Karmada
v1.17.5Orchestration & ManagementKarmada v1.17.5 is a mixed release spanning v1.17.0 (feature release) through v1.17.5 (patch releases). Major changes include removal of the deprecated --etcd-init-image flag and deprecation of two cluster lease flags and an ETCD configuration field; introduction of workload affinity/anti-affinity scheduling and ControllerPriorityQueue beta (now default); controller optimization reducing startup time from 20+ minutes to ~5 minutes at scale; Kubernetes and ETCD image defaults updated to v1.35.2 and 3.6.6-0; and several critical bug fixes addressing graceful eviction race conditions, credential rotation failures, scheduler misrouting, and certificate rotation CSR auto-approval.
breakingDeprecated --etcd-init-image flag removed
The flag `--etcd-init-image` for the `init` command has been removed in v1.17.0. Operators using this flag must remove it from their initialization scripts or configuration.
breakingCluster lease duration flags deprecated
The flags `--cluster-lease-duration` and `--cluster-lease-renew-interval-fraction` are deprecated in v1.17.0 and will be removed in a future release. Applies if karmada-controller-manager is configured with either flag.
breakingEtcd.Local.InitImage configuration field deprecated
The field `Etcd.Local.InitImage` in Karmada Init Configuration is deprecated in v1.17.0 and will be removed in a future version. Applies if Karmada initialization configuration includes this field.
enhancementMetric semantics change: 409 conflicts no longer counted as errors
The metric `work_sync_workload_duration_seconds` now excludes retriable Kubernetes 409 conflicts from error counts, improving accuracy of availability tracking and reducing false alerts from conflict retry flapping.
Key changes (14)
- Removed deprecated `--etcd-init-image` flag from init command
- Deprecated `--cluster-lease-duration` and `--cluster-lease-renew-interval-fraction` flags
- Deprecated `Etcd.Local.InitImage` configuration field
- Introduced workload affinity and anti-affinity scheduling via new `workloadAffinity` field in PropagationPolicy
- ControllerPriorityQueue promoted to Beta and enabled by default, reducing restart downtime
- Optimized dependencies-distributor controller logic, reducing startup time from 20+ minutes to ~5 minutes at scale
- karmada-operator now supports Tolerations, Affinity, priority class, and pod disruption budget in control plane CommonSettings
- Default Kubernetes and ETCD images updated (kube-apiserver/kube-controller-manager to v1.35.2, ETCD to 3.6.6-0)
- Helm chart encryption-at-rest support for karmada-apiserver
- Fixed race condition in graceful eviction that could silently drop tasks when multiple controllers modify ResourceBinding concurrently
- Fixed transient credential failures incorrectly setting cluster Ready=False without respecting ClusterFailureThreshold
- Fixed scheduler error-type propagation causing insufficient-replica bindings to be misrouted
- Fixed karmada-agent certificate rotation CSR auto-approval failure due to SignerName mismatch
- Alpine base image bumped across patch releases (3.23.3 → 3.24.1)
A weekly email arrives when a release needs action. Like the breaking changes in this release.