Changes since v0.16.7:
Actions Required Before Upgrading
(No, really, you MUST read this before you upgrade)
- Minor releases: Review the
.0release notes for each new minor version you cross; see:v0.15.0,v0.16.0. - Patch releases: Review the patch release notes leading up to this version, but only within this minor release line; see:
v0.16.1,v0.16.2,v0.16.3,v0.16.4,v0.16.5,v0.16.6,v0.16.7.
Changes by Kind
Feature
- Observability: Improved FairSharing strategy-evaluation logs by including DRS share values and emitting them at verbosity level V(4). (#11186, @PBundyra)
Bug or Regression
- ElasticJobsViaWorkloadSlices: Fixed a bug where workload slices with identical creation timestamps could be incorrectly sorted, potentially leading to quota leaks during scale-up. (#11200, @KumarADITHYA123)
- Fixed a regression where Kueue could mark newly created Workloads as finished, potentially blocking queues. The FinishOrphanedWorkloads feature gate has been downgraded to Alpha. (#11019, @mbobrovskyi)
- Fixed multi-arch image builds for importer, kueue-populator, and kueueviz backend images so runtime images
and binaries are built for the target platform, preventing wrong-architecture containers and exec format error
for non-amd64 target platforms, such as arm64, ppc64le, and s390x. (#10915, @carterpewpew) - Fixed vulnerability where two podsets with total requests exceeding max int64 would lead to integer overflow and break quota limits. (#11151, @pajakd)
- Helm: Fixed manager probe templates so periodSeconds correctly uses the configured periodSeconds value,
rather than initialDelaySeconds. (#10982, @cixuuz) - Helm: Fixed the FlowSchema priorityLevelConfiguration reference to use the Helm fullname template, preventing APF configuration from breaking when fullnameOverride or nameOverride is set. (#10981, @cixuuz)
- KueueViz: Fixed RBAC permissions for WorkloadPriorityClass objects by using the correct plural workloadpriorityclasses resource name. (#10985, @cixuuz)
- KueueViz: Fixed the LocalQueue details page to show only workloads from the selected queue. (#11213, @ManthanNimodiya)
- LeaderWorkerSet & StatefulSet: Fixed a race condition bug that could occasionally result in reverting, at the level of the Workload object, manual changes to the queue-name label for LeaderWorkerSet and StatefulSet. (#11194, @mbobrovskyi)
- Scheduling: Fixed a bug where in-flight workloads that were concurrently marked as finished (
Finished=True) or deactivated could be requeued by Kueue's scheduler, causing re-scheduling attempts which were interfering with the scheduling of other workloads. (#11021, @mbobrovskyi) - TAS: Fixed NodeHotSwap with TASReplaceNodeOnNodeTaints enabled to evaluate node taints using effective Workload tolerations, including tolerations from AdmissionCheck PodSetUpdates. (#11227, @Ladicle)
- TAS: Fixed a bug where multi-resource workloads, such as workloads requesting both CPU and memory,
could fail admission during second-pass scheduling for ProvisioningRequests or NodeHotSwap because one
resource's usage was double-counted against quota. (#11040, @cvgenesis) - TAS: Fixed cache cleanup for non-TAS Pods that reach a terminal phase without Kueue observing the expected status update, preventing stale Pod usage from remaining in the TAS cache. (#11145, @amy)
- TAS: optimize performance of building the snapshot by pre-aggregating the node usage coming from non-TAS Pods. (#11074, @jzhaojieh)