Bug: `airflow standalone` UI not displaying DAG import errors when `SIGSEGV` happens

[tatiana]

Apache Airflow version

3.0.0

Python 3.10.16 & 3.12.8

If "Other Airflow 2 version" selected, which one?

No response

What happened?

The Airflow UI (when running airflow standalone) and airflow dags list-import-errors currently do not display the exception when a SIGSEGV happens.

What you think should happen instead?

I believe the Airflow UI (when running airflow standalone) and airflow dags list-import-errors should display the exception when SIGSEGV happens.

How to reproduce

This was initially observed in astronomer/astronomer-cosmos#1754

An agnostic way to reproduce it, though, is:

Create a example_dag.py in the DAGs folder with the following content:

import signal
import os

os.kill(os.getpid(), signal.SIGSEGV)

Run:

airflow dag-processor

Observe there is an error:

dag-processor | [2025-05-07T14:50:52.962+0100] {manager.py:748} INFO -
dag-processor | ================================================================================
dag-processor | DAG File Processing Stats
dag-processor | Bundle       File Path            PID    Current Duration      # DAGs    # Errors  Last Duration    Last Run At
dag-processor | -----------  -------------------  -----  ------------------  --------  ----------  ---------------  -------------------
dag-processor | dags-folder  example_dag.py                                    0           1  0.21s            2025-05-07T13:50:22

Run:

airflow standalone

Open the Airflow UI via web browser:
http://localhost:8080/dags

And don't see any errors:

Running:

airflow dags list-import-errors

Gives:

No data found

Operating System

MacOS 15.4.1

Versions of Apache Airflow Providers

N/A

Deployment

Other

Deployment details

Local standalone

Anything else?

No response

Are you willing to submit PR?

• Yes I am willing to submit a PR!

Code of Conduct

• I agree to follow this project's Code of Conduct

[ephraimbuddy]

This issue is caused by a segmentation fault disrupting the DAG file loading process. When it occurs, the process exits immediately.

Since segmentation faults cannot be caught with try...except, one viable solution would be to load DAG file modules in a subprocess. However, I'm thinking this might increase the time it takes to process each file.
Wdyt, @jedcunningham?

[ephraimbuddy]

Reproduction:

Run airflow dag reserialize --bundle-name dags-folder with the below file in dags folder:

import signal
from airflow import DAG
import os
from airflow.decorators import task

os.kill(os.getpid(), signal.SIGSEGV)

with DAG('testbug'):
    @task
    def mytask():
        print(1)
    mytask()