Not working for clustered puma

[mariokam]

When running yabeda rails for clustered puma and does not work for streaming metrics, any ideas? We are running it via puma config bundle exec puma -C config/puma.rb

# ps aux
USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
root         1  0.7  1.3 1415016 217088 ?      Ssl  13:05   0:09 puma 3.12.0 (tcp://0.0.0.0:3000) [app]
root        22  0.1  1.3 1632116 214932 ?      Sl   13:06   0:01 puma: cluster worker 0: 1 [app]
root        27  0.0  1.1 1565552 189348 ?      Sl   13:06   0:00 puma: cluster worker 1: 1 [app]
root        32  0.4  1.6 2806432 264888 ?      Sl   13:06   0:05 puma: cluster worker 2: 1 [app]
root        37  0.0  1.1 1565552 189680 ?      Sl   13:06   0:00 puma: cluster worker 3: 1 [app]
root       307  0.0  0.0   4296   728 pts/0    Ss   13:15   0:00 sh
root       581  0.0  0.0  36644  2828 pts/0    R+   13:26   0:00 ps aux

[Envek]

Try replacing default prometheus-client with prometheus-client-mmap:

# Gemfile
gem "yabeda"
gem "yabeda-rails"
gem "prometheus-client-mmap"
gem "yabeda-prometheus"

See https://github.com/yabeda-rb/yabeda-prometheus/issues/4 for discussion and more information

[dsalahutdinov]

hey, @mariokam support of gitlab client moved to another gem https://github.com/yabeda-rb/yabeda-prometheus-mmap Try it out.

[i5okie]

Ended up using yabeda-prometheus-mmap too. Works great!

[splitice]

Unfortuantely this doesnt work currently with:

gem 'yabeda'
gem 'yabeda-rails'
gem 'yabeda-puma-plugin'    
gem 'yabeda-prometheus-mmap'

Individual workers supply their own metrics (not aggregated)

[Envek]

That's weird. Can you please create reproduction? E.g. write a Ruby script using inline bundler and publish it as a secret gist? Also please show exact versions of all related gems – maybe some dependencies got updated…

Also, just to confirm that by “Individual workers” you mean forks of main Puma process that aren't synced (so you have single /metrics endpoint that respond with metrics from different child worker process on every request). Because different Puma clusters on different machines/containers won't be synced – this is Prometheus' work to scrape each and aggregate.

[splitice]

I've solved this, it took a while but the answer was calling Prometheus client configure before fork

If you fail in that it's initialised after fork and multiple directories in /tmp created.