In our metrics processing E2E test we use Open Telemetry collector and Prometheus in a way:

Operator => OTel Collector => Prometheus.

Since OTel Collector is installed via the operator, this seems to be unstable in GitHub actions: long startup time, maybe even memory constraits.

So although this is nice practical setup that could help our users, if we are not able to stabilize this we might want to just remove OTel middle step and directly publish metrics to Prometheus.