What kubernetes/kubernetes shipped
Written by FoxPlug from the repository's own releases; not affiliated with Kubernetes. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/kubernetes/kubernetes; Kubernetes did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of November 2, 2026
What shipped
- Volume health APIs were updated to the newest KEP version, moving health reporting from the pod to pod.Status.VolumeStatus. Pull request #143644
- The ContainerStopSignals feature gate graduated to beta in v1.38 and is on by default, with a new
kubelet_pod_stop_signalsmetric. Pull request #143690 - Dynamic resize of memory-backed volumes received a beta update supporting adding and removing sizeLimit on emptyDir volumes with Memory medium. Pull request #141807
- Admission webhook metrics were promoted to beta, adding _total replacements and keeping deprecated _count aliases. Pull request #143357
- Jobs now report a SchedulingBlocked condition on status when their scheduling objects cannot be used. Pull request #143536
- Job failure condition reasons became more granular, using an optional name on each PodFailurePolicy rule. Pull request #142913
- A socket leak was fixed by closing the underlying connection when a SPDY upgrade fails. Pull request #143073
- Kube-proxy now honors conntrack max even above the per-core sizing cap. Pull request #143794
- Runtime class admission no longer fails a pod whose overhead field is an empty map. Pull request #124899
- Mount-utils now decodes kernel octal escapes in mount paths read from mount tables. Pull request #143108
Why it matters
This week brings several user-visible changes: volume health reporting moves to pod status, container stop signals move to beta, memory-backed volume resizing gets a beta update, and Job scheduling failures gain clearer status conditions. Admission webhook metrics are promoted to beta, while several bugs affecting admission, kube-proxy conntrack, SPDY connections and mount path parsing are fixed.
Changelog entry
- Updated volume health APIs to the newest KEP version, moving volume health from the pod to pod.Status.VolumeStatus. Pull request #143644
- Graduated the ContainerStopSignals feature gate to beta in v1.38, on by default, and added the
kubelet_pod_stop_signalsmetric. Pull request #143690 - Beta update for dynamic resize of memory-backed volumes, supporting adding and removing sizeLimit on emptyDir volumes with Memory medium. Pull request #141807
- Promoted admission webhook metrics to beta and introduced _total replacements while retaining deprecated _count aliases. Pull request #143357
- Jobs now set SchedulingBlocked=True and report why when their scheduling objects cannot be used. Pull request #143536
- Added an optional name to each PodFailurePolicy rule and used it in the Job failure condition reason. Pull request #142913
- Closed the underlying connection when a SPDY upgrade fails, fixing a socket leak in CLOSE_WAIT. Pull request #143073
- Kube-proxy now honors conntrack max above the per-core sizing cap. Pull request #143794
- Runtime class admission no longer fails requests with an empty overhead map. Pull request #124899
- Mount-utils now decodes kernel octal escapes in mount paths. Pull request #143108
- Promoted the CloudControllerManagerWatchBasedRoutesReconciliation feature gate to beta in 1.38, default off. Pull request #143876
- Admission policy matching now evaluates resource rules before selectors. Pull request #143859
This week: volume health APIs updated to pod.Status.VolumeStatus, ContainerStopSignals to beta, dynamic resize of memory-backed volumes beta update, admission webhook metrics to beta, and fixes for SPDY socket leaks, conntrack sizing and mount path escapes.
