cuda: keep async expert slots alive through completion - #989
Open
ZacharyZcR wants to merge 1 commit into
Open
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Fixes #933.
Async CUDA expert groups can retain borrowed zero-copy views after
moe()returns. This change adds per-slot in-flight references, acquires them before every async group issue, and releases them only after the corresponding stream-completingtake().Eviction, pilot reload, io_uring reload, and RSS reclamation now skip referenced slots. If every LRU slot is busy, cache promotion is skipped with a one-time warning instead of recycling live storage or waiting indefinitely.
A regression test covers victim selection while slots are referenced.
Verification:
make -C c test-cmake -C c -j2CUDA hardware/toolkit is not available locally; CUDA-specific compilation and runtime coverage are left to CI and maintainer hardware.