Hi Antonin,
On 1/10/18 5:38 PM, Tom Lane wrote:
> Antonin Houska <ah@cybertec.at> writes:
>> After having read the thread on your patch I think that the reason you were
>> asked to evaluate performance was that your patch can possibly make syslogger
>> a bottleneck. In contrast, my patch does not prevent user from disabling the
>> syslogger if it (the syslogger) seems to cause performance issues.
>
> Just to clarify that: we know that in workloads that emit lots of log
> output, the log collector *already is* a bottleneck; there are reports
> that some people can't use it because it's too slow. See e.g. towards
> the end of this thread:
>
> https://www.postgresql.org/message-id/flat/CABUevExztL0GORyWM9S4tR_Ft3FmJbRaxQdxj%2BBQZjpvmRurdw%40mail.gmail.com
>
> and particularly the referenced thread from 2011. (I seem to recall other
> reports but didn't have much luck finding them.)
>
> I'm quite concerned by the proposed patch, and not even so much any
> performance issues; what bothers me is that it adds complexity into a
> portion of the system where we can ill afford it. Bugs in the logging
> mechanism compromise one's ability to have any faith in tracking down
> other bugs. The difficulty of reconfiguring the logger on the fly
> is another reason to not want more configuration options for it.
> On the whole, therefore, I'd just as soon not go there --- especially
> seeing that there's been little field demand for such features.
I think this feature would be useful, especially for an extension like
pgaudit. It's a request I hear fairly frequently.
However, there doesn't seem to be consensus that this is a viable
approach, so marked as Returned with Feedback for this CF.
This may be too invasive a feature to be a good fit for the last PG11 CF
in March but I hope you keep working on the idea.
Regards,
--
-David
david@pgmasters.net