[FG:InPlacePodVerticalScaling] Tracking TODO items to address pre-beta, at beta, GA, and GA+1 #109547
Description
What would you like to be added?
This enhancement tracks various TODO items from alpha to GA for the In-Place Pod Vertical Scaling feature.
To find pending TODO items in the k/k repo, do:
git grep TODO | grep InPlacePodVerticalScalingTODOs not tracked in code:
- Add and expose a helper function to get a pod's resource requirements and allocations for use by metrics, kubectl describe, etc
Why is this needed?
These were TODO items found during review of PR #102884 and it was agreed they should not block alpha. Most need to be handled before Beta, and a few need to be addressed at GA/GA+1.
### Tasks
- [ ] pkg/apis/core/validation/validation.go:5071 - Drop this var once InPlacePodVerticalScaling goes GA and featuregate is gone.
- [ ] pkg/kubelet/container/helpers.go:128 - Remove this in GA+1 and make HashContainerWithoutResources to become Hash.
- [ ] pkg/kubelet/container/runtime.go:301 - Remove this in GA+1 and make HashWithoutResources to become Hash.
- [ ] pkg/kubelet/kubelet.go:1962 - Investigate doing this in HandlePodUpdates + periodic SyncLoop scan
- [ ] pkg/kubelet/kubelet.go:2588 - Can we recover from checkpoint failure in some way? Investigate
- [ ] pkg/kubelet/kubelet.go:2847 - Can we recover from checkpoint update failure in some way? Investigate
- [ ] pkg/kubelet/kubelet.go:2855 - Can we recover from resize status checkpoint failure in some way? Investigate
- [ ] pkg/kubelet/kubelet_pods.go:2107 - Update this to include extended resources in addition to CPU, memory, ephemeral storage. Add test case for extended resources.
- [ ] pkg/kubelet/kuberuntime/helpers_linux.go:63 - Address issue that sets min req/limit to 2m/10m before beta ( See: https://github.com/kubernetes/kubernetes/pull/102884#discussion_r662552642 )
- [ ] pkg/kubelet/kuberuntime/kuberuntime_container_linux_test.go:867 - Add unit tests for cgroup v1 & v2
- [ ] pkg/kubelet/kuberuntime/kuberuntime_manager.go:662 - Figure out best way to get enforceMemoryQoS value (parameter kubelet should know which containers it is managing #4 below) in platform-agnostic way
- [ ] pkg/scheduler/internal/queue/scheduling_queue.go:1074 - Fix this to determine when a pod is truly resized down (might need oldPod if we cannot determine from Status alone)
- [ ] test/e2e/node/pod_resize.go:85 - Can we optimize this?
- [ ] test/e2e/node/pod_resize.go:334 - Is there a better way to determine this?
- [ ] test/e2e/node/pod_resize.go:500 - Remove this check once base-OS updates to containerd>=1.6.9
- [ ] pkg/kubelet/cm/cgroup_manager_linux.go:645 - Support memory request in getCgroupMemoryConfig
- [ ] pkg/kubelet/cm/cgroup_manager_linux.go:731 - Support memory request in setCgroupMemoryConfig