Metrics declared at toplevel of program interfere with autoreload

Reproduce el problema con bar.py, el notebook de Jupyter usando IPython autoreload y prometheus_client.Gauge, y luego inspecciona el traceback de duplicate CollectorRegistry. Determina el comportamiento previsto al recargar una métrica de nivel superior; done debe significar que volver a ejecutar la celda del notebook después de editar bar.py se completa sin el error de series temporales duplicadas.

To reproduce:

1. Create a file bar.py with contents

from prometheus_client import Gauge

my_gauge = Gauge('my_gauge', 'my description')

def add(x, y):
    return x + y

2. Create a Jupyter notebook with the contents

%load_ext autoreload
%autoreload 2
from bar import add
# ---- cell break ----
add(2, 3)

and run it.
3. Edit bar.py in some way (e.g. add a product(x, y) function).
4. Re-run the add(2, 3) cell in the notebook

Expected behavior

Autoreload works fine; the cell runs without issue.

Actual behavior

After running add(2, 3) a second time, it prints this error to stderr:

[autoreload of bar failed: Traceback (most recent call last):
  File "/dev/shm/uid-21748-gid-32786/39d5b5d2-seed-nspid4026531836-ns-4026531841/lib/python3.10/site-packages/IPython/extensions/autoreload.py", line 245, in check
    superreload(m, reload, self.old_objects)
  File "/dev/shm/uid-21748-gid-32786/39d5b5d2-seed-nspid4026531836-ns-4026531841/lib/python3.10/site-packages/IPython/extensions/autoreload.py", line 394, in superreload
    module = reload(module)
  File "/dev/shm/uid-21748-gid-32786/39d5b5d2-seed-nspid4026531836-ns-4026531841/lib/python3.10/imp.py", line 315, in reload
    return importlib.reload(module)
  File "/dev/shm/uid-21748-gid-32786/39d5b5d2-seed-nspid4026531836-ns-4026531841/lib/python3.10/importlib/__init__.py", line 169, in reload
    _bootstrap._exec(spec, module)
  File "<frozen importlib._bootstrap>", line 619, in _exec
  File "<frozen importlib._bootstrap_external>", line 883, in exec_module
  File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
  File "/j/igm/user/kstaley/pub/Jupyter/play/bar.py", line 3, in <module>
    my_gauge = Gauge('my_gauge', 'my description')
  File "/dev/shm/uid-21748-gid-32786/39d5b5d2-seed-nspid4026531836-ns-4026531841/lib/python3.10/site-packages/prometheus_client/metrics.py", line 365, in __init__
    super().__init__(
  File "/dev/shm/uid-21748-gid-32786/39d5b5d2-seed-nspid4026531836-ns-4026531841/lib/python3.10/site-packages/prometheus_client/metrics.py", line 143, in __init__
    registry.register(self)
ValueError: Duplicated timeseries in CollectorRegistry: {'my_gauge'}
]

Notes

I think instantiating a metric should be idempotent. If you are passing the exact same metric name, docstring, labels, etc., you should get back the same metric object instance. This would imply having a factory function instead of directly exposing the metric classes. Maybe this behavior could be opt-in if enabling it by default is worrisome.