Redis Software for Kubernetes 8.2.6-21 (October 2026) release notes

Feature release with Kubernetes Gateway API support, certificate-based authentication for the REST API, client maintenance notifications, FIPS 140-3 on Amazon EKS with Bottlerocket, and Redis Software 8.2.6.

Redis Software for Kubernetes 8.2.6-21 is a feature release that supports Redis Software 8.2.6 and includes new features, bug fixes, and enhancements.

Highlights

• Support for exposing cluster endpoints and Active-Active databases through the Kubernetes Gateway API.
• Certificate-based authentication for the Redis Software REST API, with the option to turn off basic authentication.
• Support for Redis Software 8.2.6.

Enhancements

• Added support for exposing the cluster's REST API, Cluster Manager UI, and metrics endpoints, and Active-Active databases, through the Kubernetes Gateway API
• Added certificate-based authentication to the REST API for the operator and for users, and the option to turn off basic authentication
• Added REC support for client maintenance notifications, for databases that use the OSS Cluster API and for databases in proxy mode
• Added FIPS 140-3 support on Amazon EKS with Bottlerocket FIPS worker nodes
• Added the disconnectClientsOnPasswordRemoval setting to the REC, REDB, and REAADB, which controls whether the proxy disconnects clients whose password was removed
• Added support for custom labels on database Services, which the services rigger no longer removes
• Added Redis Software 8.2.6 support

Resolved issues

• Fixed the operator overwriting nodeAffinity and podAffinity set in redisEnterpriseAdditionalPodSpecAttributes
• Fixed the REC getting stuck in the Upgrade state after you upgraded a FIPS-enabled cluster
• Fixed changes to module arguments on an existing REDB not being applied to the database shards
• Fixed the operator failing to update the REC LDAP configuration when you switched spec.ldap.authenticationQuery from template to query
• Fixed new REAADBs being created without modules when spec.globalConfigurations.modulesList wasn't set
• Fixed ossCluster being ignored for REDBs that use the all-nodes proxy policy
• Fixed the log collector in restricted mode leaving collected resources out of the package when it couldn't read PersistentVolumes
• Fixed the log collector leaving out the error message when it failed to collect a resource

API changes

This release adds fields for Gateway API exposure, REST API authentication, client maintenance notifications, and client certificate validation. The table below summarizes the notable changes. For the complete field-level reference, see the API reference.

Supported distributions

Redis Software for Kubernetes is compatible with CNCF-conformant Kubernetes platforms. The operator follows standard Kubernetes APIs and practices and is designed to run consistently across certified Kubernetes environments.

The following table shows supported Kubernetes versions at the time of this release. For a list of platforms tested with this release, see Supported Kubernetes distributions.

Downloads

• Redis Enterprise: redislabs/redis:8.2.6-25.21 (redislabs/redis:8.2.6-25.21.minimal for minimal base image)
• Operator: redislabs/operator:8.2.6-21
• Services Rigger: redislabs/k8s-controller:8.2.6-21
• Call Home Client: redislabs/re-call-home-client:8.2.6-21

OpenShift downloads

• OLM operator bundle: 8.2.6-21.1
• Redis Enterprise: registry.connect.redhat.com/redislabs/redis-enterprise:8.2.6-25.21 (registry.connect.redhat.com/redislabs/redis-enterprise:8.2.6-25.21.minimal for minimal base image)
• Operator: registry.connect.redhat.com/redislabs/redis-enterprise-operator:8.2.6-21
• Services Rigger: registry.connect.redhat.com/redislabs/services-manager:8.2.6-21
• Call Home Client: registry.connect.redhat.com/redislabs/call-home-client:8.2.6-21

Known limitations

• Active-Active database updates can be rejected after upgrading to 8.2.6. On clusters running versions earlier than 8.2.6, a cluster master failover (for example, during a Kubernetes operator upgrade) can interrupt a change to an Active-Active database while the change is in progress. After the upgrade to 8.2.6, updates to that database can still be rejected with 409 – Another operation is already in progress. This only happens if the configuration change started within seconds of the failover. Workaround: Before you upgrade, wait until all pending Active-Active database changes have finished.

