Weekly GitHub Report for Kubernetes: February 15, 2026 - February 22, 2026 (14:51:41)
Weekly GitHub Report for Kubernetes
Thank you for subscribing to our weekly newsletter! Each week, we deliver a comprehensive summary of your GitHub project's latest activity right to your inbox, including an overview of your project's issues, pull requests, contributors, and commit activity.
Table of Contents
I. News
1.1 Recent Version Releases:
The current version of this repository is v1.32.3
1.2 Version Information:
The Kubernetes version released on March 11, 2025, introduces key updates detailed in the official CHANGELOG, with additional binary downloads available. For comprehensive information on new features and changes, users are encouraged to consult the Kubernetes announce forum and the linked CHANGELOG.
II. Issues
2.1 Top 5 Active Issues:
We consider active issues to be issues that that have been commented on most frequently within the last week. Bot comments are omitted.
-
[SIG/SCHEDULING] [KIND/FLAKE] [NEEDS-TRIAGE] Flaking UT k8s.io/kubernetes/pkg/scheduler:TestSchedulerScheduleOne/prebind_pod_cancelled_during_prebind: This issue reports a flaking unit test failure in the Kubernetes scheduler related to the test
TestSchedulerScheduleOne/prebind_pod_cancelled_during_prebind, which started occurring after a specific pull request was merged. The failure is caused by a race condition where the nominated node name (NNN) is not properly cleared on prebind failure due to asynchronous API calls and stale pod status updates, leading to inconsistent pod state and test expectations.- The comments discuss the root cause involving asynchronous API call delays and stale pod data in the scheduler's update logic, propose several approaches to fix the issue including stronger pod retrieval methods and always updating the nominated node name, and share detailed debugging insights and logs; a fix is pending review to address the flakiness.
- Number of comments this week: 11
-
[KIND/CLEANUP] [SIG/SCHEDULING] [NEEDS-TRIAGE] [AREA/WORKLOAD-AWARE] Clean up the kubernetes/kubernetes repository from v1alpha1 PriorityClass API: This issue concerns the removal of the deprecated v1alpha1 version of the PriorityClass API from the kubernetes/kubernetes repository, as the API has been generally available since 2019 and its continued presence is no longer necessary. The discussion focuses on cleaning up all references and directories related to v1alpha1 PriorityClass after a prerequisite pull request merges, ensuring a thorough and coordinated cleanup process.
- The comments reveal a collaborative effort where one contributor self-assigned the issue and sought permission to proceed, receiving guidance to wait for a related pull request to merge first. Detailed instructions were provided on the scope of the cleanup, including manual updates beyond just removing type definitions, and the conversation included sharing resources and plans for further engagement through a new Slack channel to facilitate ongoing contributions.
- Number of comments this week: 9
-
[SIG/SCALABILITY] [SIG/API-MACHINERY] [KIND/FEATURE] [TRIAGE/ACCEPTED] Interning for API Server and controller memory reduction: This issue proposes implementing object interning techniques in the Kubernetes API Server and control plane components to reduce memory usage, focusing on string interning during protobuf deserialization and deduplication of Server-Side Apply metadata. The goal is to address memory bottlenecks caused by highly replicated workloads and large resource sizes by reusing existing objects and optimizing JSON handling, supported by benchmarks demonstrating significant memory savings.
- The comments include requests for prototype commits and benchmarks on non-duplicate data, clarifications on the mutability and safety of shared objects, suggestions to convert mutable byte slices to strings for safety, volunteer contributions to improve deduplication, and acceptance of the proposal for further development.
- Number of comments this week: 9
-
[SIG/STORAGE] [KIND/FLAKE] [NEEDS-TRIAGE] [Flaking Tests][sig-storage] CSI Volumes [Driver: csi-hostpath] [Testpattern: Dynamic PV] suite: This issue reports flaking tests in the Kubernetes e2e suite related to the CSI Volumes with the csi-hostpath driver using the Dynamic PV test pattern, where several tests intermittently fail due to errors creating SnapshotMetadataService CRDs. The failures have been observed across multiple jobs since February 14, 2026, and the root cause appears to be conflicts with existing custom resource definitions during test execution.
- The comments confirm that the flaking occurs across additional test jobs beyond the initially reported one, with multiple triage links provided for further investigation, indicating ongoing recognition and tracking of the flake by the release team.
