Production Deployment¶
This document describes various aspects of the production deployments of Apache Airflow.
Production Container Images¶
Customizing or extending the Production Image¶
Before you dive-deeply in the way how the Airflow Image is build, named and why we are doing it the way we do, you might want to know very quickly how you can extend or customize the existing image for Apache Airflow. This chapter gives you a short answer to those questions.
The docker image provided (as convenience binary package) in the Apache Airflow DockerHub is a bare image that has not many external dependencies and extras installed. Apache Airflow has many extras that can be installed alongside the “core” airflow image and they often require some additional dependencies. The Apache Airflow image provided as convenience package is optimized for size, so it provides just a bare minimal set of the extras and dependencies installed and in most cases you want to either extend or customize the image.
Airflow Summit 2020’s Production Docker Image talk provides more details about the context, architecture and customization/extension methods for the Production Image.
Extending the image¶
Extending the image is easiest if you just need to add some dependencies that do not require
compiling. The compilation framework of Linux (so called build-essential
) is pretty big, and
for the production images, size is really important factor to optimize for, so our Production Image
does not contain build-essential
. If you need compiler like gcc or g++ or make/cmake etc. - those
are not found in the image and it is recommended that you follow the “customize” route instead.
How to extend the image - it is something you are most likely familiar with - simply
build a new image using Dockerfile’s FROM
directive and add whatever you need. Then you can add your
Debian dependencies with apt
or PyPI dependencies with pip install
or any other stuff you need.
You should be aware, about a few things:
The production image of airflow uses “airflow” user, so if you want to add some of the tools as
root
user, you need to switch to it withUSER
directive of the Dockerfile. Also you should remember about following the best practises of Dockerfiles to make sure your image is lean and small.
FROM apache/airflow:1.10.15
USER root
RUN apt-get update \
&& apt-get install -y --no-install-recommends \
my-awesome-apt-dependency-to-add \
&& apt-get autoremove -yqq --purge \
&& apt-get clean \
&& rm -rf /var/lib/apt/lists/*
USER airflow
PyPI dependencies in Apache Airflow are installed in the user library, of the “airflow” user, so you need to install them with the
--user
flag and WITHOUT switching to airflow user. Note also that using –no-cache-dir is a good idea that can help to make your image smaller.
FROM apache/airflow:1.10.15
RUN pip install --no-cache-dir --user my-awesome-pip-dependency-to-add
If your apt, or PyPI dependencies require some of the build-essentials, then your best choice is to follow the “Customize the image” route. However it requires to checkout sources of Apache Airflow, so you might still want to choose to add build essentials to your image, even if your image will be significantly bigger.
FROM apache/airflow:1.10.15
USER root
RUN apt-get update \
&& apt-get install -y --no-install-recommends \
build-essential my-awesome-apt-dependency-to-add \
&& apt-get autoremove -yqq --purge \
&& apt-get clean \
&& rm -rf /var/lib/apt/lists/*
USER airflow
RUN pip install --no-cache-dir --user my-awesome-pip-dependency-to-add
You can also embed your dags in the image by simply adding them with COPY directive of Airflow. The DAGs in production image are in /opt/airflow/dags folder.
Customizing the image¶
Customizing the image is an alternative way of adding your own dependencies to the image - better suited to prepare optimized production images.
The advantage of this method is that it produces optimized image even if you need some compile-time dependencies that are not needed in the final image. You need to use Airflow Sources to build such images from the official distribution folder of Apache Airflow for the released versions, or checked out from the GitHub project if you happen to do it from git sources.
The easiest way to build the image image is to use breeze
script, but you can also build such customized
image by running appropriately crafted docker build in which you specify all the build-args
that you need to add to customize it. You can read about all the args and ways you can build the image
in the #production-image-build-arguments chapter below.
Here just a few examples are presented which should give you general understanding of what you can customize.
