Skip to content

Commit

Permalink
Merge pull request #2189 from recommenders-team/staging
Browse files Browse the repository at this point in the history
Staging to main: Merge multiple Dockerfiles into a single one
  • Loading branch information
miguelgfierro authored Nov 15, 2024
2 parents 397e80a + f2abb55 commit 65fae79
Show file tree
Hide file tree
Showing 7 changed files with 360 additions and 350 deletions.
54 changes: 29 additions & 25 deletions .devcontainer/devcontainer.json
Original file line number Diff line number Diff line change
@@ -1,22 +1,18 @@
{
"name": "Recommenders",
// Version list: https://github.com/devcontainers/images/tree/main/src/base-ubuntu
// Includes: curl, wget, ca-certificates, git, Oh My Zsh!,
"image": "mcr.microsoft.com/devcontainers/base:ubuntu-24.04",
"hostRequirements": {
"cpus": 4,
"memory": "16gb",
"storage": "32gb"
},
"features": {
// https://github.com/devcontainers/features/blob/main/src/anaconda/devcontainer-feature.json
"ghcr.io/devcontainers/features/anaconda:1": {
"version": "2024.06-1"
"build": {
"dockerfile": "../tools/docker/Dockerfile",
"context": "..",
"target": "deps",
"args": {
"COMPUTE": "cpu",
"PYTHON_VERSION": "3.11"
}
},
"customizations": {
"vscode": {
// Set *default* container specific settings.json values on container create.
// Set default container specific settings.json values on container
// create
"settings": {
"[python]": {
"editor.defaultFormatter": "ms-python.black-formatter",
Expand All @@ -27,24 +23,32 @@
},
"isort.args": ["--profile", "black"],
"python.analysis.autoImportCompletions": true,
"python.defaultInterpreterPath": "/usr/local/conda/envs/Recommenders/bin/python",
// Conda env name *must* align with the one in Dockerfle
"python.defaultInterpreterPath": "/root/conda/envs/Recommenders/bin/python",
"python.testing.pytestEnabled": true,
// set the directory where all tests are
// Test directory
"python.testing.pytestArgs": ["tests"]
},
// Add the IDs of extensions you want installed when the container is created.
// VS Code extensions to install on container create
"extensions": [
"ms-python.black-formatter", // https://marketplace.visualstudio.com/items?itemName=ms-python.black-formatter
"ms-python.isort", // https://marketplace.visualstudio.com/items?itemName=ms-python.isort
"ms-python.mypy-type-checker", // https://marketplace.visualstudio.com/items?itemName=ms-python.mypy-type-checker
"ms-python.pylint", // https://marketplace.visualstudio.com/items?itemName=ms-python.pylint
"ms-python.python", // https://marketplace.visualstudio.com/items?itemName=ms-python.python
"ms-toolsai.datawrangler", // https://marketplace.visualstudio.com/items?itemName=ms-toolsai.datawrangler
"ms-toolsai.jupyter" // https://marketplace.visualstudio.com/items?itemName=ms-toolsai.jupyter
// https://marketplace.visualstudio.com/items?itemName=ms-python.black-formatter
"ms-python.black-formatter",
// https://marketplace.visualstudio.com/items?itemName=ms-python.isort
"ms-python.isort",
// https://marketplace.visualstudio.com/items?itemName=ms-python.mypy-type-checker
"ms-python.mypy-type-checker",
// https://marketplace.visualstudio.com/items?itemName=ms-python.pylint
"ms-python.pylint",
// https://marketplace.visualstudio.com/items?itemName=ms-python.python
"ms-python.python",
// https://marketplace.visualstudio.com/items?itemName=ms-toolsai.datawrangler
"ms-toolsai.datawrangler",
// https://marketplace.visualstudio.com/items?itemName=ms-toolsai.jupyter
"ms-toolsai.jupyter"
]
}
},

// Use 'postCreateCommand' to run commands after the container is created.
