Kubernetes Maintenance Support

Your Pulsar server group is configured as a group of StatefulSets, both quorum-based and Active-Active. It has some implications on performing Kubernetes maintenance activities, such as workload rebalancing, node-image upgrades, or security patching.

The basic strategy for minimizing downtime when restarting StatefulSets is:

  • Add new replacement nodes

  • Cordon current nodes to avoid node-stickiness issues

  • Process each StatefulSet in the order that Apache Pulsar recommends: deleting each pod and waiting for it to restart and fully rejoin the group