• [CHANGE] #969 Pull cass-config-builder from ghcr.io instead of docker.io where its no longer updated
• [FEATURE] #964 Add --max-concurrent-reconciles to configure reconciling concurrency (default is 1), --pprof-bind-address to enable profiling, and --reconciliation-timeout to limit reconciliation time (default is 2m)
• [ENHANCEMENT] #947 Read Events through an uncached APIReader instead of a cluster-wide cached informer in failure detection, reducing memory usage in cluster-scoped deployments.
• [ENHANCEMENT] #947 Bound the informer cache in cluster-scoped mode: label-scope Pod/StatefulSet/PodDisruptionBudget/Service informers to operator-managed objects, read Secrets/PVCs/ConfigMaps/Endpoints/EndpointSlices/StorageClasses through live API calls with metadata-only Secret watches, and strip managedFields/last-applied from cached objects. Operator memory no longer scales with cluster size.
• [ENHANCEMENT] #959 Synchronize Cassandra Services with configured container ports and remove ports that are unavailable or intended only for pod-level communication
• [ENHANCEMENT] #983 Poll the target decommissioning pod and ask it if it has finished decommission, do not trust other nodes in the ring by default. Add a new annotation cassandra.datastax.com/skip-local-decommission-check to allow old behavior
• [ENHANCEMENT] #952 Allow decommission to continue even in environments where user deletes superuser secret before decommission has finished
• [ENHANCEMENT] #977 Skip repeated legacy additionalSeeds Endpoints cleanup checks after successful cleanup, reducing API-server traffic.
• [ENHANCEMENT] #215 Mark pods where full query logging was enabled with the annotation cassandra.datastax.com/fql-enabled and only call the mgmt-api when the pod is recreated or the configuration changes.
• [BUGFIX] #971 Decommissioning last pod of a datacenter left its pod_status metric abandoned instead of being removed. And also we got an error logline which wasn't a real error
• [BUGFIX] #973 Prevent premature decommission cleanup when Cassandra metadata requests fail.
• [BUGFIX] #975 Use FailedStarts and StatefulSet revisions as part of the failureModeDetection to allow forced apply of change
• [BUGFIX] #986 Match pods to their endpoint state by host id instead of pod IP when reading the load used by the scale-down capacity check, and fail the validation when a running decommission target has no load reported instead of treating it as zero bytes used.
• [BUGFIX] #981 Enforce balance of pod bootstraps per rack even if certain StatefulSets are unable to schedule correctly.
• [BUGFIX] #997 Pod labeling could cause delay in the fast path startup, causing certain pods to never start until restarted by the operator (10 minutes timeout)
• [BUGFIX] #990 Full query logging was applied to pods of every datacenter in the cluster instead of only the target datacenter and a pod without FQL support stopped processing the remaining pods
• [BUGFIX] #998 Do not modify the shared default HTTP client or supplied transports when building mgmt-api clients, propagate the reconcile logger context correctly, and update k8ssandra-client to v0.9.2 and Vector to 0.60.0