|
| 1 | +## v0.8.0 |
| 2 | + |
| 3 | +Changes since `v0.7.0`: |
| 4 | + |
| 5 | +### Urgent Upgrade Notes |
| 6 | + |
| 7 | +#### (No, really, you MUST read this before you upgrade) |
| 8 | + |
| 9 | +- Use a single rate limiter for all API types clients. |
| 10 | + |
| 11 | + Consider adjusting `clientConnection.qps` and `clientConnection.burst` if you observe any performance degradation. (#2462, @trasc) |
| 12 | + |
| 13 | +### Feature |
| 14 | + |
| 15 | +- Add a column to workload indicating if it is finished (#2615, @highpon) |
| 16 | +- Add preempted_workloads_total metric that tracks the number of preemptions issued by a ClusterQueue) (#2538, @vladikkuzn) |
| 17 | +- Add the following events for eviction on the workload indicating the reason for eviction: |
| 18 | + - "EvictedDueToPodsReadyTimeout" |
| 19 | + - "EvictedDueToAdmissionCheck" |
| 20 | + - "EvictedDueToClusterQueueStopped" |
| 21 | + - "EvictedDueToInactiveWorkload" (renamed from InactiveWorkload) |
| 22 | + |
| 23 | + If you were watching for the typed Normal event with `InactiveWorkload` reason, use `EvictedDueToInactiveWorkload` reason one instead. (#2376, @mbobrovskyi) |
| 24 | +- AdmissionChecks: A workload with a Rejected AdmissionCheck gets deactivated (#2363, @PBundyra) |
| 25 | +- Allow stoping admission from a specific LocalQueue. (#2173, @mbobrovskyi) |
| 26 | +- Allow usage of the pod integration for pods belonging to jobs that Kueue supports, if the support for the job type is explicitly disabled (#2493, @trasc) |
| 27 | +- CLI: Add stop/resume localqueue commands (#2415, @rainfd) |
| 28 | +- CLI: Added Node Labels column on resource flavor list. (#2557, @mbobrovskyi) |
| 29 | +- CLI: Added create resourceflavor command. (#2517, @mbobrovskyi) |
| 30 | +- CLI: Added list resourceflavor command. (#2525, @mbobrovskyi) |
| 31 | +- CLI: Added resourceflavor to pass-through commands. (#2518, @mbobrovskyi) |
| 32 | +- CLI: Added version command. (#2346, @mbobrovskyi) |
| 33 | +- CLI: Adds `for` filter to list workloads. (#2238, @IrvingMg) |
| 34 | +- CLI: Adds create clusterqueue command. (#2201, @IrvingMg) |
| 35 | +- CLI: Support autocompletion (#2314, @mbobrovskyi) |
| 36 | +- CLI: Support paging on kueue CLI list commands. (#2313, @mbobrovskyi) |
| 37 | +- CLI: kubectl-kueue tar.gz archives is part of the release artifacts. (#2513, @mbobrovskyi) |
| 38 | +- Do not start Kueue when the visibility server cannot be started, but is requested. (#2636, @mbobrovskyi) |
| 39 | +- Experimental support for helm charts in the gcr.io/k8s-staging-kueue/charts/kueue repository (#2377, @IrvingMg) |
| 40 | +- Improved logging for scheduling and preemption in levels 4 and 5 (#2504, @alculquicondor) |
| 41 | +- Introduce the MultiplePreemptions flag, which allows more than one |
| 42 | + preemption to occur in the same scheduling cycle, even with overlapping |
| 43 | + FlavorResources (#2641, @gabesaba) |
| 44 | +- More granular Preemption condition reasons: PriorityReclamation, InCohortReclamation, InCohortFairSharing, InCohortReclaimWhileBorrowing (#2411, @vladikkuzn) |
| 45 | +- MultiKueue: Allow for defaulting of the spec.managedBy field for Jobs managed by MultiKueue. |
| 46 | + The defaulting is enabled by the MultiKueueBatchJobWithManagedBy feature-gate. (#2401, @vladikkuzn) |
| 47 | +- MultiKueue: Remove remote objects synchronously when the worker cluster is reachable. (#2347, @trasc) |
