Skip to content

development.md

Latest commit

 

History

History
159 lines (112 loc) · 4.03 KB

development.md

File metadata and controls

159 lines (112 loc) · 4.03 KB
Raw

Development

This guide walks through deploying both the batch-gateway-apiserver and batch-gateway-processor in a Kind(Kubernetes in Docker) cluster for local development and testing.

Prerequisites

1. Create a Kind Cluster

$ kind create cluster --name batch-gateway-dev

Switch context to the Kind cluster and verify the cluster is up:

$ kubectl cluster-info --context kind-batch-gateway-dev

$ kubectl get nodes

2. Build Container Images

From the repository root, build both images. The Makefile auto-detects Docker or Podman.

$ make image-build

To force Podman:

$ CONTAINER_TOOL=podman make image-build

This produces:

  • ghcr.io/llm-d/batch-gateway-apiserver:0.0.1
  • ghcr.io/llm-d/batch-gateway-processor:0.0.1

To use a different tag:

IMAGE_TAG=dev make image-build

3. Load Images to Kind

Kind runs in Docker/Podman, so load the local images into the cluster.

Docker

# load apiserver image
$ kind load docker-image ghcr.io/llm-d/batch-gateway-apiserver:0.0.1 --name batch-gateway-dev

# load processor image
$ kind load docker-image ghcr.io/llm-d/batch-gateway-processor:0.0.1 --name batch-gateway-dev

Podman

If kind load docker-image does not work with your Podman setup (e.g. no Docker API socket), save images to a tar and load the archive:

# save images to tar
$ podman save -o /tmp/apiserver.tar ghcr.io/llm-d/batch-gateway-apiserver:0.0.1
$ podman save -o /tmp/processor.tar ghcr.io/llm-d/batch-gateway-processor:0.0.1

# load into Kind
$ kind load image-archive /tmp/apiserver.tar --name batch-gateway-dev
$ kind load image-archive /tmp/processor.tar --name batch-gateway-dev

If you built with a custom tag (e.g. dev), use that tag in both make image-build and the load commands above.

4. Deploy with Helm

The chart defaults to apiserver only. To run both apiserver and processor using the images loaded in #3, override the values.

$ helm install batch-gateway ./charts/batch-gateway \
    --set apiserver.image.pullPolicy=IfNotPresent \
    --set apiserver.image.tag=0.0.1 \
    --set processor.enabled=true \
    --set processor.image.pullPolicy=IfNotPresent \
    --set processor.image.tag=0.0.1 \
    --namespace default

5. Verify Deployment

# Pods
$ kubectl get pods -l app.kubernetes.io/name=batch-gateway-apiserver
$ kubectl get pods -l app.kubernetes.io/name=batch-gateway-processor

# Services
$ kubectl get svc -l app.kubernetes.io/instance=batch-gateway

Check apiserver health:

# forward traffic from cluster port to your machine port
$ kubectl port-forward svc/batch-gateway-batch-gateway-apiserver 8000:8000

# In another terminal:
$ curl -s http://localhost:8000/health

6. Rebuild and Reloading images

After code changes:

Docker:

# rebuild images
$ make image-build

# load images with docker
$ kind load docker-image ghcr.io/llm-d/batch-gateway-apiserver:0.0.1 --name batch-gateway-dev
$ kind load docker-image ghcr.io/llm-d/batch-gateway-processor:0.0.1 --name batch-gateway-dev

# load images with podman
$ podman save -o /tmp/apiserver.tar ghcr.io/llm-d/batch-gateway-apiserver:0.0.1
$ podman save -o /tmp/processor.tar ghcr.io/llm-d/batch-gateway-processor:0.0.1
$ kind load image-archive /tmp/apiserver.tar --name batch-gateway-dev
$ kind load image-archive /tmp/processor.tar --name batch-gateway-dev

# reload deployment
$ kubectl rollout restart deployment -l app.kubernetes.io/instance=batch-gateway

7. Cleanup

# removes all resources
$ helm uninstall batch-gateway

# deletes the Kind cluster
$ kind delete cluster --name batch-gateway-dev