats are stalled.
//
// This check does not hold for writes done through dbeval because it runs with a global X lock.
dassert(!txn->lockState()->isLocked() || txn->getClient()->isInDirectClient());
// Next handle blocking on disk
Timer syncTimer;
auto replCoord = repl::getGlobalReplicationCoordinator();
WriteConcernOptions writeConcernWithPopulatedSyncMode =
replCoord->populateUnsetWriteConcernOptionsSyncMode(writeConcern);
switch (writeConcernWithPopulatedSyncMode.syncMode) {
case WriteConcernOptions::SyncMode::UNSET:
severe() << "Attempting to wait on a WriteConcern with an unset sync option";
fassertFailed(34410);
case WriteConcernOptions::SyncMode::NONE:
break;
case WriteConcernOptions::SyncMode::FSYNC: {
StorageEngine* storageEngine = getGlobalServiceContext()->getGlobalStorageEngine();
if (!storageEngine->isDurable()) {
result->fsyncFiles = storageEngine->flushAllFiles(true);
} else {
// We only need to commit the journal if we're durable
txn->recoveryUnit()->waitUntilDurable();
}
break;
}
case WriteConcernOptions::SyncMode::JOURNAL:
if (replCoord->getReplicationMode() != repl::ReplicationCoordinator::Mode::modeNone) {
// Wait for ops to become durable then update replication system's
// knowledge of this.
OpTime appliedOpTime = replCoord->getMyLastAppliedOpTime();
txn->recoveryUnit()->waitUntilDurable();
replCoord->setMyLastDurableOpTimeForward(appliedOpTime);
} else {
txn->recoveryUnit()->waitUntilDurable();
}
break;
}
result->syncMillis = syncTimer.millis();
// Now wait for replication
if (replOpTime.isNull()) {
// no write happened for this client yet
return Status::OK();
}
// needed to avoid incrementing gleWtimeStats SERVER-9005
if (writeConcernWithPopulatedSyncMode.wNumNodes <= 1 &&
writeConcernWithPopulatedSyncMode.wMode.empty()) {
// no desired replication check
return Status::OK();
}
// Now we wait for replication
// Note that replica set stepdowns and gle mode changes are thrown as errors
repl::ReplicationCoordinator::StatusAndDuration replStatus =
repl::getGlobalReplicationCoordinator()->awaitReplication(
txn, replOpTime, writeConcernWithPopulatedSyncMode);
if (replStatus.status == ErrorCodes::WriteConcernFailed) {
gleWtimeouts.increment();
result->err = "timeout";
result->wTimedOut = true;
}
// Add stats
result->writtenTo = repl::getGlobalReplicationCoordinator()->getHostsWrittenTo(
replOpTime,
writeConcernWithPopulatedSyncMode.syncMode == WriteConcernOptions::SyncMode::JOURNAL);
gleWtimeStats.recordMillis(durationCount<Milliseconds>(replStatus.duration));
result->wTime = durationCount<Milliseconds>(replStatus.duration);
return replStatus.status;
}
