Skip to content

fix: quarantine PD pages until transfer quiesces - #1406

Open
sufubao wants to merge 1 commit into
ModelTC:mainfrom
sufubao:fix/pd-nixl-page-release-order
Open

fix: quarantine PD pages until transfer quiesces#1406
sufubao wants to merge 1 commit into
ModelTC:mainfrom
sufubao:fix/pd-nixl-page-release-order

Conversation

@sufubao

@sufubao sufubao commented Jul 27, 2026

Copy link
Copy Markdown
Collaborator

NIXL WRITE 超时后不再立即回收传输页。Prefill 保留 source page 和 handle,继续轮询到 DONE/ERR,释放 handle 后向 Decode 返回 quiesced 状态;Decode 在此之前将 destination page 留在 quarantine,收到 quiesced error 或晚到 DONE 后再回收。

这样不会因为请求已经返回失败而复用仍可能被旧传输访问的页面。若传输永久停留在 PROC,页面会继续隔离。

验证:PD transfer lifecycle 单测 5/5 通过,Black、Flake8、compileall 和 git diff --check 通过。未运行双节点 NIXL 集群测试。

@gemini-code-assist

Copy link
Copy Markdown
Contributor

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

@sufubao
sufubao force-pushed the fix/pd-nixl-page-release-order branch from 85f2774 to 6508d71 Compare July 27, 2026 09:49
@sufubao sufubao changed the title fix: release PD transfer handle before recycling page fix: quarantine PD pages until transfer quiesces Jul 27, 2026
@sufubao
sufubao force-pushed the fix/pd-nixl-page-release-order branch from 6508d71 to c4e01ca Compare July 27, 2026 10:02
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant