Related to COO-784 and associated workaround (increasing memory limits to address OOMKill events), a customer reports significantly higher memory usage in larger clusters.

In a cluster with ~1300 namespaces, the observability-operator pod consumes up to 1.9Gi of memory on startup, before stabilising around 1Gi. The previously suggested workaround of setting a 512Mi memory limit is insufficient in such environments;  the customer required a limit of 2Gi to avoid OOMKill events.

This issue does not appear in smaller clusters, where 512Mi remains adequate.

Engineering suggested a new bug to review the behaviour/requirements specifically in larger clusters.  Observability must-gather available in case 04115625.

OpenShift 4.16, Cluster Logging Operator 6.0.6.