This builds the production image in version 3.7 with additional airflow extras from 1.10.15 PyPI package and additional apt dev and runtime dependencies.
docker build . \
--build-arg PYTHON_BASE_IMAGE="python:3.7-slim-buster" \
--build-arg PYTHON_MAJOR_MINOR_VERSION=3.7 \
--build-arg AIRFLOW_INSTALLATION_METHOD="apache-airflow" \
--build-arg AIRFLOW_VERSION="1.10.15" \
--build-arg AIRFLOW_INSTALL_VERSION="==1.10.15" \
--build-arg AIRFLOW_CONSTRAINTS_REFERENCE="constraints-1-10" \
--build-arg AIRFLOW_SOURCES_FROM="empty" \
--build-arg AIRFLOW_SOURCES_TO="/empty" \
--build-arg ADDITIONAL_AIRFLOW_EXTRAS="jdbc" \
--build-arg ADDITIONAL_PYTHON_DEPS="pandas" \
--build-arg ADDITIONAL_DEV_APT_DEPS="gcc g++" \
--build-arg ADDITIONAL_RUNTIME_APT_DEPS="default-jre-headless" \
--tag my-image
the same image can be built using breeze
(it supports auto-completion of the options):
./breeze build-image \
--production-image --python 3.7 --install-airflow-version=1.10.15 \
--additional-extras=jdbc --additional-python-deps="pandas" \
--additional-dev-apt-deps="gcc g++" --additional-runtime-apt-deps="default-jre-headless"
You can customize more aspects of the image - such as additional commands executed before apt dependencies are installed, or adding extra sources to install your dependencies from. You can see all the arguments described below but here is an example of rather complex command to customize the image based on example in this comment:
docker build . -f Dockerfile \
--build-arg PYTHON_BASE_IMAGE="python:3.7-slim-buster" \
--build-arg PYTHON_MAJOR_MINOR_VERSION=3.7 \
--build-arg AIRFLOW_INSTALLATION_METHOD="apache-airflow" \
--build-arg AIRFLOW_VERSION="1.10.15" \
--build-arg AIRFLOW_INSTALL_VERSION="==1.10.15" \
--build-arg AIRFLOW_CONSTRAINTS_REFERENCE="constraints-1-10" \
--build-arg AIRFLOW_SOURCES_FROM="empty" \
--build-arg AIRFLOW_SOURCES_TO="/empty" \
--build-arg ADDITIONAL_AIRFLOW_EXTRAS="slack" \
--build-arg ADDITIONAL_PYTHON_DEPS="apache-airflow-backport-providers-odbc \
apache-airflow-backport-providers-odbc \
azure-storage-blob \
sshtunnel \
google-api-python-client \
oauth2client \
beautifulsoup4 \
dateparser \
rocketchat_API \
typeform" \
--build-arg ADDITIONAL_DEV_APT_DEPS="msodbcsql17 unixodbc-dev g++" \
--build-arg ADDITIONAL_DEV_APT_COMMAND="curl https://packages.microsoft.com/keys/microsoft.asc | apt-key add --no-tty - && curl https://packages.microsoft.com/config/debian/10/prod.list > /etc/apt/sources.list.d/mssql-release.list" \
--build-arg ADDITIONAL_DEV_ENV_VARS="ACCEPT_EULA=Y" \
--build-arg ADDITIONAL_RUNTIME_APT_COMMAND="curl https://packages.microsoft.com/keys/microsoft.asc | apt-key add --no-tty - && curl https://packages.microsoft.com/config/debian/10/prod.list > /etc/apt/sources.list.d/mssql-release.list" \
--build-arg ADDITIONAL_RUNTIME_APT_DEPS="msodbcsql17 unixodbc git procps vim" \
--build-arg ADDITIONAL_RUNTIME_ENV_VARS="ACCEPT_EULA=Y" \
--tag my-image
Customizing images in high security restricted environments¶
You can also make sure your image is only build using local constraint file and locally downloaded wheel files. This is often useful in Enterprise environments where the binary files are verified and vetted by the security teams.
This builds below builds the production image in version 3.7 with packages and constraints used from the local
docker-context-files
rather than installed from PyPI or GitHub. It also disables MySQL client
installation as it is using external installation method.
