Status ShardingStateRecovery::recover(OperationContext* txn) {
BSONObj recoveryDocBSON;
try {
AutoGetCollection autoColl(txn, NamespaceString::kConfigCollectionNamespace, MODE_IS);
if (!Helpers::findOne(
txn, autoColl.getCollection(), RecoveryDocument::getQuery(), recoveryDocBSON)) {
return Status::OK();
}
} catch (const DBException& ex) {
return ex.toStatus();
}
const auto recoveryDocStatus = RecoveryDocument::fromBSON(recoveryDocBSON);
if (!recoveryDocStatus.isOK())
return recoveryDocStatus.getStatus();
const auto recoveryDoc = std::move(recoveryDocStatus.getValue());
log() << "Sharding state recovery process found document " << recoveryDoc.toBSON();
// Make sure the sharding state is initialized
ShardingState* const shardingState = ShardingState::get(txn);
shardingState->initialize(txn, recoveryDoc.getConfigsvr().toString());
shardingState->setShardName(recoveryDoc.getShardName());
if (!recoveryDoc.getMinOpTimeUpdaters()) {
// Treat the minOpTime as up-to-date
grid.shardRegistry()->advanceConfigOpTime(recoveryDoc.getMinOpTime());
return Status::OK();
}
log() << "Sharding state recovery document indicates there were "
<< recoveryDoc.getMinOpTimeUpdaters()
<< " metadata change operations in flight. Contacting the config server primary in order "
"to retrieve the most recent opTime.";
// Need to fetch the latest uptime from the config server, so do a logging write
Status status =
grid.catalogManager(txn)->logChange(txn,
"Sharding recovery thread",
"Sharding minOpTime recovery",
NamespaceString::kConfigCollectionNamespace.ns(),
recoveryDocBSON);
if (!status.isOK())
return status;
log() << "Sharding state recovered. New config server opTime is "
<< grid.shardRegistry()->getConfigOpTime();
// Finally, clear the recovery document so next time we don't need to recover
status = modifyRecoveryDocument(txn, RecoveryDocument::Clear, kMajorityWriteConcern);
if (!status.isOK()) {
warning() << "Failed to reset sharding state recovery document due to " << status;
}
return Status::OK();
}
Status ShardingStateRecovery::recover(OperationContext* txn) {
if (serverGlobalParams.clusterRole != ClusterRole::ShardServer) {
return Status::OK();
}
BSONObj recoveryDocBSON;
try {
AutoGetCollection autoColl(txn, NamespaceString::kConfigCollectionNamespace, MODE_IS);
if (!Helpers::findOne(
txn, autoColl.getCollection(), RecoveryDocument::getQuery(), recoveryDocBSON)) {
return Status::OK();
}
} catch (const DBException& ex) {
return ex.toStatus();
}
const auto recoveryDocStatus = RecoveryDocument::fromBSON(recoveryDocBSON);
if (!recoveryDocStatus.isOK())
return recoveryDocStatus.getStatus();
const auto recoveryDoc = std::move(recoveryDocStatus.getValue());
log() << "Sharding state recovery process found document " << recoveryDoc.toBSON();
// Make sure the sharding state is initialized
ShardingState* const shardingState = ShardingState::get(txn);
// For backwards compatibility. Shards added by v3.4 cluster should have been initialized by
// the shard identity document.
// TODO(SERER-25276): Remove this after 3.4 since 3.4 shards should always have ShardingState
// initialized by this point.
if (!shardingState->enabled()) {
shardingState->initializeFromConfigConnString(txn, recoveryDoc.getConfigsvr().toString());
shardingState->setShardName(recoveryDoc.getShardName());
}
if (!recoveryDoc.getMinOpTimeUpdaters()) {
// Treat the minOpTime as up-to-date
grid.advanceConfigOpTime(recoveryDoc.getMinOpTime());
return Status::OK();
}
log() << "Sharding state recovery document indicates there were "
<< recoveryDoc.getMinOpTimeUpdaters()
<< " metadata change operations in flight. Contacting the config server primary in order "
"to retrieve the most recent opTime.";
// Need to fetch the latest uptime from the config server, so do a logging write
Status status =
grid.catalogClient(txn)->logChange(txn,
"Sharding minOpTime recovery",
NamespaceString::kConfigCollectionNamespace.ns(),
recoveryDocBSON,
ShardingCatalogClient::kMajorityWriteConcern);
if (!status.isOK())
return status;
log() << "Sharding state recovered. New config server opTime is " << grid.configOpTime();
// Finally, clear the recovery document so next time we don't need to recover
status = modifyRecoveryDocument(txn, RecoveryDocument::Clear, kLocalWriteConcern);
if (!status.isOK()) {
warning() << "Failed to reset sharding state recovery document due to " << status;
}
return Status::OK();
}