- Number of comments this week: 3
-
[SIG/STORAGE] [KIND/FLAKE] [NEEDS-TRIAGE] [Flaking tests][sig-storage] gce-cos-master-slow - CSI Volumes [Driver: csi-hostpath] [Testpattern: Dynamic PV] suite: This issue reports flaking tests in the Kubernetes e2e suite related to the CSI Volumes driver csi-hostpath, specifically under the Dynamic PV test pattern, which have been failing intermittently since February 13, 2026. The failures involve creating snapshot-metadata resources and are characterized by errors such as exceeding the maximum allowed byte grid size, impacting the stability of the gce-cos-master-slow job.
- The comments show interest from contributors in investigating the flake by reviewing logs and attempting local reproduction, references to related discussions, and inquiries about the impact of this issue on the upcoming v1.36.0-alpha.2 release cut.
- Number of comments this week: 3
2.2 Top 5 Stale Issues:
We consider stale issues to be issues that has had no activity within the last 30 days. The team should work together to get these issues resolved and closed as soon as possible.
As of our latest update, there are no stale issues for the project this week.
2.3 Open Issues
This section lists, groups, and then summarizes issues that were created within the last week in the repository.
Issues Opened This Week: 25
Summarized Issues:
- CSI Volume Flaking Tests: Multiple flaking tests related to CSI Volumes with the csi-hostpath driver and dynamic persistent volumes have been reported, involving failures caused by attempts to create SnapshotMetadataService CRDs that already exist and exceeding byte grid size limits. These issues affect e2e tests on different platforms and test patterns, causing intermittent test failures.
- issues/137039, issues/137040
- Flaky and Failing Tests in Scheduler and PodGroup: Several flaky tests have been reported in the Kubernetes scheduler and podgroup integration tests, including race conditions during prebind phases and intermittent failures in podgroup scheduling tests. These flakes cause inconsistent pod status updates and test failures during scheduling and integration test runs.
- issues/137125, issues/137178
- Flaky Tests in Integration and E2E Suites: Flaky test failures have been observed in various integration and e2e test suites, including pod certificate requests, kubectl logs retrieval, and cloud-provider service controller tests. These flakes cause intermittent failures affecting ARM64 master jobs, deployment log retrieval, and load balancer synchronization tests.
- issues/137179, issues/137180, issues/137184
- Deprecated API and Function Removal: The removal of deprecated APIs and functions is ongoing, including the v1alpha1 PriorityClass API and the wait.NewExponentialBackoffManager function. These removals are part of cleanup efforts to maintain up-to-date and supported codebases.
- issues/137054, issues/137068
- Port Forwarding Implementation Improvement: A proposal exists to replace the manual port-forward implementation with a direct call to kubectl’s PortForwardOptions.RunPortForwardContext method. This change aims to simplify and standardize port-forwarding logic in the codebase.
- issues/137058
- Snapshot and Scheduling State Consistency: Adding a snapshot of PodGroupState before scheduling cycles is proposed to ensure consistent and stable state visibility across workload scheduling plugins. This addresses issues with live cluster state changes and duplicated assume states during scheduling.
- issues/137071
- Resource Metrics Reporting Issues: The container_swap_usage_bytes metric exposed by Kubelet always reports zero despite active swap usage, indicating a discrepancy between this metric and other correctly reporting swap usage metrics. This causes inaccurate resource usage reporting for containers.
- issues/137093
- Test Failures Due to Infrastructure and Security Issues: Some test jobs fail due to infrastructure problems like SSH command failures and timeouts, or security issues such as OpenPGP signature verification failures and deprecated SHA1 algorithm usage. These failures result in skipped tests and verification errors.
- issues/137096, issues/137097
- Memory Usage and Leak Problems: Memory leaks and high memory usage issues have been identified, including a leak in the kube-controller-manager's Service controller due to unremoved map entries and proposals to reduce memory usage via object interning and ManagedFields deduplication. These issues impact scalability and resource efficiency.
- issues/137109, issues/137151
- Dynamic Resource Scheduler Flake: A flaky test failure in the dynamic resource scheduler plugin has been reported, with no clear cause identified despite reproducibility. This flake affects the reliability of the scheduler's extended resource handling.
- issues/137113
- Scale Test Failures on GCE: Since a recent merge, the 5000-node GCE scale performance test consistently fails in multiple ClusterLoaderV2 test steps related to huge-service and load scenarios, while the AWS equivalent test remains unaffected. This indicates platform-specific scale test regressions.
- issues/137117
- Kubelet Plugin Optimization Proposal: A proposal suggests enhancing kubelet to optionally skip calls to NodePrepareResources and NodeUnprepareResources for certain device classes that do not require node-local drivers. This aims to prevent pods with ResourceClaims from being stuck in Terminating state and reduce pod startup delays.
