pedro-cf opened a new issue, #67714:
URL: https://github.com/apache/airflow/issues/67714
### Under which category would you file this issue?
Providers
### Apache Airflow version
3.2.2
### What happened and how to reproduce it?
### What happened
Graph View renders incorrectly when a `@task_group` returns task references
and **internal tasks inside the group** are wired directly to a **downstream
task outside the group**, in addition to a normal exit/join path.
Observed symptoms:
- TaskGroup layout is misaligned (e.g. vertical stacking instead of
left-to-right flow)
- Dependency edges cross/overlap the TaskGroup boundary in confusing ways
(wrap-around lines)
- Internal task dependencies inside the group may not render clearly
The DAG **executes correctly**. This is a **Graph View rendering/layout
issue only**.
### Minimal reproduction
Below is a minimal DAG. With only `entry` / `exit` wired across the group
boundary, Graph View renders correctly.
Adding these two lines breaks Graph View:
```python
group_result["a2"] >> final_task_result
group_result["a3"] >> final_task_result
```
Full DAG:
```python
from datetime import datetime
from airflow.sdk import dag, task, task_group
@dag(
dag_id="taskgroup_return_render_bug_demo",
start_date=datetime(2024, 1, 1),
schedule=None,
catchup=False,
)
def taskgroup_return_render_bug_demo():
from airflow.task.trigger_rule import TriggerRule
@task
def start():
return 1
@task(trigger_rule=TriggerRule.NONE_FAILED_MIN_ONE_SUCCESS)
def final_task():
pass
@task_group
def group_a():
@task
def task_a1():
pass
@task.branch
def branch_a():
return [
"group_a.task_a2",
"group_a.task_a3",
]
@task
def task_a2():
pass
@task
def task_a3():
pass
@task(trigger_rule=TriggerRule.NONE_FAILED_MIN_ONE_SUCCESS)
def group_done():
pass
a1 = task_a1()
branch = branch_a()
a2 = task_a2()
a3 = task_a3()
done = group_done()
a1 >> branch >> [a2, a3]
[a2, a3] >> done
return {
"entry": a1,
"branch": branch,
"a2": a2,
"a3": a3,
"exit": done,
}
start_task = start()
group_result = group_a()
final_task_result = final_task()
start_task >> group_result["entry"]
group_result["exit"] >> final_task_result
# Graph View breaks when these are present:
group_result["a2"] >> final_task_result
group_result["a3"] >> final_task_result
taskgroup_return_render_bug_demo()
```
### Steps to reproduce
1. Deploy the DAG above to Airflow 3.2.2
2. Open **Graph View** for `taskgroup_return_render_bug_demo`
3. Compare rendering:
- **Works:** comment out the two `a2` / `a3` → `final_task` lines
- **Breaks:** uncomment those two lines
### What you think should happen instead?
Graph View should render TaskGroups consistently whether or not internal
returned task references are wired to external downstream tasks.
Expected behavior:
- TaskGroup box layout should stay coherent ( sensible left-to-right /
top-to-bottom flow )
- Edges from internal tasks to external downstream tasks should route
cleanly across the group boundary, without wrap-around or overlapping lines
- Internal dependencies inside the TaskGroup should remain visible when the
group is expanded
- Rendering should match the actual dependency graph produced at parse time
Since execution is already correct, Graph View should reflect the same
structure operators see in code / Grid view, without requiring authors to avoid
legitimate dependency patterns.
### Operating System
Ubuntu 24.04.4 LTS
### Deployment
Docker-Compose
### Apache Airflow Provider(s)
_No response_
### Versions of Apache Airflow Providers
_No response_
### Official Helm Chart version
Not Applicable
### Kubernetes Version
_No response_
### Helm Chart configuration
_No response_
### Docker Image customizations
_No response_
### 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]
