Varnish 503ing on ~1/100 POSTs

[samcrawford at gmail.com]

Hi Ronan,

It'd also be worth getting output from varnishlog for that broken request if possible. I realise it could be difficult to isolate it though.

Thanks,

Sam

------Original Message------
From: Ronan Mullally
To: Sam Crawford
Cc: varnish-misc at varnish-cache.org
Subject: Re: Varnish 503ing on ~1/100 POSTs
Sent: 5 Apr 2011 16:49

Hi Sam,

On Tue, 5 Apr 2011, Sam Crawford wrote:

> I suspect you are right - Apache may be discarding the connection as you
> are sending the POST request, resulting in the 503. Incidentally, what
> are your Apache keep-alive settings? If I remember rightly, there is a
> KeepAliveTimeout (which it sounds like you've been using) and also a
> KeepAlive parameter (which controls the maximum number of requests on
> one TCP connection, irrespective of timeouts).

I've got the number of requests per session set high (thousands).  The
number of actual back-end requests is modest (10-20 per back-end per
second at most) so it won't hit that limit often.

> 1) If you disable keepalives completely at Apache, does the problem
> disappear?

Not entirely.  I did this eariler in the process but still saw occurances
(not as many as I saw with the Apache Keepalive timeout = 1, but not
significantly fewer than when it was = 5 either).

I've disabled keepalives again and will run with it that way overnight, so
far it looks like more of the same - I've seen one occurance in about 2000
POSTs.


-Ronan


Sent from my BlackBerry® wireless device