Release notes

Prometheus Exporter configuration support for Management Center: Management Center Prometheus Exporter properties can now be configured directly via the custom resource, giving Operator-managed deployments fine-grained control over Prometheus metric exposure and filtering without requiring manual Management Center configuration.

Custom logging configuration support: Custom log4j2.properties logging configurations can now be specified via a ConfigMap, allowing users to manage logging behavior directly through the custom resource.

ModuleDeploy support for Management Center: ModuleDeploy can now be used to download custom modules and add them to the Management Center classpath, enabling greater customization and extensibility.

Added support for Java 25: Hazelcast Platform Operator now supports Java 25 alongside the existing Java 17 and 21 versions.

Dynamic Diagnostic Logs are now Generally Available: First introduced as BETA in 5.16, Dynamic Diagnostic Logs are now GA. Diagnostic logging can be controlled dynamically via Hazelcast Operator without requiring a cluster restart. It supports configuring log locations, enabling or disabling diagnostic plugins, and setting an optional auto-off timer, improving operational flexibility in production environments.

Resource limits and requests configuration for init containers: CPU and memory resource limits and requests can now be configured for init containers. Previously, deployments in Kubernetes environments with enforced resource quotas could fail with a "Forbidden: failed quota" error because init containers lacked resource settings. Users can now specify these settings through the custom resource to ensure successful deployments in quota-restricted namespaces.

serializeKeys parameter support in near cache configuration for Map CR: The serializeKeys parameter can now be configured within the nearCache section of the Map custom resource. When enabled, near cache keys are stored in serialized form, which is required for correctness in certain distributed scenarios.

Management Center instance naming updated to include namespace: The Kubernetes namespace is now included as part of the Management Center instance name, preventing naming collisions when multiple Management Center instances with the same name are deployed across different namespaces.

Improved pod error reporting: The reconciler’s error handling now deduplicates repeated identical errors in logs, includes additional context in error messages to aid troubleshooting, and correctly differentiates between actual errors and transient pod initialization states.

hazelcast-resources category label added to Custom Resource Definitions: CRDs are now annotated with the hazelcast-resources Kubernetes resource category, allowing users to list all Hazelcast-managed resources using standard kubectl get hazelcast-resources commands.

Dynamic label management for Operator-managed resources: Labels applied to Hazelcast custom resources are now dynamically propagated and kept in sync across all managed Kubernetes resources. Previously, label changes were reverted by the Operator, requiring workarounds to maintain consistent labeling across resources.

Validation to prevent nativeMemory.size from exceeding pod memory limits: The configured nativeMemory.size is now validated against the memory limits defined for the Hazelcast pod, preventing misconfigured deployments that could lead to out-of-memory errors at runtime.

Fixed TLS keystore not reloaded for Hazelcast clients after secret update: Resolved an issue where clients continued to connect to the cluster using outdated TLS certificates after the TLS secret was updated. The Operator now correctly reloads the updated keystore, ensuring that certificate changes are enforced and clients using invalid certificates are rejected after a secret update.

Fixed near cache validation logic: Resolved an issue where certain invalid near cache configurations in the Map CR were accepted by the Operator’s validation logic and silently passed to the cluster, potentially causing unexpected runtime behavior. The fix ensures that invalid configurations are rejected at CR application time.

Fixed unnecessary JAR redownloads on pod restart when multiple download sources are configured: Resolved an issue where JARs were unnecessarily re-downloaded on every pod restart when multiple download options were configured for user code deployment. The fix ensures that already-downloaded artifacts are correctly detected and reused, reducing pod startup time and avoiding redundant network operations.

Fixed Lite members failing to start due to missing partition count configuration: Resolved an issue where Lite members could not start because the partition count configuration was not included in their generated Hazelcast configuration. This caused Lite member pods to fail on startup. The fix ensures partition count is consistently applied to all member types.

Fixed missing validation for maxSize parameter with PER_NODE eviction policy: Added validation to reject invalid maxSize values when the PER_NODE eviction policy is configured in the Map CR, preventing misconfigurations from being silently applied to the cluster.

Fixed Map CR update handling and recreation after deletion: Resolved two related issues with Map CR lifecycle management: updates to an existing Map CR could fail to apply configuration changes correctly, and applying a new Map CR with the same name as a previously deleted one could fail if additional configuration fields were present. The fix ensures reliable Map CR updates and recreation.

Fixed SQL catalog persistence property to prevent unintended cluster recreation from backup: Resolved an issue where the catalogPersistenceEnabled property was not handled correctly during backup restore operations, potentially causing the cluster to recreate its SQL catalog unexpectedly. The fix ensures the property is correctly preserved and respected when restoring from a backup.

Fixed incorrect field name UniqueKeyTransition: Corrected a naming inconsistency where the index configuration field was incorrectly referenced as UniqueKeyTransition instead of the correct UniqueKeyTransformation. This mismatch caused index configurations relying on this field to be silently ignored.