SERVER-28677 InitialSyncer should reset optimes on every attempt

4 files changed, 36 insertions, 0 deletions

diff --git a/src/mongo/db/repl/initial_syncer.cpp b/src/mongo/db/repl/initial_syncer.cpp
index 3923d3402b7..66742d8cada 100644
--- a/src/mongo/db/repl/initial_syncer.cpp
+++ b/src/mongo/db/repl/initial_syncer.cpp
@@ -218,6 +218,7 @@ InitialSyncer::InitialSyncer(
     uassert(ErrorCodes::BadValue, "invalid replication process", _replicationProcess);
     uassert(ErrorCodes::BadValue, "invalid getMyLastOptime function", _opts.getMyLastOptime);
     uassert(ErrorCodes::BadValue, "invalid setMyLastOptime function", _opts.setMyLastOptime);
+    uassert(ErrorCodes::BadValue, "invalid resetOptimes function", _opts.resetOptimes);
     uassert(ErrorCodes::BadValue, "invalid getSlaveDelay function", _opts.getSlaveDelay);
     uassert(ErrorCodes::BadValue, "invalid sync source selector", _opts.syncSourceSelector);
     uassert(ErrorCodes::BadValue, "callback function cannot be null", _onCompletion);
@@ -447,8 +448,12 @@ void InitialSyncer::_startInitialSyncAttemptCallback(
     LOG(2) << "Resetting sync source so a new one can be chosen for this initial sync attempt.";
     _syncSource = HostAndPort();
 
+    // Reset all optimes before a new initial sync attempt.
+    _opts.resetOptimes();
     _lastApplied = {};
     _lastFetched = {};
+
+    // Clear the oplog buffer.
     _oplogBuffer->clear(makeOpCtx().get());
 
     // Get sync source.
diff --git a/src/mongo/db/repl/initial_syncer.h b/src/mongo/db/repl/initial_syncer.h
index b438a0c97e1..8ee406d8876 100644
--- a/src/mongo/db/repl/initial_syncer.h
+++ b/src/mongo/db/repl/initial_syncer.h
@@ -81,6 +81,9 @@ struct InitialSyncerOptions {
     /** Function to update optime of last operation applied on this node */
     using SetMyLastOptimeFn = stdx::function<void(const OpTime&)>;
 
+    /** Function to reset all optimes on this node (e.g. applied & durable). */
+    using ResetOptimesFn = stdx::function<void()>;
+
     /** Function to sets this node into a specific follower mode. */
     using SetFollowerModeFn = stdx::function<bool(const MemberState&)>;
 
@@ -109,6 +112,7 @@ struct InitialSyncerOptions {
 
     GetMyLastOptimeFn getMyLastOptime;
     SetMyLastOptimeFn setMyLastOptime;
+    ResetOptimesFn resetOptimes;
     GetSlaveDelayFn getSlaveDelay;
 
     SyncSourceSelector* syncSourceSelector = nullptr;
diff --git a/src/mongo/db/repl/initial_syncer_test.cpp b/src/mongo/db/repl/initial_syncer_test.cpp
index 570c57840e5..c549bfbe391 100644
--- a/src/mongo/db/repl/initial_syncer_test.cpp
+++ b/src/mongo/db/repl/initial_syncer_test.cpp
@@ -293,6 +293,7 @@ protected:
         options.initialSyncRetryWait = Milliseconds(1);
         options.getMyLastOptime = [this]() { return _myLastOpTime; };
         options.setMyLastOptime = [this](const OpTime& opTime) { _setMyLastOptime(opTime); };
+        options.resetOptimes = [this]() { _setMyLastOptime(OpTime()); };
         options.getSlaveDelay = [this]() { return Seconds(0); };
         options.syncSourceSelector = this;
 
@@ -526,6 +527,7 @@ TEST_F(InitialSyncerTest, InvalidConstruction) {
     InitialSyncerOptions options;
     options.getMyLastOptime = []() { return OpTime(); };
     options.setMyLastOptime = [](const OpTime&) {};
+    options.resetOptimes = []() {};
     options.getSlaveDelay = []() { return Seconds(0); };
     options.syncSourceSelector = this;
     auto callback = [](const StatusWith<OpTimeWithHash>&) {};
@@ -732,6 +734,30 @@ TEST_F(InitialSyncerTest,
         << progress;
 }
 
+TEST_F(InitialSyncerTest, InitialSyncerResetsOptimesOnNewAttempt) {
+    auto initialSyncer = &getInitialSyncer();
+    auto opCtx = makeOpCtx();
+
+    _syncSourceSelector->setChooseNewSyncSourceResult_forTest(HostAndPort());
+
+    const std::uint32_t initialSyncMaxAttempts = 1U;
+    ASSERT_OK(initialSyncer->startup(opCtx.get(), initialSyncMaxAttempts));
+
+    auto net = getNet();
+    auto origOptime = OpTime(Timestamp(1000, 1), 1);
+
+    _setMyLastOptime(origOptime);
+
+    // Simulate a failed initial sync attempt
+    _simulateChooseSyncSourceFailure(net, _options.syncSourceRetryWait);
+    advanceClock(net, _options.initialSyncRetryWait);
+
+    initialSyncer->join();
+
+    // Make sure the initial sync attempt reset optimes.
+    ASSERT_EQUALS(OpTime(), _options.getMyLastOptime());
+}
+
 TEST_F(InitialSyncerTest,
        InitialSyncerReturnsCallbackCanceledIfShutdownWhileRetryingSyncSourceSelection) {
     auto initialSyncer = &getInitialSyncer();
diff --git a/src/mongo/db/repl/replication_coordinator_impl.cpp b/src/mongo/db/repl/replication_coordinator_impl.cpp
index 1dbb34349bf..18121b41c07 100644
--- a/src/mongo/db/repl/replication_coordinator_impl.cpp
+++ b/src/mongo/db/repl/replication_coordinator_impl.cpp
@@ -265,6 +265,7 @@ InitialSyncerOptions createInitialSyncerOptions(
         replCoord->setMyLastAppliedOpTime(opTime);
         externalState->setGlobalTimestamp(replCoord->getServiceContext(), opTime.getTimestamp());
     };
+    options.resetOptimes = [replCoord]() { replCoord->resetMyLastOpTimes(); };
     options.getSlaveDelay = [replCoord]() { return replCoord->getSlaveDelaySecs(); };
     options.syncSourceSelector = replCoord;
     options.replBatchLimitBytes = dur::UncommittedBytesLimit;