SERVER-52680 Start replication when leaving REMOVED state

(cherry picked from commit 73ab98a9094de18b82e596e8d1d0bf311858548b)

2 files changed, 63 insertions, 2 deletions

diff --git a/jstests/replsets/single_node_set_new_hostname.js b/jstests/replsets/single_node_set_new_hostname.js
new file mode 100644
index 00000000000..484cf65a1cf
--- /dev/null
+++ b/jstests/replsets/single_node_set_new_hostname.js
@@ -0,0 +1,50 @@
+/*
+ * When a one-node set is restarted on a different port and reconfigured with the new port, it
+ * should re-elect itself.
+ *
+ * @tags: [
+ *     requires_persistence,
+ * ]
+ */
+(function() {
+"use strict";
+
+load("jstests/replsets/rslib.js");
+
+const replTest = new ReplSetTest({nodes: 1});
+replTest.startSet();
+replTest.initiate();
+replTest.getPrimary();
+
+/*
+ * Prepare to restart the sole node on a new port, it no longer finds itself in the old config.
+ */
+const config = replTest.getReplSetConfigFromNode(0);
+const newPort = replTest.getPort(0) + 1;
+const hostname = config.members[0].host.split(":")[0];
+const newHostAndPort = `${hostname}:${newPort}`;
+
+jsTestLog("Restarting the sole node on a new port: " + newPort);
+replTest.restart(0, {port: newPort});
+let restartedNode;
+assert.soonNoExcept(() => {
+    restartedNode = new Mongo(newHostAndPort);
+    return true;
+}, `Couldn't connect to restarted node "${newHostAndPort}`);
+waitForState(restartedNode, ReplSetTest.State.REMOVED);
+
+/*
+ * Update the config to match the node's new port.
+ */
+jsTestLog("Reconfiguring the set to change the sole node's port.");
+jsTestLog(`Original config: ${tojson(config)}`);
+config.version++;
+config.members[0].host = newHostAndPort;
+jsTestLog(`New config: ${tojson(config)}`);
+// Force reconfig since the restarted node is in REMOVED state, not PRIMARY.
+assert.commandWorked(
+    restartedNode.getDB("admin").runCommand({replSetReconfig: config, force: true}));
+waitForState(restartedNode, ReplSetTest.State.PRIMARY);
+
+replTest.stopSet();
+}());
diff --git a/src/mongo/db/repl/replication_coordinator_impl.cpp b/src/mongo/db/repl/replication_coordinator_impl.cpp
index 5ddb1475f06..7f022cbb0e4 100644
--- a/src/mongo/db/repl/replication_coordinator_impl.cpp
+++ b/src/mongo/db/repl/replication_coordinator_impl.cpp
@@ -681,11 +681,11 @@ void ReplicationCoordinatorImpl::_stopDataReplication(OperationContext* opCtx) {
 
 void ReplicationCoordinatorImpl::_startDataReplication(OperationContext* opCtx,
                                                        stdx::function<void()> startCompleted) {
-    if (_startedSteadyStateReplication.load()) {
+    if (_startedSteadyStateReplication.swap(true)) {
+        // This is not the first call.
         return;
     }
 
-    _startedSteadyStateReplication.store(true);
     // Check to see if we need to do an initial sync.
     const auto lastOpTime = getMyLastAppliedOpTime();
     const auto needsInitialSync =
@@ -3031,6 +3031,17 @@ ReplicationCoordinatorImpl::_updateMemberStateFromTopologyCoordinator(WithLock l
         _cancelPriorityTakeover_inlock();
     }
 
+    // Ensure replication is running if we are no longer REMOVED.
+    if (_memberState.removed() && !newState.arbiter()) {
+        log() << "Scheduling a task to begin or continue replication";
+        _scheduleWorkAt(_replExecutor->now(),
+                        [=](const mongo::executor::TaskExecutor::CallbackArgs& cbData) {
+                            _externalState->startThreads(_settings);
+                            auto opCtx = cc().makeOperationContext();
+                            _startDataReplication(opCtx.get());
+                        });
+    }
+
     log() << "transition to " << newState << " from " << _memberState << rsLog;
     // Initializes the featureCompatibilityVersion to the latest value, because arbiters do not
     // receive the replicated version. This is to avoid bugs like SERVER-32639.