Open lfrancke opened 1 year ago
After some investigation, this largely comes down to two sources:
There are some docs on this now from kube in https://kube.rs/controllers/optimization/. in particular;
managedFields
or .status
store contributions)metadata_watcher
(also limits IO)should all help reduce the memory contribution of cached objects - although it will still generally scale with num_objects in your stores.
kube 0.92.0 also seems to have nice memory improvements: https://kube.rs/blog/2024/06/11/watcher-memory-improvements/ :)
Currently the memory required by the commons-operator scales with the amount of Pods in a K8s cluster. This can lead to more or less unbounded memory usage.
We need to investigate how we can limit this. One possibility to limits this is by leveraging the work that will be done in https://github.com/stackabletech/issues/issues/188