Status MigrationChunkClonerSourceLegacy::startClone(OperationContext* txn) {
invariant(!txn->lockState()->isLocked());
auto scopedGuard = MakeGuard([&] { cancelClone(txn); });
// Resolve the donor and recipient shards and their connection string
{
auto donorShardStatus = grid.shardRegistry()->getShard(txn, _args.getFromShardId());
if (!donorShardStatus.isOK()) {
return donorShardStatus.getStatus();
}
_donorCS = donorShardStatus.getValue()->getConnString();
}
{
auto recipientShardStatus = grid.shardRegistry()->getShard(txn, _args.getToShardId());
if (!recipientShardStatus.isOK()) {
return recipientShardStatus.getStatus();
}
auto recipientShard = recipientShardStatus.getValue();
auto shardHostStatus = recipientShard->getTargeter()->findHost(
ReadPreferenceSetting{ReadPreference::PrimaryOnly});
if (!shardHostStatus.isOK()) {
return shardHostStatus.getStatus();
}
_recipientHost = std::move(shardHostStatus.getValue());
}
// Prepare the currently available documents
Status status = _storeCurrentLocs(txn);
if (!status.isOK()) {
return status;
}
// Tell the recipient shard to start cloning
BSONObjBuilder cmdBuilder;
StartChunkCloneRequest::appendAsCommand(&cmdBuilder,
_args.getNss(),
_sessionId,
_args.getConfigServerCS(),
_donorCS,
_args.getFromShardId(),
_args.getToShardId(),
_args.getMinKey(),
_args.getMaxKey(),
_shardKeyPattern.toBSON(),
_args.getSecondaryThrottle());
auto responseStatus = _callRecipient(cmdBuilder.obj());
if (!responseStatus.isOK()) {
return responseStatus.getStatus();
}
scopedGuard.Dismiss();
return Status::OK();
}
Status MigrationChunkClonerSourceLegacy::commitClone(OperationContext* txn) {
invariant(!txn->lockState()->isLocked());
{
stdx::lock_guard<stdx::mutex> sl(_mutex);
invariant(!_cloneCompleted);
}
auto responseStatus = _callRecipient(createRecvChunkCommitRequest(_args.getNss(), _sessionId));
if (responseStatus.isOK()) {
_cleanup(txn);
return Status::OK();
}
cancelClone(txn);
return responseStatus.getStatus();
}
Status MigrationChunkClonerSourceLegacy::awaitUntilCriticalSectionIsAppropriate(
OperationContext* txn, Milliseconds maxTimeToWait) {
invariant(!txn->lockState()->isLocked());
auto scopedGuard = MakeGuard([&] { cancelClone(txn); });
const auto startTime = Date_t::now();
int iteration = 0;
while ((Date_t::now() - startTime) < maxTimeToWait) {
// Exponential sleep backoff, up to 1024ms. Don't sleep much on the first few iterations,
// since we want empty chunk migrations to be fast.
sleepmillis(1 << std::min(iteration, 10));
iteration++;
auto responseStatus = _callRecipient(BSON(kRecvChunkStatus << _args.getNss().ns()));
if (!responseStatus.isOK()) {
return {responseStatus.getStatus().code(),
str::stream()
<< "Failed to contact recipient shard to monitor data transfer due to "
<< responseStatus.getStatus().toString()};
}
BSONObj res = std::move(responseStatus.getValue());
log() << "moveChunk data transfer progress: " << res << " my mem used: " << _memoryUsed;
if (res["state"].String() == "steady") {
// Ensure all cloned docs have actually been transferred
const std::size_t locsRemaining = _cloneLocs.size();
if (locsRemaining != 0) {
return {
ErrorCodes::OperationIncomplete,
str::stream()
<< "cannot enter critical section before all data is cloned, "
<< locsRemaining
<< " locs were not transferred but to-shard thinks they are all cloned"};
}
scopedGuard.Dismiss();
return Status::OK();
}
if (res["state"].String() == "fail") {
return {ErrorCodes::OperationFailed, "Data transfer error"};
}
if (res["ns"].str() != _args.getNss().ns() || res["from"].str() != _donorCS.toString() ||
!res["min"].isABSONObj() || res["min"].Obj().woCompare(_args.getMinKey()) != 0 ||
!res["max"].isABSONObj() || res["max"].Obj().woCompare(_args.getMaxKey()) != 0) {
// This can happen when the destination aborted the migration and received another
// recvChunk before this thread sees the transition to the abort state. This is
// currently possible only if multiple migrations are happening at once. This is an
// unfortunate consequence of the shards not being able to keep track of multiple
// incoming and outgoing migrations.
return {ErrorCodes::OperationIncomplete,
"Destination shard aborted migration because a new one is running"};
}
if (_memoryUsed > 500 * 1024 * 1024) {
// This is too much memory for us to use so we're going to abort the migration
return {ErrorCodes::ExceededMemoryLimit,
"Aborting migration because of high memory usage"};
}
Status interruptStatus = txn->checkForInterruptNoAssert();
if (!interruptStatus.isOK()) {
return interruptStatus;
}
}
scopedGuard.Dismiss();
return {ErrorCodes::ExceededTimeLimit, "Timed out waiting for the cloner to catch up"};
}
