Update known issues¶

This section lists the update known issues with workarounds for the MOSK release 24.1.

[40036] Node is not removed from a cluster when its ‘Machine’ is ‘disabled’¶

During the ClusterRelease update of a MOSK cluster, a node cannot be removed from the Kubernetes cluster if the related Machine object is disabled.

As a workaround, remove the finalizer from the affected Node object.

[42463] KubePodsCrashLooping is firing during cluster update¶

During major or patch update of a MOSK cluster with StackLight enabled in non-HA mode, the KubePodsCrashLooping alert may be firing for the Grafana ReplicaSet.

Grafana relies on PostgreSQL for persistent data. In non-HA StackLight setup, PostgreSQL becomes temporarily unavailable during updates. If Grafana loses its database connection or fails to establish one during startup, Grafana fails with an error. This may cause the Grafana pod to enter the CrashLoopBackOff state. Such behavior is expected in non-HA StackLight setups. The Grafana pod will resume normal operation after PostgreSQL is restored.

To prevent the issue, deploy StackLight in HA mode.

[41810] Cluster update is stuck due to the OpenStack Controller flooding¶

Fixed in MOSK 24.1.3

The cluster update may stuck if the maximum number of the worker nodes to update simultaneously is ten or higher.

To work around the problem, set the spec.providerSpec.maxWorkerUpgradeCount to a value lower than 10. For configuration details, see Configure the parallel update of worker nodes.