Skip to content

Re-add slow Gitaly methods to the apdex using a combined apdex with mixed thresholds - second try

This is yet another try for improving the Gitaly goserver SLIs by using a combined apdex with different thresholds for fast and slow methods. Last try was !4513 (merged) which was reverted in !4530 (merged). This time I'm (re)adding a fix in the combined method fixing the aggregation when setting both the satisfied and tolerated thresholds.

As noted here, even though sli_aggregations:grpc_server_handling_seconds_bucket_rate5m was defined as a sum(...) it appears 2 aggregations defined in a rule still can't be summed without being wrapped into an aggregation for some reason.

@reprazent you mentioned in the previous MR that the double aggregation approach is not ideal because it might affect the recordings for Sidekiq. I've re-added my fix here but do you have another suggestion for a way to fix this? combined() works for the Sidekiq SLIs because they have a single thresholds, but it's still broken for double threshold cases in its current state.

Merge request reports