fs-operator

Kubernetes operator for clustered go-faster/fs — an S3-compatible object store with quorum replication (rf2.5, rf3, ec:k,m), failure-domain-aware placement, automatic rebalancing and scrub/repair.

The operator manages the full lifecycle of fs clusters through three namespaced custom resources:

Kind	Purpose
FSCluster	A whole fs cluster: nodes, racks, disks, etcd, auth, exposure, tuning.
FSBucket	An S3 bucket in a referenced cluster.
FSAccessKey	One S3 credential with bucket grants, generated or imported.

It provisions per-node StatefulSets with PVC-backed disks, maps fs racks (failure domains) onto zones, and encodes fs's operational contracts as controller logic: rolling updates one node at a time gated on cluster reconvergence, explicit schema migrations, and drain-before-remove decommissioning.

See SPEC.md for the full design and docs/ for installation and guides.

Quick start

# Install the operator (CRDs included).
helm install fs-operator oci://ghcr.io/go-faster/charts/fs-operator \
  --namespace fs-operator-system --create-namespace

# Create a 3-node dev cluster (requires a reachable etcd).
kubectl apply -f examples/01-minimal.yaml

The examples/ gallery goes from a minimal dev cluster to a zonal, multi-disk production shape.

Development

Standard kubebuilder workflow:

make manifests generate   # regenerate CRDs and deepcopy after API changes
make helm-sync-crds       # keep the owned chart's CRDs in lockstep
make test                 # unit + envtest
make lint                 # golangci-lint

The Helm chart in dist/chart is committed and hand-owned; its CRD templates and the manager RBAC are synced from config/ by hack/sync-chart.sh and CI fails on drift.

License

Apache-2.0. Copyright 2026.