potiuk commented on a change in pull request #4543: [AIRFLOW-3718] Multi-layered version of the docker image URL: https://github.com/apache/airflow/pull/4543#discussion_r248747685
########## File path: Dockerfile ########## @@ -16,26 +16,83 @@ FROM python:3.6-slim -COPY . /opt/airflow/ +SHELL ["/bin/bash", "-c"] + +# Make sure noninteractie debian install is used +ENV DEBIAN_FRONTEND=noninteractive + +# Increase the value to force renstalling of all apt-get dependencies +ENV FORCE_REINSTALL_APT_GET_DEPENDENCIES=1 + +# Install core build dependencies +RUN apt-get update \ + && apt-get install -y --no-install-recommends \ + libkrb5-dev libsasl2-dev libssl-dev libffi-dev libpq-dev git \ + && apt-get clean + + # Install useful utilities and other airflow required dependencies +RUN apt-get update \ + && apt-get install -y --no-install-recommends \ + libsasl2-dev freetds-bin build-essential default-libmysqlclient-dev apt-utils \ + curl rsync netcat locales \ + && apt-get clean ARG AIRFLOW_HOME=/usr/local/airflow -ARG AIRFLOW_DEPS="all" -ARG PYTHON_DEPS="" -ARG buildDeps="freetds-dev libkrb5-dev libsasl2-dev libssl-dev libffi-dev libpq-dev git" -ARG APT_DEPS="$buildDeps libsasl2-dev freetds-bin build-essential default-libmysqlclient-dev apt-utils curl rsync netcat locales" +RUN mkdir -p $AIRFLOW_HOME + +# Airflow extras to be installed +ARG AIRFLOW_EXTRAS="all" + +# Increase the value here to force reinstalling Apache Airflow pip dependencies +ENV FORCE_REINSTALL_ALL_PIP_DEPENDENCIES=1 + +# Speeds up building the image - cassandra driver without CYTHON saves around 10 minutes +# of build on typical machine +ARG CASS_DRIVER_NO_CYTHON_ARG="" + +# Build cassandra driver on multiple CPUs +ENV CASS_DRIVER_BUILD_CONCURRENCY=8 + +# Speeds up the installation of cassandra driver +ENV CASS_DRIVER_NO_CYTHON=${CASS_DRIVER_NO_CYTHON_ARG} + +## Airflow requires this variable be set on installation to avoid a GPL dependency. +ENV SLUGIFY_USES_TEXT_UNIDECODE yes + +# Airflow sources change frequently but dependency onfiguration won't change that often +# We copy setup.py and other files needed to perform setup of dependencies +# This way cache here will only be invalidated if any of the +# version/setup configuration change but not when airflow sources change Review comment: The dependencies should be resolved properly and minor updates should be installed as needed. It works as follows: 1) The first pip install will install the dependencies as they are at the moment the dockerfile is generated for the first time (line 74). 2) Whenever any of the sources of airflow change, the docker image at line 77 gets invalidated and docker is rebuilt starting from that line. Then we upgrade apt-get dependencies to the latest ones (line 80) and then we run `pip install` Again (without using the old cache) to see if the pip transient dependencies will resolve to different versions of related packages. In most cases, this will be no-op but in case some transient dependencies will cause different resolution of them - they will get installed at that time (line 89) 3) Whenever any of the setup.*, README, version.py, airflow script change (all files related to setup) we go even back a step - the Docker is invalidated at one of the lines 66-70 and it will continue from there - which means that completely fresh 'pip install' is run from scratch. 4) We can always force installation from the scratch by increasing value of ENV FORCE_REINSTALL_APT_GET_DEPENDENCIES (currently = 1). If we manually increase it to 2 and commit such Dockerfile to docker it will invalidate docker images at line 25 and the all the installation is run from the scratch. 5) The only thing I am not 100% sure (but about 99%) is what happens when new version of python3.6-slim gets released. I believe it will trigger the build from scratch, but that's something that we will have to confirm with DockerHub (It might depend on how they are doing caching as this part might be done differently). It's not a big problem even if this is not the case because at point 3) we do `apt-get upgrade` and at this time python version will get upgraded as well if it is released. So this is more of an optimisation question than a problem. So eventually - whenever you make any change to airflow sources, you are sure that both 'apt-get upgrade' and 'pip install' have been called. Whenever you make change to setup.py - you are sure that 'pip install' is called from scratch , whenever you force it you can do apt-get install from the scratch. I thin this fairly robust and expected behaviour. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services