我的python应用程序的Dockerfile看起来像这样-

FROM python:3.10-alpine
RUN apk update && apk upgrade
RUN pip install -U pip poetry==1.1.13
WORKDIR /app
COPY . .
RUN poetry export --without-hashes --format=requirements.txt > requirements.txt
RUN pip install -r requirements.txt
EXPOSE 8000
ENTRYPOINT [ "python" ]
CMD ["main.py"]

多阶段Docker构建与诗歌和venv

不要禁用virtualenv创建功能。virtualenv在Docker构建中发挥了作用，因为它们提供了一种优雅的方式来利用多阶段构建。简而言之，构建阶段将所有内容安装到virtualenv中，而最后阶段只是将virtualenv复制到一个小映像中。

在复制代码之前，先使用诗歌导出并安装固定的需求。这将允许您使用Docker构建缓存，并且永远不会因为更改代码中的一行而重新安装依赖项。

不要使用诗意安装来安装代码，因为它将执行可编辑的安装。相反，使用诗歌构建来构建一个轮子，然后将其安装到您的virtualenv中。(感谢PEP 517，这整个过程也可以通过简单的pip install .来执行，但由于构建隔离，您最终将安装另一个Poetry副本。)

下面是一个Dockerfile安装Flask应用到Alpine镜像的例子，它依赖于Postgres。本例使用入口点脚本激活virtualenv。但一般来说，没有入口点脚本也没问题，因为您可以在CMD指令中简单地引用/venv/bin/ Python的Python二进制文件。

Dockerfile

FROM python:3.7.6-alpine3.11 as base

ENV PYTHONFAULTHANDLER=1 \
    PYTHONHASHSEED=random \
    PYTHONUNBUFFERED=1

WORKDIR /app

FROM base as builder

ENV PIP_DEFAULT_TIMEOUT=100 \
    PIP_DISABLE_PIP_VERSION_CHECK=1 \
    PIP_NO_CACHE_DIR=1 \
    POETRY_VERSION=1.0.5

RUN apk add --no-cache gcc libffi-dev musl-dev postgresql-dev
RUN pip install "poetry==$POETRY_VERSION"
RUN python -m venv /venv

COPY pyproject.toml poetry.lock ./
RUN poetry export -f requirements.txt | /venv/bin/pip install -r /dev/stdin

COPY . .
RUN poetry build && /venv/bin/pip install dist/*.whl

FROM base as final

RUN apk add --no-cache libffi libpq
COPY --from=builder /venv /venv
COPY docker-entrypoint.sh wsgi.py ./
CMD ["./docker-entrypoint.sh"]

docker-entrypoint.sh

#!/bin/sh

set -e

. /venv/bin/activate

while ! flask db upgrade
do
     echo "Retry..."
     sleep 1
done

exec gunicorn --bind 0.0.0.0:5000 --forwarded-allow-ips='*' wsgi:app

wsgi.py

import your_app

app = your_app.create_app()

博士TL;

我已经能够使用postgres为Django项目创建诗歌。在做了一些研究后，我最终得到了以下Dockerfile:

FROM python:slim

# Keeps Python from generating .pyc files in the container
ENV PYTHONDONTWRITEBYTECODE 1
# Turns off buffering for easier container logging
ENV PYTHONUNBUFFERED 1

# Install and setup poetry
RUN pip install -U pip \
    && apt-get update \
    && apt install -y curl netcat \
    && curl -sSL https://raw.githubusercontent.com/python-poetry/poetry/master/get-poetry.py | python -
ENV PATH="${PATH}:/root/.poetry/bin"

WORKDIR /usr/src/app
COPY . .
RUN poetry config virtualenvs.create false \
  && poetry install --no-interaction --no-ansi

# run entrypoint.sh
ENTRYPOINT ["/usr/src/app/entrypoint.sh"]

这是entrypoint.sh的内容:

#!/bin/sh

if [ "$DATABASE" = "postgres" ]
then
    echo "Waiting for postgres..."

