Operations

This section is the day-2 reference for running Valkey clusters on the Momento Valkey Operator in production: what happens when things fail, how to place and size clusters, how to validate performance, and how to diagnose problems. It assumes a cluster is already provisioned; for first-time setup, see Getting started and the platform and product team guides.

• Failure modes — the flagship page. How a Valkey cluster behaves, automatically, when a replica dies, a primary dies with or without quorum or replicas, a Kubernetes node or zone is lost, or the operator itself goes down. Read this before you need it.
• Zone-aware placement — how placement.zones, nodeSelector, and zoneSpread control where pods land, what changing placement does to a live cluster, and how placement interacts with a cluster autoscaler.
• Sizing — setting resources in a ValkeyConfig so pods get Guaranteed quality of service, leaving maxmemory headroom, and choosing when to scale up versus out.
• Benchmarking — validating throughput and latency against an operator-provisioned cluster with standard Valkey tooling, in-cluster, including TLS and ACL variants.
• Troubleshooting — symptom-first diagnosis: stuck Creating, Invalid clusters, pods stuck Pending, ACL authentication failures, stalled rolling upgrades, stuck Terminating, and what to collect before contacting support.
• Uninstall — the order to remove clusters, the operator, and CRDs, and how to verify nothing is left behind.

note

Failure modes and Data durability cover related ground from different angles: failure modes describes the operator's automatic response to each event, while data durability states the storage model and exactly which events lose data. Read them together before setting availability expectations for a cluster.