Skip to content

Revert "[single_controller] feat: support resource pool split method" - #4258

Merged
wuxibin89 merged 1 commit into
mainfrom
revert-4251-split_rp
Nov 24, 2025
Merged

wuxibin89 merged 1 commit into
mainfrom
revert-4251-split_rp

Conversation

@vermouth1992

Copy link
Copy Markdown
Collaborator

Reverts #4251

@gemini-code-assist gemini-code-assist Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request reverts the resource pool split feature and replaces it with a new mechanism for resource allocation using replica_rank and replica_world_size in RayWorkerGroup. This simplifies RayResourcePool and moves the resource slicing logic into RayWorkerGroup, which seems like a good refactoring. However, I've found a critical issue in the new implementation within verl/single_controller/ray/base.py. The logic for assigning ranks to workers assumes a homogeneous cluster configuration (i.e., all nodes have the same number of processes), and it will fail in heterogeneous environments. I've provided a detailed comment and a suggested fix for this issue.

@yyDing1
yyDing1 self-requested a review November 24, 2025 06:52
@wuxibin89
wuxibin89 merged commit 65ef28d into main Nov 24, 2025
71 of 73 checks passed
@wuxibin89
wuxibin89 deleted the revert-4251-split_rp branch November 24, 2025 07:03
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants