KEP-4210: Fill in KEP for ImageGCMaximumAge

Signed-off-by: Peter Hunt <[email protected]>
kubernetes · Sep 19, 2023 · 9dacaf0 · 9dacaf0
1 parent c0b7032
commit 9dacaf0
Show file tree

Hide file tree

Showing 3 changed files with 407 additions and 0 deletions.
diff --git a/keps/prod-readiness/sig-node/4210.yaml b/keps/prod-readiness/sig-node/4210.yaml
@@ -0,0 +1,3 @@
+kep-number: 4210
+alpha:
+  approver: "@johnbelamaric"
diff --git a/keps/sig-node/4210-max-image-gc-age/README.md b/keps/sig-node/4210-max-image-gc-age/README.md
@@ -0,0 +1,370 @@
+# KEP-4210: ImageMaximumGCAge in Kubelet
+
+<!-- toc -->
+- [Release Signoff Checklist](#release-signoff-checklist)
+- [Summary](#summary)
+- [Motivation](#motivation)
+  - [Goals](#goals)
+  - [Non-Goals](#non-goals)
+- [Proposal](#proposal)
+  - [User Stories (Optional)](#user-stories-optional)
+    - [Story 1](#story-1)
+  - [Notes/Constraints/Caveats (Optional)](#notesconstraintscaveats-optional)
+  - [Risks and Mitigations](#risks-and-mitigations)
+- [Design Details](#design-details)
+  - [Test Plan](#test-plan)
+      - [Prerequisite testing updates](#prerequisite-testing-updates)
+      - [Unit tests](#unit-tests)
+      - [e2e tests](#e2e-tests)
+  - [Graduation Criteria](#graduation-criteria)
+    - [Alpha](#alpha)
+    - [Beta](#beta)
+    - [GA](#ga)
+  - [Upgrade / Downgrade Strategy](#upgrade--downgrade-strategy)
+  - [Version Skew Strategy](#version-skew-strategy)
+- [Production Readiness Review Questionnaire](#production-readiness-review-questionnaire)
+  - [Feature Enablement and Rollback](#feature-enablement-and-rollback)
+  - [Rollout, Upgrade and Rollback Planning](#rollout-upgrade-and-rollback-planning)
+  - [Monitoring Requirements](#monitoring-requirements)
+  - [Dependencies](#dependencies)
+  - [Scalability](#scalability)
+  - [Troubleshooting](#troubleshooting)
+- [Implementation History](#implementation-history)
+- [Drawbacks](#drawbacks)
+- [Alternatives](#alternatives)
+- [Infrastructure Needed (Optional)](#infrastructure-needed-optional)
+<!-- /toc -->
+
+## Release Signoff Checklist
+
+Items marked with (R) are required *prior to targeting to a milestone / release*.
+
+- [ ] (R) Enhancement issue in release milestone, which links to KEP dir in [kubernetes/enhancements] (not the initial KEP PR)
+- [ ] (R) KEP approvers have approved the KEP status as `implementable`
+- [ ] (R) Design details are appropriately documented
+- [ ] (R) Test plan is in place, giving consideration to SIG Architecture and SIG Testing input (including test refactors)
+  - [ ] e2e Tests for all Beta API Operations (endpoints)
+  - [ ] (R) Ensure GA e2e tests meet requirements for [Conformance Tests](https://github.com/kubernetes/community/blob/master/contributors/devel/sig-architecture/conformance-tests.md) 
+  - [ ] (R) Minimum Two Week Window for GA e2e tests to prove flake free
+- [ ] (R) Graduation criteria is in place
+  - [ ] (R) [all GA Endpoints](https://github.com/kubernetes/community/pull/1806) must be hit by [Conformance Tests](https://github.com/kubernetes/community/blob/master/contributors/devel/sig-architecture/conformance-tests.md) 
+- [ ] (R) Production readiness review completed
+- [ ] (R) Production readiness review approved
+- [ ] "Implementation History" section is up-to-date for milestone
+- [ ] User-facing documentation has been created in [kubernetes/website], for publication to [kubernetes.io]
+- [ ] Supporting documentation—e.g., additional design documents, links to mailing list discussions/SIG meetings, relevant PRs/issues, release notes
+
+<!--
+**Note:** This checklist is iterative and should be reviewed and updated every time this enhancement is being considered for a milestone.