Note that as a prerequisite - you need to have downloaded wheel files. In the example below we
first download such constraint file locally and then use pip download
to get the .whl files needed
but in most likely scenario, those wheel files should be copied from an internal repository of such .whl
files. Note that AIRFLOW_INSTALL_VERSION
is only there for reference, the apache airflow .whl file
in the right version is part of the .whl files downloaded.
Note that ‘pip download’ will only works on Linux host as some of the packages need to be compiled from
sources and you cannot install them providing --platform
switch. They also need to be downloaded using
the same python version as the target image.
The pip download
might happen in a separate environment. The files can be committed to a separate
binary repository and vetted/verified by the security team and used subsequently to build images
of Airflow when needed on an air-gaped system.
Preparing the constraint files and wheel files:
rm docker-context-files/*.whl docker-context-files/*.txt
curl -Lo "docker-context-files/constraints-1-10.txt" \
https://raw.githubusercontent.com/apache/airflow/constraints-1-10/constraints-3.7.txt
pip download --dest docker-context-files \
--constraint docker-context-files/constraints-1-10.txt \
apache-airflow[async,aws,azure,celery,dask,elasticsearch,gcp,kubernetes,mysql,postgres,redis,slack,ssh,statsd,virtualenv]==1.10.15
Building the image (after copying the files downloaded to the “docker-context-files” directory:
./breeze build-image \
--production-image --python 3.7 --install-airflow-version=1.10.15 \
--disable-mysql-client-installation --disable-pip-cache --install-from-local-files-when-building \
--constraints-location="/docker-context-files/constraints-1-10.txt"
or
docker build . \
--build-arg PYTHON_BASE_IMAGE="python:3.7-slim-buster" \
--build-arg PYTHON_MAJOR_MINOR_VERSION=3.7 \
--build-arg AIRFLOW_INSTALLATION_METHOD="apache-airflow" \
--build-arg AIRFLOW_VERSION="1.10.15" \
--build-arg AIRFLOW_INSTALL_VERSION="==1.10.15" \
--build-arg AIRFLOW_CONSTRAINTS_REFERENCE="constraints-1-10" \
--build-arg AIRFLOW_SOURCES_FROM="empty" \
--build-arg AIRFLOW_SOURCES_TO="/empty" \
--build-arg INSTALL_MYSQL_CLIENT="false" \
--build-arg AIRFLOW_PRE_CACHED_PIP_PACKAGES="false" \
--build-arg INSTALL_FROM_DOCKER_CONTEXT_FILES="true" \
--build-arg AIRFLOW_CONSTRAINTS_LOCATION="/docker-context-files/constraints-1-10.txt"
Customizing & extending the image together¶
You can combine both - customizing & extending the image. You can build the image first using
customize
method (either with docker command or with breeze
and then you can extend
the resulting image using FROM
any dependencies you want.
Customizing PYPI installation¶
You can customize PYPI sources used during image build by adding a docker-context-files/.pypirc file This .pypirc will never be committed to the repository and will not be present in the final production image. It is added and used only in the build segment of the image so it is never copied to the final image.
External sources for dependencies¶
In corporate environments, there is often the need to build your Container images using other than default sources of dependencies. The docker file uses standard sources (such as Debian apt repositories or PyPI repository. However, in corporate environments, the dependencies are often only possible to be installed from internal, vetted repositories that are reviewed and approved by the internal security teams. In those cases, you might need to use those different sources.
This is rather easy if you extend the image - you simply write your extension commands using the right sources - either by adding/replacing the sources in apt configuration or specifying the source repository in pip install command.
It’s a bit more involved in the case of customizing the image. We do not have yet (but we are working on it) a capability of changing the sources via build args. However, since the builds use Dockerfile that is a source file, you can rather easily simply modify the file manually and specify different sources to be used by either of the commands.
Comparing extending and customizing the image¶
Here is the comparison of the two types of building images.
