SERVER-25126 Return a different error code if the step down occurs after the write

12 files changed, 26 insertions, 24 deletions

diff --git a/src/mongo/base/error_codes.err b/src/mongo/base/error_codes.err
index 59a8c6acaaa..2ffffeae5c5 100644
--- a/src/mongo/base/error_codes.err
+++ b/src/mongo/base/error_codes.err
@@ -187,6 +187,7 @@ error_code("ClientMetadataDocumentTooLarge", 185)
 error_code("ClientMetadataCannotBeMutated", 186)
 error_code("LinearizableReadConcernError", 187)
 error_code("IncompatibleServerVersion", 188)
+error_code("PrimarySteppedDown", 189)
 
 # Non-sequential error codes (for compatibility only)
 error_code("SocketException", 9001)
diff --git a/src/mongo/client/remote_command_retry_scheduler.cpp b/src/mongo/client/remote_command_retry_scheduler.cpp
index ded4a10d964..f72874a33a4 100644
--- a/src/mongo/client/remote_command_retry_scheduler.cpp
+++ b/src/mongo/client/remote_command_retry_scheduler.cpp
@@ -95,6 +95,7 @@ const std::initializer_list<ErrorCodes::Error> RemoteCommandRetryScheduler::kAll
     ErrorCodes::HostUnreachable,
     ErrorCodes::HostNotFound,
     ErrorCodes::NetworkTimeout,
+    ErrorCodes::PrimarySteppedDown,
     ErrorCodes::InterruptedDueToReplStateChange};
 
 std::unique_ptr<RemoteCommandRetryScheduler::RetryPolicy>
diff --git a/src/mongo/db/cloner.cpp b/src/mongo/db/cloner.cpp
index a9e9d763871..3ea687fdd5d 100644
--- a/src/mongo/db/cloner.cpp
+++ b/src/mongo/db/cloner.cpp
@@ -279,7 +279,7 @@ void Cloner::copy(OperationContext* txn,
                      options);
     }
 
-    uassert(ErrorCodes::NotMaster,
+    uassert(ErrorCodes::PrimarySteppedDown,
             str::stream() << "Not primary while cloning collection " << from_collection.ns()
                           << " to "
                           << to_collection.ns()
@@ -311,7 +311,7 @@ void Cloner::copyIndexes(OperationContext* txn,
         }
     }
 
-    uassert(ErrorCodes::NotMaster,
+    uassert(ErrorCodes::PrimarySteppedDown,
             str::stream() << "Not primary while copying indexes from " << from_collection.ns()
                           << " to "
                           << to_collection.ns()
@@ -382,7 +382,7 @@ bool Cloner::copyCollection(OperationContext* txn,
     ScopedTransaction transaction(txn, MODE_IX);
     Lock::DBLock dbWrite(txn->lockState(), dbname, MODE_X);
 
-    uassert(ErrorCodes::NotMaster,
+    uassert(ErrorCodes::PrimarySteppedDown,
             str::stream() << "Not primary while copying collection " << ns << " (Cloner)",
             !txn->writesAreReplicated() ||
                 repl::getGlobalReplicationCoordinator()->canAcceptWritesFor(nss));
diff --git a/src/mongo/db/commands/mr.cpp b/src/mongo/db/commands/mr.cpp
index f25289baecd..26b40571fd0 100644
--- a/src/mongo/db/commands/mr.cpp
+++ b/src/mongo/db/commands/mr.cpp
@@ -465,8 +465,8 @@ void State::prepTempCollection() {
         OldClientWriteContext tempCtx(_txn, _config.tempNamespace);
         WriteUnitOfWork wuow(_txn);
         NamespaceString tempNss(_config.tempNamespace);
-        uassert(ErrorCodes::NotMaster,
-                "no longer master",
+        uassert(ErrorCodes::PrimarySteppedDown,
+                "no longer primary",
                 repl::getGlobalReplicationCoordinator()->canAcceptWritesFor(tempNss));
         Collection* tempColl = tempCtx.getCollection();
         invariant(!tempColl);
@@ -711,8 +711,8 @@ void State::insert(const string& ns, const BSONObj& o) {
         OldClientWriteContext ctx(_txn, ns);
         WriteUnitOfWork wuow(_txn);
         NamespaceString nss(ns);
-        uassert(ErrorCodes::NotMaster,
-                "no longer master",
+        uassert(ErrorCodes::PrimarySteppedDown,
+                "no longer primary",
                 repl::getGlobalReplicationCoordinator()->canAcceptWritesFor(nss));
         Collection* coll = getCollectionOrUassert(ctx.db(), ns);
 
