Hybrid Kubernetes Architecture

If you're already running Kubernetes, this is the natural fit. Deploy a Prefect worker as a long-running pod via Helm chart. It polls Prefect Cloud for scheduled work and spins up Kubernetes Jobs to execute your flows. Each flow run gets its own pod, isolated and resource-controlled.

The work pool defines your defaults: namespace, Docker image, CPU and memory limits, environment variables. Individual deployments can override any of these as needed. One flow requires 4GB RAM while others run fine on 512MB? Set it at the deployment level.

This flexibility extends to multi-tenant setups. Namespace isolation means different teams get different namespaces, different service accounts, different resource quotas. Prefect Cloud handles scheduling and observability while Kubernetes handles execution exactly how your platform team configured it.

The official Helm chart includes liveness probes out of the box. If a worker stops polling, Kubernetes restarts it automatically. No manual intervention. No silent failures.

For setup details, see the Kubernetes deployment guide.