From: Andrey Borodin <x4mmm@yandex-team.ru>
Message-Id: <9DF05C0C-D165-4821-80C2-FFAF47C07FF4@yandex-team.ru>
Content-Type: multipart/mixed;
	boundary="Apple-Mail=_49D10CFB-0EC0-492E-BCF3-A527712C4438"
Mime-Version: 1.0 (Mac OS X Mail 16.0 \(3864.600.51.1.1\))
Subject: Re: BUG #19490: Streaming standby on 16.14 stops applying WAL on
 MultiXactOffsetSLRU when primary is 16.8
Date: Tue, 26 May 2026 14:33:11 +0500
In-Reply-To: <ahVnc8-EjV6WlM5-@paquier.xyz>
Cc: Ayush Tiwari <ayushtiwari.slg01@gmail.com>,
 Radim Marek <radim@boringsql.com>,
 Heikki Linnakangas <hlinnaka@iki.fi>,
 Marko Tiikkaja <marko@joh.to>,
 PostgreSQL mailing lists <pgsql-bugs@lists.postgresql.org>
To: Michael Paquier <michael@paquier.xyz>
References: <19490-9c59c6a583513b99@postgresql.org>
 <46FE61C9-F273-45FD-BED7-0F8CDA6EB992@yandex-team.ru>
 <CAL9smLBMxKBCmsA9UGcmf93bT2_MsZ+POH-oHREuwKdmMU7jfQ@mail.gmail.com>
 <46DB3CAB-EA1C-41A5-9D6D-5F913A2AAF66@yandex-team.ru>
 <CAJgoLkJfFgL-V+pYB7=R81AbURTE6sMhzVHDQDhVGnfXRSJ9Wg@mail.gmail.com>
 <CAJgoLkKCu0wCwPQZSo5no=XATU-4LMK4QfKBwV928o2uKcxe=g@mail.gmail.com>
 <CAJTYsWU6tdEvVh4YKLxz7+amZ7+Wb7_s-FBjsMMeLNj1fKeSNg@mail.gmail.com>
 <ahVTk8HKRPcz4iZ-@paquier.xyz>
 <CAJTYsWWXvbBJe+WYJZcnoSTyVz9vk5ro3x2qAq_uvXvK2KwaMQ@mail.gmail.com>
 <AC1516B2-2E78-4C70-8A6E-2966F01A1058@yandex-team.ru>
 <ahVnc8-EjV6WlM5-@paquier.xyz>
Archived-At: 
 <https://www.postgresql.org/message-id/9DF05C0C-D165-4821-80C2-FFAF47C07FF4%40yandex-team.ru>
Precedence: bulk


--Apple-Mail=_49D10CFB-0EC0-492E-BCF3-A527712C4438
Content-Transfer-Encoding: 7bit
Content-Type: text/plain;
	charset=us-ascii


> On 26 May 2026, at 14:27, Michael Paquier <michael@paquier.xyz> wrote:
> 
> Hmm, okay.  I am not sure what you mean here, could you demonstrate
> your idea with a patch later?

Something like attached, not tested yet, working on an automated test.


Best regards, Andrey Borodin.

--Apple-Mail=_49D10CFB-0EC0-492E-BCF3-A527712C4438
Content-Disposition: attachment;
	filename=demo.diff
Content-Type: application/octet-stream;
	x-unix-mode=0644;
	name="demo.diff"
Content-Transfer-Encoding: 7bit

diff --git a/src/backend/access/transam/multixact.c b/src/backend/access/transam/multixact.c
index f825579e888..8899d5ac63d 100644
--- a/src/backend/access/transam/multixact.c
+++ b/src/backend/access/transam/multixact.c
@@ -888,8 +888,6 @@ RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
 	MultiXactOffset *next_offptr;
 	MultiXactOffset next_offset;
 
-	LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
-
 	/* position of this multixid in the offsets SLRU area  */
 	pageno = MultiXactIdToOffsetPage(multi);
 	entryno = MultiXactIdToOffsetEntry(multi);
@@ -907,6 +905,9 @@ RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
 	 * multixid was assigned.  If we're replaying WAL that was generated by
 	 * such a version, the next page might not be initialized yet.  Initialize
 	 * it now.
+	 *
+	 * This block runs before acquiring MultiXactOffsetSLRULock because
+	 * SimpleLruWriteAll() needs to acquire the same lock internally.
 	 */
 	if (InRecovery && next_pageno != pageno)
 	{
@@ -951,6 +952,8 @@ RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
 		{
 			elog(DEBUG1, "next offsets page is not initialized, initializing it now");
 
+			LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
+
 			/* Create and zero the page */
 			slotno = SimpleLruZeroPage(MultiXactOffsetCtl, next_pageno);
 
@@ -958,6 +961,8 @@ RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
 			SimpleLruWritePage(MultiXactOffsetCtl, slotno);
 			Assert(!MultiXactOffsetCtl->shared->page_dirty[slotno]);
 
+			LWLockRelease(MultiXactOffsetSLRULock);
+
 			/*
 			 * Remember that we initialized the page, so that we don't zero it
 			 * again at the XLOG_MULTIXACT_ZERO_OFF_PAGE record.
@@ -967,6 +972,8 @@ RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
 		}
 	}
 
+	LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
+
 	/*
 	 * Set the starting offset of this multixid's members.
 	 *

--Apple-Mail=_49D10CFB-0EC0-492E-BCF3-A527712C4438--