SERVER-37972 Create suite for running multi shard jscore with sharded transactions with kill

3 files changed, 23 insertions, 10 deletions

diff --git a/src/mongo/db/s/transaction_coordinator.cpp b/src/mongo/db/s/transaction_coordinator.cpp
index 5a1d0dfd41c..6be5d1bf69a 100644
--- a/src/mongo/db/s/transaction_coordinator.cpp
+++ b/src/mongo/db/s/transaction_coordinator.cpp
@@ -36,6 +36,7 @@
 #include "mongo/db/logical_clock.h"
 #include "mongo/db/s/transaction_coordinator_document_gen.h"
 #include "mongo/db/s/transaction_coordinator_futures_util.h"
+#include "mongo/util/fail_point_service.h"
 #include "mongo/util/log.h"
 
 namespace mongo {
@@ -205,7 +206,15 @@ Future<CoordinatorCommitDecision> TransactionCoordinator::_runPhaseOne(
 Future<void> TransactionCoordinator::_runPhaseTwo(const std::vector<ShardId>& participantShards,
                                                   const CoordinatorCommitDecision& decision) {
     return _sendDecisionToParticipants(participantShards, decision)
-        .then([this] { return _driver.deleteCoordinatorDoc(_lsid, _txnNumber); })
+        .then([this] {
+            if (getGlobalFailPointRegistry()
+                    ->getFailPoint("doNotForgetCoordinator")
+                    ->shouldFail()) {
+                return Future<void>::makeReady();
+            }
+
+            return _driver.deleteCoordinatorDoc(_lsid, _txnNumber);
+        })
         .then([this] {
             LOG(3) << "Two-phase commit completed for session " << _lsid.toBSON()
                    << ", transaction number " << _txnNumber;
diff --git a/src/mongo/db/s/transaction_coordinator_catalog.cpp b/src/mongo/db/s/transaction_coordinator_catalog.cpp
index 3fa862ddde3..38e6f95cb30 100644
--- a/src/mongo/db/s/transaction_coordinator_catalog.cpp
+++ b/src/mongo/db/s/transaction_coordinator_catalog.cpp
@@ -186,7 +186,6 @@ void TransactionCoordinatorCatalog::_remove(const LogicalSessionId& lsid, TxnNum
 
             if (MONGO_FAIL_POINT(doNotForgetCoordinator)) {
                 auto decisionFuture = coordinator->getDecision();
-                invariant(decisionFuture.isReady());
                 // Only remember a coordinator that completed successfully. We expect that the
                 // coordinator only completes with an error if the node stepped down or was shut
                 // down while coordinating the commit. If either of these occurred, a
@@ -195,7 +194,7 @@ void TransactionCoordinatorCatalog::_remove(const LogicalSessionId& lsid, TxnNum
                 // shutdown), or should find no coordinator and instead recover the decision from
                 // the local participant (if the failover or shutdown occurred before any of the
                 // coordinator's state was made durable).
-                if (decisionFuture.getNoThrow().isOK()) {
+                if (decisionFuture.isReady() && decisionFuture.getNoThrow().isOK()) {
                     _coordinatorsBySessionDefunct[lsid][txnNumber] = std::move(coordinator);
                 }
             }
diff --git a/src/mongo/s/transaction_router.cpp b/src/mongo/s/transaction_router.cpp
index 0975508bfa9..51ca27d08b1 100644
--- a/src/mongo/s/transaction_router.cpp
+++ b/src/mongo/s/transaction_router.cpp
@@ -613,17 +613,23 @@ Shard::CommandResponse TransactionRouter::_commitMultiShardTransaction(Operation
         coordinatorShard = Grid::get(opCtx)->shardRegistry()->getConfigShard();
 
         if (!_initiatedTwoPhaseCommit) {
+            SharedTransactionOptions options;
+            options.txnNumber = _txnNumber;
+            // Intentionally leave atClusterTime blank since we don't care and to minimize
+            // possibility that storage engine won't have it available.
+            Participant configParticipant(true, 0, options);
+
             // Send a fake transaction statement to the config server primary so that the config
             // server primary sets up state in memory to receive coordinateCommit.
             auto cmdResponse = coordinatorShard->runCommandWithFixedRetryAttempts(
                 opCtx,
                 ReadPreferenceSetting{ReadPreference::PrimaryOnly},
                 "dummy",
-                coordinatorIter->second.attachTxnFieldsIfNeeded(BSON("distinct"
-                                                                     << "dummy"
-                                                                     << "key"
-                                                                     << "dummy"),
-                                                                true),
+                configParticipant.attachTxnFieldsIfNeeded(BSON("distinct"
+                                                               << "dummy"
+                                                               << "key"
+                                                               << "dummy"),
+                                                          true),
                 Shard::RetryPolicy::kIdempotent);
             uassertStatusOK(Shard::CommandResponse::getEffectiveStatus(cmdResponse));
 
@@ -633,8 +639,7 @@ Shard::CommandResponse TransactionRouter::_commitMultiShardTransaction(Operation
                 opCtx,
                 ReadPreferenceSetting{ReadPreference::PrimaryOnly},
                 "admin",
-                coordinatorIter->second.attachTxnFieldsIfNeeded(BSON("abortTransaction" << 1),
-                                                                false),
+                configParticipant.attachTxnFieldsIfNeeded(BSON("abortTransaction" << 1), false),
                 Shard::RetryPolicy::kIdempotent);
             uassertStatusOK(Shard::CommandResponse::getEffectiveStatus(cmdResponse));
         }