"postCreateCommand": "conda create -n Recommenders -c conda-forge -y python=3.10 openjdk=21 pip && conda init bash && bash -c -i 'conda activate Recommenders && pip install -e .[dev,spark]' && conda config --set auto_activate_base false"
// Install Recommenders in development mode after container create
"postCreateCommand": "bash -i -c 'conda activate Recommenders && conda install -c conda-forge -y openjdk=21 && pip install -e .[dev,spark]'"
}
85 changes: 85 additions & 0 deletions SETUP.md
Original file line number Diff line number Diff line change
Expand Up @@ -145,6 +145,91 @@ git checkout staging
pip install -e .[all]
```

We also provide a [devcontainer.json](./.devcontainer/devcontainer.json)
and [Dockerfile](./tools/docker/Dockerfile) for developers to
facilitate the development on
[Dev Containers with VS Code](https://code.visualstudio.com/docs/devcontainers/containers)
and [GitHub Codespaces](https://github.com/features/codespaces).

<details>
<summary><strong><em>VS Code Dev Containers</em></strong></summary>

The typical scenario using Docker containers for development is as
follows. Say, we want to develop applications for a specific
environment, so
1. we create a contaienr with the dependencies required,
1. and mount the folder containing the code to the container,
1. then code parsing, debugging and testing are all performed against
the container.
This workflow seperates the development environment from your local
environment, so that your local environment won't be affected. The
container used here for this end is called Dev Container in the
VS Code Dev Containers extension. And the extension eases this
development workflow with Docker containers automatically without
pain.

To use VS Code Dev Containers, your local machine must have the
following applicatioins installed:
* [Docker](https://docs.docker.com/get-started/get-docker/)
* [VS Code Remote Development Extension Pack](https://marketplace.visualstudio.com/items?itemName=ms-vscode-remote.vscode-remote-extensionpack)

Then
* When you open your local Recommenders folder in VS Code, it will
detect [devcontainer.json](./.devcontainer/devcontainer.json), and
prompt you to **Reopen in Container**. If you'd like to reopen,
it will create a container with the required environment described
in `devcontainer.json`, install a VS Code server in the container,
and mount the folder into the container.
+ If you don't see the prompt, you can use the command
**Dev Containers: Reopen in Container**
* If you don't have a local clone of Recommenders, you can also use
the command **Dev Containers: Clone Repository in Container Volume**,
and type in a branch/PR URL of Recommenders you'd like to develop
on, such as https://github.com/recommenders-team/recommenders,
https://github.com/recommenders-team/recommenders/tree/staging, or
https://github.com/recommenders-team/recommenders/pull/2098. VS
Code will create a container with the environment described in
`devcontainer.json`, and clone the specified branch of Recommenders
into the container.

Once everything is set up, VS Code will act as a client to the server
in the container, and all subsequent operations on VS Code will be
performed against the container.

</details>

<details>
<summary><strong><em>GitHub Codespaces</em></strong></summary>

GitHub Codespaces also uses `devcontainer.json` and Dockerfile in the
repo to create the environment on a VM for you to develop on the Web
VS Code. To use the GitHub Codespaces on Recommenders, you can go to
[Recommenders](https://github.com/recommenders-team/recommenders)
$\to$ switch to the branch of interest $\to$ Code $\to$ Codespaces
$\to$ Create codespaces on the branch.

</details>

<details>
<summary><strong><em>devcontainer.json & Dockerfile</em></strong></summary>

[devcontainer.json](./.devcontainer/devcontainer.json) describes:
* the Dockerfile to use with configurable build arguments, such as
`COMPUTE` and `PYTHON_VERSION`.
* settings on VS Code server, such as Python interpreter path in the
container, Python formatter.
* extensions on VS Code server, such as black-formatter, pylint.
* how to create the Conda environment for Recommenders in
`postCreateCommand`

[Dockerfile](./tools/docker/Dockerfile) is used in 3 places:
* Dev containers on VS Code and GitHub Codespaces
* [Testing workflows on AzureML](./tests/README.md)
* [Jupyter notebook examples on Docker](./tools/docker/README.md)

</details>


## Test Environments

Depending on the type of recommender system and the notebook that needs to be run, there are different computational requirements.
