Skip to content

Daily: make io_uring cancellation terminal state explicit - #203

Open
yunwei37 wants to merge 11 commits into
mainfrom
daily/2026-09-16-io-uring-cancel-contract
Open

yunwei37 wants to merge 11 commits into
mainfrom
daily/2026-09-16-io-uring-cancel-contract

Conversation

@yunwei37

Copy link
Copy Markdown
Member

Summary

  • publish the 2026-09-16 bilingual Daily Report on io_uring cancellation races and terminal operation state
  • develop generation-stable terminal-state receipts, resource-retirement fences, and adversarial race/effect tests
  • record the mix-driven adjacent-systems detour while keeping the eBPF Deployment Compatibility and Lifecycle series active
  • reconcile September 15 delivery state and refresh September 16 source/SEO records
  • make no unrelated technical SEO implementation change because current evidence does not establish a concrete defect

Evidence

Primary technical evidence includes upstream liburing cancellation and synchronous-cancel documentation, current Linux io_uring/cancel.c, current Linux io_uring UAPI definitions, and liburing maintainer clarification around -EALREADY and target completion. The Google Drive weekly export set remains unchanged through 2026-09-13; finalized/partial qualifications are preserved rather than treating missing rows as zero. The September 16 public-safe brief reports a healthy homepage/robots/sitemap surface and stable DEV coverage counters.

Mix

Before this publication, the newest ten published reports are 7 eBPF-centered / 1 pure Agent / 2 adjacent. The oldest report rotating out is adjacent, so an eBPF-centered report would violate the 7-of-10 cap. This adjacent io_uring report preserves 7 / 1 / 2 without relabeling older work.

Delivery

This is the single non-draft daily PR. Final-head CI, full diff/generated-output self-review, review-thread reinspection, squash merge, exact-merge deployment, bilingual production verification, sitemap verification, and one compact merged-PR closeout comment remain the acceptance steps. No second closeout PR will be created.

Copilot AI lite review requested due to automatic review settings September 16, 2026 15:53

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟡 Changes recommended

Unresolved critical and moderate findings remain in the reports and operational records.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Publishes a bilingual Daily Report on io_uring cancellation terminal states and updates related indexes, SEO, planning, and delivery records.

Changes:

  • Adds English and Chinese research reports.
  • Updates localized research indexes.
  • Refreshes September 16 operational and topic-mix records.
File summaries
File Description
docs/research/io-uring-cancel-terminal-state.zh.md Chinese terminal-state report
docs/research/io-uring-cancel-terminal-state.md English terminal-state report
docs/research/index.zh.md Chinese index entry
docs/research/index.md English index entry
.github/seo-data/status.md Current SEO and delivery status
.github/seo-data/plan.md Daily priorities and constraints
.github/seo-data/daily/2026-09-16.md Evidence and delivery record
.github/seo-data/content-series.md Topic-mix record
.github/seo-data/block.md Source freshness record
Review details

Suppressed comments (5)

.github/seo-data/content-series.md:202

  • The archive dates the earlier io_uring programmability report to 2026-08-14 (docs/research/io-uring-bpf-programmability.md:2), not August 15. This novelty record is off by one day; use the report's actual publication date.
benchmark. This is distinct from the August 15 eBPF/io_uring programmability

docs/research/io-uring-cancel-terminal-state.md:86

  • The example labels RECV_MULTISHOT as an opcode, but liburing keeps the opcode IORING_OP_RECV and enables multishot with the IORING_RECV_MULTISHOT flag in sqe->ioprio. As written, a concrete receipt schema cannot map this value to the UAPI.
opcode = RECV_MULTISHOT

docs/research/io-uring-cancel-terminal-state.md:133

  • Reading a file after the terminal event or process/ring shutdown only observes the page-cache/filesystem view; it cannot establish stable-storage or crash-survival effects, which the preceding gap lists as a possible boundary. Scope the oracle to read visibility or add persistence/crash evidence, otherwise the benchmark can report a false effect agreement.
For network I/O, the peer can record bytes observed. For file writes, the harness can read back the target after the terminal event and after forced shutdown. For poll/multishot operations, it can count events emitted before and after the final CQE. For io-wq paths, it can use controlled blocking points to widen the `-EALREADY` race window.

docs/research/io-uring-cancel-terminal-state.zh.md:88

  • 示例把 RECV_MULTISHOT 标成了 opcode,但 liburing 的 opcode 仍是 IORING_OP_RECV,multishot 是通过 sqe->ioprio 中的 IORING_RECV_MULTISHOT flag 开启的。按现在的写法,具体的 receipt schema 无法把这个值映射到 UAPI。
opcode = RECV_MULTISHOT

docs/research/io-uring-cancel-terminal-state.zh.md:135

  • 在终态事件或进程/ring shutdown 后回读文件,只能观察 page cache/文件系统视图,无法证明稳定存储或崩溃后仍然保留的效果,而前面的缺口把这类边界列为需要区分的对象。请把 oracle 明确限定为 read visibility,或加入持久化/崩溃证据,否则 benchmark 可能错误地报告 effect agreement。
网络 I/O 可以由 peer 记录实际收到的 bytes;文件写入可以在终态后和强制 shutdown 后回读;poll/multishot 可以统计 final CQE 前后产生的事件;io-wq 路径可以设置可控 blocking point,把 `-EALREADY` 的竞争窗口放大。
  • Files reviewed: 9/9 changed files
  • Comments generated: 5
  • Review effort level: Lite

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

Comment thread .github/seo-data/status.md Resolved

Copy link
Copy Markdown
Member Author

Follow-up on Copilot's suppressed findings: the August io_uring programmability date is corrected to 2026-08-14; the invalid RECV_MULTISHOT receipt example is gone from both reports and the concrete schema now uses SEND_ZC; and the file-write benchmark oracle is explicitly limited in both languages to post-completion read visibility, with persistence/crash survival requiring a separate durability oracle. The visible Copilot threads are also all addressed and resolved.