This week in the repository: volume health APIs were updated to the newest KEP version, moving health from the pod to pod.Status.VolumeStatus; the ContainerStopSignals feature gate graduated to beta in v1.38 and is on by default, adding the kubelet_pod_stop_signals metric; dynamic resize of memory-backed volumes received a beta update that supports adding and removing sizeLimit on emptyDir volumes with Memory medium; and admission webhook metrics were promoted to beta. Jobs now report a SchedulingBlocked condition when their scheduling objects cannot be used, and Job failure condition reasons are more granular. Bug fixes include closing the underlying connection when a SPDY upgrade fails, honoring conntrack max above the per-core sizing cap in kube-proxy, not failing admission on an empty
Week of October 26, 2026
What shipped
- The DRA scheduler now keeps a claim allocation made in PreBind when unreserving, so a failed bind no longer discards it. Pull request #143450
- The watch cache no longer sends a periodic bookmark one second after the initial-events-end bookmark of a watchlist. Pull request #143595
- The SchedulerPreQueueingHints feature gate is promoted to beta and correctness bugs in the earlier PreQueueingHint work are fixed. Pull request #141732
- PodGroup names for Jobs are now UID-aware so sibling Jobs sharing a template get distinct PodGroups, and spec.scheduling is rejected with template schedulingGroup. Pull request #143127
- A data race in waitingPod.stopWithStatus in the scheduler waiting pods map is fixed. Pull request #142429
- Kubectl no longer truncates a --kuberc path that contains an equals sign. Pull request #143586
- Kubectl error messages for objects with no REST mapping no longer break onto two lines. Pull request #143558
- Kubectl kuberc SavePreference now closes the kuberc file it opens. Pull request #143381
- Kubectl GetLogVerbosity now matches the verbosity flag by prefix instead of any argument containing it. Pull request #143385
- Kubeadm shows the missing CA hint again, telling users to run kubeadm init phase certs ca. Pull request #143557
Why it matters
This week brought several scheduler fixes, including keeping DRA claim allocations through unreserve, fixing a waiting pod data race and promoting PreQueueingHint to beta. Watch cache and kubectl also got correctness fixes, from an extra bookmark after watchlist start to truncated kuberc paths and garbled error messages. Users of preemption, DRA and kubectl command line configuration should see fewer flaky binds and clearer output.
Changelog entry
- DRA scheduler: keep the claim allocation from PreBind when unreserving. Pull request #143450
- Watch cache: avoid sending a periodic bookmark one second after the watchlist initial-events-end bookmark. Pull request #143595
- Scheduler: promote the SchedulerPreQueueingHints feature gate to beta and fix narrowing bugs. Pull request #141732
- Workload-aware scheduling: use UID-aware PodGroup names so sibling Jobs sharing a template get distinct PodGroups, and reject spec.scheduling with template schedulingGroup. Pull request #143127
- Scheduler: fix a data race in waitingPod.stopWithStatus. Pull request #142429
- Kubectl: stop truncating --kuberc paths that contain an equals sign. Pull request #143586
- Kubectl: fix error messages breaking onto two lines when an object has no REST mapping. Pull request #143558
- Kubectl: close the kuberc file in kuberc SavePreference. Pull request #143381
- Kubectl: match the verbosity flag by prefix in GetLogVerbosity. Pull request #143385
- Kubeadm: show the missing CA hint again when CA files are absent. Pull request #143557
- Mount-utils: label the NeedResize format-check error as ResizeFS.NeedResize instead of ResizeFS.Resize. Pull request #140319
- Leaderelection: allow tolerance for timer jitter in TestRecoveryEnds. Pull request #143601
This week: DRA scheduler keeps PreBind allocations on unreserve, watch cache stops sending an extra bookmark 1s after the watchlist initial bookmark, PreQueueingHints goes to beta, and kubectl fixes truncated --kuberc paths, two-line errors and verbosity flag matching.
This week in the Kubernetes repository: the DRA scheduler now keeps a claim allocation stored during PreBind when unreserving, so a failed bind no longer drops it. The watch cache no longer sends a periodic bookmark a second after the watchlist initial-events-end bookmark. The SchedulerPreQueueingHints gate is promoted to beta with correctness fixes. PodGroup names for Jobs are now UID-aware so sibling Jobs sharing a template get distinct groups. kubectl fixes include --kuberc paths containing an equals sign, error messages wrapping onto two lines, verbosity flag matching by prefix, and closing the kuberc file in SavePreference. kubeadm shows the missing CA hint again.
Week of October 19, 2026
What shipped
- The ShardedListAndWatch feature gate graduates to Beta. Pull request #142702
- Watch event conversion now uses a shallow struct copy for Added and Modified events instead of a full object deep copy. Pull request #142935
- Pod hugepage limits are now kept across an in-place resize instead of being reset to zero. Pull request #141991
- A pod terminated with a failure status now keeps its failure phase when the container handles SIGTERM by exiting 0. Pull request #142761
- The kubelet env file parser now rejects unseparated hash suffixes, matching how Bash reads the assignment. Pull request #143221
- Kubectl now lets --kuberc take precedence over KUBERC when it points to the default path. Pull request #143277
- Client-go no longer emits a zero informer store resource version before initial sync. Pull request #143314
- The CRI API adds
cgroup_mount_mode, the API changes needed for writable cgroups. Pull request #139824 - The kubelet keeps a rotated log when compressing it fails at close instead of losing it. Pull request #143147
- Partially declared cpu and memory limits are now treated as unlimited in AggregateContainerLimits(). Pull request #141014
Why it matters
This week brings the ShardedListAndWatch gate to Beta and trims allocation work on the watch path in the apiserver, both aimed at how list and watch traffic is served. Several kubelet and node fixes change how pods behave around resize, termination and logging, so some clusters will see different results after upgrading. The CRI API also gains cgroup_mount_mode, opening the door to writable cgroups in the container runtime interface.
