On Fri, Sep 22, 2023 at 01:58:37PM +0900, Ryoga Yoshida wrote:
> 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.
It seems to me that you are right here. It would make sense to me to
say that force=true is equivalent to nowait=false, as in "I'm OK to
wait on the lockas I want to make sure that the stats are flushed at
this point". Currently force=true means nowait=true, as in "I'm OK to
not have the stats flushed if I cannot take the lock".
Seeing the three callers of pgstat_report_wal(), the checkpointer
wants to force its way twice, and the WAL writer does not care if they
are not flushed immediately at it loops forever in this path.
A comment at the top of pgstat_report_wal() would be nice to document
that a bit better, at least.
--
Michael
