Hi Julien, On Thu, Dec 05, 2019 at 01:54:55PM +0100, Julien Pivotto wrote: > Dear list, > > I am looking at the Prometheus metrics in HAProxy and I miss some of > them: > > - Is it possible to get a more fine grained view of the return codes > than 2xx,3xx,4xx. Indeed, an error 404 is different than 403 > (404 might be the browser trying to get a favicon and 400 an actual > error) and 503 is different than 500 (500 might be a business SOAP > FAULT)
It was planned for 2.1 already as part of the stats improvements but we got stuck with a long backlog of bugs from the important changes in 1.9 slowing down our progress. There are other metrics we need to add, like everything related to the muxes (H1/H2/FCGI). But these require a new registration mechanism, probably we'll keep fixed fields for the legacy ones and dynamic ones (i.e. only named) for new ones, and definitely the prometheus exporter will need to be able to peek from these ones. > - I see no metrics like: > - haproxy_frontend_http_responses_seconds_total > - haproxy_backend_http_responses_seconds_total > Those metrics would allow me to calculate the response time of the > http queries that go via HAProxy. That's a very good idea. I added it long ago to the halog utility already because it allowed to quickly focus on the components that degrade the user experience. We don't have such metrics internally either so for now they cannot be reported, that must be part of the counters improvements as well. Would you be willing to file a feature request on the github issue tracker for these so that we don't lose them ? Thanks, Willy