- issues/137122
- HPA External Metrics Test Coverage Expansion: The HPA External Metrics testing framework is proposed to be expanded with multiple test cases covering multiple metrics, stabilization windows, and scaling limits. This aims to improve reliability and coverage of HPA behavior under realistic scenarios.
- issues/137132
- Validation Error Message Correction: The validation error message for the length of restartPolicyRules[].exitCodes.values incorrectly reports the limit in bytes instead of the number of elements, causing misleading feedback when the array exceeds 255 items. This requires correction for accurate validation messaging.
- issues/137135
- Table Format Streaming Test Flake: A flake in table format streaming tests is caused by timing discrepancies in CreationTimestamp to Age conversion, where list and watch operations complete at slightly different times, resulting in inconsistent age values and test failures.
- issues/137148
- kubectl Describe Node Resource Display Issue: The kubectl describe node command fails to show correct pod-level resource values in the Non-terminated Pods and Allocated resources sections, even when pod-level resources are set. This leads to inaccurate resource reporting in node descriptions.
- issues/137158
2.4 Closed Issues
This section lists, groups, and then summarizes issues that were closed within the last week in the repository. This section also links the associated pull requests if applicable.
Issues Closed This Week: 8
Summarized Issues:
- Flaking Tests in Kubernetes E2E and Integration Suites: Several issues report intermittent failures in Kubernetes end-to-end and integration tests, including service type changes causing unreachable services, inconsistent log retrieval from pods, and a flaky webhook conversion test related to etcd3 watcher code. These flaking tests affect reliability and stability of CI pipelines, making debugging and consistent validation challenging.
- issues/136687, issues/136948, issues/136978
- Kubelet Certificate Request Bugs and Logging Improvements: There are issues with the kubelet incorrectly setting ownerReference apiVersion on PodCertificateRequest objects, which prevents proper garbage collection, and a request to add debug-level logging for CertificateSigningRequest creation to aid troubleshooting. These problems hinder certificate lifecycle management and reduce observability in kubelet operations.
- issues/136852, issues/136890
- API Version and Resource Usage Updates in Tests: One issue highlights the need to update Horizontal Pod Autoscaler (HPA) end-to-end tests by replacing deprecated API versions and switching from ReplicationController to ReplicaSet to maintain compatibility with current stable Kubernetes APIs. This update is necessary to keep tests aligned with evolving Kubernetes API standards.
- issues/137074
- Performance Regression in kube-proxy due to iptables Backend Change: A significant performance regression of 5-6x was observed in kube-proxy's iptables mode on AWS nodes after switching from legacy iptables to the iptables-nft backend, which adds overhead by translating iptables operations to nftables. This regression impacts network performance and efficiency in affected clusters.
- issues/137085
- Flakiness in Kubernetes Scheduler Test: The TestSchedulerScheduleOne test intermittently fails in continuous integration environments but cannot be reproduced locally despite extensive stress testing, indicating a challenging-to-diagnose flakiness in the scheduler package. This issue complicates reliable scheduler validation and CI stability.
- issues/137118
2.5 Issue Discussion Insights
This section will analyze the tone and sentiment of discussions within this project's open and closed issues that occurred within the past week. It aims to identify potentially heated exchanges and to maintain a constructive project environment.
Based on our analysis, there are no instances of toxic discussions in the project's open or closed issues from the past week.
III. Pull Requests
3.1 Open Pull Requests
This section provides a summary of pull requests that were opened in the repository over the past week. The top three pull requests with the highest number of commits are highlighted as 'key' pull requests. Other pull requests are grouped based on similar characteristics for easier analysis. Up to 25 pull requests are displayed in this section, while any remaining pull requests beyond this limit are omitted for brevity.
Pull Requests Opened This Week: 83
Key Open Pull Requests
1. Add placement generator plugin interfaces and logic for running them: This pull request adds the necessary interfaces and implementation logic for placement generator plugins in Kubernetes, enabling the framework to run these plugins to generate placement arrays as part of the topology-aware workload scheduling enhancement described in the related KEP.
- URL: pull/137083
- Associated Commits: 4916f, 4444f, 1b2b0, 8175e, d30b9, 5797c, c1a8c, 6bba5, f7eba, c6e03, 4d1a2, 26dc8, 4efdb, 6d3fa, dc6ce, f5e9b, d76d0
2. DRA device taints: graduate to beta: This pull request promotes the Device Resource Allocation (DRA) device taints feature to beta status by enabling it by default, adds DeviceTaintRule support to the v1beta2 API while keeping it off by default due to the API group's status, and maintains backward compatibility with the deprecated v1beta1 API.
