[C++][Packaging] TST: arrow-flight-test started hanging on windows

[h-vetinari]

Describe the bug, including details regarding any error messages, version, and platform.

In conda-forge, I'm seeing hangs in all windows jobs for arrow v22+; at first I thought this was some flake, but they've been quite consistently reproducible (though also not 100%? 🤔). Interestingly, v21 does not seem affected. I don't really have more logs than that the test just never finishes, and eventually gets killed by the timeout (note the difference in timestamps).

2026-08-09T22:51:16.4468360Z  80/105 Test  #80: arrow-flight-internals-test ..................   Passed    0.20 sec
2026-08-09T22:51:16.4473910Z         Start  81: arrow-flight-test
2026-08-10T03:54:39.4204986Z Traceback (most recent call last):
[...]
2026-08-10T03:54:39.4326015Z   File "C:\Miniforge\Lib\site-packages\conda_build\utils.py", line 330, in _execute
2026-08-10T03:54:39.4327097Z     time.sleep(time_int)
2026-08-10T03:54:39.4327826Z KeyboardInterrupt

Component(s)

C++, Packaging

[VedantRalekar]

Hi @h-vetinari, Thanks for reporting this. I’m interested in investigating this Windows/Flight test hang.

Could you provide a bit more information to help narrow down the regression?

• Which exact Arrow versions are affected?
• What Windows version/environment is being used in the conda-forge jobs?
• Is there a CI job/log link where the hang can be reproduced?
• Does arrow-flight-test also hang when run locally on Windows?
• Do you know the last working Arrow 21 build and the first affected Arrow 22 build?

The CI logs suggest that arrow-flight-test starts successfully and then hangs until the job timeout, so having the full job details would be especially helpful.

Thanks!

[raulcd]

@h-vetinari that's interesting, on 22.0.0 I did some refactoring in order to support Dictionary replacement moving from our custom logic to reuse our IPC logic so it is potentially related to those. Adding here the issues related to flight for 22.0.0:

• https://github.com/apache/arrow/issues?q=is%3Aissue%20state%3Aclosed%20label%3A%22Component%3A%20FlightRPC%22%20milestone%3A22.0.0

[tadeja]

I've checked verbose repeated logs added by @raulcd on conda-forge/arrow-cpp-feedstock#2054
https://github.com/conda-forge/arrow-cpp-feedstock/actions/runs/31808481642/job/94793059241#step:7:5850

81: [  PASSED  ] 94 tests.
...
    Test #81: arrow-flight-test ................***Timeout 120.02 sec

and it somewhat resembles the rare abseil-cpp sync primitives issue on Windows worked around with #50085 (comment) (debugged via process dump) except that reproduced under parallel Ctest... Is there a chance to retrieve process dump from conda-forge ?