From ae703673f00b275089cb756de5ab30a45ab4caa7 Mon Sep 17 00:00:00 2001
From: Jordi Serra Torrens <jordi.serra-torrens@mongodb.com>
Date: Fri, 3 May 2024 11:06:12 +0000
Subject: [PATCH] Repro BF-33079

---
 jstests/repro.js                              | 43 +++++++++++++++++++
 .../db/s/drop_collection_coordinator.cpp      |  5 +++
 2 files changed, 48 insertions(+)
 create mode 100644 jstests/repro.js

diff --git a/jstests/repro.js b/jstests/repro.js
new file mode 100644
index 00000000000..c2c920f784b
--- /dev/null
+++ b/jstests/repro.js
@@ -0,0 +1,43 @@
+import {configureFailPoint} from "jstests/libs/fail_point_util.js";
+import {moveOutSessionChunks, removeShard} from "jstests/sharding/libs/remove_shard_util.js";
+
+let st = new ShardingTest({shards: 2, configShard: true, enableBalancer: true});
+
+let coll = st.s.getDB('test')['foo'];
+
+// Create sharded collection with chunks on both shards.
+assert.commandWorked(
+    st.s.adminCommand({enableSharding: 'test', primaryShard: st.shard1.shardName}));
+assert.commandWorked(st.s.adminCommand({shardCollection: coll.getFullName(), key: {x: 'hashed'}}));
+
+// Start collection drop. Make it after committing the drop on the configsvr but before contacting
+// the shards to perform the local drop.
+let fp = configureFailPoint(st.rs1.getPrimary(),
+                            "dropCollectionCoordinatorHangBeforeSendDropCollParticipant");
+let awaitDrop = startParallelShell(() => {
+    assert(db.getSiblingDB('test')['foo'].drop());
+}, st.s.port);
+
+fp.wait();
+
+// Transition to dedicated config server.
+jsTest.log("--DEBUG-- Transitioning to dedicated config server");
+removeShard(st, st.shard0.shardName);
+jsTest.log("--DEBUG-- Transitioned to dedicated config server");
+
+// Let drop finish.
+fp.off();
+awaitDrop();
+
+// Shard the collection again.
+assert.commandWorked(st.s.adminCommand({shardCollection: coll.getFullName(), key: {x: 'hashed'}}));
+
+// Transition back to catalog shard (TODO: Why is it needed? The metadata check always runs on
+// configsvr doesn't it?).
+assert.commandWorked(st.s.adminCommand({transitionFromDedicatedConfigServer: 1}));
+
+// Check metadata consistency.
+let inconsistencies = st.s.getDB('test').checkMetadataConsistency().toArray();
+assert.eq(0, inconsistencies.length, tojson(inconsistencies));
+
+st.stop();
diff --git a/src/mongo/db/s/drop_collection_coordinator.cpp b/src/mongo/db/s/drop_collection_coordinator.cpp
index 69e42d0a99c..cbbb9480102 100644
--- a/src/mongo/db/s/drop_collection_coordinator.cpp
+++ b/src/mongo/db/s/drop_collection_coordinator.cpp
@@ -79,6 +79,7 @@
 #include "mongo/s/grid.h"
 #include "mongo/s/sharding_state.h"
 #include "mongo/util/decorable.h"
+#include "mongo/util/fail_point.h"
 #include "mongo/util/future_impl.h"
 #include "mongo/util/out_of_line_executor.h"
 #include "mongo/util/uuid.h"
@@ -87,6 +88,8 @@
 
 namespace mongo {
 
+MONGO_FAIL_POINT_DEFINE(dropCollectionCoordinatorHangBeforeSendDropCollParticipant);
+
 void DropCollectionCoordinator::dropCollectionLocally(OperationContext* opCtx,
                                                       const NamespaceString& nss,
                                                       bool fromMigrate,
@@ -337,6 +340,8 @@ void DropCollectionCoordinator::_commitDropCollection(
 
     const auto primaryShardId = ShardingState::get(opCtx)->shardId();
 
+    dropCollectionCoordinatorHangBeforeSendDropCollParticipant.pauseWhileSet();
+
     // We need to send the drop to all the shards because both movePrimary and
     // moveChunk leave garbage behind for sharded collections.
     auto participants = Grid::get(opCtx)->shardRegistry()->getAllShardIds(opCtx);
-- 
2.34.1