| 48 | +- MultiKueue: Use batch/Job `spec.managedBy` field (#2331, @trasc) |
| 49 | +- Multikueue: Batch reconcile events for remote workloads. (#2380, @trasc) |
| 50 | +- ProvisioningRequest: Support for ProvisioningRequest's condition `BookingExpired` (#2445, @PBundyra) |
| 51 | +- ProvisioningRequets: Support for ProvisioningRequest's condition `CapacityRevoked`. ProvisioningRequests objects persist until the corresponding Job or the Workload is deleted (#2196, @PBundyra) |
| 52 | + |
| 53 | +### Documentation |
| 54 | + |
| 55 | +- Added details documentation for kubectl-kueue plugin. (#2613, @mbobrovskyi) |
| 56 | +- Improve the documentation for the waitForPodsReady (#2541, @mimowo) |
| 57 | + |
| 58 | +### Bug or Regression |
| 59 | + |
| 60 | +- Added raycluster roles to manifests.yaml (#2618, @mbobrovskyi) |
| 61 | +- CLI: Fixed no Auth Provider found for name "oidc" error. (#2602, @Kavinraja-G) |
| 62 | +- Fix check that prevents preemptions when a workload requests 0 for a resource that is at nominal or over it. (#2520, @alculquicondor) |
| 63 | +- Fix for the scenario when a workload doesn't match some resource flavors due to affinity or taints |
| 64 | + could cause the workload to be continuously retried. (#2407, @KunWuLuan) |
| 65 | +- Fix missing fairSharingStatus in ClusterQueue (#2424, @mbobrovskyi) |
| 66 | +- Fix missing metric cluster_queue_status (#2474, @mbobrovskyi) |
| 67 | +- Fix panic that could occur when a ClusterQueue is deleted while Kueue was updating the ClusterQueue status. (#2461, @mbobrovskyi) |
| 68 | +- Fix panic when there is not enough quota to assign flavors to a Workload in the cohort, when FairSharing is enabled. (#2439, @mbobrovskyi) |
| 69 | +- Fix performance issue in logging when processing LocalQueues. (#2485, @alexandear) |
| 70 | +- Fix race condition on delete workload from queue manager. (#2460, @mbobrovskyi) |
| 71 | +- Fix race condition on requeue workload. (#2509, @mbobrovskyi) |
| 72 | +- Fix race condition on run garbage collection in multikueuecluster reconciler. (#2479, @mbobrovskyi) |
| 73 | +- Fix the validation messages, to report the new value rather than old, for the following immutable labels: `kueue.x-k8s.io/queue-name`, `kueue.x-k8s.io/prebuilt-workload-name`, and `kueue.x-k8s.io/priority-class`. (#2544, @xuxianzhang) |
| 74 | +- Fixed issue that prevented restoring the startTime and pod template when evicting a batch/v1 Job, if any API errors happened in the process (#2567, @mbobrovskyi) |
| 75 | +- MultiKueue: Do not reject a JobSet if the corresponding cluster queue doesn't exist (#2425, @vladikkuzn) |
| 76 | +- MultiKueue: Skip garbage collection for disconnected clients which could occasionally result in panic. (#2369, @trasc) |
| 77 | +- Show weightedShare in ClusterQueue status.fairSharing even if the value is zero (#2521, @alculquicondor) |
| 78 | +- Skip duplicate Tolerations when an admission check introduces a toleration that the job also set. (#2498, @trasc) |
| 79 | + |
| 80 | +### Other (Cleanup or Flake) |
| 81 | + |
| 82 | +- Importer: corrects the field name `observedFirstIn` in logs. (#2500, @alexandear) |
| 83 | +- Use Patch instead of Update on jobframework multikueue adapters to prevent the risk of dropping fields. (#2590, @mbobrovskyi) |
| 84 | +- Use Patch instead of Update on jobframework to prevent the risk of dropping fields. (#2553, @mbobrovskyi) |
0 commit comments