Re: keep-alive time out error code

From: Willy Tarreau <w#1wt.eu>
Date: Wed, 19 May 2010 21:58:31 +0200


Hi Matt,

On Wed, May 19, 2010 at 02:31:14PM +0100, Matt wrote:
> Hi all,
>
> I'm currently seeing requests hitting a 10s http-request time out in my logs
> but not logging the cR code. I'm thinking this is because the time out is
> actually the http-keepalive one which I haven't set and therefore is using
> the http-request time out.
>
> Do people think its a good idea to log a code for when the http-keepalive
> time out strikes? or is this seen as such a common thing that it's not worth
> doing?

It is not expected, and I took care of ensuring that the keep-alive timeouts were *not* logged. There is a special state in the session for that specific purpose. Would you care to send me your config so that I can test it here, maybe something has escaped my radar and need to be fixed ?

> I'm going to insert the http-keepalive time out so I can be sure it
> is separated from the http-request time outs.

anyway I suggest to use a separate keep-alive timeout, because most of the time you can ensure a maximum keep-alive request ratio even with very short timeouts (eg: 100-200 ms), which will still ensure that your idle connections will not be monopolized for a long time. However, when the first byte of request comes in, you switch to http-request timeout which should be higher (5-10s is normally fine).

Still I'm interested to know how to reproduce your logging issue !

Thanks,
Willy Received on 2010/05/19 21:58

This archive was generated by hypermail 2.2.0 : 2010/05/19 22:00 CEST