diff --git a/src/mongo/db/exec/update.cpp b/src/mongo/db/exec/update.cpp
index 72c84ac6f7e..5041ff1ceb3 100644
--- a/src/mongo/db/exec/update.cpp
+++ b/src/mongo/db/exec/update.cpp
@@ -1010,7 +1010,7 @@ Status UpdateStage::restoreUpdateState() {
         !repl::getGlobalReplicationCoordinator()->canAcceptWritesFor(nsString);
 
     if (userInitiatedWritesAndNotPrimary) {
-        return Status(ErrorCodes::NotMaster,
+        return Status(ErrorCodes::PrimarySteppedDown,
                       str::stream() << "Demoted from primary while performing update on "
                                     << nsString.ns());
     }
diff --git a/src/mongo/db/ops/update.cpp b/src/mongo/db/ops/update.cpp
index c808d9aab34..5e0763f9eac 100644
--- a/src/mongo/db/ops/update.cpp
+++ b/src/mongo/db/ops/update.cpp
@@ -94,7 +94,7 @@ UpdateResult update(OperationContext* txn, Database* db, const UpdateRequest& re
                 !repl::getGlobalReplicationCoordinator()->canAcceptWritesFor(nsString);
 
             if (userInitiatedWritesAndNotPrimary) {
-                uassertStatusOK(Status(ErrorCodes::NotMaster,
+                uassertStatusOK(Status(ErrorCodes::PrimarySteppedDown,
                                        str::stream() << "Not primary while creating collection "
                                                      << nsString.ns()
                                                      << " during upsert"));
diff --git a/src/mongo/db/ops/write_ops_exec.cpp b/src/mongo/db/ops/write_ops_exec.cpp
index 95b73b51873..52db4a5674a 100644
--- a/src/mongo/db/ops/write_ops_exec.cpp
+++ b/src/mongo/db/ops/write_ops_exec.cpp
@@ -167,7 +167,7 @@ private:
 };
 
 void assertCanWrite_inlock(OperationContext* txn, const NamespaceString& ns) {
-    uassert(ErrorCodes::NotMaster,
+    uassert(ErrorCodes::PrimarySteppedDown,
             str::stream() << "Not primary while writing to " << ns.ns(),
             repl::ReplicationCoordinator::get(txn->getServiceContext())->canAcceptWritesFor(ns));
     CollectionShardingState::get(txn, ns)->checkShardVersionOrThrow(txn);
diff --git a/src/mongo/db/query/get_executor.cpp b/src/mongo/db/query/get_executor.cpp
index 4742f915d0d..71f0753aa0a 100644
--- a/src/mongo/db/query/get_executor.cpp
+++ b/src/mongo/db/query/get_executor.cpp
@@ -704,7 +704,7 @@ StatusWith<unique_ptr<PlanExecutor>> getExecutorDelete(OperationContext* txn,
         !repl::getGlobalReplicationCoordinator()->canAcceptWritesFor(nss);
 
     if (userInitiatedWritesAndNotPrimary) {
-        return Status(ErrorCodes::NotMaster,
+        return Status(ErrorCodes::PrimarySteppedDown,
                       str::stream() << "Not primary while removing from " << nss.ns());
     }
 
@@ -870,7 +870,7 @@ StatusWith<unique_ptr<PlanExecutor>> getExecutorUpdate(OperationContext* txn,
         !repl::getGlobalReplicationCoordinator()->canAcceptWritesFor(nsString);
 
     if (userInitiatedWritesAndNotPrimary) {
-        return Status(ErrorCodes::NotMaster,
+        return Status(ErrorCodes::PrimarySteppedDown,
                       str::stream() << "Not primary while performing update on " << nsString.ns());
     }
 
diff --git a/src/mongo/db/repl/replication_coordinator_impl.cpp b/src/mongo/db/repl/replication_coordinator_impl.cpp
index 6f4c438b255..acf4179bff8 100644
--- a/src/mongo/db/repl/replication_coordinator_impl.cpp
+++ b/src/mongo/db/repl/replication_coordinator_impl.cpp
@@ -1625,9 +1625,9 @@ ReplicationCoordinator::StatusAndDuration ReplicationCoordinatorImpl::_awaitRepl
     }
 
     if (replMode == modeReplSet && !_memberState.primary()) {
-        return StatusAndDuration(
-            Status(ErrorCodes::NotMaster, "Not master while waiting for replication"),
-            Milliseconds(timer->millis()));
+        return StatusAndDuration(Status(ErrorCodes::PrimarySteppedDown,
+                                        "Primary stepped down while waiting for replication"),
+                                 Milliseconds(timer->millis()));
     }
 
