Hey Willy,
On 2026-05-22 20:32, Willy Tarreau wrote:
Hi Christian,
On Fri, May 22, 2026 at 04:01:39PM +0200, Christian Ruppert wrote:
Hi list,
I am not entirely sure yet whether it's a bug or intended. If it is
intended, it might make sense to just add another option for it.
So like I said, it's about haproxy_frontend_http_responses_total via
PROMEX.
In this case, we have some weird values for our backends that we use
for
baselining, rate-limits etc.
haproxy_frontend_http_responses_total counts any frontend/listen
request.
The problem is, it also counts requests to monitor-uri. Technically,
I'd say
that's totally fine. Since those are not represented in the logs, I'd
have
expected them to be not counted in PROMEX either. This might already
explain
our weird values.
Would it make sense to add an option, to not count those? Or is it a
bug /
not intended and just needs to be fixed? I'd open a GitHub issue then
I
guess.
I don't know if it would be easy actually, because I *think* that the
request is counted when it is received and parsed in the mux, then it
is matched against the configured URI later at an upper layer. In this
case, it would require to duplicate the check for the monitor URI in
all muxes at the lower layers just for the sake of counting or not
counting it.
I'm wondering if instead you couldn't subtract the number of locally
responded requests (I think they're in the "intercepted" counter in
the stats, not sure about the promex counter name). They might also
account for redirects or returns though (even if I'm never fan of
subtracting stats for obvious reasons).
Willy
Thanks! That could indeed work. Might work just fine I guess. I'll give
it a try when I am back at work in about 2w. Have a nice weekend! :)
--
Regards,
Christian Ruppert
#gernePerDu / #CallMeByMyFirstName
