$ the-wire · showcase
Master drops containerd 1.7, StorageClass validation goes stable
By RepoJournal · Filed · About Kubernetes · Composed from the cited sources · methodology
14 people shipped this
caesarsage
@Caesarsage
1 cited
carlory
@carlory
1 cited
pohly
@pohly
1 cited
troychiu
@troychiu
1 cited
itzPranshul
@itzPranshul
1 cited
my-git9
@my-git9
Kyle-Law
@Kyle-Law
kubernetes-prow[bot]
@kubernetes-prow[bot]
k8s-infra-ci-robot
@k8s-infra-ci-robot
Mario Jason Braganza
@jasonbraganza
Harshal Patil
@harche
Kubernetes Publisher
@k8s-publishing-bot
serathius
@serathius
liyuerich
@liyuerich
Kubernetes master's node jobs move to containerd 2.x ahead of the kubelet fallback removal, while StorageClass declarative validation graduates from beta to stable and DRA's structured allocator gets a lookup optimization.
sig-node: remove containerd 1.7 jobs from master kubernetes/test-infra
sig-node pulled the containerd 1.7 jobs from master because the containerd 1.7 runtime in COS 117 does not implement the RuntimeConfig RPC, which fails kubelet startup now that kubelet is losing its fallback to the configured cgroup driver. The fallback removal is planned for Kubernetes 1.38 and containerd 1.7 reaches EOL in September 2026, so existing release-branch coverage keeps 1.7 while ma...
storage: graduate StorageClass declarative validation to stable kubernetes/kubernetes
The declarative validation markers on StorageClass (provisioner, parameters, reclaimPolicy, volumeBindingMode) lost their +k8s:beta(since: "1.37") prefix, and the superseded hand-written checks in ValidateStorageClassUpdate and validateProvisioner are gone. Anyone maintaining custom validation or conversion logic around StorageClass should expect the generated rules to be the single source of t...
DRA: optimize allocated shared device lookups in structured allocator kubernetes/kubernetes
The structured allocator no longer re-scans for allocated shared devices on every lookup, with the publishing bot already carrying the change into the dynamic-resource-allocation staging repo. It is internal bookkeeping, but shared-device allocation is where DRA allocation cost concentrates at scale.
e2e podlogs: fix handling of network issues kubernetes/kubernetes
pohly's fix targets two flake sources in the e2e Pods watch loop: the periodic ticker never called check(), so missed watch events went unnoticed, and a closed result channel made receives return immediately, producing a busy loop of reconnect and List calls. Both now route through a RetryWatcher, which should quiet apiserver-unreachable flakes in podlogs tests.
Repoint DRA links after docs reorganization kubernetes/website
The docs reorganization that created the new Resource Management pages left inbound DRA links pointing at the old locations, and this repoints them; the author states the work was done with AI assistance. Small on its own, but broken deep links are how readers end up on stale DRA documentation.
