Re: stats explanation

From: Willy Tarreau <w#1wt.eu>
Date: Sun, 10 Feb 2008 14:24:16 +0100


On Wed, Feb 06, 2008 at 02:40:50PM -0800, rkmr.em#gmail.com wrote:
> On Feb 6, 2008 4:24 AM, Willy Tarreau <w#1wt.eu> wrote:
> > if you have connection errors (Conn) and "option redispatch" is set,
> > haproxy tries to reconnecto to another server.
> I have this feature enabled!
>
> > If you have response
> > errors (Resp) or security errors (Sec), it means that the response
> > either was not valid or was blocked. In this case, the request was
> > served but the response was not delivered to the user.
> How do I debug this error? what kind of errors are these? are these 502 errors?

Yes, but also 504 (gateway timeout) if the server never completes its response. Checking the flags and return code in the logs would help a lot of course. Aside this, you can always sniff the server->proxy traffic to ensure that the server's responses are always valid. But in my experience, I've never had such a behaviour with a "normal" server. Please ensure that you are not using response mangling/checking keywords such as rspirep, rspdeny, etc...

> > Check errors
> > indicate individual health-check failures (possibly non-fatal), and
> check errors are served from other backends right?

no, check errors are never "served". Checks are sent to the servers to ensure they are alive, and check errors are monitored to detect that a server is failing. As long as your servers don't transition from UP to DOWN, your users will never know there were check errors.

Regards,
willy Received on 2008/02/10 14:24

This archive was generated by hypermail 2.2.0 : 2008/02/10 14:30 CET