[For discussion][DeviceMesh] Use a shared_state to cache pg per layout, root_mesh and rank_map #166010
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…h and rank_map [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/166010
Note: Links to docs will display an error until the docs builds have been completed. ❌ 5 New FailuresAs of commit 54abae7 with merge base a5f3035 ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
the
oncall: distributed
…t, root_mesh and rank_map" cc H-Huang awgu wanchaol fegin wz337 wconstab d4l3k pragupta msaroufim dcci [ghstack-poisoned]
fduwjj
changed the title
…g per layout, root_mesh and rank_map" We want to create a shared_state to store root_mesh, rank_map and pg caches. We can add more into it down the road, so that it becomes a singleton for bookkeeping and also align with our original proposal to move toward the idea of mesh universe. cc H-Huang awgu wanchaol fegin wz337 wconstab d4l3k pragupta msaroufim dcci [ghstack-poisoned]
wconstab
reviewed
Oct 27, 2025
wconstab
reviewed
Oct 27, 2025
…g per layout, root_mesh and rank_map" We want to create a shared_state to store root_mesh, rank_map and pg caches. We can add more into it down the road, so that it becomes a singleton for bookkeeping and also align with our original proposal to move toward the idea of mesh universe. cc H-Huang awgu wanchaol fegin wz337 wconstab d4l3k pragupta msaroufim dcci [ghstack-poisoned]
…g per layout, root_mesh and rank_map" We want to create a shared_state to store root_mesh, rank_map and pg caches. We can add more into it down the road, so that it becomes a singleton for bookkeeping and also align with our original proposal to move toward the idea of mesh universe. cc H-Huang awgu wanchaol fegin wz337 wconstab d4l3k pragupta msaroufim dcci [ghstack-poisoned]
lw
reviewed
Oct 28, 2025
Contributor
lw
left a comment
lw
left a comment
There was a problem hiding this comment.
Thanks!
I think this PR is a great starting point for the discussion around PG caching, and defining a clear mental model for how different DeviceMesh should share state.
torch/distributed/device_mesh.py
Outdated
Comment on lines
530
to
531
| def _get_root_mesh(self) -> "DeviceMesh": | ||
| return self._root_mesh if self._root_mesh else self | ||
| return not_none(self._shared_state.get_root_mesh()) |
Contributor
There was a problem hiding this comment.
This is the only instance where we're accessing the root mesh of the shared state. In principle, once we introduce the concept of shared state, we can get rid of the concept of root meshes.
What is preventing us from simply removing this _get_root_mesh method?
If there's any internal usage of _get_root_mesh, could we find ways of codemod it away so that we can fully remove this method?
Contributor
Author
There was a problem hiding this comment.
We might need more PRs to remove _get_root_mesh not in this PR.
…g per layout, root_mesh and rank_map" We want to create a shared_state to store root_mesh, rank_map and pg caches. We can add more into it down the road, so that it becomes a singleton for bookkeeping and also align with our original proposal to move toward the idea of mesh universe. cc H-Huang awgu wanchaol fegin wz337 wconstab d4l3k pragupta msaroufim dcci [ghstack-poisoned]
…g per layout, root_mesh and rank_map" We want to create a shared_state to store root_mesh, rank_map and pg caches. We can add more into it down the road, so that it becomes a singleton for bookkeeping and also align with our original proposal to move toward the idea of mesh universe. cc H-Huang awgu wanchaol fegin wz337 wconstab d4l3k pragupta msaroufim dcci [ghstack-poisoned]
…g per layout, root_mesh and rank_map" We want to create a shared_state to store root_mesh, rank_map and pg caches. We can add more into it down the road, so that it becomes a singleton for bookkeeping and also align with our original proposal to move toward the idea of mesh universe. cc H-Huang awgu wanchaol fegin wz337 wconstab d4l3k pragupta msaroufim dcci [ghstack-poisoned]
This was referenced Nov 14, 2025
…g per layout, root_mesh and rank_map" We want to create a shared_state to store root_mesh, rank_map and pg caches. We can add more into it down the road, so that it becomes a singleton for bookkeeping and also align with our original proposal to move toward the idea of mesh universe. cc H-Huang awgu wanchaol fegin wz337 wconstab d4l3k pragupta msaroufim dcci [ghstack-poisoned]
…g per layout, root_mesh and rank_map" We want to create a shared_state to store root_mesh, rank_map and pg caches. We can add more into it down the road, so that it becomes a singleton for bookkeeping and also align with our original proposal to move toward the idea of mesh universe. cc H-Huang awgu wanchaol fegin wz337 wconstab d4l3k pragupta msaroufim dcci [ghstack-poisoned]
kwen2501
reviewed
Nov 22, 2025
Comment on lines
+1065
to
+1066
| non_ep_mesh = global_mesh._unflatten(0, (2, 2, 2), ("dp", "cp", "tp")) | ||
| ep_mesh = global_mesh._unflatten(0, (2, 2, 2), ("dp", "ep", "ep_tp")) |
Collaborator
There was a problem hiding this comment.
What's the difference between these two lines vs a user giving multiple names to a dimension? Or a complex name such as "cp_or_ep"?
Contributor
Author
There was a problem hiding this comment.
I think this is to generate a use case where we want to test PG cache. The cache is per layout so when multiple names are assigned to one dimension, the PG will be shared.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Stack from ghstack (oldest at bottom):
To avoid creating extra PGs when not needed, (for example when calling unflatten many times and some dims share the same layout), we want to create a pg cache mechanism. We cache a PG by a pair of layout (_MeshLayout) and its pg_option, so if users flatten or unflatten into the same layout+pg option (by default pg_option will be None if not set, that way the cache key falls back to layout itself) we will not create process group if already created.
Also to further consolidate the bookkeeping of DM, we created a device_type, shared_state to store root_mesh, rank_map and pg caches. This way all shared info for a given device mesh universe now becomes a singleton.
cc @H-Huang @awgu @wanchaol @fegin @wz337 @wconstab @d4l3k @pragupta @msaroufim @dcci