Horton

[PATCH] BUG FIX: inconsistent page found in BRIN_REGULAR_PAGE

Karina Litskevich <litskevichkarina@gmail.com>

View thread View on postgr.es
From: Karina Litskevich <litskevichkarina@gmail.com>
To: Alvaro Herrera <alvherre@alvh.no-ip.org>
Cc: 王海洋 <wanghaiyang.001@bytedance.com>, Kyotaro Horiguchi <horikyota.ntt@gmail.com>, pgsql-bugs@lists.postgresql.org
Date: 2022-12-23T13:54:56Z
Lists: pgsql-bugs

Attachments

Hi Alvaro!

Is there any special reason to read WAL records until the last inserted 
record?

+my $end_lsn = $whiskey->lsn('insert');
+
+my ($ret, $out, $err) = $whiskey->psql(
+    'postgres', qq{
+    select count(*) from pg_get_wal_records_info('$start_lsn', '$end_lsn')
+    where resource_manager = 'BRIN' AND
+    record_type ILIKE '%revmap%'
+    });
+cmp_ok($out, '>=', 1);

It seems that in some rare situations on slower machines this test can 
fail. If
any background process inserts a WAL record before lsn('insert') and 
this record
isn't flushed before pg_get_wal_records_info('$start_lsn', '$end_lsn'),
pg_get_wal_records_info('$start_lsn', '$end_lsn') ends with ERROR 
"cannot accept
future end LSN" as it works only if record with end LSN is inserted.

I attached two patches with two ways of fixing this minor issue.

Best regards,
Karina Litskevich
Postgres Professional: http://postgrespro.com/

Commits

  1. Fix end LSN determination in recently added test

  2. BRIN: mask BRIN_EVACUATE_PAGE for WAL consistency checking