Repeated calls to patch_all() eventually seem to infinite loop

[palfrey]

In ddtrace 2.14.4, if you call ddtrace.patch_all() many times, nothing bad seems to happen. In 2.15.1, at some point, something seems to get stuck in an infinite loop. This occurred for us because part of the app setup in our FastAPI app did patch_all() and so pytest called it many times.

A workaround is something like the following

_PATCH_ALL_DONE = False

def setup_datadog() -> None:
    global _PATCH_ALL_DONE
    if _PATCH_ALL_DONE is True:
        return
    ddtrace.patch_all()
    _PATCH_ALL_DONE = True

[romainkomorndatadog]

@palfrey , are you using our pytest integration (eg: pytest --ddtrace) and using the --ddtrace-patch-all flag?

[palfrey]

@palfrey , are you using our pytest integration (eg: pytest --ddtrace) and using the --ddtrace-patch-all flag?

Nope. But having read https://ddtrace.readthedocs.io/en/stable/integrations.html#pytest I don't think that's what we want. We call patch_all in our app code and we want to test the code that does that, not call patch_all outside of our code.

[romainkomorndatadog]

@palfrey , thanks for the update, I think what you're doing makes sense.

@mabdinur , I don't think this is a CI app issue. It seems more like an issue with ddtrace.patch_all() not being idempotent.

[mabdinur]

@palfrey I wasn't able to reproduce this error using the following steps:

1. Install ddtrace==2.15.1, pytest==8.3.3, fastapi==0.115.5
2. Run the test below(ex: pytest test_file.py -k test_patch_all):

import ddtrace

def test_patch_all():
    for _ in range(10):
        ddtrace.patch_all()

        import fastapi

        app = fastapi.FastAPI()

Can you modify the test above to reproduce the error you are seeing in your testsuite?

[palfrey]

for _ in range(10):

Up to about 20 seems fine. At 21, 3.7s runtime; 22, 7.2s; 23, 13.72s, and I presume something in the range of doubling as that goes on.