Context:

The watcher controller is wathing all secrets and configmaps from all namespaces in the hub cluster. On most events, it will then do some "heavy" processing on the manifestworks to verify whether the resource is listed in them to then trigger a reconciliation for the impacted managed clusters.
This doesn't scale, even in the demo environment where it consumes ~1GB of memory and >2CPUs. It was still increasing when I stopped it.
Cf solutions discussion: https://redhat-internal.slack.com/archives/C0686U6KPU5/p1765358312880979

On top of that, the addon is deployed with CPU and memory limits that can't be removed by the user. It results in the addon manager constantly crashlooping, OOMKilled.

Acceptance:

Resources usage is reduced to a manageable level.