Kubernetes example

In this example you’ll deploy NativeLink to a local Kubernetes cluster using Helm and run Bazel builds against it.

Requirements

A local Kubernetes cluster. Either:
- Docker Desktop with Kubernetes enabled (Settings > Kubernetes > Enable Kubernetes), or
- minikube
Helm 3
kubectl
Bazel (to run builds against the cluster)

Install NativeLink

Pull the Helm chart from the NativeLink OCI registry and install it:

helm install nativelink \
    oci://public.ecr.aws/b1l4l6w7/tracemachina/nativelink \
    --version 1.0.0-rc1 \
    -n nativelink \
    --create-namespace

Verify the deployment

Wait for the pods to become ready:

kubectl get pods -n nativelink -w

You should see pods for the CAS, scheduler, and worker components:

NAME                                    READY   STATUS    RESTARTS   AGE
nativelink-cas-...                      1/1     Running   0          60s
nativelink-scheduler-...                1/1     Running   0          60s
nativelink-worker-...                   1/1     Running   0          60s

Explore the deployment

View logs for individual components:

# CAS (Content Addressable Storage) logs
kubectl logs deploy/nativelink-cas -n nativelink

# Scheduler logs
kubectl logs deploy/nativelink-scheduler -n nativelink

# Worker logs
kubectl logs deploy/nativelink-worker -n nativelink

Build against NativeLink

Port-forward the NativeLink services to your local machine:

# In a separate terminal — keep this running
kubectl port-forward svc/nativelink-cas 50051:50051 -n nativelink &
kubectl port-forward svc/nativelink-scheduler 50052:50052 -n nativelink &

Now run a Bazel build against the cluster:

bazel build \
    --remote_cache=grpc://localhost:50051 \
    --remote_executor=grpc://localhost:50052 \
    //...

Add these flags to a .bazelrc.user file in your workspace root so you don’t have to pass them on every invocation:

build --remote_cache=grpc://localhost:50051
build --remote_executor=grpc://localhost:50052

Make sure your .bazelrc imports it:

try-import %workspace%/.bazelrc.user

The key output to look for:

INFO: 11 processes: 9 internal, 2 remote.

This tells you the build ran against the cluster. Clean the local cache and rebuild to verify caching works:

bazel clean && bazel build \
    --remote_cache=grpc://localhost:50051 \
    --remote_executor=grpc://localhost:50052 \
    //...

You should now see cache hits:

INFO: 11 processes: 2 remote cache hit, 9 internal.

Customize the deployment

To customize the Helm chart values, first extract the default values:

helm show values \
    oci://public.ecr.aws/b1l4l6w7/tracemachina/nativelink \
    --version 1.0.0-rc1 > nativelink-values.yaml

Edit nativelink-values.yaml to your needs, then upgrade the release:

helm upgrade nativelink \
    oci://public.ecr.aws/b1l4l6w7/tracemachina/nativelink \
    --version 1.0.0-rc1 \
    -n nativelink \
    -f nativelink-values.yaml

See the Helm chart documentation for details on configuring cloud object stores (S3, GCS, Azure Blob), Redis, ingress, metrics, and more.

Clean up

When you’re done testing, remove the NativeLink deployment:

helm uninstall nativelink -n nativelink
kubectl delete namespace nativelink