SERVER-41779: reconstructPreparedTransactions should use readSource kNoTimestamp

(cherry picked from commit f1dcaea4a97903fa7c785f31c55f484d275a5aed)

6 files changed, 29 insertions, 10 deletions

diff --git a/src/mongo/db/catalog_raii.cpp b/src/mongo/db/catalog_raii.cpp
index d3e4c1f5e60..3b2b4ed417a 100644
--- a/src/mongo/db/catalog_raii.cpp
+++ b/src/mongo/db/catalog_raii.cpp
@@ -199,7 +199,9 @@ ConcealCollectionCatalogChangesBlock::~ConcealCollectionCatalogChangesBlock() {
     CollectionCatalog::get(_opCtx).onOpenCatalog(_opCtx);
 }
 
-ReadSourceScope::ReadSourceScope(OperationContext* opCtx)
+ReadSourceScope::ReadSourceScope(OperationContext* opCtx,
+                                 RecoveryUnit::ReadSource readSource,
+                                 boost::optional<Timestamp> provided)
     : _opCtx(opCtx), _originalReadSource(opCtx->recoveryUnit()->getTimestampReadSource()) {
 
     if (_originalReadSource == RecoveryUnit::ReadSource::kProvided) {
@@ -207,7 +209,7 @@ ReadSourceScope::ReadSourceScope(OperationContext* opCtx)
     }
 
     _opCtx->recoveryUnit()->abandonSnapshot();
-    _opCtx->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kUnset);
+    _opCtx->recoveryUnit()->setTimestampReadSource(readSource, provided);
 }
 
 ReadSourceScope::~ReadSourceScope() {
diff --git a/src/mongo/db/catalog_raii.h b/src/mongo/db/catalog_raii.h
index 8522f33c92b..5461d4af7e9 100644
--- a/src/mongo/db/catalog_raii.h
+++ b/src/mongo/db/catalog_raii.h
@@ -228,7 +228,9 @@ private:
  */
 class ReadSourceScope {
 public:
-    ReadSourceScope(OperationContext* opCtx);
+    ReadSourceScope(OperationContext* opCtx,
+                    RecoveryUnit::ReadSource readSource = RecoveryUnit::ReadSource::kUnset,
+                    boost::optional<Timestamp> provided = boost::none);
     ~ReadSourceScope();
 
 private:
diff --git a/src/mongo/db/repl/initial_syncer.cpp b/src/mongo/db/repl/initial_syncer.cpp
index bf5f20021f9..8cd4a86de6b 100644
--- a/src/mongo/db/repl/initial_syncer.cpp
+++ b/src/mongo/db/repl/initial_syncer.cpp
@@ -101,6 +101,9 @@ MONGO_FAIL_POINT_DEFINE(rsSyncApplyStop);
 // Failpoint which causes the initial sync function to hang afte cloning all databases.
 MONGO_FAIL_POINT_DEFINE(initialSyncHangAfterDataCloning);
 
+// Failpoint which causes the initial sync function to fail and hang before starting a new attempt.
+MONGO_FAIL_POINT_DEFINE(failAndHangInitialSync);
+
 namespace {
 using namespace executor;
 using CallbackArgs = executor::TaskExecutor::CallbackArgs;
@@ -1384,6 +1387,12 @@ void InitialSyncer::_finishInitialSyncAttempt(const StatusWith<OpTimeAndWallTime
     _stats.initialSyncAttemptInfos.emplace_back(
         InitialSyncer::InitialSyncAttemptInfo{runTime, result.getStatus(), _syncSource});
 
+    if (MONGO_FAIL_POINT(failAndHangInitialSync)) {
+        log() << "failAndHangInitialSync fail point enabled.";
+        MONGO_FAIL_POINT_PAUSE_WHILE_SET(failAndHangInitialSync);
+        result = Status(ErrorCodes::InternalError, "failAndHangInitialSync fail point enabled");
+    }
+
     if (result.isOK()) {
         // Scope guard will invoke _finishCallback().
         return;
diff --git a/src/mongo/db/repl/replication_coordinator_impl.cpp b/src/mongo/db/repl/replication_coordinator_impl.cpp
index 74112da3178..1ae959a1cda 100644
--- a/src/mongo/db/repl/replication_coordinator_impl.cpp
+++ b/src/mongo/db/repl/replication_coordinator_impl.cpp
@@ -1220,10 +1220,6 @@ void ReplicationCoordinatorImpl::_setMyLastAppliedOpTimeAndWallTime(
     _opTimeWaiterList.signalIf_inlock(
         [opTime](Waiter* waiter) { return waiter->opTime <= opTime; });
 
-    if (opTime.isNull()) {
-        return;
-    }
-
     // Update the local snapshot before updating the stable timestamp on the storage engine. New
     // transactions reading from the local snapshot should start before the oldest timestamp is
     // advanced to avoid races.
@@ -1232,6 +1228,10 @@ void ReplicationCoordinatorImpl::_setMyLastAppliedOpTimeAndWallTime(
     // Notify the oplog waiters after updating the local snapshot.
     signalOplogWaiters();
 
+    if (opTime.isNull()) {
+        return;
+    }
+
     // Add the new applied optime to the list of stable optime candidates and then set the last
     // stable optime. Stable optimes are used to determine the last optime that it is safe to revert
     // the database to, in the event of a rollback via the 'recover to timestamp' method. If we are
diff --git a/src/mongo/db/repl/transaction_oplog_application.cpp b/src/mongo/db/repl/transaction_oplog_application.cpp
index dae9faa2f7d..20c01c432e7 100644
--- a/src/mongo/db/repl/transaction_oplog_application.cpp
+++ b/src/mongo/db/repl/transaction_oplog_application.cpp
@@ -397,8 +397,11 @@ void reconstructPreparedTransactions(OperationContext* opCtx, repl::OplogApplica
         log() << "Hit skipReconstructPreparedTransactions failpoint";
         return;
     }
-    // Read the transactions table with its own snapshot and read timestamp.
-    ReadSourceScope readSourceScope(opCtx);
+    // Read the transactions table and the oplog collection without a timestamp.
+    // The below DBDirectClient read uses AutoGetCollectionForRead which could implicitly change the
+    // read source to kLastApplied. So we need to explicitly set the read source to kNoTimestamp to
+    // force reads in this scope to be untimestamped.
+    ReadSourceScope readSourceScope(opCtx, RecoveryUnit::ReadSource::kNoTimestamp);
 
     DBDirectClient client(opCtx);
     const auto cursor = client.query(NamespaceString::kSessionTransactionsTableNamespace,
diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_snapshot_manager.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_snapshot_manager.cpp
index 32299a71648..7216bc1727b 100644
--- a/src/mongo/db/storage/wiredtiger/wiredtiger_snapshot_manager.cpp
+++ b/src/mongo/db/storage/wiredtiger/wiredtiger_snapshot_manager.cpp
@@ -50,7 +50,10 @@ void WiredTigerSnapshotManager::setCommittedSnapshot(const Timestamp& timestamp)
 
 void WiredTigerSnapshotManager::setLocalSnapshot(const Timestamp& timestamp) {
     stdx::lock_guard<stdx::mutex> lock(_localSnapshotMutex);
-    _localSnapshot = timestamp;
+    if (timestamp.isNull())
+        _localSnapshot = boost::none;
+    else
+        _localSnapshot = timestamp;
 }
 
 boost::optional<Timestamp> WiredTigerSnapshotManager::getLocalSnapshot() {