+-->
+
+[kubernetes.io]: https://kubernetes.io/
+[kubernetes/enhancements]: https://git.k8s.io/enhancements
+[kubernetes/kubernetes]: https://git.k8s.io/kubernetes
+[kubernetes/website]: https://git.k8s.io/website
+
+## Summary
+
+Add an option ImageMaximumGCAge, which allows an admin to specify a time after which unused images will be garbage collected
+by the Kubelet, regardless of disk usage.
+
+## Motivation
+
+Currently, all image garbage collection the Kubelet is triggered by disk usage going over a threshold (ImageGCLowThresholdPercent).
+However, there are cases that additional conditions could be considered useful. One such condition is maximum age of an image.
+If an image is unused for a long time (the exact amount of time will be decided, but on the order of weeks is what comes to mind),
+then it is not likely to be used again.
+
+One such condition that can be imagined are clusters with automatic upgrades. If a cluster goes through an upgrade process,
+it will likely have images cached from the old release (old kube-apiserver/etcd/etc). While these images would eventually get removed
+through the disk usage condition, they will needlessly occupy disk space before that.
+
+### Goals
+
+- Introduce an option to the Kubelet ImageMaximumGCAge
+
+### Non-Goals
+
+- Introduce other conditions for image garbage collection to trigger.
+    - A WG was put together in SIG-Node to collect other GC use cases, and no other conditions were identified.
+
+## Proposal
+
+Kubelet has three different configuration fields for image garbage collection:
+- ImageMinimumGCAge: the youngest an image can be to be qualified for garbage collection
+- ImageGCLowThresholdPercent: The lowest disk usage will be before garbage collection begins
+- ImageGCHighThresholdPercent: The highest disk usage will be before garbage collection runs each GC period.
+
+Between each of these options, there is a common thread: image garbage collection only triggers when disk usage has reached a certain threshold.
+In other words, there are no alternative conditions the Kubelet will begin triggering collection. Luckily, this condition is the most important:
+the primary goal of garbage collection is to ensure images don't clutter the disk too much and cause it to fill up needlessly.
+However, having the Kubelet be purely reactive means that images will clutter the disk and cause it to fill up. While there aren't reported cases
+where this causes issues, it is inefficient with disk space and can cause the Kubelet to scramble to save disk space when the threshold is met.
+
+An additional approach is to define a way for an admin to request images are cleaned up after they're unused for a certain period of time.
+This would reduce the frequency of the disk usage hitting the level, and provide an admin more flexibility in how garbage collection is defined.
+
+The proposal of this KEP is to add an option to the KubeletConfiguration object that looks like:
+```
+	// imageMaximumGCAge is the maximum age an image can be unused before it is garbage collected.
+	ImageMaximumGCAge metav1.Duration
+```
+
+To begin, this option will be set to the maximum duration (math.MaxInt64), which effectively disables.
+In the future, a more reasonable default may be chosen.
+
+### User Stories (Optional)
+
+#### Story 1
+
+- As a cluster admin, I would like my unused images to be garbage collected in a timely manner, and not occupy disk space forever.
+
+### Notes/Constraints/Caveats (Optional)
+
+<!--
+What are the caveats to the proposal?
+What are some important details that didn't come across above?
+Go in to as much detail as necessary here.
+This might be a good place to talk about core concepts and how they relate.
+-->
+
+### Risks and Mitigations
+
+- If set incorrectly the ImageMaximumGCAge option could cause unneeded image pulls. For instance, if a Cron job ran
+once a wekk, but the ImageMaximumGCAge was set to less than a week, that image would get pulled every week, causing needless
+traffic from the registry
+    - Proper documentation on this is the best way to mitigate this risk.
+    - Defining a good default for this value will be similarly tricky.