Extending the image |
Customizing the image |
|
---|---|---|
Produces optimized image |
No |
Yes |
Use Airflow Dockerfile sources to build the image |
No |
Yes |
Requires Airflow sources |
No |
Yes |
You can build it with Breeze |
No |
Yes |
Allows to use non-default sources for dependencies |
Yes |
No [1] |
- [1] When you combine customizing and extending the image, you can use external sources
in the “extend” part. There are plans to add functionality to add external sources option to image customization. You can also modify Dockerfile manually if you want to use non-default sources for dependencies.
Using the production image¶
The PROD image entrypoint works as follows:
In case the user is not “airflow” (with undefined user id) and the group id of the user is set to 0 (root), then the user is dynamically added to /etc/passwd at entry using USER_NAME variable to define the user name. This is in order to accommodate the OpenShift Guidelines
The
AIRFLOW_HOME
is set by default to/opt/airflow/
- this means that DAGs are in default in the/opt/airflow/dags
folder and logs are in the/opt/airflow/logs
The working directory is
/opt/airflow
by default.If
AIRFLOW__CORE__SQL_ALCHEMY_CONN
variable is passed to the container and it is either mysql or postgres SQL alchemy connection, then the connection is checked and the script waits until the database is reachable. IfAIRFLOW__CORE__SQL_ALCHEMY_CONN_CMD
variable is passed to the container, it is evaluated as a command to execute and result of this evaluation is used asAIRFLOW__CORE__SQL_ALCHEMY_CONN
. The_CMD
variable takes precedence over theAIRFLOW__CORE__SQL_ALCHEMY_CONN
variable.If no
AIRFLOW__CORE__SQL_ALCHEMY_CONN
variable is set then SQLite database is created in ${AIRFLOW_HOME}/airflow.db and db reset is executed.If first argument equals to “bash” - you are dropped to a bash shell or you can executes bash command if you specify extra arguments. For example:
docker run -it apache/airflow:master-python3.6 bash -c "ls -la"
total 16
drwxr-xr-x 4 airflow root 4096 Jun 5 18:12 .
drwxr-xr-x 1 root root 4096 Jun 5 18:12 ..
drwxr-xr-x 2 airflow root 4096 Jun 5 18:12 dags
drwxr-xr-x 2 airflow root 4096 Jun 5 18:12 logs
If first argument is equal to “python” - you are dropped in python shell or python commands are executed if you pass extra parameters. For example:
> docker run -it apache/airflow:master-python3.6 python -c "print('test')"
test
If first argument equals to “airflow” - the rest of the arguments is treated as an airflow command to execute. Example:
docker run -it apache/airflow:master-python3.6 airflow webserver
If there are any other arguments - they are simply passed to the “airflow” command
> docker run -it apache/airflow:master-python3.6 version
2.0.0.dev0
If
AIRFLOW__CELERY__BROKER_URL
variable is passed and airflow command with scheduler, worker of flower command is used, then the script checks the broker connection and waits until the Celery broker database is reachable. IfAIRFLOW__CELERY__BROKER_URL_CMD
variable is passed to the container, it is evaluated as a command to execute and result of this evaluation is used asAIRFLOW__CELERY__BROKER_URL
. The_CMD
variable takes precedence over theAIRFLOW__CELERY__BROKER_URL
variable.
