Requirements & sizing
What you need before deploying Eddytor, and how to size the engine and datastores.
Eddytor's footprint is small - the heavy lifting is your object storage, which Eddytor connects to rather than hosts. Here's what to have ready.
What you need
| Path | Prerequisites |
|---|---|
| Docker Compose | A host with Docker + the Compose plugin, and outbound access to pull from ghcr.io/nordalf. |
| Kubernetes | A cluster + kubectl and helm ≥ 3.8 (for OCI charts). Nothing Eddytor-specific. |
Plus, for anything you keep:
- Postgres - Eddytor's metadata store. Bundled for evaluation; bring a managed/backed-up instance for production. See Bundled vs external.
- Object storage - an S3 / GCS / Azure Blob bucket where the Delta tables live. See Connect storage.
Sizing the parts
| Part | Default request | Notes |
|---|---|---|
| Server | ~200m CPU / 512Mi (×2 replicas) | Control plane; light. |
| Engine | ~1 CPU / 2Gi (×2 replicas) | Does the I/O and query work - scale this under load. |
| Postgres | small | Metadata only, not your table data. A Burstable/B2s-class managed instance is plenty to start. |
On Kubernetes, three 4-vCPU nodes give comfortable headroom for the defaults. For a small cluster, trim replicas:
--set engine.replicas=1 --set server.replicas=1 --set engine.autoscaling.enabled=falseScaling model
The engine is stateless - add replicas as query/IO load grows and the server fans work across them. The server and Postgres rarely need scaling. See Scaling the engine.