     if (writeConcern.wMode.empty()) {
@@ -1651,11 +1651,10 @@ ReplicationCoordinator::StatusAndDuration ReplicationCoordinatorImpl::_awaitRepl
         }
 
         if (replMode == modeReplSet && !_getMemberState_inlock().primary()) {
-            return StatusAndDuration(Status(ErrorCodes::NotMaster,
-                                            "Not master anymore while waiting for replication"
-                                            " - this most likely means that a step down"
-                                            " occurred while waiting for replication"),
-                                     elapsed);
+            return StatusAndDuration(
+                Status(ErrorCodes::PrimarySteppedDown,
+                       "Not primary anymore while waiting for replication - primary stepped down"),
+                elapsed);
         }
 
         if (writeConcern.wTimeout != WriteConcernOptions::kNoTimeout &&
diff --git a/src/mongo/db/repl/replication_coordinator_impl_test.cpp b/src/mongo/db/repl/replication_coordinator_impl_test.cpp
index bd3267cb172..659096ed571 100644
--- a/src/mongo/db/repl/replication_coordinator_impl_test.cpp
+++ b/src/mongo/db/repl/replication_coordinator_impl_test.cpp
@@ -655,7 +655,7 @@ TEST_F(ReplCoordTest, NodeReturnsNotMasterWhenRunningAwaitReplicationAgainstASec
     // Node should fail to awaitReplication when not primary.
     ReplicationCoordinator::StatusAndDuration statusAndDur =
         getReplCoord()->awaitReplication(txn.get(), time, writeConcern);
-    ASSERT_EQUALS(ErrorCodes::NotMaster, statusAndDur.status);
+    ASSERT_EQUALS(ErrorCodes::PrimarySteppedDown, statusAndDur.status);
 }
 
 TEST_F(ReplCoordTest, NodeReturnsOkWhenRunningAwaitReplicationAgainstPrimaryWithWZero) {
@@ -1321,7 +1321,7 @@ TEST_F(ReplCoordTest, NodeReturnsNotMasterWhenSteppingDownBeforeSatisfyingAWrite
     ASSERT_OK(getReplCoord()->setLastAppliedOptime_forTest(2, 2, time1));
     getReplCoord()->stepDown(txn.get(), true, Milliseconds(0), Milliseconds(1000));
     ReplicationCoordinator::StatusAndDuration statusAndDur = awaiter.getResult();
-    ASSERT_EQUALS(ErrorCodes::NotMaster, statusAndDur.status);
+    ASSERT_EQUALS(ErrorCodes::PrimarySteppedDown, statusAndDur.status);
     awaiter.reset();
 }
 
diff --git a/src/mongo/s/client/shard_remote.cpp b/src/mongo/s/client/shard_remote.cpp
index 006647b748c..6ae61ab1e19 100644
--- a/src/mongo/s/client/shard_remote.cpp
+++ b/src/mongo/s/client/shard_remote.cpp
@@ -137,7 +137,8 @@ void ShardRemote::updateReplSetMonitor(const HostAndPort& remoteHost,
         return;
 
     if (ErrorCodes::isNotMasterError(remoteCommandStatus.code()) ||
-        (remoteCommandStatus == ErrorCodes::InterruptedDueToReplStateChange)) {
+        (remoteCommandStatus == ErrorCodes::InterruptedDueToReplStateChange) ||
+        (remoteCommandStatus == ErrorCodes::PrimarySteppedDown)) {
         _targeter->markHostNotMaster(remoteHost);
     } else if (ErrorCodes::isNetworkError(remoteCommandStatus.code())) {
         _targeter->markHostUnreachable(remoteHost);
diff --git a/src/mongo/s/write_ops/batch_downconvert.cpp b/src/mongo/s/write_ops/batch_downconvert.cpp
index 1ece78ecdd5..2d792351b60 100644
--- a/src/mongo/s/write_ops/batch_downconvert.cpp
+++ b/src/mongo/s/write_ops/batch_downconvert.cpp
@@ -90,7 +90,7 @@ Status extractGLEErrors(const BSONObj& gleResponse, GLEErrors* errors) {
                // 2.6 Error codes
                ||
                code == ErrorCodes::NotMaster || code == ErrorCodes::UnknownReplWriteConcern ||
-               code == ErrorCodes::WriteConcernFailed) {
+               code == ErrorCodes::WriteConcernFailed || code == ErrorCodes::PrimarySteppedDown) {
         // Write concern errors that get returned as regular errors (result may not be ok: 1.0)
         errors->wcError.reset(new WriteConcernErrorDetail());
         errors->wcError->setErrCode(ErrorCodes::fromInt(code));