SERVER-49044 Make AsyncRequestSender not retry remote command requests with startTransaction=true

2 files changed, 76 insertions, 3 deletions

diff --git a/jstests/sharding/mongos_not_retry_commands_in_transactions.js b/jstests/sharding/mongos_not_retry_commands_in_transactions.js
new file mode 100644
index 00000000000..7fd6d8d5ad7
--- /dev/null
+++ b/jstests/sharding/mongos_not_retry_commands_in_transactions.js
@@ -0,0 +1,74 @@
+/*
+ * Tests that mongos doesn't retry commands with startTransaction=true.
+ * @tags: [requires_fcv_46]
+ */
+(function() {
+'use strict';
+
+const setCommandToFail = (nodeConnection, command, namespace) => {
+    return nodeConnection.adminCommand({
+        configureFailPoint: 'failCommand',
+        mode: {times: 1},
+        data: {
+            errorCode: ErrorCodes.InterruptedDueToReplStateChange,
+            failCommands: [command],
+            namespace,
+            failInternalCommands: true
+        }
+    });
+};
+
+const kDbName = "testDb";
+const kCollName = "testColl";
+const kNs = `${kDbName}.${kCollName}`;
+
+const kDoc0 = {
+    _id: 0
+};
+const kDoc1 = {
+    _id: 1
+};
+
+let transactionNumber = 1;
+
+const st = new ShardingTest({
+    mongos: 1,
+    shards: 1,
+    rs: {nodes: 1},
+});
+
+// Initializes test and inserts dummy document
+jsTest.log("Inserting test document.");
+const mongosDB = st.s0.startSession().getDatabase(kDbName);
+const primaryConnection = st.rs0.getPrimary();
+
+assert.commandWorked(mongosDB.runCommand({
+    insert: kCollName,
+    documents: [kDoc0],
+}));
+
+// Set the failCommand failpoint to make the next 'find' command fail once due to a failover.
+// Start a transaction & execute a find command.
+// It should fail once due to the 'failCommand' failpoint and should not be retried.
+jsTest.log(
+    "Testing that mongos doesn't retry the read command with startTransaction=true on replication set failover.");
+assert.commandWorked(setCommandToFail(primaryConnection, "find", kNs));
+
+assert.commandFailedWithCode(mongosDB.runCommand({
+    find: kCollName,
+    filter: kDoc0,
+    startTransaction: true,
+    txnNumber: NumberLong(transactionNumber++),
+    stmtId: NumberInt(0),
+    autocommit: false
+}),
+                             ErrorCodes.InterruptedDueToReplStateChange);
+
+jsTest.log("Testing that mongos retries retryable writes on failover.");
+assert.commandWorked(setCommandToFail(primaryConnection, "insert", kNs));
+
+assert.commandWorked(mongosDB.runCommand(
+    {insert: kCollName, documents: [kDoc1], txnNumber: NumberLong(transactionNumber++)}));
+
+st.stop();
+})();
\ No newline at end of file
diff --git a/src/mongo/s/async_requests_sender.cpp b/src/mongo/s/async_requests_sender.cpp
index 36f91459d29..7cca30f0b99 100644
--- a/src/mongo/s/async_requests_sender.cpp
+++ b/src/mongo/s/async_requests_sender.cpp
@@ -272,9 +272,9 @@ auto AsyncRequestsSender::RemoteData::handleResponse(RemoteCommandOnAnyCallbackA
         }
 
         shard->updateReplSetMonitor(failedTargets.front(), status);
-
+        bool isStartingTransaction = _cmdObj.getField("startTransaction").booleanSafe();
         if (!_ars->_stopRetrying && shard->isRetriableError(status.code(), _ars->_retryPolicy) &&
-            _retryCount < kMaxNumFailedHostRetryAttempts) {
+            _retryCount < kMaxNumFailedHostRetryAttempts && !isStartingTransaction) {
 
             LOGV2_DEBUG(4615637,
                         1,
@@ -284,7 +284,6 @@ auto AsyncRequestsSender::RemoteData::handleResponse(RemoteCommandOnAnyCallbackA
                         "shardId"_attr = _shardId,
                         "hosts"_attr = failedTargets,
                         "error"_attr = redact(status));
-
             ++_retryCount;
             _shardHostAndPort.reset();
             // retry through recursion