On Wed, Jan 13, 2021 at 9:58 AM Christopher Faulet <[email protected]> wrote: > The first 2 patches are ok. Relying on fill_info is indeed cleaner. However, > for > the last three patches, it is more annoying. The result is clean of course but > the fill_* functions will then be called one time for each metric for a given > frontend/backend/server because of the Prometheus output format. For instance, > if you have a backend with 100 servers, fill_sv_stats will be called 100 times > for each metric, thus approximately 5000 times. > For now, I will only merge the first two patch if it's ok for you. For the > other, I guess the stats api may be changed to fill all metric or a specific > one.
indeed I did not realise this impact compared to the stats API, and it will not likely work for big clusters with thousands of servers. I might try to have a closer look and see whether I can adapt things. Thanks a lot for your input. -- William
