javi-cortes opened a new issue, #57743:
URL: https://github.com/apache/airflow/issues/57743
### Apache Airflow version
3.1.1
### If "Other Airflow 2/3 version" selected, which one?
_No response_
### What happened?
Hi!
I’m using Airflow 3.1.0 / 3.1.1 with the KubernetesExecutor, and I’m
overriding the pod via executor_config in default_args.
This works perfectly when applied at task level, but when I set it at DAG
level (via default_args), the UI (FastAPI) returns a 500 Internal Server Error
because the response serializer can’t handle the V1Pod object.
Error from api-server logs:
`
INFO: 10.101.11.64:44642 - "GET
/api/v2/dags/pip_freeze_dag_override/details HTTP/1.1" 500 Internal Server Error
ERROR: Exception in ASGI application
Traceback (most recent call last):
[...]
File
"/home/airflow/.local/lib/python3.12/site-packages/fastapi/routing.py", line
188, in serialize_response
return field.serialize(
File
"/home/airflow/.local/lib/python3.12/site-packages/fastapi/_compat.py", line
152, in serialize
return self._type_adapter.dump_python(
File
"/home/airflow/.local/lib/python3.12/site-packages/pydantic/type_adapter.py",
line 572, in dump_python
return self.serializer.to_python(
pydantic_core._pydantic_core.PydanticSerializationError: Unable to serialize
unknown type: <class 'kubernetes.client.models.v1_pod.V1Pod'>
`
Task runs fine when pod_override is passed directly to the operator.
UI / API call fails (/api/v2/dags/{dag_id}/details) because V1Pod is not
JSON-serializable by Pydantic.
Scheduler logs show a proper V1Pod object being used if the task is
triggered manually.
The serialization failure happens when the UI (or API) calls the
/api/v2/dags/{dag_id}/details endpoint, which is implemented as:
`
@dags_router.get("/{dag_id}/details")
def get_dag_details(...):
dag = get_latest_version_of_dag(dag_bag, dag_id, session)
dag_model = session.get(DagModel, dag_id)
for key, value in dag.__dict__.items():
if not key.startswith("_") and not hasattr(dag_model, key):
setattr(dag_model, key, value)
return dag_model
`
The returned dag_model is then serialized into a DAGDetailsResponse Pydantic
schema.
When the DAG contains executor_config={"pod_override": <V1Pod>}, the
serializer hits a PydanticSerializationError because V1Pod isn’t
JSON-serializable.
If you sanitize it into a dict, the scheduler later fails, since
PodGenerator.from_obj() rejects non-V1Pod objects.
PodGenerator only deserializes YAML templates (via
deserialize_model_file()), so there’s currently no code path that handles this
correctly for dicts returned by the API.
### What you think should happen instead?
The API response should serialize cleanly even when
executor_config["pod_override"] contains a V1Pod.
Alternatively, Airflow could auto-sanitize or exclude V1Pod fields when
building the DAG details response.
### How to reproduce
Minimal Reproducible Example:
`from airflow import DAG
from airflow.operators.bash import BashOperator
from datetime import datetime
from kubernetes import client as k8s
default_args = {
"executor_config": {
"pod_override": k8s.V1Pod(
spec=k8s.V1PodSpec(
containers=[
k8s.V1Container(
name="base",
image="busybox",
image_pull_policy="IfNotPresent",
)
],
restart_policy="Never",
),
)
}
}
with DAG(
dag_id="example_pod_override_dag",
start_date=datetime(2024, 1, 1),
schedule=None,
catchup=False,
default_args=default_args, # ← DAG-level
) as dag:
BashOperator(
task_id="pip_freeze",
bash_command="python -m pip --version && python -m pip freeze",
# inherits executor_config from default_args
)
`
### Operating System
Linux, Ubuntu 24.04
### Versions of Apache Airflow Providers
3.1.0
### Deployment
Official Apache Airflow Helm Chart
### Deployment details
k8s official helm chart
### Anything else?
_No response_
### Are you willing to submit PR?
- [ ] Yes I am willing to submit a PR!
### Code of Conduct
- [x] I agree to follow this project's [Code of
Conduct](https://github.com/apache/airflow/blob/main/CODE_OF_CONDUCT.md)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