- URL: pull/137170
3. [WIP] EvictionRequest controller: This pull request implements a new EvictionRequest controller within the kube-controller-manager to manage the graceful eviction of pods, introducing the EvictionRequest API and related features as part of the Kubernetes enhancement proposal referenced.
- URL: pull/137133
Other Open Pull Requests
- Podresources API fixes and improvements: This work-in-progress pull request addresses multiple issues in the podresources API lane by partially reverting a previous PR to restore backward compatibility and improving end-to-end tests. It also removes redundant tests, filters out inactive pods in the Get() method, reorganizes helper functions, adds additional tests for general availability, and fixes gRPC connection flakiness.
- EvictionRequest API introduction: This pull request introduces a new declarative EvictionRequest API with a .spec.evictionInterceptors field for Pods, enabling coordinated and graceful eviction through requesters and interceptors. This feature is aligned with the related Kubernetes Enhancement Proposal (KEP).
- GPU sharing feature implementation: This pull request adds fine-grained, dynamic GPU sharing by implementing a custom scheduler plugin and device manager extension. It allows multiple pods to share fractional GPU resources efficiently via the
gpushare.com/vgpuresource, improving GPU utilization and scheduling flexibility for smaller AI and inference workloads.
- Declarative validation rules graduation: This pull request graduates several declarative validation rules for the
scheduling.x-k8s.io/v1alpha1API group from alpha to stable status by promoting validation tags and removing legacy handwritten validation functions and tests. The changes follow the migration strategy outlined in KEP-5073.
- Port-forward implementation refactor: This pull request replaces the manual port-forward implementation with kubectl's
PortForwardOptions.RunPortForward(), introducing acustomPortForwarderstruct that wraps client-go'sPortForwarder. This reduces code duplication, improves maintainability, and fixes blocking and stopping issues in port-forwarding behavior.
- Optionalorrequired linter enablement: Multiple pull requests enable the optionalorrequired linter for various API groups including autoscaling, apps, apidiscovery, and flowcontrol. These changes address lint errors by properly tagging fields as optional or required without introducing user-facing changes or breaking compatibility.
- Apiserver startup and controller fixes: This pull request fixes an apiserver startup failure during upgrades on large clusters with MultiCIDRServiceAllocator enabled by modifying the RepairIPAddress controller to retry on Forbidden errors until the namespace informer finishes syncing. This ensures the controller succeeds instead of failing the PostStartHook.
- ServiceAccount metrics promotion: This pull request promotes deprecated alpha ServiceAccount metrics to beta status, improving their stability and usability in Kubernetes.
- Etcd library update proof of concept: This work-in-progress pull request updates etcd library versions by excluding the gogo protobuf dependency and involves pinned replacements of etcd modules along with related dependency and linting script updates.
- Storage/cacher metrics promotion: This pull request promotes storage/cacher metrics in the Kubernetes apiserver to BETA stability by adding unit tests, updating documentation and test data, and introducing a global function for testing Gauge metrics.
- Apiserver watch metrics promotion: This pull request promotes
apiserver_watch_events_totalandapiserver_watch_events_sizesmetrics to BETA status, includes tests for these metrics, and updates related documentation and stable metrics listings.
- Validation-gen linting enhancements: This pull request implements advanced linting rules in the
validation-gentool to enforce declarative validation marker compliance with stability and structural requirements from KEP 5073. It integrates linting into the code generation pipeline, removes redundant tag rules, and fixes existing API lint errors.
- Kube-apiserver connection recovery improvements: This pull request improves kube-apiserver by enabling recovery from connections mistakenly made to incorrect servers during APIService availability checks. It enforces optional TLS validation, invalidates cached bad connections on unexpected responses or network errors, and includes tests for these behaviors.
- Image pull record key and interface clarification: This pull request fixes a bug by changing the primary key in image pull records from a vague image reference to a stable image ID and renames
GetImageRef()toGetImageID()to clarify the kubelet interface and ensure consistent querying of image pull data.
- Noreference linter rule enablement: This pull request enables the
noreferencerule in the Kube API Linter to enforce using "Ref"/"Refs" instead of "Reference"/"References" in field names, while temporarily adding exceptions for certain fields to be renamed later.
