Repository navigation
Does controller-runtime support preloading resources when selecting a leader? #2600
Description
The controller-manager stores a significant number of objects, and the overhead of fetching and deserializing these objects from the API server during each upgrade process is unavoidable. Restarting the controller-manager may take several minutes to complete the recovery process. My question is, how can we preload the required resources into memory before the controller-runtime selects the leader? Does controller-runtime support this space-time trade-off strategy?
Activity
duanmengkk commentedon Nov 29, 2023
/kind support
alvaroaleman commentedon Dec 3, 2023
No, this is currently not possible. Someone attempted it, but it didn't get to the point of being mergeable: #2005
duanmengkk commentedon Dec 5, 2023
OK, do we have any progress or discussions about this plan? Our cluster scale has reached several hundred thousand nodes, so we want our controller-runtime's work replicas to switch within seconds in case of failures.
alvaroaleman commentedon Dec 5, 2023
Right.
You won't get second range fail over one way or another because the old lease needs to expire first. But feel free to take a stab at it.
k8s-triage-robot commentedon Mar 5, 2024
The Kubernetes project currently lacks enough contributors to adequately respond to all issues.
This bot triages un-triaged issues according to the following rules:
lifecycle/staleis appliedlifecycle/stalewas applied,lifecycle/rottenis appliedlifecycle/rottenwas applied, the issue is closedYou can:
/remove-lifecycle stale/closePlease send feedback to sig-contributor-experience at kubernetes/community.
/lifecycle stale
duanmengkk commentedon Mar 6, 2024
/close
k8s-ci-robot commentedon Mar 6, 2024
@duanmengkk: Closing this issue.
Details
In response to this:
Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.