Horton

Re: Error on pgbench logs

Michael Paquier <michael@paquier.xyz>

View thread View on postgr.es
From: Michael Paquier <michael@paquier.xyz>
To: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Cc: rulyox@gmail.com, pgsql-hackers@lists.postgresql.org
Date: 2021-06-09T03:46:27Z
Lists: pgsql-hackers 
On Tue, Jun 08, 2021 at 06:59:04PM +0900, Kyotaro Horiguchi wrote:
> The cause is that the time unit is changed to usec but the patch
> forgot to convert agg_interval into the same unit in doLog. I tempted
> to change it into pg_time_usec_t but it seems that it is better that
> the unit is same with other similar variables like duration.

As the option remains in seconds, I think that it is simpler to keep
it as an int, and do the conversion where need be.  It would be good
to document that agg_interval is in seconds where the variable is
defined.

-       while (agg->start_time + agg_interval <= now)
+       while (agg->start_time + agg_interval * 1000000 <= now)
In need of a cast with (int64), no?

The other things are "progress" and "duration".  These look correctly
handled to me.
--
Michael

Commits

  1. Fix pgbench timestamp bugs.

  2. pgbench: Improve time logic.

  3. Make [U]INT64CONST safe for use in #if conditions.

  4. Teach libpq to detect integer overflow in the row count of a PGresult.