Support arm builds

.github/workflows/docker.yml

@@ -32,6 +32,15 @@ jobs:
       !contains(github.event.pull_request.title, 'ci skip')
 
     steps:
+      # Setup docker to build for multiple platforms, see:
+      # https://github.com/docker/build-push-action/tree/master#usage
+      # https://github.com/docker/build-push-action/blob/master/docs/advanced/multi-platform.md
+      - name: Set up QEMU (for docker buildx)
+        uses: docker/setup-qemu-action@27d0a4f181a40b142cce983c5393082c365d1480 # dependabot updates to latest release
+
+      - name: Set up Docker Buildx (for multi-arch builds)
+        uses: docker/setup-buildx-action@0d135e0c2fc0dba0729c1a47ecfcf5a3c7f8579e # dependabot updates to latest release
+
       - name: Clone Main Repo
         uses: actions/checkout@v2
         with:
@@ -48,18 +57,22 @@ jobs:
           make -C main dev-env
 
       - name: Build Docker Images
-        run: make -C main build-test-all
+        run: make -C main build-all-multi
         env:
           # Full logs for CI build
           BUILDKIT_PROGRESS: plain
 
+      - name: Test Docker Images
+        run: make -C main test-all
+
       - name: Clone Wiki
         uses: actions/checkout@v2
         with:
           repository: ${{github.repository}}.wiki
           path: wiki
 
       - name: Run Post-Build Hooks
+        id: hook-all
         run: make -C main hook-all
 
       - name: Push Wiki to GitHub
@@ -78,4 +91,4 @@ jobs:
 
       - name: Push Images to DockerHub
         if: github.ref == 'refs/heads/master' || github.ref == 'refs/heads/main'
-        run: make -C main push-all
+        run: make -C main push-all-multi

Makefile

@@ -7,6 +7,22 @@ SHELL:=bash
 OWNER?=jupyter
 
 # Need to list the images in build dependency order
+
+# Images supporting the following architectures:
+# - linux/amd64
+# - linux/arm64
+MULTI_IMAGES:= \
+	base-notebook \
+	minimal-notebook
+# Images that can only be built on the amd64 architecture (aka. x86_64)
+AMD64_ONLY_IMAGES:= \
+	r-notebook \
+	scipy-notebook \
+	tensorflow-notebook \
+	datascience-notebook \
+	pyspark-notebook \
+	all-spark-notebook
+# All of the images
 ALL_IMAGES:= \
 	base-notebook \
 	minimal-notebook \
@@ -26,7 +42,7 @@ export DOCKER_BUILDKIT:=1
 help:
 	@echo "jupyter/docker-stacks"
 	@echo "====================="
-	@echo "Replace % with a stack directory name (e.g., make build/minimal-notebook)"
+	@echo "Replace % with a stack directory name (e.g., make build-multi/minimal-notebook)"
 	@echo
 	@grep -E '^[a-zA-Z0-9_%/-]+:.*?## .*$$' $(MAKEFILE_LIST) | sort | awk 'BEGIN {FS = ":.*?## "}; {printf "\033[36m%-30s\033[0m %s\n", $$1, $$2}'
 
