Deployment Overview
Yardrail is a two-binary system — yardrail-frontend (web UI, API, gRPC server) and yardrail-worker (Temporal poller, CWL workflow execution). Both connect to shared infrastructure: PostgreSQL, Temporal, and S3-compatible storage.
The worker supports two container executors that determine how CWL tool containers are run. Your choice of executor determines which deployment path to follow.
Deployment paths
| Docker Compose | Kubernetes (Helm) | |
|---|---|---|
| Executor | docker | kubernetes |
| Best for | Local development, single-server | Production, multi-tenant |
| How tools run | Docker containers on the host daemon | Kubernetes Pods with S3 CSI volumes |
| Infrastructure | Docker Engine, MinIO (bundled) | K8s cluster, S3, CSI driver |
| Tool isolation | Docker network mode none | Cluster NetworkPolicy |
| Image pulling | Docker daemon pulls and caches | Kubelet pulls per-node |
| Scaling | Single worker process | HPA on worker Deployment |
| Security model | Docker socket access (root-equivalent) | RBAC-scoped ServiceAccount |
Docker Compose
The fastest way to get Yardrail running. The docker-compose.yml in the repository bundles
PostgreSQL, MinIO, Temporal, and both Yardrail services. The worker runs CWL tools as sibling
Docker containers on the host daemon.
Best for local development and small single-server deployments where simplicity matters more than isolation.
Get started with Docker Compose →
Kubernetes
The production deployment path. Both services are packaged as a Helm chart. The worker runs CWL tools as Kubernetes Pods with S3 CSI inline volumes — no Docker socket needed, no root-equivalent access, and workflow Pods are first-class cluster citizens subject to RBAC, resource quotas, and network policies.
Best for production deployments where security, scalability, and operational visibility matter.
Shared dependencies
Both paths require these external services. Docker Compose bundles them; Kubernetes deployments bring their own.
| Dependency | Purpose | Notes |
|---|---|---|
| PostgreSQL v18+ | Frontend persistence, Temporal history | Migrations applied automatically on frontend startup |
| Temporal Server | Durable workflow execution | Single task queue, encrypted payloads |
| S3-compatible storage | Workflow artifacts, avatar uploads | MinIO for dev, AWS S3 / compatible for production |
Architecture deep dives
For details on how each executor works internally:
- Docker Executor Architecture — Docker SDK, tar-streaming, multiplexed stream protocol
- Kubernetes Executor Architecture — Pod lifecycle, CSI volumes, SPDY attach