- StrictIPCIDRValidation feature gate beta: This pull request updates the
StrictIPCIDRValidationfeature gate to Beta status in kube-apiserver, enabling stricter validation that disallows IP or CIDR values with extraneous leading zeros or ambiguous subnet/mask semantics as specified in KEP-4858.
- Kube-apiserver and kube-aggregator metrics promotion: This pull request promotes a set of kube-apiserver and kube-aggregator metrics from Alpha to Beta stability, maintaining existing metric names and label sets to ensure compatibility for metric consumers.
- Device metadata feature in kubeletplugin: This pull request adds an
enable-device-metadataflag to the kubeletplugin that uses CDI mounts to automatically populate a metadata file for each device claim, supported by a new DeviceMetadata API schema for reliable metadata consumption.
- Uniquemarkers linter enablement: This pull request enables the uniquemarkers linter in Kubernetes to satisfy static analysis requirements without changing API semantics, behavior, or compatibility, addressing issue #136878.
- Upstream Go unit tests support: This pull request adds support for running upstream Go unit tests within the downstream Kubernetes project, enabling tests for the logtools linter and an etcd client package by adjusting test execution for vendored upstream packages and workspace changes.
- Dual-stack end-to-end test fixes: This pull request fixes and cleans up dual-stack end-to-end tests by correcting logic to assert that in a dual-stack cluster, at least one node has an IPv4 IP and at least one node has an IPv6 IP, instead of requiring every node to have both.
- Coscheduling plugin introduction: This pull request introduces the Coscheduling plugin to the kube-scheduler framework, adding a DesiredCount mechanism that delays scheduling of pod groups until a specified number of pods are present. It also includes a centralized PodGroupPolicyEvaluator utility to unify scheduling logic between Coscheduling and GangScheduling plugins.
3.2 Closed Pull Requests
This section provides a summary of pull requests that were closed in the repository over the past week. The top three pull requests with the highest number of commits are highlighted as 'key' pull requests. Other pull requests are grouped based on similar characteristics for easier analysis. Up to 25 pull requests are displayed in this section, while any remaining pull requests beyond this limit are omitted for brevity.
Pull Requests Closed This Week: 58
Key Closed Pull Requests
1. e2e: node: fix existing podresources tests: This pull request aims to fix and clean up existing end-to-end podresources tests in the Kubernetes node component by removing redundant tests, improving test failure diagnostics with GinkgoHelper calls, and correcting the handling and filtering of terminated pods in the Get() API calls to ensure accurate test expectations and functionality.
- URL: pull/137111
2. KEP-4671: Introduce Workload Scheduling Cycle: This pull request introduces a Workload Scheduling Cycle into the Kubernetes scheduler's main loop to initiate scheduling when any pod from a pod group is dequeued, laying the groundwork for enhanced workload-aware scheduling while still requiring additional features like gang scheduling plugin logic, observability, and test coverage.
- URL: pull/136618
3. Update/kubectl in kustomize to v5.8.1: This pull request upgrades the kubectl kustomize functionality by updating the kustomize dependency from version 5.7.1 to 5.8.1, including a cleanup of dependencies such as the removal of the unwanted github.com/pkg/errors package.
- URL: pull/136892
Other Closed Pull Requests
3.3 Pull Request Discussion Insights
This section will analyze the tone and sentiment of discussions within this project's open and closed pull requests that occurred within the past week. It aims to identify potentially heated exchanges and to maintain a constructive project environment.
Based on our analysis, there are no instances of toxic discussions in the project's open or closed pull requests from the past week.
IV. Contributors
4.1 Contributors
Active Contributors:
We consider an active contributor in this project to be any contributor who has made at least 1 commit, opened at least 1 issue, created at least 1 pull request, or made more than 2 comments in the last month.
If there are more than 10 active contributors, the list is truncated to the top 10 based on contribution metrics for better clarity.
| Contributor | Commits | Pull Requests | Issues | Comments |
|---|---|---|---|---|
| pohly | 65 | 13 | 2 | 25 |
| brejman | 35 | 4 | 0 | 35 |
| atiratree | 12 | 1 | 0 | 58 |
| dims | 44 | 1 | 0 | 1 |
| macsko | 10 | 1 | 0 | 34 |
| lalitc375 | 17 | 6 | 0 | 11 |
| yongruilin | 31 | 1 | 0 | 0 |
| danwinship | 15 | 6 | 0 | 8 |
| Argh4k | 12 | 2 | 0 | 15 |
| liggitt | 12 | 4 | 0 | 12 |
Access Last Week's Newsletter: