fix: skip spare ranks gracefully in end-to-end test

Spare ranks would call sys.exit(0) during NTP initialization,
which pytest treats as a failure. Now spare ranks skip the test
gracefully before that happens.

Author: daiyaanarfeen

tests/unit_tests/distributed/test_nonuniform_tp.py

@@ -408,9 +408,18 @@ def test_ntp_end_to_end_with_8_gpus(self):
             non_active_ranks_per_dp={(0, 0, 0): [2, 3]},  # DP=0: GPUs 2,3 are spares
         )
 
+        # Check if this rank is a spare (will exit during initialization)
+        # Spare ranks: DP=0 with tp_rank=2,3
+        is_spare = dp_rank == 0 and tp_rank in [2, 3]
+
         # Reconfigure process groups for NTP
+        # Note: spare ranks will call sys.exit(0) in initialize_nonuniform_tp_process_groups
         from megatron.core.distributed.nonuniform_tp import initialize_nonuniform_tp_process_groups
 
+        if is_spare:
+            # For spare ranks in test, just mark as passed and exit gracefully
+            pytest.skip(f"Rank {rank} is a spare rank, skipping test gracefully")
+
         initialize_nonuniform_tp_process_groups(ddp_config)
 
         # After reconfiguration, check TP size