Changelog entry
- Graduated ShardedListAndWatch to Beta. Pull request #142702
- Replaced the full object deep copy in watch event conversion with a shallow struct copy for Added and Modified events. Pull request #142935
- Kept Pod hugepage limits across an in-place resize. Pull request #141991
- Kept the failure phase when a terminated container exits 0. Pull request #142761
- Kubelet now rejects unseparated hash suffixes in env files. Pull request #143221
- Kubelet keeps the rotated log when compressing it fails at close. Pull request #143147
- Fixed --kuberc flag precedence when pointing to the default path. Pull request #143277
- Client-go no longer emits a zero informer store resource version before sync. Pull request #143314
- Added
cgroup_mount_modeto the CRI API. Pull request #139824 - Treated partially declared cpu and memory limits as unlimited in AggregateContainerLimits(). Pull request #141014
- Kept watch cache snapshots when a consistency check passes. Pull request #143321
- Bumped the etcd SDK to 3.7.3. Pull request #143358
This week: ShardedListAndWatch graduates to Beta. Also in: shallow-copy watch events, hugepage limits kept across in-place resize, failure phase kept when a container exits 0, CRI cgroup_mount_mode added, and a kuberc flag precedence fix in kubectl.
This week in the project: the ShardedListAndWatch feature gate graduates to Beta, and watch event conversion now uses a shallow struct copy for Added and Modified events instead of a full object deep copy. Pods keep their hugepage limits across an in-place resize, and a pod terminated with a failure status keeps its failure phase when the container exits 0 on SIGTERM. The CRI API adds cgroup_mount_mode for writable cgroups, client-go stops emitting a zero informer store resource version before sync, and kubectl fixes --kuberc precedence over KUBERC at the default path. The kubelet also keeps a rotated log when compression fails at close and rejects unseparated hash suffixes in env files.
Week of October 12, 2026
What shipped
- Volume binding now compares storage capacities with Quantity.Cmp so large storage sizes above 2^63 no longer overflow the CSI capacity check. Pull request #142088
- The scheduler now restores pointer-based mutation when namespace selectors expand into explicit namespace sets, fixing negative namespace selector evaluation. Pull request #141563
- The attach-detach controller now clears MountedByNode immediately when a node is deleted, allowing immediate detach instead of waiting for the 6-minute force-detach timeout. Pull request #136029
- Pod group preemption now clears the low priority nominated pods count correctly, since the cluster wide candidate name is cluster. Pull request #142684
- MarkVolumesAsReportedInUse now marks volumes synchronously, removing a race in the volume reported-in-use path. Pull request #142815
- The metrics-server addon was updated to v0.9.0. Pull request #142963
- The kubelet pod resource allocation checkpoint format now stores the entire PodSpec, with startup migration code for existing checkpoints. Pull request #141039
- Kubeadm no longer uses client-go/kubernetes/fake, replacing it with a local roundtripper to avoid inflating the binary. Pull request #142742
- Kube-proxy configuration validation now reports errors using public v1alpha1 JSON and YAML field names instead of internal Go type and field names. Pull request #142583
- Tico88612 was added as a reviewer for component-base/metrics. Pull request #143125
Why it matters
This week brings several correctness fixes that users feel directly: large storage capacity comparisons no longer overflow in CSI scheduling, negative namespace selectors evaluate correctly again, volumes detach sooner after node deletion, and volume reported-in-use marking no longer races. Alongside these, the project updated its metrics-server addon, improved kube-proxy validation messages to use public field names, and moved the kubelet pod resource allocation checkpoint to store the full PodSpec with migration code.
