Allow recording error rates in Gitlab::Metrics::Sli
In &525 (closed) we started by adding a framework to define SLIs in the application. The first thing we needed this for was the rails_request_apdex
with a customizable duration.
Our sli definition tooling in the runbooks repository separates an error
and apdex
component:
-
apdex
is dealt with as a success ratio: higher is better 100% is best -
error
is dealt with as an error ratio: lower is better, 0% is best
To capture the error
component of an SLI, we currently have no support in the setup in Gitlab::Metrics::Sli
or the runbooks side of this. But having that would allow us to record more granular information on errors for endpoints.