From 536870ce9d135f90c2d8cd0b189b7a48dcd731f3 Mon Sep 17 00:00:00 2001
From: Jordi Serra Torrens <jordi.serra-torrens@mongodb.com>
Date: Fri, 26 Aug 2022 14:35:28 +0000
Subject: [PATCH] Repro SERVER-69181

---
 jstests/sharding/repro-server-69181.js      | 59 +++++++++++++++++++++
 src/mongo/db/s/move_primary_coordinator.cpp |  5 +-
 2 files changed, 63 insertions(+), 1 deletion(-)
 create mode 100644 jstests/sharding/repro-server-69181.js

diff --git a/jstests/sharding/repro-server-69181.js b/jstests/sharding/repro-server-69181.js
new file mode 100644
index 00000000000..1f603faae22
--- /dev/null
+++ b/jstests/sharding/repro-server-69181.js
@@ -0,0 +1,59 @@
+(function() {
+'use strict';
+
+load('jstests/libs/fail_point_util.js');  // For configureFailPoint
+load('jstests/libs/parallel_shell_helpers.js');
+
+const dbName = 'test';
+const collName = 'foo';
+const ns = dbName + '.' + collName;
+
+const st = new ShardingTest({shards: 2});
+
+const db = st.s.getDB(dbName);
+const unshardedColl = db['foo'];
+
+// Create the database, with shard0 as db-primary.
+assert.commandWorked(
+    st.s.adminCommand({enableSharding: dbName, primaryShard: st.shard0.shardName}));
+
+let hangBeforeCleanStaleData = configureFailPoint(st.shard0, 'hangBeforeCleanStaleData');
+
+// Create an unsharded collection on the db.
+assert.commandWorked(unshardedColl.insert({x: 0}));
+
+// Start move primary
+const awaitMovePrimary = startParallelShell(
+    funWithArgs(function(dbName, toShardName) {
+        assert.commandWorked(db.adminCommand({movePrimary: dbName, to: toShardName}));
+    }, dbName, st.shard1.shardName), st.s.port);
+
+// Wait for movePrimary to commit, and block before proceeding with cleaning the unsharded
+// collections on the source shard.
+hangBeforeCleanStaleData.wait();
+
+// Drop the database from the new primary (that is, shard1).
+assert.commandWorked(db.dropDatabase());
+
+// Now recreate the db on shard0 as db-primary.
+assert.commandWorked(
+    st.s.adminCommand({enableSharding: dbName, primaryShard: st.shard0.shardName}));
+
+// Do some writes on an unsharded collection on that db. Note, this collection needs to have the
+// same name as one that existed before movePrimary.
+assert.commandWorked(unshardedColl.insert({x: 1}));
+assert.commandWorked(unshardedColl.insert({x: 2}));
+assert.commandWorked(unshardedColl.insert({x: 3}));
+assert.eq(3, unshardedColl.find().itcount());
+
+// Let the movePrimary coordinator proceed with cleaning "stale" data.
+hangBeforeCleanStaleData.off();
+
+// Wait for the cleanup procedure to finish.
+awaitMovePrimary();
+
+// Check whether we lost documents.
+assert.eq(3, unshardedColl.find().itcount()); // This will fail! Writes were lost.
+
+st.stop();
+})();
diff --git a/src/mongo/db/s/move_primary_coordinator.cpp b/src/mongo/db/s/move_primary_coordinator.cpp
index a19a5ebb286..e552628fae6 100644
--- a/src/mongo/db/s/move_primary_coordinator.cpp
+++ b/src/mongo/db/s/move_primary_coordinator.cpp
@@ -40,12 +40,14 @@
 #include "mongo/logv2/log.h"
 #include "mongo/s/client/shard_registry.h"
 #include "mongo/s/grid.h"
+#include "mongo/util/fail_point.h"
 
 #define MONGO_LOGV2_DEFAULT_COMPONENT ::mongo::logv2::LogComponent::kSharding
 
-
 namespace mongo {
 
+MONGO_FAIL_POINT_DEFINE(hangBeforeCleanStaleData);
+
 void MovePrimaryCoordinator::appendCommandInfo(BSONObjBuilder* cmdInfoBuilder) const {
     stdx::lock_guard lk{_docMutex};
     cmdInfoBuilder->append("request", BSON(_doc.kToShardIdFieldName << _doc.getToShardId()));
@@ -109,6 +111,7 @@ ExecutorFuture<void> MovePrimaryCoordinator::_runImpl(
             uassertStatusOK(movePrimarySourceManager.clone(opCtx));
             uassertStatusOK(movePrimarySourceManager.enterCriticalSection(opCtx));
             uassertStatusOK(movePrimarySourceManager.commitOnConfig(opCtx));
+            hangBeforeCleanStaleData.pauseWhileSet();
             uassertStatusOK(movePrimarySourceManager.cleanStaleData(opCtx));
         })
         .onError([this, anchor = shared_from_this()](const Status& status) {
-- 
2.17.1