diff options
Diffstat (limited to 'src/mongo/db/repl/initial_syncer.cpp')
| -rw-r--r-- | src/mongo/db/repl/initial_syncer.cpp | 100 |
1 files changed, 56 insertions, 44 deletions
diff --git a/src/mongo/db/repl/initial_syncer.cpp b/src/mongo/db/repl/initial_syncer.cpp index 02fb4af0be5..d8a2f5104d7 100644 --- a/src/mongo/db/repl/initial_syncer.cpp +++ b/src/mongo/db/repl/initial_syncer.cpp @@ -548,6 +548,7 @@ void InitialSyncer::_setUp_inlock(OperationContext* opCtx, std::uint32_t initial _stats.initialSyncStart = _exec->now(); _stats.maxFailedInitialSyncAttempts = initialSyncMaxAttempts; _stats.failedInitialSyncAttempts = 0; + _stats.exec = std::weak_ptr<executor::TaskExecutor>(_exec); _allowedOutageDuration = Seconds(initialSyncTransientErrorRetryPeriodSeconds.load()); } @@ -1393,53 +1394,63 @@ void InitialSyncer::_lastOplogEntryFetcherCallbackForStopTimestamp( std::shared_ptr<OnCompletionGuard> onCompletionGuard) { OpTimeAndWallTime resultOpTimeAndWallTime = {OpTime(), Date_t()}; { - stdx::lock_guard<Latch> lock(_mutex); - auto status = _checkForShutdownAndConvertStatus_inlock( - result.getStatus(), "error fetching last oplog entry for stop timestamp"); - if (_shouldRetryError(lock, status)) { - auto scheduleStatus = - (*_attemptExec) - ->scheduleWork([this, - onCompletionGuard](executor::TaskExecutor::CallbackArgs args) { - // It is not valid to schedule the retry from within this callback, - // hence we schedule a lambda to schedule the retry. - stdx::lock_guard<Latch> lock(_mutex); - // Since the stopTimestamp is retrieved after we have done all the work of - // retrieving collection data, we handle retries within this class by - // retrying for 'initialSyncTransientErrorRetryPeriodSeconds' (default 24 - // hours). This is the same retry strategy used when retrieving collection - // data, and avoids retrieving all the data and then throwing it away due to - // a transient network outage. - auto status = _scheduleLastOplogEntryFetcher_inlock( - [=](const StatusWith<mongo::Fetcher::QueryResponse>& status, - mongo::Fetcher::NextAction*, - mongo::BSONObjBuilder*) { - _lastOplogEntryFetcherCallbackForStopTimestamp(status, - onCompletionGuard); - }, - kInitialSyncerHandlesRetries); - if (!status.isOK()) { - onCompletionGuard->setResultAndCancelRemainingWork_inlock(lock, status); - } - }); - if (scheduleStatus.isOK()) + { + stdx::lock_guard<Latch> lock(_mutex); + auto status = _checkForShutdownAndConvertStatus_inlock( + result.getStatus(), "error fetching last oplog entry for stop timestamp"); + if (_shouldRetryError(lock, status)) { + auto scheduleStatus = + (*_attemptExec) + ->scheduleWork( + [this, onCompletionGuard](executor::TaskExecutor::CallbackArgs args) { + // It is not valid to schedule the retry from within this callback, + // hence we schedule a lambda to schedule the retry. + stdx::lock_guard<Latch> lock(_mutex); + // Since the stopTimestamp is retrieved after we have done all the + // work of retrieving collection data, we handle retries within this + // class by retrying for + // 'initialSyncTransientErrorRetryPeriodSeconds' (default 24 hours). + // This is the same retry strategy used when retrieving collection + // data, and avoids retrieving all the data and then throwing it + // away due to a transient network outage. + auto status = _scheduleLastOplogEntryFetcher_inlock( + [=](const StatusWith<mongo::Fetcher::QueryResponse>& status, + mongo::Fetcher::NextAction*, + mongo::BSONObjBuilder*) { + _lastOplogEntryFetcherCallbackForStopTimestamp( + status, onCompletionGuard); + }, + kInitialSyncerHandlesRetries); + if (!status.isOK()) { + onCompletionGuard->setResultAndCancelRemainingWork_inlock( + lock, status); + } + }); + if (scheduleStatus.isOK()) + return; + // If scheduling failed, we're shutting down and cannot retry. + // So just continue with the original failed status. + } + if (!status.isOK()) { + onCompletionGuard->setResultAndCancelRemainingWork_inlock(lock, status); return; - // If scheduling failed, we're shutting down and cannot retry. - // So just continue with the original failed status. - } - if (!status.isOK()) { - onCompletionGuard->setResultAndCancelRemainingWork_inlock(lock, status); - return; - } + } - auto&& optimeStatus = parseOpTimeAndWallTime(result); - if (!optimeStatus.isOK()) { - onCompletionGuard->setResultAndCancelRemainingWork_inlock(lock, - optimeStatus.getStatus()); - return; + auto&& optimeStatus = parseOpTimeAndWallTime(result); + if (!optimeStatus.isOK()) { + onCompletionGuard->setResultAndCancelRemainingWork_inlock(lock, + optimeStatus.getStatus()); + return; + } + resultOpTimeAndWallTime = optimeStatus.getValue(); } - resultOpTimeAndWallTime = optimeStatus.getValue(); + // Release the _mutex to write to disk. + auto opCtx = makeOpCtx(); + _replicationProcess->getConsistencyMarkers()->setMinValid( + opCtx.get(), resultOpTimeAndWallTime.opTime, true); + + stdx::lock_guard<Latch> lock(_mutex); _initialSyncState->stopTimestamp = resultOpTimeAndWallTime.opTime.getTimestamp(); // If the beginFetchingTimestamp is different from the stopTimestamp, it indicates that @@ -2186,9 +2197,10 @@ void InitialSyncer::Stats::append(BSONObjBuilder* builder) const { builder->appendNumber("maxFailedInitialSyncAttempts", static_cast<long long>(maxFailedInitialSyncAttempts)); + auto e = exec.lock(); if (initialSyncStart != Date_t()) { builder->appendDate("initialSyncStart", initialSyncStart); - auto elapsedDurationEnd = Date_t::now(); + auto elapsedDurationEnd = e ? e->now() : Date_t::now(); if (initialSyncEnd != Date_t()) { builder->appendDate("initialSyncEnd", initialSyncEnd); elapsedDurationEnd = initialSyncEnd; |