v8-0001-Fix-DROP-SUBSCRIPTION-deadlock-with-walsender-pro_v16_17_18.patch

From e0b849ffc5a93e190e89adeefdbad7ae8fa52618 Mon Sep 17 00:00:00 2001
From: Dilip Kumar <dilipkumarb@google.com>
Date: Mon, 18 Aug 2025 11:32:21 +0000
Subject: [PATCH v8] Fix DROP SUBSCRIPTION deadlock with walsender process

The DROP SUBSCRIPTION command caused a deadlock with the walsender
process because of a lock conflict on the pg_subscription catalog.
This happened when the command was trying to drop a replication slot
by connecting to a newly created database with uninitialized caches.

The DROP SUBSCRIPTION command would first acquire an AccessExclusiveLock
on pg_subscription to prevent new workers from starting. However, the
connection to the new database needed to acquire an AccessShareLock on
the same catalog during its initialization. This created a deadlock
where DROP SUBSCRIPTION held its lock and was waiting for the new
connection to complete, while the new connection was simultaneously
blocked from starting because it couldn't acquire its required lock.

This commit resolves the deadlock by having DROP SUBSCRIPTION acquire a
less restrictive RowExclusiveLock on the catalog instead.

To address the original concern of orphaned workers, a new check is
implemented.  The replication worker now takes a shared object lock on the
subscription itself.

If a worker starts for a subscription that no longer exists, it immediately
detects this condition and exits. This ensures that no orphan workers are
created without the need for an overly broad AccessExclusiveLock on the
system catalog.
---
 src/backend/commands/subscriptioncmds.c  |  8 ++++---
 src/backend/replication/logical/worker.c |  7 ++++++
 src/test/subscription/t/100_bugs.pl      | 30 ++++++++++++++++++++++++
 3 files changed, 42 insertions(+), 3 deletions(-)

diff --git a/src/backend/commands/subscriptioncmds.c b/src/backend/commands/subscriptioncmds.c
index e1260fc0e90..aebfaecd9d8 100644
--- a/src/backend/commands/subscriptioncmds.c
+++ b/src/backend/commands/subscriptioncmds.c
@@ -1491,10 +1491,12 @@ DropSubscription(DropSubscriptionStmt *stmt, bool isTopLevel)
 	bool		must_use_password;
 
 	/*
-	 * Lock pg_subscription with AccessExclusiveLock to ensure that the
-	 * launcher doesn't restart new worker during dropping the subscription
+	 * The launcher may concurrently start a new worker for this subscription.
+	 * During initialization, the worker checks for subscription validity and
+	 * exits if the subscription has already been dropped. See
+	 * InitializeLogRepWorker.
 	 */
-	rel = table_open(SubscriptionRelationId, AccessExclusiveLock);
+	rel = table_open(SubscriptionRelationId, RowExclusiveLock);
 
 	tup = SearchSysCache2(SUBSCRIPTIONNAME, MyDatabaseId,
 						  CStringGetDatum(stmt->subname));
diff --git a/src/backend/replication/logical/worker.c b/src/backend/replication/logical/worker.c
index 4df0a594dc9..9b5c641941f 100644
--- a/src/backend/replication/logical/worker.c
+++ b/src/backend/replication/logical/worker.c
@@ -4492,6 +4492,13 @@ InitializeApplyWorker(void)
 	StartTransactionCommand();
 	oldctx = MemoryContextSwitchTo(ApplyContext);
 
+	/*
+	 * Lock the subscription to prevent it from being concurrently dropped,
+	 * then re-verify its existence. After the initialization, the worker will
+	 * be terminated gracefully if the subscription is dropped.
+	 */
+	LockSharedObject(SubscriptionRelationId, MyLogicalRepWorker->subid, 0,
+					 AccessShareLock);
 	MySubscription = GetSubscription(MyLogicalRepWorker->subid, true);
 	if (!MySubscription)
 	{
diff --git a/src/test/subscription/t/100_bugs.pl b/src/test/subscription/t/100_bugs.pl
index 6744998b620..87643b8e620 100644
--- a/src/test/subscription/t/100_bugs.pl
+++ b/src/test/subscription/t/100_bugs.pl
@@ -568,4 +568,34 @@ is($result, 't',
 $node_publisher->stop('fast');
 $node_subscriber->stop('fast');
 
+# BUG #18988
+# The bug happened due to a self-deadlock between the DROP SUBSCRIPTION
+# command and the walsender process for accessing pg_subscription. This
+# occurred when DROP SUBSCRIPTION attempted to remove a replication slot by
+# connecting to a newly created database whose caches are not yet
+# initialized.
+#
+# The bug is fixed by reducing the lock-level during DROP SUBSCRIPTION.
+$node_publisher->start();
+
+$publisher_connstr = $node_publisher->connstr . ' dbname=regress_db';
+$node_publisher->safe_psql(
+	'postgres', qq(
+	CREATE DATABASE regress_db;
+	CREATE SUBSCRIPTION regress_sub1 CONNECTION '$publisher_connstr' PUBLICATION regress_pub WITH (connect=false);
+));
+
+my ($ret, $stdout, $stderr) =
+  $node_publisher->psql('postgres', q{DROP SUBSCRIPTION regress_sub1});
+
+isnt($ret, 0, "replication slot does not exist: exit code not 0");
+like(
+	$stderr,
+	qr/ERROR:  could not drop replication slot "regress_sub1" on publisher/,
+	"could not drop replication slot: error message");
+
+$node_publisher->safe_psql('postgres', "DROP DATABASE regress_db");
+
+$node_publisher->stop('fast');
+
 done_testing();
-- 
2.51.0.rc1.167.g924127e9c0-goog