MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII;
 format=flowed
Content-Transfer-Encoding: 7bit
Date: Fri, 22 Sep 2023 13:58:37 +0900
From: Ryoga Yoshida <bt23yoshidar@oss.nttdata.com>
To: pgsql-hackers@lists.postgresql.org
Subject: Doesn't pgstat_report_wal() handle the argument "force" incorrectly
User-Agent: Roundcube Webmail/1.4.11
Message-ID: <f87a4d7be70530606b864fd1df91718c@oss.nttdata.com>
Archived-At: <https://www.postgresql.org/message-id/f87a4d7be70530606b864fd1df91718c%40oss.nttdata.com>
Precedence: bulk

Hi,

pgstat_report_wal() calls pgstat_flush_wal() and pgstat_flush_io(). When 
calling them, pgstat_report_wal() specifies its argument "force" as the 
argument of them, as follows. But according to the code of 
pgstat_flush_wal() and pgstat_flush_io(), their argument is "nowait" and 
its meaning seems the opposite of "force". This means that, even when 
checkpointer etc calls pgstat_report_wal() with force=true to forcibly 
flush the statistics, pgstat_flush_wal() and pgstat_flush_io() skip 
flushing the statistics if they fail to acquire the lock immediately 
because they are called with nowait=true. This seems unexpected behavior 
and a bug.
void
pgstat_report_wal(bool force)
{
	pgstat_flush_wal(force);

	pgstat_flush_io(force);
}

BTW, pgstat_report_stat() treats "nowait" and "force" as the opposite 
one, as follows.
/* don't wait for lock acquisition when !force */
nowait = !force;

Ryoga Yoshida