Tips for running Temporal on Kubernetes

Temporal provides durability as a service for distributed applications, handling state persistence, retries, and recovery automatically. Deploy Temporal on Kubernetes using official Helm charts, but disable bundled dependencies and connect to external production databases. Workers should be containerized with both workflow definitions and worker processes together, deployed as standard Kubernetes deployments. For autoscaling, avoid CPU/memory metrics and instead use Task Queue backlog, schedule-to-start latency, and worker task slot utilization. KEDA enables automatic scaling based on queue depth. Set GOMAXPROCS to match Pod CPU limits to prevent throttling and maintain consistent latency. Load test iteratively on dedicated clusters before production changes.