Production image build arguments¶
The following build arguments (--build-arg
in docker build command) can be used for production images:
Build argument |
Default value |
Description |
---|---|---|
|
|
Base python image. |
|
|
major/minor version of Python (should match base image). |
|
|
version of Airflow. |
|
|
the repository from which PIP dependencies are pre-installed. |
|
|
the branch from which PIP dependencies are pre-installed initially. |
|
If not empty, it will override the source of the constraints with the specified URL or file. Note that the file has to be in docker context so it’s best to place such file in one of the folders included in .dockerignore. |
|
|
|
Reference (branch or tag) from GitHub
where constraints file is taken from
It can be |
|
(see Dockerfile) |
Default extras with which airflow is installed. |
|
|
If set to true, Airflow is installed
from PyPI. if you want to install
Airflow from self-build package
you can set it to false, put package in
|
|
|
Allows to pre-cache airflow PIP packages from the GitHub of Apache Airflow This allows to optimize iterations for Image builds and speeds up CI builds. |
|
|
If set to true, Airflow, providers and
all dependencies are installed from
from locally built/downloaded
.whl and .tar.gz files placed in the
|
|
Optional additional extras with which airflow is installed. |
|
|
Optional python packages to extend the image with some extra dependencies. |
|
|
(see Dockerfile) |
Dev apt command executed before dev deps are installed in the Build image. |
|
Additional Dev apt command executed
before dev dep are installed
in the Build image. Should start with
|
|
|
(see Dockerfile) |
Dev APT dependencies installed in the Build image. |
|
Additional apt dev dependencies installed in the Build image. |
|
|
Additional env variables defined when installing dev deps. |
|
|
(see Dockerfile) |
Runtime apt command executed before deps are installed in the Main image. |
|
Additional Runtime apt command executed
before runtime dep are installed
in the Main image. Should start with
|
|
|
(see Dockerfile) |
Runtime APT dependencies installed in the Main image. |
|
Additional apt runtime dependencies installed in the Main image. |
|
|
Additional env variables defined when installing runtime deps. |
|
|
|
Airflow’s HOME (that’s where logs and sqlite databases are stored). |
|
|
Airflow user UID. |
|
|
Airflow group GID. Note that most files
created on behalf of airflow user belong
to the |
|
|
Home directory of the Airflow user. |
|
|
Number of processors to use for cassandra PIP install (speeds up installing in case cassandra extra is used). |
|
|
Whether MySQL client should be installed The mysql extra is removed from extras if the client is not installed. |
There are build arguments that determine the installation mechanism of Apache Airflow for the production image. There are three types of build:
From local sources (by default for example when you use
docker build .
)You can build the image from released PyPi airflow package (used to build the official Docker image)
You can build the image from any version in GitHub repository(this is used mostly for system testing).
Build argument |
Default |
What to specify |
---|---|---|
|
|
Should point to the installation method of Apache Airflow. It can be
|
|
Optional - might be used for package installation of different Airflow version for example”==1.10.15”. For consistency, you should also set``AIRFLOW_VERSION`` to the same value AIRFLOW_VERSION is embedded as label in the image created. |
|
|
|
Reference (branch or tag) from GitHub where constraints file is taken from.
It can be |
|
|
In case of Airflow 2.0 it should be “www”, in case of Airflow 1.10 series it should be “www_rbac”. |
|
|
Sources of Airflow. Set it to “.” when you install airflow from local sources. |
|
|
Target for Airflow sources. Set to “/opt/airflow” when you want to install airflow from local sources. |
This builds production image in version 3.6 with default extras from the local sources (master version of 2.0 currently):
docker build .
This builds the production image in version 3.7 with default extras from 1.10.15 tag and constraints taken from constraints-1-10-12 branch in GitHub.
docker build . \
--build-arg PYTHON_BASE_IMAGE="python:3.7-slim-buster" \
--build-arg PYTHON_MAJOR_MINOR_VERSION=3.7 \
--build-arg AIRFLOW_INSTALLATION_METHOD="https://github.com/apache/airflow/archive/1.10.15.tar.gz#egg=apache-airflow" \
--build-arg AIRFLOW_CONSTRAINTS_REFERENCE="constraints-1-10" \
--build-arg AIRFLOW_BRANCH="v1-10-test" \
--build-arg AIRFLOW_SOURCES_FROM="empty" \
--build-arg AIRFLOW_SOURCES_TO="/empty"
This builds the production image in version 3.7 with default extras from 1.10.15 PyPI package and constraints taken from 1.10.15 tag in GitHub and pre-installed pip dependencies from the top of v1-10-test branch.
