fix(rollout): make dynamic refills granular#2217
Closed
EazyReal wants to merge 1 commit into
Closed
Conversation
Allow positive refill sizes smaller than the target rollout batch, and cancel unfinished non-partial generation tasks before waiting for SGLang to drain. This prevents one rejected prompt group from launching a full replacement wave or keeping the drain loop alive through new agent turns.
EazyReal
force-pushed
the
maxwill/upstream-granular-rollout-refill
branch
from
July 18, 2026 02:14
f6f503e to
e3eca96
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Problem
Dynamic sampling currently requires
over_sampling_batch_size >= rollout_batch_size, even though refill generation can safely request only the number of groups still missing. This forces oversized refill batches.Stopping a non-partial rollout also waits for pending agent tasks until after SGLang is drained. Those tasks can keep issuing requests while the drain is in progress, delaying shutdown.
Change
over_sampling_batch_size