+
+## Design Details
+
+Add an option to the Kubelet configuration:
+```
+	// imageMaximumGCAge is the maximum age an image can be unused before it is garbage collected.
+	ImageMaximumGCAge metav1.Duration
+```
+
+This option will be wired down to the Kubelet's [image manager](https://github.com/kubernetes/kubernetes/blob/d5690f12b69a/pkg/kubelet/images/image_gc_manager.go),
+similarly to the other garbage collection fields.
+
+The Kubelet's image manager already keeps track of the last time an image was used through the `lastUsed` field in the
+[imageRecord](https://github.com/kubernetes/kubernetes/blob/d5690f12b69a/pkg/kubelet/images/image_gc_manager.go#L153) structure.
+So a comparison can be made in the realImageGCManager's function
+[GarbageCollect](https://github.com/kubernetes/kubernetes/blob/d5690f12b69a/pkg/kubelet/images/image_gc_manager.go#L288) to garbage collect
+the images that are older than the specified image age.
+
+### Test Plan
+
+[x] I/we understand the owners of the involved components may require updates to
+existing tests to make this code solid enough prior to committing the changes necessary
+to implement this enhancement.
+
+##### Prerequisite testing updates
+
+<!--
+Based on reviewers feedback describe what additional tests need to be added prior
+implementing this enhancement to ensure the enhancements have also solid foundations.
+-->
+
+##### Unit tests
+
+- `pkg/kubelet/images`: `2023-09-14` - `84.2`
+
+Additional tests will be added to pkg/kubelet/images to unit test the new field and verify it works along with the other GC options.
+
+##### e2e tests
+
+- `test/e2e_node/garbage_collector_test.go`
+
+Additional tests will be added to this file to cover the garbage collection e2e.
+
+### Graduation Criteria
+
+
+#### Alpha
+
+- Configuration field added to the Kubelet (disabled by default)
+- Feature supported by Kubelet Image Manager
+- Unit tests and e2e tests added
+- Add a metric `kubelet_image_garbage_collected_total` which tracks the number of images the kubelet is GC'ing through any mechanism.
+
+#### Beta
+
+- Gather feedback from users
+- Depending on PRR review and SIG-Node opt-in, this feature could start as a disabled Beta.
+
+#### GA
+
+- Addition of conformance tests
+- Some examples of real-world usage
+- Allowing time for feedback
+
+### Upgrade / Downgrade Strategy
+
+This option is purely contained within the Kubelet, so the only concern is the flag is added to the configuration of the newer
+Kubelet and then downgraded.
+
+There's nothing the Kubernetes community can do to prevent this, and admins should ensure their configuration fields will function with
+the processes they run.
+
+### Version Skew Strategy
+
+Version skew is not a worry assuming the internal Kubelet changes are synchronized with the configuration changes.
+
+## Production Readiness Review Questionnaire
+
+
+### Feature Enablement and Rollback
+
+###### How can this feature be enabled / disabled in a live cluster?
+
+- [ ] Feature gate (also fill in values in `kep.yaml`)
+  - Feature gate name:
+  - Components depending on the feature gate:
+- [x] Other
+  - Describe the mechanism:
+    - Kubelet configuration field
+  - Will enabling / disabling the feature require downtime of the control
+    plane?
+    - No, it's a kubelet only configuration
+  - Will enabling / disabling the feature require downtime or reprovisioning
+    of a node?
+    - Potentially, to restart the Kubelet.
+
+###### Does enabling the feature change any default behavior?
+
+Yes, when the configuration option is set images will potentially be garbage collected earlier.
+
+###### Can the feature be disabled once it has been enabled (i.e. can we roll back the enablement)?
+
+Yes, given a restart of the Kubelet.
+
+###### What happens if we reenable the feature if it was previously rolled back?
+
+- Nothing unexpected.
+
+###### Are there any tests for feature enablement/disablement?
+
+There will be a test to verify when the Kubelet configuration option is disabled that the image isn't GC'd early.
+
+### Rollout, Upgrade and Rollback Planning
+
+###### How can a rollout or rollback fail? Can it impact already running workloads?