    while ! nc -z $SQL_HOST $SQL_PORT; do
      sleep 0.1
    done

    echo "PostgreSQL started"
fi

python manage.py migrate

exec "$@"

详细解释

需要注意以下几点:

I have decide to use slim instead of alpine as tag for the python image because even though alpine images are supposed to reduce the size of Docker images and speed up the build, with Python, you can actually end up with a bit larger image and that takes a while to build (read this article for more info). Using this configuration builds containers faster than using the alpine image because I do not need to add some extra packages to install Python packages properly. I am installing poetry directly from the URL provided in the documentation. I am aware of the warnings provided by sobolevn. However, I consider that it is better in the long term to use the lates version of poetry by default than relying on an environment variable that I should update periodically. Updating the environment variable PATH is crucial. Otherwise, you will get an error saying that poetry was not found. Dependencies are installed directly in the python interpreter of the container. It does not create poetry to create a virtual environment before installing the dependencies.

如果你需要这个Dockerfile的alpine版本:

FROM python:alpine

# Keeps Python from generating .pyc files in the container
ENV PYTHONDONTWRITEBYTECODE 1
# Turns off buffering for easier container logging
ENV PYTHONUNBUFFERED 1

# Install dev dependencies
RUN apk update \
    && apk add curl postgresql-dev gcc python3-dev musl-dev openssl-dev libffi-dev

# Install poetry
RUN pip install -U pip \
    && curl -sSL https://raw.githubusercontent.com/python-poetry/poetry/master/get-poetry.py | python -
ENV PATH="${PATH}:/root/.poetry/bin"

WORKDIR /usr/src/app
COPY . .
RUN poetry config virtualenvs.create false \
  && poetry install --no-interaction --no-ansi

# run entrypoint.sh
ENTRYPOINT ["/usr/src/app/entrypoint.sh"]

注意，alpine版本需要一些依赖关系postgresql-dev gcc python3-dev musl-dev openssl-dev libffi-dev才能正常工作。

下面是一个剥离的例子，其中首先是一个带有依赖关系的层(仅在这些依赖关系发生变化时构建)，然后将一个带有完整源代码的层添加到图像中。将poetry设置为安装到全局站点包中会留下一个配置工件，该配置工件也可以被删除。

FROM python:alpine

WORKDIR /app

COPY poetry.lock pyproject.toml ./
RUN pip install --no-cache-dir --upgrade pip \
 && pip install --no-cache-dir poetry \
 \
 && poetry config settings.virtualenvs.create false \
 && poetry install --no-dev \
 \
 && pip uninstall --yes poetry \

COPY . ./

有两个项目，你可以看到如何正确地做到这一点，或者你可以使用这些项目来构建你自己的图像，因为它们只是基础图像:

https://github.com/max-pfeiffer/uvicorn-poetry https://github.com/max-pfeiffer/uvicorn-gunicorn-poetry

base image的Dockerfile: https://github.com/max-pfeiffer/uvicorn-poetry/blob/main/build/Dockerfile

ARG OFFICIAL_PYTHON_IMAGE
FROM ${OFFICIAL_PYTHON_IMAGE}

ENV PYTHONUNBUFFERED=1 \
    PYTHONDONTWRITEBYTECODE=1 \
    PIP_NO_CACHE_DIR=off \
    PIP_DISABLE_PIP_VERSION_CHECK=on \
    PIP_DEFAULT_TIMEOUT=100 \
    POETRY_VERSION=1.1.11 \
    POETRY_HOME="/opt/poetry" \
    POETRY_VIRTUALENVS_IN_PROJECT=true \
    PYTHONPATH=/application_root \
    VIRTUAL_ENVIRONMENT_PATH="/application_root/.venv"

ENV PATH="$POETRY_HOME/bin:$VIRTUAL_ENVIRONMENT_PATH/bin:$PATH"

# https://python-poetry.org/docs/#osx--linux--bashonwindows-install-instructions
RUN apt-get update \
    && apt-get install --no-install-recommends -y \
        build-essential \
        curl \
    && curl -sSL https://raw.githubusercontent.com/sdispater/poetry/master/get-poetry.py | python - \
    && apt-get purge --auto-remove -y \
      build-essential \
      curl

