Skip to content

Exclude workhorse from service aggregation

Bob Van Landuyt requested to merge bvl/exclude-workhorse-from-service-aggregation into master

Exclude workhorse from service aggregation

This excludes the workhorse component from service aggregations in favour of the rails request component.

The rails request component has more appropriate apdex targets defined inside the application itself. Which makes it more appropriate to use here.

The disadvantage of this approach is that any 503 errors that workhorse might encounter because a puma worker is not available would not be recognized as errors for the service. I prefer this tradeoff, because when we would not have enough workers available like this, this would also be noticeable in the apdex. For GitLab.com this is unlikely because we have 1:1 workhorse-puma in the same pod.

This only affects the service aggregation that rolls up into the SLA for GitLab.com, it does not affect alerting.

I noticed this when looking at the availability numbers for git in the context of gitlab-com/gl-infra/production#15997 (closed)

Edited by Bob Van Landuyt

Merge request reports