Skip to content

[pull] master from kubernetes:master - #1924

Open
pull[bot] wants to merge 9016 commits into
next-stack:masterfrom
kubernetes:master
Open

pull[bot] wants to merge 9016 commits into
next-stack:masterfrom
kubernetes:master

Conversation

@pull

@pull pull Bot commented Oct 24, 2021 •

Copy link
Copy Markdown

See Commits and Changes for more details.


Created by pull[bot]

Can you help keep this open source service alive? 💖 Please sponsor : )

@pull pull Bot added the ⤵️ pull label Oct 24, 2021
kubernetes-prow Bot and others added 29 commits September 9, 2026 14:06
chore(networking): bump aws-cni to version 1.23.1
kOps 1.36 switched gossip workers to bootstrap through the NLB, exposing
an existing Terraform bug: the VPC task ID is unset, so NLB discovery
finds no IPs and falls back to an unresolvable gossip hostname.

Use the VPC ID returned by AWS for the NLB.
…eature-gates

cert-manager: add feature gate options for webhook and cainjector
Promote the SELinuxMount feature flag from opt-in to enabled-by-default.
Instead of being gated only by the flag, SELinux mount support in the
AWS EBS and GCP PD CSI drivers is now rendered when the cluster sets
spec.containerd.selinuxEnabled: true.

This avoids polluting CSI driver pods with SELinux hostPath mounts on
clusters that don't use SELinux, while making the feature work out of
the box on clusters that do. On SELinux-enforcing nodes, eligible
volumes are then mounted with -o context instead of recursively
relabeled, speeding up pod startup for large volumes.

The SELinuxMount flag is kept as an escape hatch: set
KOPS_FEATURE_FLAGS=-SELinuxMount to disable the feature
unconditionally.
It will check that SELinux propagates where appropriate and CSI drivers are
installed with SELinuxMount support.
k/k PR 140532 replaced the runtime SkipUnlessProviderIs
call in test/e2e/node/ssh.go with a declarative framework.WithProvider
tag, which renamed the test from

  [sig-node] SSH should SSH to all nodes and run commands

to

  [sig-node] SSH [Provider:gce,aws,local,azure] should SSH to all nodes and run commands

Our skip pattern was `SSH.should.SSH.to.all.nodes.and.run.commands`, which
no longer matches, so the test started running again on clusters with
private or dual-stack subnets.

It fails there because the framework SSHes from the pod running the test
binary to the node addresses reported by the API. Private-topology nodes
are only reachable via the bastion, and IPv6 clusters only report IPv6
node addresses that the IPv4-only Prow pod cannot reach, so
NodeSSHHosts() returns zero hosts and the test fails with
`dial tcp: missing address`.

Drop the leading `SSH.` anchor so the pattern is not coupled to the
container node's tag list.
Fix SSH skip regex after upstream provider tag rename
build kubetest2 binaries and publish them to the staging bucket
Add availability zones for ap-east-2 (Taipei), ap-southeast-5
(Malaysia), ap-southeast-6 (New Zealand), ap-southeast-7 (Thailand)
and mx-central-1 (Mexico) to the well-known AWS zone list used for
shell completion of `kops create cluster --zones`.

kOps already works in these regions because the region is validated
against ec2:DescribeRegions at runtime; only the completion list was
out of date.
Got the IG creation working.
Kubelet is coming up on the etcd IG.
Currently having issues with getting the etcd to attach to peers.
fixed goimports
Fixed ptr to to new on count.
Fixed image rewrite to work for all of control plane.
Fixed docs to reflect changes.
Fixed role direct equality check.
Fix node check for bastion.
Fix KubControllerManager to KubeControllerManager.
Fixed experimental roles test.
Fixed nodelabel tests.
Fix label test.
The v1alpha2 fuzzer cleared CloudProvider.AWS.BinariesLocation and
CloudProvider.GCE.BinariesLocation because v1alpha2 had nowhere to put
them. #18740 gave them a home: CloudConfiguration.AWSBinariesLocation
and GCEBinariesLocation, written in Convert_kops_ClusterSpec_To_
v1alpha2_ClusterSpec and read back in the reverse direction.

The fixup outlived that, so the field the round trip was meant to cover
is cleared before it is ever exercised. Dropping it restores the
coverage for a field that has already been lost in conversion once.

TestRoundTripTypes passes over 50 seeds without the fixup.

Signed-off-by: SaiPisey2 <piseysai0202@gmail.com>
zones: add newer AWS regions to --zones completion
hakman and others added 30 commits October 25, 2026 15:48
Add spec.networking.kindnet.image, a full container image reference.
The addon uses it when it is set and
registry.k8s.io/networking/kindnet:<version> otherwise. Add a bootstrap
channel builder case with a custom image and document the field.
spec.containerd.env lists environment variables for the containerd
service, in the EnvVar form etcd-manager's env already uses. The
processes containerd starts, such as CNI plugins and runtime shims,
inherit them.
Reject names that are not C identifiers, duplicate names and
CONTAINERD_OPTS, which nodeup builds from the other containerd fields.
The containerd unit loads the file as an EnvironmentFile. Values are
double-quoted with \, ", ` and $ escaped, so systemd passes them to
containerd unchanged.
Keep AWS-specific NodeClass selection in the caller while preserving the rendered output.
feat(kindnet): add an image field to the kindnet networking spec
Update AWS Load Balancer Controller to v3.6.2
feat(containerd): add env to the containerd spec
Re-derive the manifest from upstream deploy/ at v0.6.19, which adds the
eks.amazonaws.com/skip-pod-identity-webhook objectSelector entry.

admissionReviewVersions stays at v1beta1: switching to v1 together with
the image would reject responses from the old v0.6.17 pods during the
rollout.
aws: update pod-identity-webhook to v0.6.19
Reject unsupported providers instead of rendering an AWS NodeClass reference.
parameterize the Karpenter NodeClass group and kind
kops-controller imports pkg/commands for ConfigBuilder, so this CLI-only file pulled kubectl, cli-runtime and kustomize into the controller. Moving it next to the other kops commands removes 119 packages and about 55 MB from the kops-controller binary.
refactor: move the helpers command from pkg/commands to cmd/kops
bare-metal e2e: bound apt-get update and artifact ssh connect time
chore(channels): bump alpha to October patches
fix(assets): reject conflicting hashes for the same file download URL
Add Packages (*PackagesConfig) to KubeletConfigSpec to allow overriding the kubelet binary URL and hash per architecture on a cluster-wide or per-InstanceGroup basis.

Signed-off-by: Arnaud Meukam <ameukam@gmail.com>
feat(kubelet): support custom kubelet binary package
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.