From 800d15a94d5b3f90b5e467f80378408e76d9ee62 Mon Sep 17 00:00:00 2001
From: Jordi Serra Torrens <jordi.serra-torrens@mongodb.com>
Date: Mon, 16 May 2022 13:20:03 +0000
Subject: [PATCH] Repro BF-25230

---
 jstests/sharding/repro-bf-25230.js            | 62 +++++++++++++++++++
 .../db/s/migration_destination_manager.cpp    |  3 +
 2 files changed, 65 insertions(+)
 create mode 100644 jstests/sharding/repro-bf-25230.js

diff --git a/jstests/sharding/repro-bf-25230.js b/jstests/sharding/repro-bf-25230.js
new file mode 100644
index 00000000000..caacfee63dc
--- /dev/null
+++ b/jstests/sharding/repro-bf-25230.js
@@ -0,0 +1,62 @@
+(function() {
+"use strict";
+
+load('jstests/libs/parallel_shell_helpers.js');
+load("jstests/libs/fail_point_util.js");
+
+var st = new ShardingTest({shards: {rs0: {nodes: 1}, rs1: {nodes: 2}}});
+
+const dbName = "test";
+const collName = "foo";
+const ns = dbName + "." + collName;
+
+const db = st.s.getDB(dbName);
+
+assert.commandWorked(
+    st.s.adminCommand({enableSharding: dbName, primaryShard: st.shard0.shardName}));
+assert.commandWorked(st.s.adminCommand({shardCollection: ns, key: {x: 1}}));
+assert.commandWorked(st.s.adminCommand({split: ns, middle: {x: 0}}));
+
+// Start a migration. Make it hang on the recipient side after it has released the recipient
+// critical section but before it has deleted the recovery document.
+let fp = configureFailPoint(st.rs1.getPrimary(), 'migrationRecipientHangBeforeDeleteRecoveryDoc');
+
+const awaitMigration = startParallelShell(
+    funWithArgs(function(ns, toShard) {
+        assert.commandWorked(db.adminCommand({moveChunk: ns, find: {x: 0}, to: toShard}));
+    }, ns, st.shard1.shardName), st.s.port);
+
+fp.wait();
+
+// Set up a prepared txn
+const hangBeforeWritingDecisionFp = configureFailPoint(st.rs1.getPrimary(), 'hangBeforeWritingDecision');
+const awaitTxn = startParallelShell(
+    funWithArgs(function(dbName, collName) {
+        var session = db.getMongo().startSession();
+        var sessionDb = session.getDatabase(dbName)
+        session.startTransaction()
+        assert.commandWorked(sessionDb[collName].insert({x: 1}));
+        assert.commandWorked(sessionDb[collName].insert({x: -1}));
+        session.commitTransaction();
+    }, dbName, collName), st.s.port);
+
+hangBeforeWritingDecisionFp.wait();
+jsTest.log("--XXXX-- Prepared txn");
+
+// Stepdown the recipient
+assert.commandWorked(st.rs1.getPrimary().adminCommand({replSetStepDown: 60, force: true}));
+sleep(1*1000);
+fp.off();
+hangBeforeWritingDecisionFp.off();
+
+// Check that a new recipient primary managed to stepup
+jsTest.log("--XXXX-- Waiting for new recipient primary to stepup");
+st.rs1.getPrimary();
+jsTest.log("--XXXX-- There's a new recipient primary node");
+
+awaitTxn();
+jsTest.log("--XXXX-- Committed txn");
+
+awaitMigration();
+st.stop();
+})();
diff --git a/src/mongo/db/s/migration_destination_manager.cpp b/src/mongo/db/s/migration_destination_manager.cpp
index 5d6976a3676..f8063e03ddc 100644
--- a/src/mongo/db/s/migration_destination_manager.cpp
+++ b/src/mongo/db/s/migration_destination_manager.cpp
@@ -303,6 +303,7 @@ MONGO_FAIL_POINT_DEFINE(failMigrationOnRecipient);
 MONGO_FAIL_POINT_DEFINE(failMigrationReceivedOutOfRangeOperation);
 MONGO_FAIL_POINT_DEFINE(migrationRecipientFailPostCommitRefresh);
 
+MONGO_FAIL_POINT_DEFINE(migrationRecipientHangBeforeDeleteRecoveryDoc);
 }  // namespace
 
 MigrationDestinationManager::MigrationDestinationManager() = default;
@@ -1851,6 +1852,8 @@ void MigrationDestinationManager::awaitCriticalSectionReleaseSignalAndCompleteMi
     // persistence to be majority committed.
     CatalogCacheLoader::get(opCtx).waitForCollectionFlush(opCtx, _nss);
 
+    migrationRecipientHangBeforeDeleteRecoveryDoc.pauseWhileSet(opCtx);
+
     // Delete the recovery document
     migrationutil::deleteMigrationRecipientRecoveryDocument(opCtx, *_migrationId);
 }
-- 
2.17.1