Changelog entry
- Volumebinding: compare storage capacities with Quantity.Cmp to avoid overflow for storage sizes larger than 2^63. Pull request #142088
- Scheduler: fix negative namespace selector evaluation by restoring pointer-based mutation during expansion. Pull request #141563
- Fix clearing NNN for low priority pods in pod group preemption. Pull request #142684
- Fix MarkVolumesAsReportedInUse race by marking volumes synchronously. Pull request #142815
- Ad-controller: allow immediate detach when a node is deleted. Pull request #136029
- Addons: update metrics-server to v0.9.0. Pull request #142963
- Use public v1alpha1 field names in kube-proxy configuration validation errors. Pull request #142583
- Kubelet: migrate pod resource allocation checkpoint format to store the entire PodSpec. Pull request #141039
- Kubeadm: remove usage of client-go/kubernetes/fake. Pull request #142742
- Bump images and versions to Go 1.26.9 and update distroless-iptables. Pull request #143120
- Replace remaining gsutil usages with gcloud storage in cluster and test/e2e_node. Pull request #142558
- Add tico88612 as component-base/metrics reviewer. Pull request #143125
This week: storage capacity comparisons no longer overflow above 2^63, negative namespace selectors evaluate correctly again, volumes detach sooner after node deletion, and volume reported-in-use marking no longer races. Also: metrics-server v0.9.0 and clearer kube-proxy validati
This week in Kubernetes: volume binding now compares storage capacities with Quantity.Cmp, so storage sizes above 2^63 no longer overflow the CSI capacity check. The scheduler restores pointer-based mutation when namespace selectors expand, fixing negative namespace selector evaluation. The attach-detach controller clears MountedByNode immediately when a node is deleted, allowing detach without waiting for the force-detach timeout, and volume reported-in-use marking no longer races. Also shipped: metrics-server updated to v0.9.0, kube-proxy validation errors reported with public v1alpha1 field names, kubeadm no longer pulling the fake client into its binary, and kubelet checkpoint files now storing the entire PodSpec with migration code.
Week of October 5, 2026
What shipped
- Kubectl configuration format kuberc promoted to stable with new kubectl.config.k8s.io/v1 version and deprecation of v1beta1. Pull request #142619
- Prometheus client library updated to 1.25.0 with minimum Go version now 1.26. Pull request #142689
- Go version upgraded to 1.27.2 for security fixes. Pull request #142873
- API server fix for large identity header sets backported to release-1.35. Pull request #142772
- Pod group runtime hierarchies validation added to kube-scheduler. Pull request #142719
- Deployment controller now surfaces old ReplicaSet scale-down errors for workqueue retry. Pull request #142650
- Webhook authorizer extended to support conditions as part of conditional authorization implementation. Pull request #137633
- Kubelet now overlays local allocatable onto uninitialized API nodes created by external controllers. Pull request #142234
- Kubectl events command fixed to keep multi-line event messages inside their table rows. Pull request #142149
- API server metrics refactored to emit
apiserver_resource_size_estimate_bytesandapiserver_resource_objectswith explicit timestamps. Pull request #142007
Why it matters
This week brings stability improvements to core components, backports of critical security fixes across multiple release branches, and advances in authorization and scheduling features. The promotion of kuberc to stable and Go 1.27.2 security upgrade are particularly important for users managing cluster configurations and deploying production workloads.
Changelog entry
- Kuberc promoted to stable: kubectl.config.k8s.io/v1 introduced, v1beta1 deprecated Pull request #142619
- Prometheus/client_golang updated to 1.25.0 with Go 1.26 minimum requirement Pull request #142689
- Go version upgraded to 1.27.2 security release Pull request #142873
- API server: large identity header sets support backported to release-1.35 Pull request #142772
- API server: large identity header sets support backported to release-1.34 Pull request #142773
- API server: large identity header sets support backported to release-1.36 Pull request #142771
- API server: large identity header sets support backported to release-1.37 Pull request #142770
- Deployment controller surfaces scale-down errors from old ReplicaSets for retry Pull request #142650
- Kubelet overlays local allocatable onto uninitialized API nodes from external controllers Pull request #142234
- Scheduler validates pod group runtime hierarchies Pull request #142719
- Webhook authorizer supports conditions for conditional authorization Pull request #137633
- Kubectl events preserves multi-line messages in table output Pull request #142149
This week: kuberc config format reaches stable status, Go 1.27.2 security update integrated, pod group scheduling validation added, and multiple core controller bug fixes for deployments and kubelet.
Kubernetes shipped significant updates this week: kuberc configuration format promoted to stable, Go runtime updated to 1.27.2 for security, new pod group runtime hierarchy validation in the scheduler, and critical fixes for deployment controllers and kubelet node initialization. These changes strengthen core reliability and security across multiple cluster components.