@@ -40,7 +56,59 @@ build/%: ## build the latest image for a stack using the system's architecture
 	@docker images $(OWNER)/$(notdir $@):latest --format "{{.Size}}"
 	@echo "::endgroup::"
 build-all: $(foreach I, $(ALL_IMAGES), build/$(I)) ## build all stacks
-build-test-all: $(foreach I, $(ALL_IMAGES), build/$(I) test/$(I)) ## build and test all stacks
+
+# Limitations on docker buildx build (using docker/buildx 0.5.1):
+#
+# 1. Can't --load and --push at the same time
+#
+# 2. Can't --load multiple platforms
+#
+# What does it mean to --load?
+#
+# - It means that the built image can be referenced by `docker` CLI, for example
+#   when using the `docker tag` or `docker push` commands.
+#
+# Workarounds due to limitations:
+#
+# 1. We always build a dedicated image using the current system architecture
+#    named as OWNER/<stack>-notebook so we always can reference that image no
+#    matter what during tests etc.
+#
+# 2. We always also build a multi-platform image during build-multi that will be
+#    inaccessible with `docker tag` and `docker push` etc, but this will help us
+#    test the build on the different platform and provide cached layers for
+#    later.
+#
+# 3. We let push-multi refer to rebuilding a multi image with `--push`.
+#
+#    We can rely on the cached layer from build-multi now even though we never
+#    tagged the multi image.
+#
+# Outcomes of the workaround:
+#
+# 1. We can keep using the previously defined Makefile commands that doesn't
+#    include `-multi` suffix as before.
+#
+# 2. Assuming we have setup docker/dockerx properly to build in arm64
+#    architectures as well, then we can build and publish such images via the
+#    `-multi` suffix without needing a local registry.
+#
+# 3. If we get dedicated arm64 runners, we can test everything separately
+#    without needing to update this Makefile, and if all tests succeeds we can
+#    do a publish job that creates a multi-platform image for us.
+#
+build-multi/%: DARGS?=
+build-multi/%: ## build the latest image for a stack on both amd64 and arm64
+	@echo "::group::Build $(OWNER)/$(notdir $@) (system's architecture)"
+	docker buildx build $(DARGS) --rm --force-rm -t $(OWNER)/$(notdir $@):latest ./$(notdir $@) --build-arg OWNER=$(OWNER)
+	@echo -n "Built image size: "
+	@docker images $(OWNER)/$(notdir $@):latest --format "{{.Size}}"
+	@echo "::endgroup::"
+
+	@echo "::group::Build $(OWNER)/$(notdir $@) (amd64,arm64)"
+	docker buildx build $(DARGS) --rm --force-rm -t build-multi-tmp-cache/$(notdir $@):latest ./$(notdir $@) --build-arg OWNER=$(OWNER) --platform "linux/amd64,linux/arm64"
+	@echo "::endgroup::"
+build-all-multi: $(foreach I, $(MULTI_IMAGES), build-multi/$(I)) $(foreach I, $(AMD64_ONLY_IMAGES), build/$(I)) ## build all stacks
 
 
 
@@ -117,6 +185,13 @@ push/%: ## push all tags for a jupyter image
 	@echo "::endgroup::"
 push-all: $(foreach I, $(ALL_IMAGES), push/$(I)) ## push all tagged images
 
+push-multi/%: DARGS?=
+push-multi/%: ## push all tags for a jupyter image that support multiple architectures
+	@echo "::group::Push $(OWNER)/$(notdir $@) (amd64,arm64)"
+	docker buildx build $(DARGS) --rm --force-rm $($(subst -,_,$(notdir $@))_EXTRA_TAG_ARGS) -t $(OWNER)/$(notdir $@):latest ./$(notdir $@) --build-arg OWNER=$(OWNER) --platform "linux/amd64,linux/arm64"
+	@echo "::endgroup::"
+push-all-multi: $(foreach I, $(MULTI_IMAGES), push-multi/$(I)) $(foreach I, $(AMD64_ONLY_IMAGES), push/$(I)) ## push all tagged images
+
 
 
 run/%: DARGS?=
@@ -134,5 +209,4 @@ test/%: ## run tests against a stack (only common tests or common tests + specif
 	@if [ ! -d "$(notdir $@)/test" ]; then TEST_IMAGE="$(OWNER)/$(notdir $@)" pytest -m "not info" test; \
 	else TEST_IMAGE="$(OWNER)/$(notdir $@)" pytest -m "not info" test $(notdir $@)/test; fi
 	@echo "::endgroup::"
-
 test-all: $(foreach I, $(ALL_IMAGES), test/$(I)) ## test all stacks

README.md

@@ -105,6 +105,18 @@ This change is tracked in the issue [#1217](https://github.com/jupyter/docker-st
 - [Jupyter Website](https://jupyter.org)
 - [Images on DockerHub](https://hub.docker.com/u/jupyter)
 
