SERVER-41615: set flag to denote that the coordinator has recovered.

(cherry picked from commit 398471a4fde3cdd7ca57f6bbf601ad66998b1150)

6 files changed, 56 insertions, 5 deletions

diff --git a/src/mongo/db/s/single_transaction_coordinator_stats.cpp b/src/mongo/db/s/single_transaction_coordinator_stats.cpp
index 1ef6209f3c9..585e1c97188 100644
--- a/src/mongo/db/s/single_transaction_coordinator_stats.cpp
+++ b/src/mongo/db/s/single_transaction_coordinator_stats.cpp
@@ -96,6 +96,10 @@ void SingleTransactionCoordinatorStats::setDeletingCoordinatorDocStartTime(
     _deletingCoordinatorDocStartWallClockTime = curWallClockTime;
 }
 
+void SingleTransactionCoordinatorStats::setRecoveredFromFailover() {
+    _hasRecoveredFromFailover = true;
+}
+
 Microseconds SingleTransactionCoordinatorStats::getDurationSinceCreation(
     TickSource* tickSource, TickSource::Tick curTick) const {
     invariant(_createTime);
@@ -198,6 +202,7 @@ void SingleTransactionCoordinatorStats::reportMetrics(BSONObjBuilder& parent,
 
     invariant(_createTime);
     parent.append("commitStartTime", _createWallClockTime);
+    parent.append("hasRecoveredFromFailover", _hasRecoveredFromFailover);
 
     if (_writingParticipantListStartTime) {
         const auto statValue = getWritingParticipantListDuration(tickSource, curTick);
@@ -243,5 +248,4 @@ void SingleTransactionCoordinatorStats::reportLastClient(BSONObjBuilder& parent)
     parent.append("appName", _lastClientInfo.appName);
     parent.append("clientMetadata", _lastClientInfo.clientMetadata);
 }
-
 }  // namespace mongo
diff --git a/src/mongo/db/s/single_transaction_coordinator_stats.h b/src/mongo/db/s/single_transaction_coordinator_stats.h
index 1856bf9b054..a6d8be9b54f 100644
--- a/src/mongo/db/s/single_transaction_coordinator_stats.h
+++ b/src/mongo/db/s/single_transaction_coordinator_stats.h
@@ -281,6 +281,10 @@ public:
         invariant(client);
         _lastClientInfo.update(client);
     }
+    /*
+     * Marks this transaction coordinator has having recovered from failure.
+     */
+    void setRecoveredFromFailover();
 
 private:
     Date_t _createWallClockTime;
@@ -307,6 +311,7 @@ private:
     TickSource::Tick _endTime{0};
 
     LastClientInfo _lastClientInfo;
+    bool _hasRecoveredFromFailover = false;
 };
 
 }  // namespace mongo
diff --git a/src/mongo/db/s/transaction_coordinator.cpp b/src/mongo/db/s/transaction_coordinator.cpp
index 7cfe3f90231..903ac4a71fc 100644
--- a/src/mongo/db/s/transaction_coordinator.cpp
+++ b/src/mongo/db/s/transaction_coordinator.cpp
@@ -145,7 +145,6 @@ TransactionCoordinator::TransactionCoordinator(OperationContext* operationContex
 
                 _step = Step::kWritingParticipantList;
 
-                // TODO: The duration will be meaningless after failover.
                 _transactionCoordinatorMetricsObserver->onStartWritingParticipantList(
                     ServerTransactionCoordinatorsMetrics::get(_serviceContext),
                     _serviceContext->getTickSource(),
@@ -184,7 +183,6 @@ TransactionCoordinator::TransactionCoordinator(OperationContext* operationContex
 
                 _step = Step::kWaitingForVotes;
 
-                // TODO: The duration will be meaningless after failover.
                 _transactionCoordinatorMetricsObserver->onStartWaitingForVotes(
                     ServerTransactionCoordinatorsMetrics::get(_serviceContext),
                     _serviceContext->getTickSource(),
@@ -226,7 +224,6 @@ TransactionCoordinator::TransactionCoordinator(OperationContext* operationContex
 
                 _step = Step::kWritingDecision;
 
-                // TODO: The duration will be meaningless after failover.
                 _transactionCoordinatorMetricsObserver->onStartWritingDecision(
                     ServerTransactionCoordinatorsMetrics::get(_serviceContext),
                     _serviceContext->getTickSource(),
@@ -259,7 +256,6 @@ TransactionCoordinator::TransactionCoordinator(OperationContext* operationContex
 
