-
|
Encountering an issue when running a Keycloak cluster with 2 pods on Kubernetes, using the default DNS_PING protocol for pod discovery. Problem description: After redeploying Keycloak with 2 pods, the cluster forms but logs show repeated warnings like: |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 5 replies
-
|
Thank you for reaching out. Please provide the keycloak version you use, the configuration options you used and the complete startup log of an instance with that message. |
Beta Was this translation helpful? Give feedback.
-
|
v25, 2025-08-08 11:22:32,051 INFO [org.infinispan.LIFECYCLE] (non-blocking-thread--p2-t5) [Context=org.infinispan.CONFIG] ISPN100010: Finished rebalance with members [sample-pod-3, , sample-pod-2, sample-pod-1], topology id 258 |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for the version and the excerpt from the log. The version you are using is longer supported. Please upgrade to the latest version of Keycloak and see if this problem persists. I didn't see the configuration I've asked for. Looking at the logs you might be using a deployment, where we advocate for a StatefulSet. Maybe that will resolve it. While we recommend using the Keycloak Operator, you find in the latest docs an example on how to use a manually set up StatefulSet. |
Beta Was this translation helpful? Give feedback.
-
|
@ahus1 Why do you suggest statefulset rather than deployment with regard to this issue |
Beta Was this translation helpful? Give feedback.
-
|
Keycloak holds some information about authentication sessions in memory, and ensures that there is a backup in one other node for that information. To avoid losing data, scaling down one node at a time, and doing rolling updates one node at a time is the recommended way, which is the default for a StatefulSet. Also when starting Keycloak, starting the nodes one-by-one is a good way to ensure that a cluster forms correctly all the time. We have taken some measures in 26.3.4 that this shouldn't be a problem any more at least not when you use the built-in |
Beta Was this translation helpful? Give feedback.
-
|
Currently we are using deployment approach with rolling update set to scale down one pod at a time. Cluster is forming. Also dns_ping was used for node discovery instead of jdbc_ping. Is this approach ok to proceed |
Beta Was this translation helpful? Give feedback.
Thank you for the version and the excerpt from the log.
The version you are using is longer supported. Please upgrade to the latest version of Keycloak and see if this problem persists.
I didn't see the configuration I've asked for. Looking at the logs you might be using a deployment, where we advocate for a StatefulSet. Maybe that will resolve it. While we recommend using the Keycloak Operator, you find in the latest docs an example on how to use a manually set up StatefulSet.