• SSO configuration does not work with IPv6 or dual-stack (IPv4/IPv6) clusters.

• Upgrades from versions earlier than 7.4.2-2 are not supported. If you use an earlier version, upgrade to 7.4.2-2 before upgrading to this version.

• Missing endpoint for admission endpoint (rare). Restart the operator pod.

• The REDB redisVersion field cannot be used for memcached databases.

• Modifying the database suffix for an Active-Active database while the services rigger is terminating causes a loop. The services rigger deletes and recreates ingress or route resources repeatedly. Wait for the services rigger pod to finish terminating before modifying the suffix.

• REAADB changes might fail with "gateway timeout" errors, especially on OpenShift. Retry the operation.

• Creating two databases with the same name in the Redis Enterprise Cluster Manager UI deletes the service and makes the database unavailable. Avoid duplicate database names. The admission controller prevents duplicate names when you create databases through the Kubernetes operator.

• Installing the operator bundle produces the warning Warning: would violate PodSecurity "restricted: v1.24". Ignore this warning. Red Hat documentation identifies this issue as benign.

• RERC resources must have a unique name. The rec-name/rec-namespace string must differ from all other participating clusters in the Active-Active database.

• Admission does not block REAADB resources with a shardCount that exceeds the license quota. Correct the REAADB configuration and reapply.

• The Active-Active controller supports only global database options. Location-specific configuration is not supported.

• Removing an Active-Active setup might leave services or routes undeleted. Delete the services or routes manually.

• Setting autoUpgrade to true can cause unexpected database upgrades when redisUpgradePolicy is also true. Contact support if your deployment is affected.

• Using the previous quick start guide causes REDB creation issues due to an unrecognized memory field name. Use the current version of the Deploy Redis Enterprise Software for Kubernetes guide.

• PVC size does not work with decimal values. Use integer values for the PVC size.

• REC might report error states on initial startup. No workaround exists. Ignore the errors.

• HashiCorp Vault integration does not support Gesher. No workaround exists. Gesher support is deprecated.

• REC clusters fail to start on Kubernetes clusters with unsynchronized clocks. Use NTP to synchronize the underlying Kubernetes nodes.

• Deleting an OpenShift project with a deployed REC might hang. When an REC cluster is deployed in a project (namespace) with REDB resources, you must delete the REDB resources before deleting the REC. The project deletion hangs until you delete the REDB resources. Delete the REDB resources first, then delete the REC, and then delete the project.

• Clusters must be named "rec" in OLM-based deployments. When you deploy the operator through OLM, the security context constraints (SCC) are bound to a specific service account name ("rec"). The deployment fails if the cluster has a different name.

• Readiness probe does not correctly report failures. The StatefulSet readiness probe does not mark a node as "not ready" when running rladmin status on node failure.

• Internal DNS and Kubernetes DNS might conflict. DNS conflicts can occur between the cluster mdns_server and Kubernetes DNS. This affects only DNS resolution from within cluster nodes for Kubernetes DNS names.

• Kubernetes-based 5.4.10 clusters might negatively affect existing 5.4.6 clusters. Upgrade your clusters to the latest version.

• Node CPU usage is reported instead of pod CPU usage. The reported CPU usage is for the Kubernetes worker node hosting the REC pod, not the pod itself.

• An unreachable cluster shows status as running. When a cluster is unreachable, the status remains running instead of showing an error.

• Long cluster names cause routes to be rejected. A cluster name longer than 20 characters causes route rejection because the host part of the domain name exceeds 63 characters. Limit the cluster name to 20 characters or fewer.

• Cluster CR (REC) errors are not reported after invalid updates. A cluster CR specification error is not reported if you update two or more invalid CR resources in sequence.