[Bug] Preventing side effects in distributed tests by destroying process group …

[drivanov]

in test_molecule_gpt_datase.

This PR addresses a test isolation issue by explicitly destroying the distributed process group at the end of the test_molecule_gpt_dataset. This is necessary because:

The test initializes a distributed environment (e.g., via torch.distributed backend implicitly or via PyG internals).

If the process group is not properly destroyed, subsequent tests may fail with errors like:

        group = group or _get_default_group()
        opts = AllgatherOptions()
        opts.asyncOp = async_op
>       work = group.allgather([tensor_list], [tensor], opts)
E       RuntimeError: No backend type associated with device type cpu

This issue was observed specifically on GB200 machines when running multiple CI tests in sequence.

For instance, this bug manifests when running:

pytest /opt/pyg/pytorch_geometric/test/datasets/test_molecule_gpt_dataset.py::test_molecule_gpt_dataset \
        /opt/pyg/pytorch_geometric/test/metrics/test_link_pred_metric.py

but does not appear when running the second test (test_link_pred_metric.py) independently or before the first.

This change ensures proper resource cleanup in CI and avoids test interference across modules that may share distributed state.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 85.58%. Comparing base (c211214) to head (d955001).
Report is 55 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #10314      +/-   ##
==========================================
- Coverage   86.11%   85.58%   -0.53%     
==========================================
  Files         496      498       +2     
  Lines       33655    34146     +491     
==========================================
+ Hits        28981    29223     +242     
- Misses       4674     4923     +249     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[xnuohz]

how about moving to test/conftest.py

[drivanov]

Done.
@xnuohz: Thanks, it was a good suggestion!