Re: Timings and HTTP keep-alives

On Thu, Apr 8, 2010 at 3:11 PM, Willy Tarreau <w#1wt.eu> wrote:

> That's what I first thought too, but I quickly realized that there is
> no way to know because the client remains idle and suddenly sends a
> request. So it is not different from the first request case, except
> that during the first request, there is an active connection from the
> client, so the client knows it has to talk and does it ASAP, meaning
> there is almost never any delay.
>

Yes, I see; one just has to remember this when interpreting the logs.

>
> > For clarity in the docs (since they are already so comprehensive),
> > I might suggest mentioning in section on HTTP log format that
> > Tq and Tt are cumulative for the connection (while Tw, Tc, and Tr
> > are of course per request).
>
> No, Tq is included in Tt. If you have a case where this is not true,
> then it's a bug.
>
>

Yes, as you say, I do always see Tt > Tq. It's just that the 1.4 doc says:   "Tq is the total time in milliseconds spent waiting for the client to send   a full HTTP request... It should always be very small..." so I was initially surprised to see such large values of Tq and Tt, and had to infer what it meant. (Such large values on the other timers would be cause for alarm.) I now look at Tt minus Tq to monitor server performance.

Regards,
Josh Received on 2010/04/09 00:26