refactoring: reimplement Docker strategy (#3162)

* setup base images

* add cpu flavor

* use the same Dockerfile for cpu and gpu

* better naming, add docs

* add docker workflow

* add missing image input

* change cwd for bake

* also push api images

* try conditional tagging for releases

* revert testing code

* update docker readme

* document variable override

* use Python 3.10

* allow empty HAYSTACK_EXTRAS

* Apply suggestions from code review

Co-authored-by: Sara Zan <[email protected]>

* remove repo description step, can't make it work so far

* add docs to the last step as it's tricky

* manage tags for the newest images

* tests are passing, checking in the last bit

Co-authored-by: Sara Zan <[email protected]>

.github/workflows/docker_release.yml

@@ -0,0 +1,93 @@
+name: Release Docker images
+
+on:
+  workflow_dispatch:
+  push:
+    branches:
+      - main
+    tags:
+      - v*
+
+env:
+  DOCKER_REPO_NAME: deepset/haystack
+
+jobs:
+  build-and-push:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v3
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v2
+
+      - name: Login to DockerHub
+        uses: docker/login-action@v1
+        with:
+          username: ${{ secrets.DOCKER_HUB_USER }}
+          password: ${{ secrets.DOCKER_HUB_TOKEN }}
+
+      - name: Docker meta
+        id: meta
+        uses: docker/metadata-action@v4
+        with:
+          images: $DOCKER_REPO_NAME
+
+      - name: Build base images
+        uses: docker/bake-action@v2
+        env:
+          IMAGE_TAG_SUFFIX: ${{ steps.meta.outputs.version }}
+        with:
+          workdir: docker
+          targets: base
+          push: true
+
+      - name: Build api images
+        uses: docker/bake-action@v2
+        env:
+          IMAGE_TAG_SUFFIX: ${{ steps.meta.outputs.version }}
+          BASE_IMAGE_TAG_SUFFIX: ${{ steps.meta.outputs.version }}
+        with:
+          workdir: docker
+          targets: api
+          push: true
+
+      - name: Get latest version of Haystack
+        id: latest-version
+        uses: pozetroninc/github-action-get-latest-release@master
+        if: startsWith(github.ref, 'refs/tags/')
+        with:
+          repository: ${{ github.repository }}
+          excludes: prerelease, draft
+
+      - name: Compare current version with latest
+        uses: madhead/semver-utils@latest
+        id: version
+        if: startsWith(github.ref, 'refs/tags/')
+        with:
+          # Version being built
+          version: ${{  github.ref_name }}
+          # Compare to latest
+          compare-to: ${{ steps.latest-version.outputs.release }}
+
+      - name: Use latest
+        if: steps.version.outputs.comparison-result == '>'
+        run: |
+          echo ${{ steps.version.outputs.comparison-result }};
+          echo ${{ steps.latest-version.outputs.release }};
+
+      # This step should only run when we release a new minor, so
+      # that we can tag the most recent image without the version number.
+      # For example, if the previous step builds `deepset/haystack:cpu-1.8.0`,
+      # this builds `deepset/haystack:cpu`
+      - name: Build api images no version in tag
+        uses: docker/bake-action@v2
+        if: steps.version.outputs.comparison-result == '>'
+        env:
+          IMAGE_TAG_SUFFIX: ${{ steps.meta.outputs.version }}
+          BASE_IMAGE_TAG_SUFFIX: ${{ steps.meta.outputs.version }}
+        with:
+          workdir: docker
+          targets: api-latest
+          push: true

docker/Dockerfile.api

@@ -0,0 +1,13 @@
+ARG base_image_tag
+
+FROM deepset/haystack:${base_image_tag}
+
+# Create a folder for the /file-upload API endpoint with write permissions
+RUN mkdir -p /opt/file-upload && chmod 777 /opt/file-upload
+
+# Tell rest_api which folder to use for uploads
+ENV FILE_UPLOAD_PATH="/opt/file-upload"
+
+EXPOSE 8000
+
+CMD ["gunicorn", "rest_api.application:app",  "-b", "0.0.0.0", "-k", "uvicorn.workers.UvicornWorker", "--workers", "1", "--timeout", "180"]

docker/Dockerfile.base

@@ -0,0 +1,38 @@
+ARG build_image
+ARG base_immage
+
+FROM $build_image AS build-image
+
+ARG haystack_version
+ARG haystack_extras
+ARG torch_scatter
+
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    build-essential gcc git curl \
+    tesseract-ocr libtesseract-dev poppler-utils
+
+# Install PDF converter
+RUN curl -O https://dl.xpdfreader.com/xpdf-tools-linux-4.04.tar.gz && \
+    tar -xvf xpdf-tools-linux-4.04.tar.gz && \
+    cp xpdf-tools-linux-4.04/bin64/pdftotext /opt && \
+    rm -rf xpdf-tools-linux-4.04
+
+# Shallow clone Haystack repo, we'll install from the local sources
+RUN git clone --depth=1 --branch=${haystack_version} https://github.com/deepset-ai/haystack.git /opt/haystack
+WORKDIR /opt/haystack
+
+# Use a virtualenv we can copy over the next build stage
+RUN python -m venv --system-site-packages /opt/venv
+ENV PATH="/opt/venv/bin:$PATH"
+
+RUN pip install --upgrade pip && \
+    pip install --no-cache-dir .${haystack_extras} && \
+    pip install --no-cache-dir ./rest_api && \
+    pip install --no-cache-dir torch-scatter -f $torch_scatter
+
+FROM $base_immage AS final
+
+COPY --from=build-image /opt/venv /opt/venv
+COPY --from=build-image /opt/pdftotext /usr/local/bin
+
+ENV PATH="/opt/venv/bin:$PATH"