Expand Down
23 changes: 20 additions & 3 deletions tests/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -63,9 +63,26 @@ GitHub workflows `azureml-unit-tests.yml`, `azureml-cpu-nightly.yml`, `azureml-g

There are three scripts used with each workflow, all of them are located in [ci/azureml_tests](./ci/azureml_tests/):

* `submit_groupwise_azureml_pytest.py`: this script uses parameters in the workflow yml to set up the AzureML environment for testing using the AzureML SDK.
* `run_groupwise_pytest.py`: this script uses pytest to run the tests of the libraries and notebooks. This script runs in an AzureML workspace with the environment created by the script above.
* `test_groups.py`: this script defines the groups of tests. If the tests are part of the unit tests, the total compute time of each group should be less than 15min. If the tests are part of the nightly builds, the total time of each group should be less than 35min.
* [`submit_groupwise_azureml_pytest.py`](./ci/azureml_tests/submit_groupwise_azureml_pytest.py):
this script uses parameters in the workflow yml to set up the
AzureML environment for testing using the AzureML SDK.
* [`run_groupwise_pytest.py`](./ci/azureml_tests/run_groupwise_pytest.pyy):
this script uses pytest to run the tests of the libraries and
notebooks. This script runs in an AzureML workspace with the
environment created by the script above.
* [`aml_utils.py`](./ci/azureml_tests/aml_utils.py): this script
defines several utility functions using
[the AzureML Python SDK v2](https://learn.microsoft.com/en-us/azure/machine-learning/concept-v2?view=azureml-api-2).
These functions are used by the scripts above to set up the compute and
the environment for the tests on AzureML. For example, the
environment with all dependencies of Recommenders is created by the
function `get_or_create_environment` via the [Dockerfile](../tools/docker/Dockerfile).
More details on Docker support can be found at [tools/docker/README.md](../tools/docker/README.md).
* [`test_groups.py`](./ci/azureml_tests/test_groups.py): this script
defines the groups of tests. If the tests are part of the unit
tests, the total compute time of each group should be less than
15min. If the tests are part of the nightly builds, the total time
of each group should be less than 35min.

## How to contribute tests to the repository

Expand Down
110 changes: 34 additions & 76 deletions tests/ci/azureml_tests/aml_utils.py
Original file line number Diff line number Diff line change
Expand Up @@ -8,14 +8,15 @@
* https://learn.microsoft.com/en-us/azure/machine-learning/reference-migrate-sdk-v1-mlflow-tracking?view=azureml-api-2&tabs=aml%2Ccli%2Cmlflow
"""
import pathlib
import tempfile
import re

from azure.ai.ml import MLClient, command
from azure.ai.ml.entities import AmlCompute, BuildContext, Environment, Workspace
from azure.ai.ml.exceptions import JobException
from azure.core.exceptions import ResourceExistsError
from azure.identity import DefaultAzureCredential


def get_client(subscription_id, resource_group, workspace_name):
"""
Get the client with specified AzureML workspace, or create one if not existing.
Expand Down Expand Up @@ -61,9 +62,8 @@ def get_or_create_environment(
environment_name,
use_gpu,
use_spark,
conda_pkg_jdk,
conda_openjdk_version,
python_version,
commit_sha,
):
"""
AzureML requires the run environment to be setup prior to submission.