-## Architectures
+## CPU Architectures
 
-Currently published containers only support x86, some containers may support cross-building with docker buildx.
+All published containers support amd64 (x86_64). The base-notebook and
+minimal-notebook containers also support arm64. The ambition is to have all
+containers support both amd64 and arm64.
+
+### Caveats for arm64 images
+
+- The manifests we publish in this projects wiki as well as the image tags for
+  the multi platform images that also support arm, are all based on the amd64
+  version even though details about the installed packages versions could differ
+  between architectures. For the status about this, see
+  [#1401](https://github.com/jupyter/docker-stacks/issues/1401).
+- Only the amd64 images are actively tested currently. For the status about
+  this, see [#1402](https://github.com/jupyter/docker-stacks/issues/1402).

docs/index.rst

@@ -44,10 +44,23 @@ Docker destroys the container after notebook server exit, but any files written
 
     docker run --rm -p 10000:8888 -e JUPYTER_ENABLE_LAB=yes -v "${PWD}":/home/jovyan/work jupyter/datascience-notebook:33add21fab64
 
-Architectures
------------
-Currently published containers only support x86, some containers may support cross-building with docker buildx.
+CPU Architectures
+-----------------
+
+All published containers support amd64 (x86_64). The base-notebook and
+minimal-notebook containers also support arm64. The ambition is to have all
+containers support both amd64 and arm64.
+
+Caveats for arm64 images
+^^^^^^^^^^^^^^^^^^^^^^^^
 
+- The manifests we publish in this projects wiki as well as the image tags for
+  the multi platform images that also support arm, are all based on the amd64
+  version even though details about the installed packages versions could differ
+  between architectures. For the status about this, see
+  [#1401](https://github.com/jupyter/docker-stacks/issues/1401).
+- Only the amd64 images are actively tested currently. For the status about
+  this, see [#1402](https://github.com/jupyter/docker-stacks/issues/1402).
 
 Table of Contents
 -----------------

tagging/github_set_env.py

@@ -0,0 +1,11 @@
+# Copyright (c) Jupyter Development Team.
+# Distributed under the terms of the Modified BSD License.
+import os
+
+
+def github_set_env(env_name, env_value):
+    if not os.environ.get("GITHUB_ACTIONS") or not os.environ.get("GITHUB_ENV"):
+        return
+
+    with open(os.environ["GITHUB_ENV"], "a") as f:
+        f.write(f"{env_name}={env_value}\n")

tagging/tag_image.py

@@ -6,6 +6,7 @@
 from plumbum.cmd import docker
 from .docker_runner import DockerRunner
 from .get_taggers_and_manifests import get_taggers_and_manifests
+from .github_set_env import github_set_env
 
 
 logger = logging.getLogger(__name__)
@@ -15,21 +16,31 @@ def tag_image(short_image_name: str, owner: str) -> None:
     """
     Tags <owner>/<short_image_name>:latest with the tags reported by all taggers
     for the given image.
+
+    Tags are in a GitHub Actions environment also saved to environment variables
+    in a format making it easy to append them.
     """
     logger.info(f"Tagging image: {short_image_name}")
     taggers, _ = get_taggers_and_manifests(short_image_name)
 
     image = f"{owner}/{short_image_name}:latest"
 
     with DockerRunner(image) as container:
+        tags = []
         for tagger in taggers:
             tagger_name = tagger.__name__
             tag_value = tagger.tag_value(container)
+            tags.append(tag_value)
             logger.info(
                 f"Applying tag tagger_name: {tagger_name} tag_value: {tag_value}"
             )
             docker["tag", image, f"{owner}/{short_image_name}:{tag_value}"]()
 
+        if tags:
+            env_name = f'{short_image_name.replace("-", "_")}_EXTRA_TAG_ARGS'
+            docker_build_tag_args = "-t " + " -t ".join(tags)
+            github_set_env(env_name, docker_build_tag_args)
+
 
 if __name__ == "__main__":
     logging.basicConfig(level=logging.INFO)