From 9fcb1f6911dc7904163515f45b361a1333a55310 Mon Sep 17 00:00:00 2001
From: "A. Jesse Jiryu Davis" <jesse@mongodb.com>
Date: Mon, 2 Nov 2020 17:05:22 -0500
Subject: SERVER-33747 Fix crash when arbiter restarts and enters REMOVED

(cherry picked from commit 72aacd4ffaf6500777a8a51f87b0797f8ea8ad0b)
---
 jstests/replsets/arbiter_new_hostname.js           | 46 ++++++++++++++++++++++
 src/mongo/db/repl/replication_coordinator_impl.cpp |  7 +++-
 src/mongo/db/repl/replication_coordinator_impl.h   |  2 +
 .../replication_coordinator_impl_heartbeat.cpp     |  2 +-
 4 files changed, 55 insertions(+), 2 deletions(-)
 create mode 100644 jstests/replsets/arbiter_new_hostname.js

diff --git a/jstests/replsets/arbiter_new_hostname.js b/jstests/replsets/arbiter_new_hostname.js
new file mode 100644
index 00000000000..4850cacead9
--- /dev/null
+++ b/jstests/replsets/arbiter_new_hostname.js
@@ -0,0 +1,46 @@
+/*
+ * An arbiter that is stopped and restarted on a different port and rejoins the
+ * replica set should enter removed state and should not start data replication.
+ *
+ * @tags: [
+ *     requires_fcv_49,
+ * ]
+ */
+(function() {
+    "use strict";
+    const replTest = new ReplSetTest({name: 'test', nodes: 3});
+    replTest.startSet();
+    const nodes = replTest.nodeList();
+    let config = {
+        "_id": "test",
+        "members": [
+            {"_id": 0, "host": nodes[0]},
+            {"_id": 1, "host": nodes[1]},
+            {"_id": 2, "host": nodes[2], arbiterOnly: true}
+        ]
+    };
+    replTest.initiate(config);
+
+    let primary = replTest.getPrimary();
+    replTest.awaitReplication();
+    replTest.awaitSecondaryNodes();
+
+    const arbiterId = 2;
+    const newPort = replTest.getPort(arbiterId) + 1;
+    jsTestLog("Restarting the arbiter node on a new port: " + newPort);
+    replTest.stop(arbiterId);
+    replTest.start(arbiterId, {port: newPort}, true);
+
+    jsTestLog("Reconfiguring the set to change the arbiter's port.");
+    config = replTest.getReplSetConfigFromNode();
+    jsTestLog(`Original config: ${tojson(config)}`);
+
+    const hostname = config.members[arbiterId].host.split(":")[0];
+    config.version++;
+    config.members[arbiterId].host = hostname + ":" + newPort;
+    jsTestLog(`New config: ${tojson(config)}`);
+    assert.commandWorked(primary.getDB("admin").runCommand({replSetReconfig: config}));
+    replTest.awaitReplication();
+    replTest.awaitNodesAgreeOnConfigVersion();
+    replTest.stopSet();
+}());
diff --git a/src/mongo/db/repl/replication_coordinator_impl.cpp b/src/mongo/db/repl/replication_coordinator_impl.cpp
index a6a4d0084bb..dc7ba40996a 100644
--- a/src/mongo/db/repl/replication_coordinator_impl.cpp
+++ b/src/mongo/db/repl/replication_coordinator_impl.cpp
@@ -623,7 +623,7 @@ void ReplicationCoordinatorImpl::_finishLoadLocalConfig(
         _performPostMemberStateUpdateAction(action);
     }
 
-    if (!isArbiter) {
+    if (!isArbiter && myIndex.getValue() != -1) {
         _externalState->startThreads(_settings);
         _startDataReplication(opCtx.get());
     }
@@ -652,6 +652,11 @@ void ReplicationCoordinatorImpl::_stopDataReplication(OperationContext* opCtx) {
 
 void ReplicationCoordinatorImpl::_startDataReplication(OperationContext* opCtx,
                                                        stdx::function<void()> startCompleted) {
+    if (_startedSteadyStateReplication.load()) {
+        return;
+    }
+
+    _startedSteadyStateReplication.store(true);
     // Check to see if we need to do an initial sync.
     const auto lastOpTime = getMyLastAppliedOpTime();
     const auto needsInitialSync =
diff --git a/src/mongo/db/repl/replication_coordinator_impl.h b/src/mongo/db/repl/replication_coordinator_impl.h
index 0127e2807aa..105269b8855 100644
--- a/src/mongo/db/repl/replication_coordinator_impl.h
+++ b/src/mongo/db/repl/replication_coordinator_impl.h
@@ -1383,6 +1383,8 @@ private:
     // here so we can update our term to match as part of finishing stepdown.
     boost::optional<long long> _pendingTermUpdateDuringStepDown;  // (M)
 
+    AtomicWord<bool> _startedSteadyStateReplication{false};
+
     // If we're in terminal shutdown.  If true, we'll refuse to vote in elections.
     bool _inTerminalShutdown = false;  // (M)
 };
diff --git a/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp b/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp
index 1934ad0c40b..36293c867ba 100644
--- a/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp
+++ b/src/mongo/db/repl/replication_coordinator_impl_heartbeat.cpp
@@ -550,7 +550,7 @@ void ReplicationCoordinatorImpl::_heartbeatReconfigStore(
             }
         }
 
-        if (!isArbiter && isFirstConfig) {
+        if (!isArbiter && myIndex.isOK() && myIndex.getValue() != -1) {
             shouldStartDataReplication = true;
         }
 
-- 
cgit v1.2.1