Expand All @@ -77,81 +77,39 @@ def get_or_create_environment(
added to the conda environment, else False
use_spark (bool): True if PySpark packages should be
added to the conda environment, else False
conda_pkg_jdk (str): "openjdk=8" by default
python_version (str): python version, such as "3.9"
commit_sha (str): the commit that triggers the workflow
conda_openjdk_version (str): "21" by default
python_version (str): python version, such as "3.11"
"""
conda_env_name = "reco"
conda_env_yml = "environment.yml"
condafile = fr"""
name: {conda_env_name}
channels:
- conda-forge
dependencies:
- python={python_version}
- {conda_pkg_jdk}
- pip
- pip:
- recommenders[dev{",gpu" if use_gpu else ""}{",spark" if use_spark else ""}]@git+https://github.com/recommenders-team/recommenders.git@{commit_sha}
"""
# See https://github.com/Azure/AzureML-Containers/blob/master/base/cpu/openmpi4.1.0-ubuntu22.04
image = "mcr.microsoft.com/azureml/openmpi4.1.0-ubuntu22.04"
# See https://github.com/Azure/AzureML-Containers/blob/master/base/gpu/openmpi4.1.0-cuda11.8-cudnn8-ubuntu22.04
dockerfile = fr"""# syntax=docker/dockerfile:1
FROM nvcr.io/nvidia/cuda:12.5.1-devel-ubuntu22.04
SHELL ["/bin/bash", "-c"]
USER root:root
ENV NVIDIA_VISIBLE_DEVICES all
ENV NVIDIA_DRIVER_CAPABILITIES compute,utility
ENV LANG=C.UTF-8 LC_ALL=C.UTF-8
ENV DEBIAN_FRONTEND noninteractive
RUN apt-get update && \
apt-get install -y wget git-all && \
apt-get clean -y && \
rm -rf /var/lib/apt/lists/*
# Install Conda
ENV CONDA_PREFIX /opt/miniconda
RUN wget -qO /tmp/miniconda.sh https://repo.anaconda.com/miniconda/Miniconda3-py311_24.5.0-0-Linux-x86_64.sh && \
bash /tmp/miniconda.sh -bf -p ${{CONDA_PREFIX}} && \
${{CONDA_PREFIX}}/bin/conda update --all -c conda-forge -y && \
${{CONDA_PREFIX}}/bin/conda clean -ay && \
rm -rf ${{CONDA_PREFIX}}/pkgs && \
rm /tmp/miniconda.sh && \
find / -type d -name __pycache__ | xargs rm -rf
# Create Conda environment
COPY {conda_env_yml} /tmp/{conda_env_yml}
RUN ${{CONDA_PREFIX}}/bin/conda env create -f /tmp/{conda_env_yml}
# Activate Conda environment
ENV CONDA_DEFAULT_ENV {conda_env_name}
ENV CONDA_PREFIX ${{CONDA_PREFIX}}/envs/${{CONDA_DEFAULT_ENV}}
ENV PATH="${{CONDA_PREFIX}}/bin:${{PATH}}" LD_LIBRARY_PATH="${{CONDA_PREFIX}}/lib:$LD_LIBRARY_PATH"
"""

with tempfile.TemporaryDirectory() as tmpdir:
tmpdir = pathlib.Path(tmpdir)
dockerfile_path = tmpdir / "Dockerfile"
condafile_path = tmpdir / conda_env_yml
build = BuildContext(path=tmpdir, dockerfile_path=dockerfile_path.name)

with open(dockerfile_path, "w") as file:
file.write(dockerfile)
with open(condafile_path, "w") as file:
file.write(condafile)

try:
client.environments.create_or_update(
Environment(
name=environment_name,
image=None if use_gpu else image,
build=build if use_gpu else None,
conda_file=None if use_gpu else condafile_path,
)
compute = "gpu" if use_gpu else "cpu"
extras = (
"[dev" + (",gpu" if use_gpu else "") + (",spark" if use_spark else "") + "]"
)
dockerfile = pathlib.Path("tools/docker/Dockerfile")

# Docker's --build-args is not supported by AzureML Python SDK v2 as shown
# in [the issue #33902](https://github.com/Azure/azure-sdk-for-python/issues/33902)
# so the build args are configured by regex substituion
text = dockerfile.read_text()
text = re.sub(r"(ARG\sCOMPUTE=).*", rf'\1"{compute}"', text)
text = re.sub(r"(ARG\sEXTRAS=).*", rf'\1"{extras}"', text)
text = re.sub(r"(ARG\sGIT_REF=).*", r'\1""', text)
text = re.sub(r"(ARG\sJDK_VERSION=).*", rf'\1"{conda_openjdk_version}"', text)
text = re.sub(r"(ARG\sPYTHON_VERSION=).*", rf'\1"{python_version}"', text)
dockerfile.write_text(text)

try:
client.environments.create_or_update(
Environment(
name=environment_name,
build=BuildContext(
# Set path for Docker to access to Recommenders root
path=".",
dockerfile_path=dockerfile,
),
)
except ResourceExistsError:
pass
)
except ResourceExistsError:
pass


def run_tests(
Expand Down
Loading

0 comments on commit 65fae79

Please sign in to comment.