On 11/8/2004 12:03 PM, D'Arcy J.M. Cain wrote:
> I checked the FAQ and docs but haven't found anything definitive. This
> is my SQL test script:
>
> SELECT pg_backend_pid();
> SELECT * FROM pg_stat_activity order by procpid;
>
> When I run psql reading that I find that my backend procpid is not in
> the list. I know that I can see it if I can introduce a little sleep (1
> second) between the connection and the reading of pg_stat_activity.
That is because the way your backend gets the information from the
pgstat daemon process is by reading a file, which is at maximum written
by that process every 500 msec. You will hardly ever see your own query.
>
> My question is this. Is there a more reliable way to see if a
> particular backend is still alive? I am trying to do a locking system
> and this is necessary to make it work. I think that in actual sessions
> I will be OK but my unit test fails most of the time because of this.
All these statistics are unreliable "by design", as explained numerous
times on this list. In short, a backend should never be slowed down
because the pgstat process can't swallow the stats fast enough. That is
the reason for using UDP in the first place. So the kernel is allowed to
drop stats packets, but not to block a backend.
Jan
--
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me. #
#================================================== JanWieck@Yahoo.com #