@yunwei37 yunwei37 left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reviewed the head 65a5aa6b as a content change: research page on io_uring
cancellation terminal state, index.md entry, and seo-data bookkeeping
(block, content-series, daily/2026-09-16, plan, status).

Diff-shaped edits throughout, page indexed. Checks green.

AI-generated response; a maintainer will review and follow up later

@yunwei37 yunwei37 left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reviewed at the current head, the merge of master that resolved the conflict reported on this pull request.

The generated index files were resolved as a union with no duplicate headings.

Verification and the resolution per file are in the pull request conversation
comment; state is mergeable_state=blocked, which is the approval gate. A
maintainer's approval and merge are what remain.

AI-generated response; a maintainer will review and follow up later

@yunwei37
yunwei37 force-pushed the daily/2026-09-16-io-uring-cancel-contract branch 2 times, most recently from 212a87a to f9f2eb5 Compare November 15, 2026 08:34

@yunwei37 yunwei37 left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Reviewed head f9f2eb5b93: the previous review point (02889a50ba) moved onto today's main. The branch's own content is unchanged.

Evidence: for every path this pull request changes (5 files), the git blob at this head equals the blob at 02889a50. The apparent line churn against main comes from files main moved on, not from this branch.

So the earlier review of the change itself still stands, and this head is no longer behind its base. Merge readiness is unchanged: it waits on a maintainer's decision.

AI patrol review

@yunwei37

Copy link
Copy Markdown
Member Author

Verdict: ready to merge

Head f9f2eb5b carries a patrol review of its change (see the review above/below on this pull request), and every check on this head is green. Nothing in the change is left for the patrol to fix.

Merge is the maintainer's to make: eunomia.dev is not in the small-repository group, so the patrol does not merge it.

AI patrol review

@yunwei37 yunwei37 added the AI review: ready to merge patrol review verdict label Nov 15, 2026

@yunwei37 yunwei37 left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Verdict: ready to merge

The review above stands: this head's content is blob-identical to the previously reviewed revision, so only the rebase onto today's main moved. Checks green, mergeable, level with the base.

AI patrol review

@yunwei37
yunwei37 force-pushed the daily/2026-09-16-io-uring-cancel-contract branch from f9f2eb5 to b52d84c Compare November 15, 2026 19:08

@yunwei37 yunwei37 left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Verdict: ready to merge once CI is green

Reviewed head b52d84ca. Every changed file is blob-identical to the previously reviewed revision (f9f2eb5b), including the two commits that narrowed the cancellation oracle after the earlier review (dce8443e, ef2c147d) and their Chinese counterparts: docs/research/io-uring-cancel-terminal-state{,.zh}.md, both index edits, .github/seo-data/daily/2026-09-16.md. Merge base unchanged at f8a2cce8.

Content is unchanged, and its sharpest claim still holds: IORING_OP_ASYNC_CANCEL reports completion of the cancellation request, not of the target, so a retry loop has to observe the target's own terminal CQE (or -ECANCELED) before treating the resource as retired. The file-write scope added in the follow-ups keeps that claim from being read as covering every opcode.

Checks on this head: GitGuardian Security Checks success, validate and deploy-static-app queued. The verdict turns to merge-ready when those report green.

AI patrol review

@yunwei37
yunwei37 force-pushed the daily/2026-09-16-io-uring-cancel-contract branch from b52d84c to 9aee014 Compare November 16, 2026 03:27

@yunwei37 yunwei37 left a comment

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Verdict: ready to merge once CI is green

Reviewed at head 9aee0140, which is the previously reviewed b52d84ca moved onto today's base. Nothing in the change itself moved: for the range merge-base(main, head)..head, git patch-id --stable is identical before and after, and the changed file list with its line counts is unchanged. The earlier review (5912544411) therefore stands as written, and so does its verdict.

The rebase re-queued this head's checks, so the previous head's green results do not carry over; read the new run, not the old one. Merge stays with a maintainer.

AI patrol review

@yunwei37 yunwei37 added the AI review: merge after CI patrol review verdict label Nov 16, 2026
@yunwei37

Copy link
Copy Markdown
Member Author

Verdict: ready to merge

This is the verdict for head 9aee0140. The branch was replayed onto current main; its diff against the base has the same patch-id as the reviewed head, so no line of the change moved — only the base did. All checks on this head are green.

Merge is the maintainer's to make: eunomia.dev is not in the small-repository group, so the patrol does not merge it.

AI patrol review

@yunwei37 yunwei37 removed the AI review: merge after CI patrol review verdict label Nov 16, 2026

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants