Skip to content

Sentry: configure uWSGI worker respawning to be less disastrous and have more web pods running

Adeline Yeungrequested to merge
ayeung/sentry-drop-events-tweaks into master

What

Trying to rectify https://gitlab.com/gitlab-com/gl-infra/reliability/-/issues/24028

Why

  • The default settings for uWSGI provided by Sentry are ridiculous and not fit for production use. It results in all uWSGI workers across all web pods getting respawned at the same time once a day.
  • This is bad because the relay and worker pods need to hit the web pods to query the API for project information in order to process events. Events don't get processed if all the web workers are down.
  • Given how important being able to access the API is for event processing, we need more web pods than the 3 we currently have. Let's see how we go with 5

Merge request reports