Re: Fix safe_wal_size for slots without restart_lsn

Xuneng Zhou <xunengzhou@gmail.com>

From: Xuneng Zhou <xunengzhou@gmail.com>
To: pgsql-hackers <pgsql-hackers@lists.postgresql.org>
Date: 2026-05-28T01:01:01Z
Lists: pgsql-hackers

Attachments

On Wed, May 27, 2026 at 6:55 PM Xuneng Zhou <xunengzhou@gmail.com> wrote:
>
> Hi,
>
> Currently, pg_replication_slots can report a non-null safe_wal_size
> for a replication slot that has never reserved WAL, if
> max_slot_wal_keep_size is finite.
>
> Such a slot has restart_lsn = NULL and wal_status = NULL. safe_wal_size is
> derived from restart_lsn, so it's not meaningful in this state. The
> current code handles WALAVAIL_REMOVED but not WALAVAIL_INVALID_LSN
> before computing safe_wal_size, leading arithmetic on InvalidXLogRecPtr.
>
> Fix this by returning NULL for safe_wal_size when WAL availability is
> WALAVAIL_INVALID_LSN.
>

Fixed the indentation issue in v1 and updated the comment and doc. Not
sure this small issue warrants an additional test, still added one in
case someone wants it.


-- 
Regards,
Xuneng Zhou
HighGo Software Co., Ltd.