+
+- Invalid configuration configured.
+- Even in the case where the ImageMaximumGCAge is set to 0, the Kubelet will only GC images when their corresponding containers are
+removed, so no running workloads can be affected.
+
+###### What specific metrics should inform a rollback?
+
+- `kubelet_image_garbage_collected_total` metric drastically (100x) increasing, indicating thrashing of the GC manager and
+  images being pulled.
+
+###### Were upgrade and rollback tested? Was the upgrade->downgrade->upgrade path tested?
+
+They will be, there should be no side effects.
+
+###### Is the rollout accompanied by any deprecations and/or removals of features, APIs, fields of API types, flags, etc.?
+
+No.
+
+### Monitoring Requirements
+
+###### How can an operator determine if the feature is in use by workloads?
+
+- Verify the Kubelet Configuration with the Kubelet's configz endpoint
+- Monitor the `kubelet_image_garbage_collected_total`, and expect a slight increase.
+
+###### How can someone using this feature know that it is working for their instance?
+
+- [x] Other (treat as last resort)
+  - `kubelet_image_garbage_collected_total` metric increases when an image ages out.
+
+###### What are the reasonable SLOs (Service Level Objectives) for the enhancement?
+
+- The eventual default value should increase the average `kubelet_image_garbage_collected_total` by no more than 10x
+    - TODO: On what clusters?
+
+###### What are the SLIs (Service Level Indicators) an operator can use to determine the health of the service?
+
+- [x] Metrics
+  - Metric name: `kubelet_image_garbage_collected_total`
+  - Components exposing the metric: Kubelet
+
+###### Are there any missing metrics that would be useful to have to improve observability of this feature?
+
+- A metric for each different GC trigger (disk usage vs time based).
+
+### Dependencies
+
+###### Does this feature depend on any specific services running in the cluster?
+
+Just Kubelet
+
+### Scalability
+
+###### Will enabling / using this feature result in any new API calls?
+
+- Kubelet will call `RemoveImage` to the CRI implementation when an image should be garbage collected,
+  which could happen more frequently/faster.
+
+###### Will enabling / using this feature result in introducing new API types?
+
+No
+
+###### Will enabling / using this feature result in any new calls to the cloud provider?
+
+No
+
+###### Will enabling / using this feature result in increasing size or count of the existing API objects?
+
+KubeletConfiguration will gain an additional int64
+
+###### Will enabling / using this feature result in increasing time taken by any operations covered by existing SLIs/SLOs?
+
+No
+
+###### Will enabling / using this feature result in non-negligible increase of resource usage (CPU, RAM, disk, IO, ...) in any components?
+
+- Potentially, depending on the age chosen, there could be more CPU used to do the image removal.
+  - The frequency of the image removal will be a tradeoff for existing disk space
+
+###### Can enabling / using this feature result in resource exhaustion of some node resources (PIDs, sockets, inodes, etc.)?
+
+- Not likely, it's intended to prevent resource exhaustion of disk
+
+### Troubleshooting
+
+###### How does this feature react if the API server and/or etcd is unavailable?
+
+- N/A
+
+###### What are other known failure modes?
+
+- The Kubelet could thrash with itself in a image pull/remove cycle if the value is set too low.
+
+###### What steps should be taken if SLOs are not being met to determine the problem?
+
+- Set a minimum value this field could be.
+
+## Implementation History
+
+
+2023-09-18: KEP opened, targeted at Alpha
+
+## Drawbacks
+
+- It could be considered unnecessary, as the disk usage based garbage collection already covers this use case, albeit slower.
+
+## Alternatives
+
+- Add a Kubelet garbage collection plugin system
+    - Too complicated, probably won't be needed.
+    - The Image GC WG off of SIG-Node brainstormed use cases and could only think of one other--image GC based on pod priority,
+      which we deemed not useful enough to pursue a generic solution.
+- Delegate responsibility down to CRI
+    - Would cause code duplication between CRI implementaions, out of scope for this.
+
+## Infrastructure Needed (Optional)
+
+N/A