On Wed, 11 May 2022 at 10:06, Augustin Husson <[email protected]> wrote:
> Dear Prometheus developers, > > I'm currently working on trying to optimize the synchronization of the > targets in the Prometheus Discovery. (Issue is here > <https://github.com/prometheus/prometheus/issues/8014>) > > And after a couple of investigations and tests on a real environment, we > figured out that actually the usage of the method Set of the > labels.Builder > <https://github.com/prometheus/prometheus/blob/main/model/labels/labels.go#L388> > cost a lot, and more generally the global usage of labels.Builder cost. > > In this PR <https://github.com/prometheus/prometheus/pull/10662> that is > supposed to fix the issue mentioned above, I started to replace the usage > of labels.Builder by a map and it immediately optimized the code. The > reason is because it reduces the complexity of the different methods. Like > for the method Set, we moved from o(n2) to o(n). > > So my question is: Would it be ok to change the internal implementation of > labels.Builder by using a map instead of multiple different slices ? It > won't change the API exposed, just the internal implementation. > A large part of the performance gains of Prometheus 2.x is attributed to it not using maps, so I would be extremely cautious about changing it for the sake of SD which is not the hot path. n is generally small for ingestion, so complexity alone can be misleading. When I last looked at the SD code the main performance issue looked to be that it had not been converted to using labels.Builder yet, and the continuous conversions from a map was what the main problem looked to be. Along with other low hanging fruit. I believe it will help to optimize smoothly (for example) the method > relabel.Process > <https://github.com/prometheus/prometheus/blob/main/model/relabel/relabel.go#L193> > This isn't on a hot path these days, as it's only run the first time we see a given sample. Brian > > Cheers, > Augustin. > > -- > You received this message because you are subscribed to the Google Groups > "Prometheus Developers" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To view this discussion on the web visit > https://groups.google.com/d/msgid/prometheus-developers/CAOJizGe0AxLGw5G9BM1Z6Menamq%3DxTMHzFXjj73fTqM5%2BZgz5A%40mail.gmail.com > <https://groups.google.com/d/msgid/prometheus-developers/CAOJizGe0AxLGw5G9BM1Z6Menamq%3DxTMHzFXjj73fTqM5%2BZgz5A%40mail.gmail.com?utm_medium=email&utm_source=footer> > . > -- Brian Brazil www.robustperception.io -- You received this message because you are subscribed to the Google Groups "Prometheus Developers" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/prometheus-developers/CAHJKeLqfcDUm7pMr890h2cXv6xdSwvibh_93jc0nvSKqMzfoGw%40mail.gmail.com.
