Fix prefix cache OOB in prepare_prefill when num_cached_tokens grows during allocate#233
Open
yuanmouya-prog wants to merge 1 commit into
Open
Conversation
…during allocate When prefix caching hits during block_manager.allocate(), num_cached_tokens can jump from 0 to N*block_size. But the scheduler already computed num_scheduled_tokens based on the old num_cached_tokens=0 before calling allocate(). This causes prepare_prefill to compute end = start + seqlen_q exceeding num_tokens, leading to block_table index out of range. The fix clamps seqlen_q to not exceed the remaining uncached tokens. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Bug
When prefix caching hits during
block_manager.allocate(),num_cached_tokenscan jump from 0 to N*block_size. But the scheduler already computednum_scheduled_tokensbased on the oldnum_cached_tokens=0before callingallocate(). This causesprepare_prefillto computeend = start + seqlen_qexceedingnum_tokens, leading toblock_tableindex out of range.Reproduce
High concurrency (512 requests) + KV cache full + prefix cache hit → preempt → re-prefill → 100% crash with
IndexError: list index out of rangeatmodel_runner.py:155.Fix
One line: clamp
seqlen_qto not exceed remaining uncached tokens.