docker/README.md

@@ -0,0 +1,49 @@
+# Haystack Docker image
+
+Haystack is an end-to-end framework that enables you to build powerful and production-ready
+pipelines for different search use cases. The Docker image comes with a web service
+configured to serve Haystack's `rest_api` to ease pipelines' deployments in containerized
+environments.
+
+Start the Docker container binding the TCP port `8000` locally:
+```sh
+docker run -p 8000:8000 deepset/haystack
+```
+
+If you need the container to access other services available in the host:
+```sh
+docker run -p 8000:8000 --network="host" deepset/haystack
+```
+
+## Image variants
+
+The Docker image comes in two variants:
+- `haystack:cpu-<version>`: this image is smaller but doesn't support GPU
+- `haystack:gpu-<version>`: this image comes with the Cuda runtime and is capable of running on GPUs
+
+
+## Image development
+
+Images are built with BuildKit and we use `bake` to orchestrate the process.
+You can build a specific image by simply run:
+```sh
+docker buildx bake gpu
+```
+
+You can override any `variable` defined in the `docker-bake.hcl` file and build custom
+images, for example if you want to use a branch from the Haystack repo:
+```sh
+HAYSTACK_VERSION=mybranch_or_tag BASE_IMAGE_TAG_SUFFIX=latest docker buildx bake gpu --no-cache
+```
+
+# License
+
+View [license information](https://github.com/deepset-ai/haystack/blob/main/LICENSE) for
+the software contained in this image.
+
+As with all Docker images, these likely also contain other software which may be under
+other licenses (such as Bash, etc from the base distribution, along with any direct or
+indirect dependencies of the primary software being contained).
+
+As for any pre-built image usage, it is the image user's responsibility to ensure that any
+use of this image complies with any relevant licenses for all software contained within.

docker/docker-bake.hcl

@@ -0,0 +1,94 @@
+variable "HAYSTACK_VERSION" {
+  default = "main"
+}
+
+variable "GITHUB_REF" {
+  default = ""
+}
+
+variable "IMAGE_NAME" {
+  default = "deepset/haystack"
+}
+
+variable "IMAGE_TAG_SUFFIX" {
+  default = "local"
+}
+
+variable "BASE_IMAGE_TAG_SUFFIX" {
+  default = "local"
+}
+
+variable "HAYSTACK_EXTRAS" {
+  default = ""
+}
+
+group "base" {
+  targets = ["base", "base-gpu"]
+}
+
+group "api" {
+  targets = ["cpu", "gpu"]
+}
+
+group "api-latest" {
+  targets = ["cpu-latest", "gpu-latest"]
+}
+
+group "all" {
+  targets = ["base", "base-gpu", "cpu", "gpu"]
+}
+
+target "docker-metadata-action" {}
+
+target "base" {
+  dockerfile = "Dockerfile.base"
+  tags = ["${IMAGE_NAME}:base-${IMAGE_TAG_SUFFIX}"]
+  args = {
+    build_image = "python:3.10-slim"
+    base_immage = "python:3.10-slim"
+    haystack_version = "${HAYSTACK_VERSION}"
+    haystack_extras = notequal("",HAYSTACK_EXTRAS) ? "${HAYSTACK_EXTRAS}" : "[docstores,crawler,preprocessing,ocr,onnx,beir]"
+    torch_scatter = "https://data.pyg.org/whl/torch-1.12.0+cpu.html"
+  }
+}
+
+target "base-gpu" {
+  dockerfile = "Dockerfile.base"
+  tags = ["${IMAGE_NAME}:base-gpu-${IMAGE_TAG_SUFFIX}"]
+  args = {
+    build_image = "pytorch/pytorch:1.12.1-cuda11.3-cudnn8-runtime"
+    base_immage = "pytorch/pytorch:1.12.1-cuda11.3-cudnn8-runtime"
+    haystack_version = "${HAYSTACK_VERSION}"
+    haystack_extras = notequal("",HAYSTACK_EXTRAS) ? "${HAYSTACK_EXTRAS}" : "[docstores-gpu,crawler,preprocessing,ocr,onnx-gpu,beir]"
+    torch_scatter = "https://data.pyg.org/whl/torch-1.12.1%2Bcu113.html"
+  }
+}
+
+target "cpu" {
+  dockerfile = "Dockerfile.api"
+  tags = ["${IMAGE_NAME}:cpu-${IMAGE_TAG_SUFFIX}"]
+  args = {
+    base_image_tag = "base-${BASE_IMAGE_TAG_SUFFIX}"
+  }
+}
+
+target "cpu-latest" {
+  inherits = ["cpu"]
+  tags = ["${IMAGE_NAME}:cpu"]
+}
+
+target "gpu" {
+  dockerfile = "Dockerfile.api"
+  tags = ["${IMAGE_NAME}:gpu-${IMAGE_TAG_SUFFIX}"]
+  args = {
+    base_image_tag = "base-gpu-${BASE_IMAGE_TAG_SUFFIX}"
+  }
+  platforms = [
+    "linux/amd64"
+  ]
+}
+
+target "gpu-latest" {
+  inherits = ["gpu"]
+  tags = ["${IMAGE_NAME}:gpu"]
+}