docker build . \
--build-arg PYTHON_BASE_IMAGE="python:3.7-slim-buster" \
--build-arg PYTHON_MAJOR_MINOR_VERSION=3.7 \
--build-arg AIRFLOW_INSTALLATION_METHOD="apache-airflow" \
--build-arg AIRFLOW_VERSION="1.10.15" \
--build-arg AIRFLOW_INSTALL_VERSION="==1.10.15" \
--build-arg AIRFLOW_BRANCH="v1-10-test" \
--build-arg AIRFLOW_CONSTRAINTS_REFERENCE="constraints-1.10.15" \
--build-arg AIRFLOW_SOURCES_FROM="empty" \
--build-arg AIRFLOW_SOURCES_TO="/empty"
This builds the production image in version 3.7 with additional airflow extras from 1.10.15 PyPI package and additional python dependencies and pre-installed pip dependencies from 1.10.15 tagged constraints.
docker build . \
--build-arg PYTHON_BASE_IMAGE="python:3.7-slim-buster" \
--build-arg PYTHON_MAJOR_MINOR_VERSION=3.7 \
--build-arg AIRFLOW_INSTALLATION_METHOD="apache-airflow" \
--build-arg AIRFLOW_VERSION="1.10.15" \
--build-arg AIRFLOW_INSTALL_VERSION="==1.10.15" \
--build-arg AIRFLOW_BRANCH="v1-10-test" \
--build-arg AIRFLOW_CONSTRAINTS_REFERENCE="constraints-1.10.15" \
--build-arg AIRFLOW_SOURCES_FROM="empty" \
--build-arg AIRFLOW_SOURCES_TO="/empty" \
--build-arg ADDITIONAL_AIRFLOW_EXTRAS="mssql,hdfs" \
--build-arg ADDITIONAL_PYTHON_DEPS="sshtunnel oauth2client"
This builds the production image in version 3.7 with additional airflow extras from 1.10.15 PyPI package and additional apt dev and runtime dependencies.
docker build . \
--build-arg PYTHON_BASE_IMAGE="python:3.7-slim-buster" \
--build-arg PYTHON_MAJOR_MINOR_VERSION=3.7 \
--build-arg AIRFLOW_INSTALLATION_METHOD="apache-airflow" \
--build-arg AIRFLOW_VERSION="1.10.15" \
--build-arg AIRFLOW_INSTALL_VERSION="==1.10.15" \
--build-arg AIRFLOW_CONSTRAINTS_REFERENCE="constraints-1-10" \
--build-arg AIRFLOW_SOURCES_FROM="empty" \
--build-arg AIRFLOW_SOURCES_TO="/empty" \
--build-arg ADDITIONAL_AIRFLOW_EXTRAS="jdbc" \
--build-arg ADDITIONAL_DEV_APT_DEPS="gcc g++" \
--build-arg ADDITIONAL_RUNTIME_APT_DEPS="default-jre-headless"
More details about the images¶
You can read more details about the images - the context, their parameters and internal structure in the IMAGES.rst document.
Kerberos-authenticated workers¶
Apache Airflow has a built-in mechanism for authenticating the operation with a KDC (Key Distribution Center).
Airflow has a separate command airflow kerberos
that acts as token refresher. It uses the pre-configured
Kerberos Keytab to authenticate in the KDC to obtain a valid token, and then refreshing valid token
at regular intervals within the current token expiry window.
Each request for refresh uses a configured principal, and only keytab valid for the principal specified is capable of retrieving the authentication token.
The best practice to implement proper security mechanism in this case is to make sure that worker
workloads have no access to the Keytab but only have access to the periodically refreshed, temporary
authentication tokens. This can be achieved in docker environment by running the airflow kerberos
command and the worker command in separate containers - where only the airflow kerberos
token has
access to the Keytab file (preferably configured as secret resource). Those two containers should share
a volume where the temporary token should be written by the airflow kerberos
and read by the workers.
In the Kubernetes environment, this can be realized by the concept of side-car, where both Kerberos token refresher and worker are part of the same Pod. Only the Kerberos side-car has access to Keytab secret and both containers in the same Pod share the volume, where temporary token is written by the side-care container and read by the worker container.
This concept is implemented in the development version of the Helm Chart that is part of Airflow source code.