`Cannot create DagRun for DAG because the dag is not serialized` when using `Dag.test`

[multimeric]

Apache Airflow version

3.1.6

If "Other Airflow 3 version" selected, which one?

No response

What happened?

Traceback (most recent call last):
  File "/Users/milton.m/Programming/AirflowDagTest/main.py", line 27, in <module>
    my_dag_instance.test()
  File "/Users/milton.m/Programming/AirflowDagTest/.venv/lib/python3.12/site-packages/airflow/sdk/definitions/dag.py", line 1245, in test
    dr: DagRun = get_or_create_dagrun(
                 ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/milton.m/Programming/AirflowDagTest/.venv/lib/python3.12/site-packages/airflow/models/dagrun.py", line 2175, in get_or_create_dagrun
    dr = dag.create_dagrun(
         ^^^^^^^^^^^^^^^^^^
  File "/Users/milton.m/Programming/AirflowDagTest/.venv/lib/python3.12/site-packages/airflow/utils/session.py", line 98, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/milton.m/Programming/AirflowDagTest/.venv/lib/python3.12/site-packages/airflow/serialization/serialized_objects.py", line 3342, in create_dagrun
    orm_dagrun = _create_orm_dagrun(
                 ^^^^^^^^^^^^^^^^^^^
  File "/Users/milton.m/Programming/AirflowDagTest/.venv/lib/python3.12/site-packages/airflow/utils/session.py", line 98, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/Users/milton.m/Programming/AirflowDagTest/.venv/lib/python3.12/site-packages/airflow/serialization/serialized_objects.py", line 2379, in _create_orm_dagrun
    raise AirflowException(f"Cannot create DagRun for DAG {dag.dag_id} because the dag is not serialized")
airflow.exceptions.AirflowException: Cannot create DagRun for DAG my_dag because the dag is not serialized

What you think should happen instead?

Dag.test should appropriately serialize the DAG to prevent this issue. I can actually see some logic for doing this already in the code, but it doesn't seem to work:

airflow/task-sdk/src/airflow/sdk/definitions/dag.py

Line 1233 in d6a009b

scheduler_dag = SerializedDAG.deserialize_dag(SerializedDAG.serialize_dag(self))

How to reproduce

Here's the minimal example I can come up with:

import tempfile
import os
import pendulum

with tempfile.NamedTemporaryFile(suffix=".sqlite3", delete=False) as tmpfile:
    conn_str = f"sqlite:///{tmpfile.name}"
    os.environ["AIRFLOW__DATABASE__SQL_ALCHEMY_CONN"] = conn_str
    from airflow.utils.db import upgradedb

    upgradedb()

    from airflow.sdk import task, dag

    @dag(start_date=pendulum.datetime(2024, 1, 1))
    def my_dag():
        @task
        def hello_task():
            print("Hello, Airflow!")

        hello_task()

    my_dag_instance = my_dag()

    my_dag_instance.test()

Operating System

MacOS 15.6

Versions of Apache Airflow Providers

None are relevant

Deployment

Virtualenv installation

Deployment details

A simple pyproject.toml with:

[project]
name = "airflowdagtest"
version = "0.1.0"
description = "Add your description here"
readme = "README.md"
requires-python = ">=3.12"
dependencies = [
    "apache-airflow>=3.1.6",

Anything else?

The Dag.test code seems to be changed in the current main branch, but installing commit 1e2e29d doesn't solve the issue.

This does work fine in airflow 3.0.6, and is broken in 3.1.2, so this bug was introduced relatively recently. This also makes me think that it wasn't deliberate.

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

[jayaprajapatii]

Hii,
I traced the issue to Dag.test passing a non-serialized DAG into create_dagrun.

Before proceeding with a fix, I want to confirm the intended behavior:
Should Dag.test always serialize the DAG before invoking 'create_dagrun' or is there a preferred place earlier/later in the call chain where this should be enforced ?

[multimeric]

Hi @jayaprajapatii, would you mind pointing out this part of the code or even just opening a draft PR so that I can see how it might be fixed? This is just so I can create a workaround to this bug before it gets fixed here.

[multimeric]

Here's a test script that demonstrates the behaviour:

Details

import os
import tempfile

with tempfile.NamedTemporaryFile(suffix=".sqlite3", delete=False) as tmpfile:
    os.environ["AIRFLOW__DATABASE__SQL_ALCHEMY_CONN"] = f"sqlite:///{tmpfile.name}"

from airflow.utils.db import upgradedb

upgradedb()

from airflow.sdk import dag, task


@dag
def demo_dag():
    @task
    def return_number() -> int:
        return 2


demo_dag().test()

I tested this with 3.0.6 and it was not present, then with 3.1.0 and it was, so it must have been 3.1.0 that introduced this issue.

Then I git bisected the repo to find when the bug was introduced. It was 7ea7fcd ie #50825

[BernardoGBS1]

I'm working on this and seem to have found the issue. I'll submit a fix within the week.

[multimeric]

Hi @ephraimbuddy, could you explain the motivation for #50825? Was it intended to prevent in-memory DAGs from being tested?

[flamafre]

HI all, really glad to see that this is being fixed. I have just upgraded to Airflow version 3.1.8, and I have the same issue.

[ephraimbuddy]

Hi @ephraimbuddy, could you explain the motivation for #50825? Was it intended to prevent in-memory DAGs from being tested?

It was done because of dag versioning. We want to ensure that the serialized dag exist because it's used in scheduler and several other places. Without serialized dag, we won't know the version to create the dagrun for.

[multimeric]

Without serialized dag, we won't know the version to create the dagrun for

How can we test unserialized DAGs then? Can we just add a default version to them?

[laurenhall1526-netizen]

Thanks for the context, that helps. I get the need for serialized DAGs with versioning, but this does break simple dag.test() workflows. Maybe we could fall back to a default version when the DAG isn’t serialized (just for testing), instead of failing? Curious if that would work without impacting scheduler behavior.

…

On Mon, Mar 30, 2026, 6:05 PM Michael ***@***.***> wrote: *multimeric* left a comment (apache/airflow#60860) <#60860 (comment)> Without serialized dag, we won't know the version to create the dagrun for How can we test unserialized DAGs then? Can we just add a default version to them? — Reply to this email directly, view it on GitHub <#60860?email_source=notifications&email_token=B7IHA46ZXVLKX6XLJGN34B34TL4UXA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIMJVHA3DSMJYGQYKM4TFMFZW63VKON2WE43DOJUWEZLEUVSXMZLOOSWGM33PORSXEX3DNRUWG2Y#issuecomment-4158691840>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/B7IHA45OWWLZBNBKTIIAUKT4TL4UXAVCNFSM6AAAAACSLVPBWKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DCNJYGY4TCOBUGA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>