Re: idle connections

Thank you for the informative post.

We believe this was caused by a network issue yesterday after checking our
network speeds. We were maxing our ethernet port at 100 mbps, i believe some
apache processes were hanging because of this issue, having no available
bandwidth left. Considering an upgrade to a gbps port and we will see how
this will effect.

On Tue, Oct 6, 2009 at 9:28 AM, Steve Crawford <
scrawford(at)pinpointresearch(dot)com> wrote:

> Aras Angelo wrote:
>
>> ...
>> When i kill the earliest idle process the others stop too. So i dont know
>> whats wrong really. All our apps use the same footer, with pg_close() at the
>> end....
>> ...
>> I have done most of the things you guys suggested, so it seems to me that
>> something between php-apache-postgresql is not doing good, and it only
>> effects us at peak times, so just wondering if killing processes every 2-3
>> minutes, would do harm on our setup....
>>
>
> Are you killing web processes or postgresql processes? If web, you will
> probably end up impacting at least some of you users directly. If
> PostgreSQL, then you will undoubtedly end up yanking the rug out from under
> a process that expected a working database connection. So yes, you might
> cause harm. Worse still, you may not solve the problem.
>
> When I hear symptoms like this, especially when they tend to happen under
> load, I start looking for a process that is slowing/blocking the other
> processes.
>
> Consider a fairly typical PHP script. It might start out opening a database
> connection and then do various queries interspersed with processing. If one
> process hogs sufficient resources or is holding some resource or lock that
> slows or blocks the other PHP processes then you could easily see numerous
> idle PG processes. Kill the "offending" process and the others will often
> rapidly complete thus releasing their connections.
>
> If you have enough traffic to justify three webservers then you probably
> have enough traffic to deplete your available pg connections in seconds, not
> minutes so keep digging.
>
> Cheers,
> Steve
>
>