Repository navigation
[TRTLLM-17125][test] Check every re-prefill after repeat preemption - #21215
Merged
asfiyab-nvidia merged 2 commits intoOct 28, 2026
Merged
asfiyab-nvidia merged 2 commits into
asfiyab-nvidia merged 2 commits into
Conversation
When a request is preempted again, forget its earlier re-prefill so the replayed prompt is checked on every re-prefill, not just the first, and the request must decode after its latest one. Signed-off-by: Asfiya Baig <asfiyab@nvidia.com>
Under aggregated Helix without a tier below GPU, a generation request that finds no pages and no victim sits out the step without suspending. Log it at debug level with the request ID so stalls can be traced. Signed-off-by: Asfiya Baig <asfiyab@nvidia.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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Description
Fixes a gap in
test_agg_helix_preempts_identically_and_re_prefills_correctly: the replayed prompt was only checked on a request's first re-prefill. When a request was preempted again, its later re-prefills were skipped, because the request was already marked as re-prefilled.On a repeat preemption the test now forgets the earlier re-prefill, so every replay is checked (its prompt is every token the request held when it was last preempted, in full on every rank), and the request must decode after its latest re-prefill. In
chunked_staggered_arrivals, request 2 is preempted three times; its second and third re-prefills were previously unchecked.Test-only. Raised in review of #21122, which carries the same fix for
main.Test Coverage
tests/unittest/_torch/executor/kv_cache/test_helix_agg_preemption.py::test_agg_helix_preempts_identically_and_re_prefills_correctly(GPU)PR Checklist
Please review the following before submitting your PR:
PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.
PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.
Test cases are provided for new code paths (see test instructions)
If PR introduces API changes, an appropriate PR label is added - either
api-compatibleorapi-breaking. Forapi-breaking, includeBREAKINGin the PR title.Any new dependencies have been scanned for license and vulnerabilities
CODEOWNERS updated if ownership changes
Documentation updated as needed
Update tava architecture diagram if there is a significant design change in PR.
The reviewers assigned automatically/manually are appropriate for the PR.
Please check this after reviewing the above items as appropriate for this PR.
GitHub Bot Help
To see a list of available CI bot commands, please comment
/bot help.