BatchProcessor and PeriodicExportingMetricReader cause TypeError: 'NoneType' object is not callable on os.fork() after being garbage collected

[pranaysb]

Describe your environment

OS: macOS
Python version: Python 3.12, 3.13 (tested); root cause is a plain os.register_at_fork + weakref interaction, not version-specific
SDK version: 1.44.0
API version: 1.44.0

What happened?

If a BatchProcessor or PeriodicExportingMetricReader is instantiated and subsequently garbage collected, any future calls to os.fork() will print a TypeError: 'NoneType' object is not callable stack trace to sys.stderr.
This happens because the after_in_child callbacks registered with os.register_at_fork in their __init__ methods do not check if the weak reference has died before calling it.

Steps to Reproduce

import os
import sys
from opentelemetry.sdk._shared_internal import BatchProcessor
class DummyExporter:
    def export(self, batch): pass
    def shutdown(self): pass
class DummyMetrics:
    def register_queue_size(self, callback): pass
    def drop_items(self, count): pass
    def finish_items(self, count, error): pass
def reproduce():
    processor = BatchProcessor(
        exporter=DummyExporter(),
        schedule_delay_millis=5000,
        max_export_batch_size=512,
        export_timeout_millis=30000,
        max_queue_size=2048,
        exporting="Span",
        metrics=DummyMetrics()
    )
    # Shutdown allows the worker thread to exit and the processor to be GC'd
    processor.shutdown()
    del processor
    
    pid = os.fork()
    if pid == 0:
        os._exit(0)
    else:
        os.waitpid(pid, 0)
reproduce()

Expected Result

The script executes cleanly without any errors printed to stderr, because the processor object has already been garbage collected.

Actual Result

Exception ignored in: <function BatchProcessor.__init__.<locals>.<lambda> at 0x104f5ab60>
Traceback (most recent call last):
  File "/opentelemetry-sdk/src/opentelemetry/sdk/_shared_internal/__init__.py", line 119, in <lambda>
    os.register_at_fork(after_in_child=lambda: weak_reinit()())  
TypeError: 'NoneType' object is not callable

Additional context

The fix is to safely check if the weak reference is alive before calling it, similar to how it is properly handled in ConcurrentMultiSpanProcessor in opentelemetry/sdk/trace/__init__.py. This prevents log spam in multi-process applications (like celery workers or gunicorn) that dynamically create and destroy pipelines.

Would you like to implement a fix?

Yes

Tip

_{React with 👍 to help prioritize this issue. Please use comments to provide useful context, avoiding +1 or me too, to help us triage it. Learn more here.}

[saisharan0103]

This appears to hit the at-fork child callback path in opentelemetry-sdk/src/opentelemetry/sdk/_shared_internal/__init__.py, especially BatchProcessor.__init__, where the registered callback can outlive the processor through a weakref. The concrete failure mode is that after shutdown() and GC, a later os.fork() can try to call through a dead weak reference and print TypeError: 'NoneType' object is not callable to stderr. For a PR, I would start by guarding the weakref dereference before invoking the child reinit path, then mirror the same pattern for PeriodicExportingMetricReader.