Support multi-platform builds (#81)

.circleci/config.yml

@@ -78,18 +78,13 @@ jobs:
       - checkout
       - pre_docker_commands
       - run:
-          name: "Build Images"
+          name: "Build and Push Images"
           environment:
             CIRCLE_TAG: <<pipeline.git.revision>>
-          command: |
-            DOCKER_REPO_BASE="gcr.io/run-ai-staging/fake-gpu-operator"
-            make images DOCKER_REPO_BASE=${DOCKER_REPO_BASE} DOCKER_TAG=${CIRCLE_TAG}
-      - run:
-          name: "Push to Google Cloud"
           command: |
             TAG=<<pipeline.git.revision>>
             DOCKER_REPO_BASE="gcr.io/run-ai-staging/fake-gpu-operator"
-            make push-all DOCKER_REPO_BASE=${DOCKER_REPO_BASE} DOCKER_TAG=${TAG}
+            make images DOCKER_REPO_BASE=${DOCKER_REPO_BASE} DOCKER_TAG=${TAG} SHOULD_PUSH=true
   deploy_to_production:
     executor: runai-executor
     steps:
@@ -102,8 +97,7 @@ jobs:
           command: |
             TAG=${CIRCLE_TAG/v/''}
             DOCKER_REPO_BASE="gcr.io/run-ai-staging/fake-gpu-operator"
-            make images DOCKER_REPO_BASE=${DOCKER_REPO_BASE} DOCKER_TAG=${TAG}
-            make push-all DOCKER_REPO_BASE=${DOCKER_REPO_BASE} DOCKER_TAG=${TAG}
+            make images DOCKER_REPO_BASE=${DOCKER_REPO_BASE} DOCKER_TAG=${TAG} SHOULD_PUSH=true
   chart_build_staging:
     working_directory: ~/go/src/github.com/run-ai/fake-gpu-operator
     docker:

Dockerfile

@@ -1,42 +1,43 @@
-FROM golang:1.22.1 as common-builder
+FROM --platform=$BUILDPLATFORM golang:1.22.1 as common-builder
 WORKDIR $GOPATH/src/github.com/run-ai/fake-gpu-operator
 COPY go.mod .
 COPY go.sum .
 RUN go mod download
 COPY Makefile .
 COPY internal/common ./internal/common
+ARG TARGETOS TARGETARCH
 
 FROM common-builder as device-plugin-builder
 COPY ./cmd/device-plugin/ ./cmd/device-plugin/
 COPY ./internal/deviceplugin/ ./internal/deviceplugin/
-RUN make build COMPONENT=device-plugin
+RUN --mount=type=cache,target=/root/.cache/go-build make build OS=$TARGETOS ARCH=$TARGETARCH COMPONENT=device-plugin
 
 FROM common-builder as status-updater-builder
 COPY ./cmd/status-updater/ ./cmd/status-updater/
 COPY ./internal/status-updater/ ./internal/status-updater/
-RUN --mount=type=cache,target=/root/.cache/go-build make build COMPONENT=status-updater
+RUN --mount=type=cache,target=/root/.cache/go-build make build OS=$TARGETOS ARCH=$TARGETARCH COMPONENT=status-updater
 
 FROM common-builder as status-exporter-builder
 COPY ./cmd/status-exporter/ ./cmd/status-exporter/
 COPY ./internal/ ./internal/
-RUN --mount=type=cache,target=/root/.cache/go-build make build COMPONENT=status-exporter
+RUN --mount=type=cache,target=/root/.cache/go-build make build OS=$TARGETOS ARCH=$TARGETARCH COMPONENT=status-exporter
 
 FROM common-builder as topology-server-builder
 COPY ./cmd/topology-server/ ./cmd/topology-server/
-RUN --mount=type=cache,target=/root/.cache/go-build make build COMPONENT=topology-server
+RUN --mount=type=cache,target=/root/.cache/go-build make build OS=$TARGETOS ARCH=$TARGETARCH COMPONENT=topology-server
 