Week of September 28, 2026
What shipped
- Kubelet now filters out scoped link-local IPv6 nameservers that pods cannot use when applying the Default DNS policy. Pull request #140880
- Etcd request metrics now separately track updateGet and deleteGet operations and include error reason labels. Pull request #142484
- Watch cache snapshots are now streamed without allocating an item slice for each watchlist. Pull request #142522
- Quantity Dec values now behave as value types instead of reference types, preventing unintended mutations. Pull request #142594
- Quantity comparison and conversion methods are now read-only and thread-safe. Pull request #142581
- Kubectl label now writes missing label warnings to stderr instead of stdout to preserve valid JSON output. Pull request #142436
- HPA resource utilization ratio computation no longer overflows when summing large int64 values. Pull request #142192
- HPAGeneration feature gate is promoted to GA and locked to enabled. Pull request #142386
- Kubectl kuberc set now clears the allowlist when policy is set to AllowAll or DenyAll. Pull request #142564
- DRA ResourceSlice now validates that device-level node selectors use exactly one term as documented. Pull request #141213
Why it matters
This week's changes fix multiple bugs affecting kubelet DNS configuration, quantity handling, metrics, and kubectl output reliability. Several improvements target memory efficiency in watch operations and correctness in API validation, while feature gate promotions signal maturation of HPA functionality.
Changelog entry
- Kubelet: Filter scoped link-local IPv6 DNS servers from pod resolver configuration Pull request #140880
- Metrics: Separate etcd deleteGet and updateGet operations and add error reason labels Pull request #142484
- Cacher: Stream watch cache snapshots without item slice allocation Pull request #142522
- Apimachinery: Make Quantity Dec behave as value type to prevent mutations Pull request #142594
- Apimachinery: Fix Quantity.Cmp and AsDec to be read-only and thread-safe Pull request #142581
- Kubectl: Write missing label warnings to stderr instead of stdout Pull request #142436
- Autoscaling: Fix HPA resource utilization ratio computation int64 overflow Pull request #142192
- Featuregate: Promote HPAGeneration to GA and lock to enabled Pull request #142386
- Kubectl: Clear kuberc allowlist when policy is AllowAll or DenyAll Pull request #142564
- Dra: Validate ResourceSlice device node selector has exactly one term Pull request #141213
- Apimachinery: Fix DeepDerivative to properly validate slice lengths Pull request #137264
- Scheduling: Fix pod deletion latency metric unit from nanoseconds to seconds Pull request #141742
This week: kubelet DNS filtering, safer Quantity types, watch cache streaming, HPA metrics separation, and kubectl stderr fixes. Bug fixes improve reliability across node, API machinery, and CLI components.
Kubernetes this week includes improvements to DNS handling in kubelet, fixes for Quantity value semantics to prevent unintended mutations, and safer etcd metrics tracking. kubectl now correctly routes warnings to stderr, preventing JSON corruption in piped commands. Watch cache streaming avoids unnecessary allocations. HPA gains maturity with feature gate promotion to GA. These changes address reliability, performance, and correctness across core components.
Week of September 21, 2026
What shipped
- Kubeadm now supports ML-DSA-44, ML-DSA-65, and ML-DSA-87 encryption algorithms for new clusters using ClusterConfiguration. Pull request #142037
- Konnectivity network proxy updated to v0.37.0 with CVE fixes and improvements to connection handling and agent data-plane communication. Pull request #142400
- CertificateSigningRequests now support ML-DSA with updates to cluster-signing flags and a new CertificateSigningRequestMLDSA feature gate. Pull request #142196
- Clients can now omit metadata.managedFields from API responses by adding drop=metadata.managedFields to the Accept header behind the ManagedFieldsAlpha feature gate. Pull request #139561
- CompositePodGroup.Spec.SchedulingPolicy.Gang.MinGroupCount is now mutable across v1alpha3 and v1beta1 versions for dynamic workload resizing. Pull request #141023
- Watch validation added using minimal and maximal revision tracking to ensure watches return consistent event tuples. Pull request #142381
- Validation-gen now supports time.Duration bounds in +k8s:minimum and +k8s:maximum annotations with quoted Go duration strings. Pull request #142368
- Callers can now choose the self-signed key algorithm in client-go/util/cert.GenerateSelfSignedCertKeyWithOptions. Pull request #142350
- DRA ResourceSlice controller can now optionally refuse to publish capacities and attributes with driver domain to prevent conflicts. Pull request #142218
- Kubernetes v1.34.12 released. Release
Why it matters
This week brings support for post-quantum cryptography through ML-DSA integration across kubeadm and certificate signing, improving security for future threats. The addition of server-side opt-out for managedFields and mutable workload scheduling parameters addresses operational needs for API efficiency and dynamic resource management. Multiple performance and correctness improvements across watch validation, caching, and device management strengthen the core platform.
