void CollectionCloner::wait() {
// If a fetcher is inactive, wait() has no effect.
_listIndexesFetcher.wait();
_findFetcher.wait();
waitForDbWorker();
}
void CollectionCloner::_handleARMResultsCallback(
const executor::TaskExecutor::CallbackArgs& cbd,
std::shared_ptr<OnCompletionGuard> onCompletionGuard) {
auto setResultAndCancelRemainingWork = [this](std::shared_ptr<OnCompletionGuard> guard,
Status status) {
stdx::lock_guard<stdx::mutex> lock(_mutex);
guard->setResultAndCancelRemainingWork_inlock(lock, status);
return;
};
if (!cbd.status.isOK()) {
// Wait for active inserts to complete.
waitForDbWorker();
Status newStatus = cbd.status.withContext(str::stream() << "Error querying collection '"
<< _sourceNss.ns());
setResultAndCancelRemainingWork(onCompletionGuard, cbd.status);
return;
}
// Pull the documents from the ARM into a buffer until the entire batch has been processed.
bool lastBatch;
{
UniqueLock lk(_mutex);
auto nextBatchStatus = _bufferNextBatchFromArm(lk);
if (!nextBatchStatus.isOK()) {
if (_options.uuid && (nextBatchStatus.code() == ErrorCodes::OperationFailed ||
nextBatchStatus.code() == ErrorCodes::CursorNotFound)) {
// With these errors, it's possible the collection was dropped while we were
// cloning. If so, we'll execute the drop during oplog application, so it's OK to
// just stop cloning. This is only safe if cloning by UUID; if we are cloning by
// name, we have no way to detect if the collection was dropped and another
// collection with the same name created in the interim.
_verifyCollectionWasDropped(lk, nextBatchStatus, onCompletionGuard, cbd.opCtx);
} else {
onCompletionGuard->setResultAndCancelRemainingWork_inlock(lk, nextBatchStatus);
}
return;
}
// Check if this is the last batch of documents to clone.
lastBatch = _arm->remotesExhausted();
}
// Schedule the next document batch insertion.
auto&& scheduleResult = _scheduleDbWorkFn([=](const executor::TaskExecutor::CallbackArgs& cbd) {
_insertDocumentsCallback(cbd, lastBatch, onCompletionGuard);
});
if (!scheduleResult.isOK()) {
Status newStatus = scheduleResult.getStatus().withContext(
str::stream() << "Error cloning collection '" << _sourceNss.ns() << "'");
setResultAndCancelRemainingWork(onCompletionGuard, scheduleResult.getStatus());
return;
}
MONGO_FAIL_POINT_BLOCK(initialSyncHangCollectionClonerAfterHandlingBatchResponse, nssData) {
const BSONObj& data = nssData.getData();
auto nss = data["nss"].str();
// Only hang when cloning the specified collection, or if no collection was specified.
if (nss.empty() || _destNss.toString() == nss) {
while (MONGO_FAIL_POINT(initialSyncHangCollectionClonerAfterHandlingBatchResponse) &&
!_isShuttingDown()) {
log() << "initialSyncHangCollectionClonerAfterHandlingBatchResponse fail point "
"enabled for "
<< _destNss.toString() << ". Blocking until fail point is disabled.";
mongo::sleepsecs(1);
}
}
}
// If the remote cursors are not exhausted, schedule this callback again to handle
// the impending cursor response.
if (!lastBatch) {
Status scheduleStatus = _scheduleNextARMResultsCallback(onCompletionGuard);
if (!scheduleStatus.isOK()) {
setResultAndCancelRemainingWork(onCompletionGuard, scheduleStatus);
return;
}
}
}
