It is time to deploy your DAG in production. To do this, first, you need to make sure that the Airflow is itself production-ready. Let’s see what precautions you need to take.
Airflow comes with an
SQLite backend by default. This allows the user to run Airflow without any external
database. However, such a setup is meant to be used for testing purposes only; running the default setup
in production can lead to data loss in multiple scenarios. If you want to run production-grade Airflow,
make sure you configure the backend to be an external database
such as PostgreSQL or MySQL.
You can change the backend using the following config
[database] sql_alchemy_conn = my_conn_string
Once you have changed the backend, airflow needs to create all the tables required for operation.
Create an empty DB and give Airflow’s user permission to
Once that is done, you can run -
airflow db migrate
migrate keeps track of migrations already applied, so it’s safe to run as often as you need.
Prior to Airflow version 2.7.0,
airflow db upgrade was used to apply migrations,
however, it has been deprecated in favor of
airflow db migrate.
SequentialExecutor by default. However, by its
nature, the user is limited to executing at most one task at a time.
Sequential Executor also pauses
the scheduler when it runs a task, hence it is not recommended in a production setup. You should use the
LocalExecutor for a single machine.
For a multi-node setup, you should use the Kubernetes executor or
the Celery executor.
Once you have configured the executor, it is necessary to make sure that every node in the cluster contains the same configuration and DAGs. Airflow sends simple instructions such as “execute task X of DAG Y”, but does not send any DAG files or configuration. You can use a simple cronjob or any other mechanism to sync DAGs and configs across your nodes, e.g., checkout DAGs from git repo every 5 minutes on all nodes.
If you are using disposable nodes in your cluster, configure the log storage to be a distributed file system
(DFS) such as
GCS, or external services such as Stackdriver Logging, Elasticsearch or
Amazon CloudWatch. This way, the logs are available even after the node goes down or gets replaced.
See Logging for Tasks for configurations.
The logs only appear in your DFS after the task has finished. You can view the logs while the task is running in UI itself.
Airflow comes bundled with a default
airflow.cfg configuration file.
You should use environment variables for configurations that change across deployments
e.g. metadata DB, password, etc. You can accomplish this using the format
Some configurations such as the Airflow Backend connection URI can be derived from bash commands as well:
sql_alchemy_conn_cmd = bash_command_to_run
Airflow users occasionally report instances of the scheduler hanging without a trace, for example in these issues:
To mitigate these issues, make sure you have a health check set up that will detect when your scheduler has not heartbeat in a while.
Production Container Images¶
We provide a Docker Image (OCI) for Apache Airflow for use in a containerized environment. Consider using it to guarantee that software will always run the same no matter where it’s deployed.
Helm Chart for Kubernetes¶
Helm provides a simple mechanism to deploy software to a Kubernetes cluster. We maintain an official Helm chart for Airflow that helps you define, install, and upgrade deployment. The Helm Chart uses our official Docker image and Dockerfile that is also maintained and released by the community.
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
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 sidecar, where both Kerberos token refresher and worker are part of the same Pod. Only the Kerberos sidecar has access to Keytab secret and both containers in the same Pod share the volume, where temporary token is written by the sidecar container and read by the worker container.
This concept is implemented in the Helm Chart for Apache Airflow.
Secured Server and Service Access on Google Cloud¶
This section describes techniques and solutions for securely accessing servers and services when your Airflow environment is deployed on Google Cloud, or you connect to Google services, or you are connecting to the Google API.
IAM and Service Accounts¶
You should not rely on internal network segmentation or firewalling as our primary security mechanisms. To protect your organization’s data, every request you make should contain sender identity. In the case of Google Cloud, the identity is provided by the IAM and Service account. Each Compute Engine instance has an associated service account identity. It provides cryptographic credentials that your workload can use to prove its identity when making calls to Google APIs or third-party services. Each instance has access only to short-lived credentials. If you use Google-managed service account keys, then the private key is always held in escrow and is never directly accessible.
If you are using Kubernetes Engine, you can use Workload Identity to assign an identity to individual pods.
For more information about service accounts in the Airflow, see Google Cloud Connection
Impersonate Service Accounts¶
If you need access to other service accounts, you can impersonate other service accounts to exchange the token with the default identity to another service account. Thus, the account keys are still managed by Google and cannot be read by your workload.
It is not recommended to generate service account keys and store them in the metadata database or the secrets backend. Even with the use of the backend secret, the service account key is available for your workload.
Access to Compute Engine Instance¶
If you want to establish an SSH connection to the Compute Engine instance, you must have the network address
of this instance and credentials to access it. To simplify this task, you can use
ComputeEngineHook support authorization with
Google OS Login service. It is an extremely robust way to manage Linux access properly as it stores
short-lived ssh keys in the metadata service, offers PAM modules for access and sudo privilege checking
and offers the
nsswitch user lookup into the metadata service as well.
It also solves the discovery problem that arises as your infrastructure grows. You can use the instance name instead of the network address.
Access to Amazon Web Service¶
Thanks to the Web Identity Federation, you can exchange the Google Cloud Platform identity to the Amazon Web Service identity, which effectively means access to Amazon Web Service platform. For more information, see: Google Cloud to AWS authentication using Web Identity Federation