Skip to content
BigFleet Providers

Install & deploy

The OCI provider ships as a container image and a Helm chart. Run one release per region; never scale a release past replicas: 1 (the provider is the single, region-scoped CapacityProvider process for its region).

Prerequisites

  • A Kubernetes cluster to run the provider in (OKE or anywhere it can reach the OCI API and your offerings).
  • The provider’s identity authorized — a dynamic group + IAM policy (Instance Principal / Workload Identity) or an ~/.oci/config Secret. See Credentials & auth.
  • A compartment OCID, a subnet OCID, and a base image OCID in the target region.
  • An offerings file describing the capacity the provider may provision (see Configuration).

1. The image

Build the image from the repository root with providers/oracle-cloud/deploy/Dockerfile — it must build from the repo root so the multi-module replace => ../.. resolves providerkit):

Terminal window
docker build -f providers/oracle-cloud/deploy/Dockerfile \
  -t <your-registry>/bigfleet-oracle-cloud:0.1.0 .
docker push <your-registry>/bigfleet-oracle-cloud:0.1.0

Then set image.repository/image.tag to wherever you pushed it. (This repo does not publish an official image; if your own release pipeline publishes one, point the chart at that instead.) The result is distroless/static:nonroot (uid 65532, no shell, read-only rootfs) and exposes the gRPC port (9000) and the metrics/health port (9090).

2. Install the chart

One release per region:

Terminal window
helm install oci-eu-frankfurt-1 providers/oracle-cloud/deploy/helm \
  --namespace bigfleet --create-namespace \
  --set image.tag=0.1.0 \
  --set region=eu-frankfurt-1 \
  --set provider=oci-eu-frankfurt-1 \
  --set oci.compartment=ocid1.compartment.oc1..bbbb \
  --set oci.subnet=ocid1.subnet.oc1..dddd \
  --set oci.image=ocid1.image.oc1..eeee \
  --set oci.auth=instance_principal \
  --set-file offerings.content=offerings.eu-frankfurt-1.json

This renders a hardened Deployment (replicas: 1, Recreate, liveness /healthz + readiness /readyz), a Service exposing grpc (BigFleet dials this) and metrics, a ServiceAccount, a ConfigMap for the offerings, and optionally a PVC for durable state.

Persist fence marks, the idempotency map, inventory, and cluster bindings across restarts by pointing --state at a PersistentVolume:

Terminal window
--set state.enabled=true \
--set state.persistence.enabled=true \
--set state.persistence.size=1Gi

Without it the store is in-memory: a restart re-seeds inventory from OCI via Describe (recovered from the bigfleet-managed / bigfleet-machine-id freeform tags), but in-flight transitions surface as FAILED for the shard to re-drive.

4. Point BigFleet at it

BigFleet shards are the gRPC client; configure a shard’s --provider-addr (Helm shard.provider.addr) at this release’s grpc Service. The provider never dials the shard.

Backend selection

--oci-backend is auto by default: it uses the real OCI backend when both --region and --compartment are set, and the in-memory fake otherwise (dev / credential-free certification). Force it with --oci-backend=oci|fake.

See Configuration for every flag and the offerings schema, and the deploy README for the Dockerfile, chart, and Terraform.