$ the-wire · showcase
Quantity becomes a value type, HPA overflow fixed
By RepoJournal · Filed · About Kubernetes · Composed from the cited sources · methodology
![kubernetes-prow[bot]](https://avatars.githubusercontent.com/in/992787?v=4)
14 people shipped this
upodroid
@upodroid
1 cited
Kubernetes Publisher
@k8s-publishing-bot
1 cited
thc1006
@thc1006
1 cited
KR-Ravindra
@KR-Ravindra
1 cited
Sascha Grunert
@saschagrunert
1 cited
ekam-walia
@ekam-walia
kubernetes-prow[bot]
@kubernetes-prow[bot]
k8s-infra-ci-robot
@k8s-infra-ci-robot
Joe Betz
@jpbetz
ksadura
@ksadura
serathius
@serathius
liggitt
@liggitt
Champbreed
@Champbreed
BenjaminBraunDev
@BenjaminBraunDev
apimachinery makes Quantity stop mutating in place, the HPA stops wrapping at MaxInt64, and the rest of the stack is routine test, infra, and archive work.
Merge pull request #142594 from jpbetz/quantity-arith-copy-on-write kubernetes/apimachinery
Quantity's arithmetic now copies on write so Dec behaves as a value type, following the earlier fix that stopped Cmp and AsDec from modifying the receiver. If you hold a Quantity across a call or a cache, operations such as Add, Sub, Mul, and Neg no longer write through to your caller's value.
HPA: compute resource utilization ratio without int64 overflow kubernetes/kubernetes
GetResourceUtilizationRatio accumulated per-pod metrics and requests as raw int64, multiplied by 100, and narrowed to int32, all three steps wrapping silently; near math.MaxInt64 the HPA reported a currentUtilization of 0 and a negative raw average and scaled the wrong way or not at all. The totals are now accumulated with math/big, and the PR also carries a testable refactor of the ratio helper.
increase the frequency of the kubernetes build job kubernetes/test-infra
The kubernetes build job takes 22 minutes today and 15 minutes with the signing step skipped, so upodroid is raising its frequency; the change is on hold pending a kubernetes/release change. Expect the build signal on master to refresh more often, at the cost of unsigned artifacts.
test/e2e/dra: keep recorded gRPC calls intact across appends and resets kubernetes/kubernetes
The DRA test driver's stream interceptor could leave a recorded Err nil even when the handler returned an error, because recordGRPCStream wrote the stream error through a pointer to the last element of the call slice and any call recorded while the stream was open could reallocate that slice. Recorded gRPC calls now survive appends and resets, which matters if you are debugging DRA failures off...
Allow the OCI referrers API on registry-sandbox.k8s.io kubernetes/k8s.io
Cloud Armor answered /v2/<name>/referrers/<digest> with 404 even though archeio already routes the OCI referrers API to the signature upstream. The referrers path now sits behind a new allow_referrers variable on the oci-proxy module and is switched on for registry-sandbox.k8s.io only; production keeps its current expression. Around it: Prow advanced to v20261001-ae4792128, CPU manager checkpoi...
