my web app opens a mysql and a memcached connection when php script
starts, and then close the database and memcached connection right
before php script finishes execution.

Based on the logging info I have, what happened was that there was
something wrong with mysql. queries were taking very long, so both
mysql and memcached connection stayed open for seconds. After I
changed memcached connection to 5000, the mysql issued occurred again
the next morning and I found out there were about 3500 memcached
'current connections', which is much higher than peak time around 500.
Imagine 3500 connections needed when I had it 1024 in config. So I
believe it crashed because of too many connections.

On Feb 27, 11:27 pm, dormando <[email protected]> wrote:
> On Sat, 27 Feb 2010, Beier wrote:
> > the memcached sever is hosted on a 64 bit Ubuntu8.0.4 distribution on
> > Amazon EC2, it was installed from source
>
> > max_connection was the default 1024. I suspect that the connection
> > might went over 1024 (at peak time the server deals with about 500
> > hits / sec), so now I set it to 5000 and it has been working well so
> > far. But I'm expecting 'too many connections' error instead of
> > event_queue error, could it be libevent's problem?
>
> Smells like a libevent bug... or perhaps a new bug from us. Not sure; will
> research this more.
>
> You can tell if you're hitting mas conns by looking at the
> listen_disabled_num counter. If you hit max conns and it's crashing every
> time instead of limiting new conns, you've got a bug we're interested in
> fixing.
>
> Any chance you could test that out real quick?

Reply via email to