Hi guys, First, thanks a lot, that's *exactly* the type of feedback I was waiting for.
On Tue, Aug 23, 2016 at 12:40:58PM +0200, Olivier Doucet wrote: > 2016-08-23 12:22 GMT+02:00 Conrad Hoffmann <[email protected]>: (...) > > So with that reasoning, I guess I would love to see: > > - all new variables created that you mentioned (%tr, %Ti, %Th, not sure > > about %TR) > > - make %tr the default as logged date The TR would simply be the replacement for the current Tq, it would only report the request time without the idle nor handshake times. I initially wanted to change Tq until I realized it would mess up with the logs people already have based on Tq (and lots of people are currently post-processing such logs doing some computation). For this reason I preferred to introduce a new field which becomes the default one for HTTP logs. > Conrad could not tell better what I had in mind. > When dealing with HTTP trafic, we sometimes need to correlate data with > HTTP logs (that we gather with the webserver itself, not haproxy). Having > the exact same date on both tools is really helpful. Absolutely. And it's even more important with keep-alive. When you see pauses of 5-10s and requests taking that much time, it's hard to correlate request times to what the server sees. > And I'm all for having detailed informations if needed, but less info by > default. > Improving the documentation to explain all this will be an important part > of this job. I think so as well. Thanks very much for your insights, thus I think we're on the right track. Willy