Changelog entry
- Kubeadm: add ML-DSA-44, ML-DSA-65, and ML-DSA-87 support to ClusterConfiguration.EncryptionAlgorithm Pull request #142037
- Kube-apiserver: update konnectivity network proxy to v0.37.0 with CVE fixes and performance improvements Pull request #142400
- Certificates: add ML-DSA support to CertificateSigningRequests with CertificateSigningRequestMLDSA feature gate Pull request #142196
- Api: support drop=metadata.managedFields in Accept header to omit managedFields from responses Pull request #139561
- Scheduling: make CompositePodGroup.Spec.SchedulingPolicy.Gang.MinGroupCount mutable Pull request #141023
- Apiserver: add watch validation using revision tracking Pull request #142381
- Validation-gen: support time.Duration in +k8s:minimum/maximum annotations Pull request #142368
- DRA: include extended resource claims in pod claim iteration and taint eviction Pull request #142289
- Kubelet/devicemanager: preserve in-flight device reservations across rebuilds Pull request #137534
- Api: document platform-specific semantics of Node.status.volumesAttached[].devicePath Pull request #141861
This week: ML-DSA post-quantum cryptography support in kubeadm and certificates, managedFields opt-out for API responses, mutable workload scheduling, watch validation improvements, and four patch releases.
Kubernetes shipped significant updates this week: post-quantum cryptography support via ML-DSA for kubeadm and certificate signing; server-side control over metadata.managedFields in API responses; mutable workload group sizing for dynamic scheduling; improved watch validation; and refinements to device management and performance. Four patch releases (v1.34.12, v1.35.9, v1.36.5, v1.37.1) are available.
Week of September 14, 2026
What shipped
- IP/CIDR Validation feature graduates to GA, removing the feature gate requirement. Pull request #141979
- PodCertificateRequest and pod certificate projections now support ML-DSA algorithms for signing PKCS#10 CSRs. Pull request #142108
- Kubectl explain command gains shell autocompletion support. Pull request #140058
- HPA now correctly excludes pod overhead from pod-level resource request calculations. Pull request #142154
- ResourceQuota validation no longer incorrectly rejects updates that preserve previously stored values. Pull request #142163
- StatefulSet restore from ControllerRevision now preserves only the.spec field to prevent unintended modifications. Pull request #142159
- Pod sorting by creation time now uses UID as a tie-breaker to ensure deterministic ordering. Pull request #141211
- Kubelet no longer incorrectly inherits probe state from replaced containers after restart. Pull request #141487
- LimitRanger no longer applies min/max checks to unchanged PersistentVolumeClaim requests on update. Pull request #142170
- Pod resize end-to-end tests now properly wait for resize actuation before continuing. Pull request #142106
Why it matters
Multiple bug fixes address correctness issues in resource management, scheduling, and validation that could affect production workloads. New features like kubectl autocompletion and ML-DSA support expand functionality for operators and security use cases.
Changelog entry
- IP/CIDR Validation feature gate removed as feature graduates to GA Pull request #141979
- Added ML-DSA algorithm support to PodCertificateRequest and pod certificate projections Pull request #142108
- Kubectl explain command now supports shell autocompletion Pull request #140058
- Fixed HPA pod-level resource request calculations to exclude pod overhead Pull request #142154
- Fixed ResourceQuota validation to accept stored values during updates Pull request #142163
- StatefulSet now restores only.spec from ControllerRevision to prevent unintended field modifications Pull request #142159
- Added UID tie-breaker to PodsByCreationTime sort for deterministic ordering Pull request #141211
- Kubelet no longer inherits probe state from replaced containers after restart Pull request #141487
- LimitRanger skips min/max checks on unchanged PersistentVolumeClaim requests during updates Pull request #142170
- Fixed container memory resize validation error message reporting wrong variable Pull request #141100
This week: IP/CIDR Validation reaches GA, pod overhead handling fixed in HPA, kubectl explain gains autocompletion, plus fixes for resource validation and StatefulSet restoration.
Notable updates to Kubernetes this week include IP/CIDR Validation graduating to GA, improvements to pod resource calculations in HPA that now properly exclude overhead, shell autocompletion for kubectl explain, and critical bug fixes for ResourceQuota validation and StatefulSet restoration from ControllerRevision.