Azure / Azure/azure-functions-python-worker

[Python 3.14 proxy worker] Missing sys.modules["__main__"] breaks multiprocessing spawn

Open
#1,903 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
357
Forks
116
Avg merge
32m
Merged PRs (30d)
1

Description

### Description

The Azure Functions Python 3.14 proxy worker removes `__main__` from
`sys.modules` before invoking application code.

As a result, Python's standard `multiprocessing` spawn context fails during
`Process.start()`, before the child process or its target function begins
executing.

This is not specific to any third-party package. The minimal reproduction uses
only `azure-functions` and the Python standard library.

### Environment

- Azure Functions Python runtime: 3.14
- Python: 3.14.6
- Azure Functions Core Tools: 4.12.0
- Functions runtime: 4.1048.200.26180
- Programming model: Python v2
- `azure-functions` application package: 2.2.0
- Reproduced locally inside the Functions worker
- The same failure was observed after deploying the application to Azure Functions on Python 3.14

The same application code worked on the Python 3.12 Functions runtime.

### Minimal reproduction

```python
import multiprocessing
import sys

import azure.functions as func

app = func.FunctionApp()

def child_process() -> None:
return None

@app.route(route="multiprocessing-repro")
def multiprocessing_repro(req: func.HttpRequest) -> func.HttpResponse:
main_present = "__main__" in sys.modules

context = multiprocessing.get_context("spawn")
process = context.Process(target=child_process)
process.start()
process.join()

return func.HttpResponse(
f"__main__ present: {main_present}; exit code: {process.exitcode}"
)
```

Start the application using Core Tools with Python 3.14 and invoke
`/api/multiprocessing-repro`.

### Actual behavior

`"__main__" in sys.modules` is false during the function invocation, and
`Process.start()` raises:

```text
Traceback (most recent call last):
File ".../multiprocessing/process.py", line 121, in start
self._popen = self._Popen(self)
File ".../multiprocessing/context.py", line 289, in _Popen
return Popen(process_obj)
File ".../multiprocessing/popen_spawn_posix.py", line 32, in __init__
super().__init__(process_obj)
File ".../multiprocessing/popen_fork.py", line 19, in __init__
self._launch(process_obj)
File ".../multiprocessing/popen_spawn_posix.py", line 42, in _launch
prep_data = spawn.get_preparation_data(process_obj._name)
File ".../multiprocessing/spawn.py", line 164, in get_preparation_data
main_module = sys.modules["__main__"]
KeyError: "__main__"
```

The failure occurs before the child process starts.

### Expected behavior

The Functions worker should retain a valid `__main__` module, or otherwise
initialize the application environment so that standard-library
`multiprocessing` spawn works inside a function invocation.

### Investigation

The Python 3.14 worker uses `proxy_worker`. Its dependency cleanup appears to
remove modules from `sys.modules` based on their file location. It excludes
modules whose names begin with `proxy_worker`, but does not appear to exclude
`__main__`.

That appears to leave the invocation environment without the module required
by `multiprocessing.spawn.get_preparation_data()`.

Launching an explicit Python subprocess with an importable module works in the
same Functions-host process, confirming that the failure is specifically in
the `multiprocessing` spawn preparation path.

### Impact

Any application or dependency that uses the standard-library spawn context can
fail under the Python 3.14 Functions worker. In our application this prevented
all isolated document conversions from starting, regardless of document type.

### Related issue

Related historical issue: #1094 reported the same missing-`__main__` worker
condition on Python 3.9 through a different caller.

This report concerns the Python 3.14 proxy worker and reproduces through:

```python
multiprocessing.get_context("spawn").Process(...).start()
```

It is therefore related behavior, but not the same caller or worker
implementation.

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.