The informers used in CAP Operator currently caches Secrets (and some other resources) across all namespaces (as we are a cluster wide operator) and is prone to high memory usage, especially when large number of secrets are present in the cluster.
While in most real clusters this has been tested with thousands of secrets - having several very large secrets (like mentioned in SAP/cap-operator-lifecycle/issues/205) does have it's limits.
We did do an evaluation with using namespace scoped informers ~2 years ago, however decided against it eventually due to complexity.
This issue is opened to re-evaluate if that change or something better (if available now) can be used -or- to at least document the recommendation to have a higher resource requirement on clusters with a very high number of secrets (especially large ones).
cap-operator-team
The informers used in CAP Operator currently caches Secrets (and some other resources) across all namespaces (as we are a cluster wide operator) and is prone to high memory usage, especially when large number of secrets are present in the cluster.
While in most real clusters this has been tested with thousands of secrets - having several very large secrets (like mentioned in SAP/cap-operator-lifecycle/issues/205) does have it's limits.
We did do an evaluation with using namespace scoped informers ~2 years ago, however decided against it eventually due to complexity.
This issue is opened to re-evaluate if that change or something better (if available now) can be used -or- to at least document the recommendation to have a higher resource requirement on clusters with a very high number of secrets (especially large ones).
cap-operator-team