Problem
The StatefulSet uses podManagementPolicy: Parallel, which starts and stops all pods simultaneously during scaling operations. For a Raft-based cluster like Typesense, this is dangerous: if multiple pods go down at the same time (e.g. during a scale-down or a correlated failure), quorum is lost.
OrderedReady ensures pods are started and stopped one at a time, preserving quorum throughout scaling and recovery operations.
Proposed Change
Change the default to OrderedReady, or expose it as a configurable value:
# values.yaml
podManagementPolicy: OrderedReady
# statefulset.yaml
podManagementPolicy: {{ .Values.podManagementPolicy | default "OrderedReady" }}Note
updateStrategy.rollingUpdate.maxUnavailable only governs Helm-triggered rollouts. It does not prevent multiple pods from being restarted simultaneously when liveness probes fail or during scaling. podManagementPolicy: OrderedReady is the correct mechanism to enforce ordered restarts at the StatefulSet level.
Problem
The StatefulSet uses
podManagementPolicy: Parallel, which starts and stops all pods simultaneously during scaling operations. For a Raft-based cluster like Typesense, this is dangerous: if multiple pods go down at the same time (e.g. during a scale-down or a correlated failure), quorum is lost.OrderedReadyensures pods are started and stopped one at a time, preserving quorum throughout scaling and recovery operations.Proposed Change
Change the default to
OrderedReady, or expose it as a configurable value:Note
updateStrategy.rollingUpdate.maxUnavailableonly governs Helm-triggered rollouts. It does not prevent multiple pods from being restarted simultaneously when liveness probes fail or during scaling.podManagementPolicy: OrderedReadyis the correct mechanism to enforce ordered restarts at the StatefulSet level.