On Thu, Aug 6, 2009 at 16:33, Tom Lane<tgl@sss.pgh.pa.us> wrote:
> Magnus Hagander <magnus@hagander.net> writes:
>> On Thu, Aug 6, 2009 at 16:20, Tom Lane<tgl@sss.pgh.pa.us> wrote:
>>> Well, it seems like you could get 90% of the way there just by filtering
>>> on the PID --- watching the bgwriter, walwriter, and archiver should
>>> cover this use-case reasonably well.
>
>> Right. But that's pretty hard to do automated, since they will get a
>> new pid whenever the database is restarted. Which is hopefully not
>> very often, but still an issue. Plus, it's hard to do any kind of
>> historical look at things.
>
> I don't think there'd be much logical difficulty in having an output
> field (ie, CSV column or log_line_prefix escape) that represents a
> classification of the PID, say as "postmaster, backend, AV worker,
> AV launcher, bgwriter, ...". It would only require changing things
> in one place, whereas your original proposal seemed mighty open-ended.
Good point. That *would* probably take care of much of the need. The
downside is aggressive filtering that way would get rid of important
messages coming out of a single backend, like out of disk space.
Meaning it would still not be possible to filter on the difference
between ERROR: syntax error in query and ERROR: out of disk space? But
it'd be an improvement still.
-- Magnus HaganderMe: http://www.hagander.net/Work: http://www.redpill-linpro.com/
