Document a way to paralellize multiarch builds

[yajo]

Behaviour

Recently I had this multiarch build that took 1:27h to report a build failure in one arch.

It seems GH runners have 2 CPU, which makes a simple multiarch build such as that one (using 5 platforms) awfully slow.

My guess (which might be wrong) is that it should be possible to use 1 runner per arch and merge the resulting images in a single multiarch one.

Steps to reproduce this issue

1. Create a multiarch build.

Expected behaviour

It shouldn't be extremely slow.

Actual behaviour

It is extremely slow.

Configuration

• Repository URL (if public): https://github.com/Tecnativa/ci-base
• Build URL (if public): https://github.com/Tecnativa/ci-base/runs/1735026697?check_suite_focus=true, building Tecnativa/ci-base@cfa2c6d.

name: ci

on:
  pull_request:
  push:
    branches:
      - main
    tags:
      - "*"
  workflow_dispatch:
    inputs:
      pytest_addopts:
        description:
          Extra options for pytest; use -vv for full details; see
          https://docs.pytest.org/en/latest/example/simple.html#how-to-change-command-line-options-defaults
        required: false

env:
  IMAGE_NAME: tecnativa/ci-base
  LANG: "en_US.utf-8"
  LC_ALL: "en_US.utf-8"
  PIP_CACHE_DIR: ${{ github.workspace }}/.cache.~/pip
  PIPX_HOME: ${{ github.workspace }}/.cache.~/pipx
  POETRY_CACHE_DIR: ${{ github.workspace }}/.cache.~/pypoetry
  POETRY_VIRTUALENVS_IN_PROJECT: "true"
  PYTEST_ADDOPTS: ${{ github.event.inputs.pytest_addopts }}
  PYTHONIOENCODING: "UTF-8"

jobs:
  build-push-test-deploy:
    strategy:
      matrix:
        python: [3.9]
    runs-on: ubuntu-latest
    services:
      registry:
        image: registry:2
        ports:
          - 5000:5000
    steps:
      - uses: actions/checkout@v2
      # Build and push
      - uses: actions/cache@v2
        with:
          path: |
            /tmp/.buildx-cache
          key:
            buildx|${{ secrets.CACHE_DATE }}|${{ runner.os }}|${{
            hashFiles('Dockerfile') }}
      - uses: docker/setup-qemu-action@v1
      - name: Set up QEMU
        uses: docker/setup-qemu-action@v1
      - name: Set up Docker Buildx
        id: buildx
        uses: docker/setup-buildx-action@v1
        with:
          driver-opts: network=host
          install: true
      - name: Docker meta for local images
        id: docker_meta_local
        uses: crazy-max/ghaction-docker-meta@v1
        with:
          images: localhost:5000/${{ env.IMAGE_NAME }}
          tag-edge: true
          tag-semver: |
            {{version}}
            {{major}}.{{minor}}
      - name: Build and push to local registry
        uses: docker/build-push-action@v2
        with:
          cache-from: type=local,src=/tmp/.buildx-cache
          cache-to: type=local,dest=/tmp/.buildx-cache,mode=max
          context: .
          file: ./Dockerfile
          labels: ${{ steps.docker_meta_local.outputs.labels }}
          push: true
          tags: ${{ steps.docker_meta_local.outputs.tags }}
          # Limit to platforms supported by base image
          platforms: |
            linux/amd64
            linux/arm/v7
            linux/arm64/v8
            linux/ppc64le
            linux/s390x
      # Test (only in local arch)
      - name: Install python
        uses: actions/setup-python@v2
        with:
          python-version: ${{ matrix.python }}
      - uses: actions/cache@v2
        with:
          path: |
            .cache.~
            .venv
            ~/.local/bin
          key:
            venv|${{ secrets.CACHE_DATE }}|${{ runner.os }}|${{ matrix.python }}|${{
            hashFiles('pyproject.toml', 'poetry.lock') }}
      - run: pip install poetry
      - name: Patch $PATH
        run: echo "$HOME/.local/bin" >> $GITHUB_PATH
      - run: poetry install
      - run: poetry run pytest --image localhost:5000/${{ env.IMAGE_NAME }}
      # Deploy (only outside of pull requests)
      - name: Docker meta for public images
        id: docker_meta_public
        if: github.event_name != 'pull_request'
        uses: crazy-max/ghaction-docker-meta@v1
        with:
          images: ghcr.io/${{ env.IMAGE_NAME }}
          tag-edge: true
          tag-semver: |
            {{version}}
            {{major}}.{{minor}}
      - name: Login to GitHub Container Registry
        uses: docker/login-action@v1
        if: github.event_name != 'pull_request'
        with:
          registry: ghcr.io
          username: ${{ secrets.BOT_LOGIN }}
          password: ${{ secrets.BOT_TOKEN }}
      - name: Push to public registry
        uses: docker/build-push-action@v2
        if: github.event_name != 'pull_request'
        with:
          cache-from: type=local,src=/tmp/.buildx-cache
          cache-to: type=local,dest=/tmp/.buildx-cache,mode=max
          context: .
          file: ./Dockerfile
          labels: ${{ steps.docker_meta_public.outputs.labels }}
          push: true
          tags: ${{ steps.docker_meta_public.outputs.tags }}
          # Limit to platforms supported by base image
          platforms: |
            linux/amd64
            linux/arm/v7
            linux/arm64/v8
            linux/ppc64le
            linux/s390x

Logs

logs_41.zip

---

@joao-p-marques @Tecnativa TT25794

[yajo]

Well, Tecnativa/ci-base#2 finally passed and built after 4:34h. There's gotta be a faster way... 🤔

[crazy-max]

@yajo

It seems GH runners have 2 CPU, which makes a simple multiarch build such as that one (using 5 platforms) awfully slow.

You could use a self-hosted runner but in the near future you will be able to configure CPU cores, RAM, disk space for the runner (see github/roadmap#161). You could also use multiple native nodes through self-hosted runners.

My guess (which might be wrong) is that it should be possible to use 1 runner per arch and merge the resulting images in a single multiarch one.

No need to do that, multi-platform build is already parallelized.

[pkit]

No need to do that, multi-platform build is already parallelized.

That's completely false for this action, it builds each image sequentially.