[Distributed] Propagate semantic group names to PyTorch process groups (#32900)

Co-authored-by: jipengtian <jipengtian@xiaohongshu.com> Co-authored-by: Claude <noreply@anthropic.com>
2026-08-06 20:38:12 -07:00
parent c54dc4582f
commit f9e6888b5a
2 changed files with 78 additions and 1 deletions
@@ -348,12 +348,14 @@ class GroupCoordinator:
                    backend="mooncake",
                    pg_options=dev_opts,
                    timeout=subgroup_timeout,
+                    group_desc=f"{group_name}:device",
                )
                cpu_group = torch.distributed.new_group(
                    ranks,
                    backend="mooncake-cpu",
                    pg_options=cpu_opts,
                    timeout=subgroup_timeout,
+                    group_desc=f"{group_name}:cpu",
                )
            else:
                active_ranks = torch.ones(
@@ -366,11 +368,15 @@ class GroupCoordinator:
                    backend=torch_distributed_backend,
                    pg_options=pg_options,
                    timeout=subgroup_timeout,
+                    group_desc=f"{group_name}:device",
                )
                # a group with `gloo` backend, to allow direct coordination
                # between processes through the CPU.
                cpu_group = torch.distributed.new_group(
-                    ranks, backend="gloo", timeout=gloo_timeout
+                    ranks,
+                    backend="gloo",
+                    timeout=gloo_timeout,
+                    group_desc=f"{group_name}:cpu",
                )
            if self.rank in ranks:
                self.ranks = ranks