During Airflow operation, variables or configurations are used that contain particularly sensitive information. This guide provides ways to protect this data.
The following are particularly protected:
Variables. See the Variables Concepts documentation for more information.
Connections. See the Connections Concepts documentation for more information.
Masking sensitive data¶
Airflow will by default mask Connection passwords and sensitive Variables and keys from a Connection's extra (JSON) field when they appear in Task logs, in the Variable and in the Rendered fields views of the UI.
It does this by looking for the specific value appearing anywhere in your output. This means that if you
have a connection with a password of
a, then every instance of the letter a in your logs will be replaced
To disable masking you can setting hide_sensitive_var_conn_fields to false.
The automatic masking is triggered by Connection or Variable access. This means that if you pass a sensitive value via XCom or any other side-channel it will not be masked when printed in the downstream task.
Sensitive field names¶
When masking is enabled, Airflow will always mask the password field of every Connection that is accessed by a task.
It will also mask the value of a Variable, or the field of a Connection's extra JSON blob if the name contains any words in ('password', 'secret', 'passwd', 'authorization', 'api_key', 'apikey', 'access_token'). This list can also be extended:
[core] sensitive_var_conn_names = comma,separated,sensitive,names
Adding your own masks¶
If you want to mask an additional secret that is already masked by one of the above methods, you can do it in
your DAG file or operator's
execute function using the
mask_secret function. For example:
@task def my_func(): from airflow.utils.log.secrets_masker import mask_secret mask_secret("custom_value") ...
class MyOperator(BaseOperator): def execute(self, context): from airflow.utils.log.secrets_masker import mask_secret mask_secret("custom_value") ...
The mask must be set before any log/output is produced to have any effect.