COPY ./scripts/start_uvicorn.sh /application_server/
RUN chmod +x /application_server/start_uvicorn.sh

COPY ./scripts/pytest_entrypoint.sh ./scripts/black_entrypoint.sh /entrypoints/
RUN chmod +x /entrypoints/pytest_entrypoint.sh
RUN chmod +x /entrypoints/black_entrypoint.sh

EXPOSE 80

CMD ["/application_server/start_uvicorn.sh"]

样例工程图片的Dockerfile: https://github.com/max-pfeiffer/uvicorn-poetry/blob/main/examples/fast_api_multistage_build/Dockerfile

ARG BASE_IMAGE_NAME_AND_TAG=pfeiffermax/uvicorn-poetry:1.0.1-python3.9.8-slim-bullseye
FROM ${BASE_IMAGE_NAME_AND_TAG} as base-image

WORKDIR /application_root

# install [tool.poetry.dependencies]
# this will install virtual environment into /.venv because of POETRY_VIRTUALENVS_IN_PROJECT=true
# see: https://python-poetry.org/docs/configuration/#virtualenvsin-project
COPY ./poetry.lock ./pyproject.toml /application_root/
RUN poetry install --no-interaction --no-root --no-dev

FROM base-image as test-base-image
ENV LOG_LEVEL="debug"

COPY --from=base-image $VIRTUAL_ENVIRONMENT_PATH $VIRTUAL_ENVIRONMENT_PATH

# install [tool.poetry.dev-dependencies]
RUN poetry install --no-interaction --no-root

COPY /app /application_root/app/
COPY /tests /application_root/tests/

# image for running pep8 checks
FROM test-base-image as black-test-image

ENTRYPOINT /entrypoints/black_entrypoint.sh $0 $@

CMD ["--target-version py39", "--check", " --line-length 80", "app"]

# image for running unit tests
FROM test-base-image as unit-test-image

ENTRYPOINT /entrypoints/pytest_entrypoint.sh $0 $@

# You need to use pytest-cov as pytest plugin. Makes life very simple.
# tests directory is configured in pyproject.toml
# https://github.com/pytest-dev/pytest-cov
CMD ["--cov=app", "--cov-report=xml:/test_coverage_reports/unit_tests_coverage.xml"]

FROM base-image as development-image
ENV RELOAD="true" \
    LOG_LEVEL="debug"

COPY --from=base-image $VIRTUAL_ENVIRONMENT_PATH $VIRTUAL_ENVIRONMENT_PATH

# install [tool.poetry.dev-dependencies]
RUN poetry install --no-interaction --no-root

COPY . /application_root/

FROM base-image as production-image

COPY --from=base-image $VIRTUAL_ENVIRONMENT_PATH $VIRTUAL_ENVIRONMENT_PATH

# This RUN statement fixes an issue while running the tests with GitHub Actions.
# Tests work reliable locally on my machine or running GitHub Actions using act.
# There is a bug with multistage builds in GitHub Actions which I can also reliable reproduce
# see: https://github.com/moby/moby/issues/37965
# Will also check if I can fix that annoying issue with some tweaks to docker build args
# see: https://gist.github.com/UrsaDK/f90c9632997a70cfe2a6df2797731ac8
RUN true

COPY /app /application_root/app/

这是最适合我的最小配置:

FROM python:3.7

ENV PIP_DISABLE_PIP_VERSION_CHECK=on

RUN pip install poetry

WORKDIR /app
COPY poetry.lock pyproject.toml /app/

RUN poetry config virtualenvs.create false
RUN poetry install --no-interaction

COPY . /app

注意，它不像@sobolevn的配置那样安全。

作为一个小问题，我将补充一下pyproject是否可以进行可编辑安装。Toml项目，可以删除一两行:

FROM python:3.7

ENV PIP_DISABLE_PIP_VERSION_CHECK=on

WORKDIR /app
COPY poetry.lock pyproject.toml /app/

RUN pip install -e .

COPY . /app