Review whether targetAverageValue is still the right choice for HPA
HPA is currently configured with a targetAverageValue=1
which is hardcoded in the gitlab chart](https://gitlab.com/gitlab-org/charts/gitlab/-/blob/47a533e61216a3338caec731ff7d264a4d7e0b17/charts/gitlab/charts/webservice/values.yaml#L91-92).
As we found in production#2940 (closed), this is probably not sufficient for pod scaling and if we don't scale pods we run the risk of hitting memory limits, especially for workhorse.
Use this issue to investigate:
- Whether
targetAverageValue
is still the right choice overtargetAverageUtilization
, and whether it should be configurable in the chart - How
targetAverageValue
works if there are multiple containers in the pod - Whether we should take memory into account in addition to cpu
Edited by Amy Phillips