                 _step = Step::kWaitingForDecisionAcks;
 
-                // TODO: The duration will be meaningless after failover.
                 _transactionCoordinatorMetricsObserver->onStartWaitingForDecisionAcks(
                     ServerTransactionCoordinatorsMetrics::get(_serviceContext),
                     _serviceContext->getTickSource(),
@@ -342,6 +338,8 @@ void TransactionCoordinator::continueCommit(const TransactionCoordinatorDocument
     if (!_reserveKickOffCommitPromise())
         return;
 
+    _transactionCoordinatorMetricsObserver->onRecoveryFromFailover();
+
     _participants = std::move(doc.getParticipants());
     if (doc.getDecision()) {
         _participantsDurable = true;
diff --git a/src/mongo/db/s/transaction_coordinator_metrics_observer.cpp b/src/mongo/db/s/transaction_coordinator_metrics_observer.cpp
index 42b82236430..dffdfdcba19 100644
--- a/src/mongo/db/s/transaction_coordinator_metrics_observer.cpp
+++ b/src/mongo/db/s/transaction_coordinator_metrics_observer.cpp
@@ -51,6 +51,12 @@ void TransactionCoordinatorMetricsObserver::onCreate(
     serverTransactionCoordinatorsMetrics->incrementTotalCreated();
 }
 
+
+void TransactionCoordinatorMetricsObserver::onRecoveryFromFailover() {
+    _singleTransactionCoordinatorStats.setRecoveredFromFailover();
+}
+
+
 void TransactionCoordinatorMetricsObserver::onStartWritingParticipantList(
     ServerTransactionCoordinatorsMetrics* serverTransactionCoordinatorsMetrics,
     TickSource* tickSource,
diff --git a/src/mongo/db/s/transaction_coordinator_metrics_observer.h b/src/mongo/db/s/transaction_coordinator_metrics_observer.h
index 9d73e64f6c7..535c865da0b 100644
--- a/src/mongo/db/s/transaction_coordinator_metrics_observer.h
+++ b/src/mongo/db/s/transaction_coordinator_metrics_observer.h
@@ -50,6 +50,11 @@ public:
                   Date_t curWallClockTime);
 
     /**
+     * Called when the transaction coordinator has recovered and continues the commit.
+     */
+    void onRecoveryFromFailover();
+
+    /**
      * Updates relevant metrics when a transaction coordinator is about to write the participant
      * list.
      */
diff --git a/src/mongo/db/s/transaction_coordinator_test.cpp b/src/mongo/db/s/transaction_coordinator_test.cpp
index 4d7487db52d..f91977bf16e 100644
--- a/src/mongo/db/s/transaction_coordinator_test.cpp
+++ b/src/mongo/db/s/transaction_coordinator_test.cpp
@@ -2324,6 +2324,39 @@ TEST_F(TransactionCoordinatorMetricsTest,
     checkServerStatus();
 }
 
+TEST_F(TransactionCoordinatorMetricsTest, RecoveryFromFailureIndicatedInReportState) {
+    TransactionCoordinator coordinator(
+        operationContext(),
+        _lsid,
+        _txnNumber,
+        std::make_unique<txn::AsyncWorkScheduler>(getServiceContext()),
+        Date_t::max());
+
+    const auto assertRecoveryFlag = [&coordinator](bool expectedFlagValue) {
+        BSONObjBuilder builder;
+        coordinator.reportState(builder);
+        auto reportDoc = builder.obj();
+        auto coordinatorDoc = reportDoc.getObjectField("twoPhaseCommitCoordinator");
+        ASSERT_EQ(coordinatorDoc.getBoolField("hasRecoveredFromFailover"), expectedFlagValue);
+    };
+
+    assertRecoveryFlag(false);
+
+    TransactionCoordinatorDocument coordinatorDoc;
+    coordinatorDoc.setParticipants(kTwoShardIdList);
+    coordinator.continueCommit(coordinatorDoc);
+
+    assertRecoveryFlag(true);
+
+    assertPrepareSentAndRespondWithSuccess();
+    assertPrepareSentAndRespondWithSuccess();
+
+    assertCommitSentAndRespondWithSuccess();
+    assertCommitSentAndRespondWithSuccess();
+
+    coordinator.onCompletion().get();
+}
+
 TEST_F(TransactionCoordinatorMetricsTest, ClientInformationIncludedInReportState) {
     const auto expectedAppName = std::string("Foo");
     associateClientMetadata(getClient(), expectedAppName);