 FROM common-builder as nvidia-smi-builder
 COPY ./cmd/nvidia-smi/ ./cmd/nvidia-smi/
-RUN make build COMPONENT=nvidia-smi
-
-FROM common-builder as preloader-builder 
-COPY ./cmd/preloader/ ./cmd/preloader/
-RUN make build-preloader
+RUN --mount=type=cache,target=/root/.cache/go-build make build OS=$TARGETOS ARCH=$TARGETARCH COMPONENT=nvidia-smi
 
 FROM common-builder as mig-faker-builder
 COPY ./cmd/mig-faker/ ./cmd/mig-faker/
 COPY ./internal/ ./internal/
-RUN --mount=type=cache,target=/root/.cache/go-build make build COMPONENT=mig-faker
+RUN --mount=type=cache,target=/root/.cache/go-build make build OS=$TARGETOS ARCH=$TARGETARCH COMPONENT=mig-faker
+
+FROM common-builder as preloader-builder 
+COPY ./cmd/preloader/ ./cmd/preloader/
+RUN make build-preloader
 
 FROM jupyter/minimal-notebook as jupyter-notebook
 COPY --from=nvidia-smi-builder /go/src/github.com/run-ai/fake-gpu-operator/bin/nvidia-smi /bin/

Makefile

@@ -7,8 +7,16 @@ DOCKER_TAG=0.0.0-dev
 DOCKER_IMAGE_NAME=${DOCKER_REPO_FULL}:${DOCKER_TAG}
 NAMESPACE=gpu-operator
 
+SHOULD_PUSH?=false
+DOCKER_BUILDX_PUSH_FLAG=$(if $(filter true,$(SHOULD_PUSH)),--push,)
+DOCKER_BUILDX_PLATFORMS?=linux/amd64,linux/arm64
+DOCKER_BUILDX_BUILDER?=fgo-multi-platform
+
+OS?=linux
+ARCH?=amd64
+
 build:
-	env GOOS=linux GOARCH=amd64 go build -o ${BUILD_DIR}/ ./cmd/${COMPONENT}
+	env GOOS=${OS} GOARCH=${ARCH} go build -o ${BUILD_DIR}/ ./cmd/${COMPONENT}
 .PHONY: build
 
 build-preloader:
@@ -20,8 +28,12 @@ clean:
 	rm -rf ${BUILD_DIR}
 .PHONY: clean
 
-image:
-	DOCKER_BUILDKIT=1 docker build -t ${DOCKER_IMAGE_NAME} --target ${COMPONENT} .
+init-buildx:
+	docker buildx inspect fgo-multi-platform > /dev/null || docker buildx create --name=fgo-multi-platform
+.PHONY: init-buildx
+
+image: init-buildx
+	docker buildx --builder=fgo-multi-platform build -t ${DOCKER_IMAGE_NAME} --target ${COMPONENT} --platform ${DOCKER_BUILDX_PLATFORMS} ${DOCKER_BUILDX_PUSH_FLAG} .
 .PHONY: image
 
 images:
@@ -33,35 +45,10 @@ images:
 	make image COMPONENT=jupyter-notebook
 .PHONY: images
 
-push:
-	docker push ${DOCKER_IMAGE_NAME}
-.PHONY: push
-
-push-all:
-	make push COMPONENT=device-plugin
-	make push COMPONENT=status-updater
-	make push COMPONENT=status-exporter
-	make push COMPONENT=topology-server
-	make push COMPONENT=mig-faker
-	make push COMPONENT=jupyter-notebook
-.PHONY: push-all
-
 restart: 
 	kubectl delete pod -l component=${COMPONENT} --force -n ${NAMESPACE}
 .PHONY: restart
 
-deploy: image push
-.PHONY: deploy
-
-deploy-all:
-	make image push COMPONENT=device-plugin
-	make image push COMPONENT=status-updater
-	make image push COMPONENT=status-exporter
-	make image push COMPONENT=topology-server
-	make image push COMPONENT=mig-faker
-	make image push COMPONENT=jupyter-notebook
-.PHONY: deploy-all
-
 image-test:
 	mkdir -p /tmp/artifacts/test-results
 	mkdir -p /tmp/artifacts/test-results/unit-tests