Reduce Rollout Manager Load with Distributed Session Merging#66
Open
odp wants to merge 16 commits into
Open
Conversation
This reverts commit cf6de42.
odp
changed the title
…er to match previous implementation
This was referenced Jul 15, 2026
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Changes made, sorted by importance:
Moved trajectory/sample merging out of the rollout manager and into the distributed Ray actor session servers, so merging is parallelized instead of done one-by-one centrally.
Added Ray actor-based session servers with multiple backends, so rollout collection work can be spread across session server actors.
Reduced rollout-manager CPU and memory pressure by returning already-merged/compacted session data from the session servers.
Avoided unnecessary tokenizer decode work during sample construction;
sample.responseis left as""andsample.metadata["response_decoded"] = False.Updated merge logic to respect
sample.metadata["response_decoded"], so it does not decode observation tokens unless responses are actually decoded.Hardened the session client/server path with explicit HTTP timeouts, retries, bounded concurrency, and better logging.