Skip to content

test(logging): stabilize recovery checks and prepare pre-release - #457

Merged
wimi321 merged 1 commit into
mainfrom
release/next-2026-09-13.2
Sep 13, 2026
Merged

test(logging): stabilize recovery checks and prepare pre-release#457
wimi321 merged 1 commit into
mainfrom
release/next-2026-09-13.2

Conversation

@wimi321

@wimi321 wimi321 commented Sep 13, 2026

Copy link
Copy Markdown
Owner

原因与修复

#456 合并后的 Windows 主线 CI(34728678679)发现 LoggingRuntimeTest 的恢复断言偶发失败,发布门禁正确停止,未创建公开版本。

测试使用的旧 awaitIdle() 最多轮询约 400ms,超时没有返回值。慢写入时队列尚未排空,测试已开始提交恢复事件并检查 recovered。本次改用现有带超时返回值的提交序号完成屏障,并强制断言成功;生产日志、队列容量与恢复规则不变。

  • 增加 0ms / 600ms 两种写入延迟。旧逻辑在受控 600ms 延迟下稳定失败于相同恢复断言,修复后两种场景通过。
  • 明确验证写线程被阻塞时屏障不能通过。
  • 验证排空前旧事件不能伪报恢复,排空后新事件确实落盘、恢复状态正确且不增加丢弃计数。
  • 保留所有原始断言,不跳过或自动重试失败测试。

发布

新增 next-2026-09-13.2 的六语固定格式说明与发布请求,功能范围与 #456 相同。旧 .1 请求保留审计记录,不复用失败流程、不移动 tag。#449 仍为草稿并排除。

CI 和干净本机完整回归通过后合并;精确合并 SHA 再通过主线 CI、四个平台构建、签名/公证及资产来源与摘要校验后才公开 pre-release。正式版、官网 stable 和 R2 stable 不变。

验证

  • 受控慢写入红/绿测试:原断言失败,修复后通过。
  • LoggingRuntimeTest:28 项,0 失败、0 错误。
  • 六语发布说明验证、git diff --check:通过。
  • 完整 portable 门禁正在运行;GitHub Windows/Linux 检查须全部通过。
  • Windows hosted CI 不等于所有真实显卡和桌面 UI 验收,不作该声明。

@wimi321

wimi321 commented Sep 13, 2026

Copy link
Copy Markdown
Owner Author

最终验收:干净提交的完整 portable 26/26 门禁通过;4,139 项测试,0 失败、0 错误、46 项既有环境条件跳过;shaded JAR 与日志打包冒烟通过,工作树干净。GitHub Windows/Linux Java 以及全部脚本与 ci-required 通过(34729121541)。按该已验证提交合并,随后由主线精确 SHA 的门禁进行全平台 pre-release 构建与公开校验,不提前发布。

@wimi321
wimi321 merged commit c1857c2 into main Sep 13, 2026
6 checks passed

@qiyi71w qiyi71w left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

P1 — Logging recovery wait is correct, but .2 six-language notes still drop RTX 30 from the TensorRT optional audience (same as #456 SPEC-1).

Standards

No findings

Spec

SPEC-1 — scope: IN_SCOPE; status: open

The .2 notes retarget #456 by tag/URL only and keep exclusive RTX 20 / GTX 16 TensorRT wording in all six languages (before-download bullets and TensorRT table labels), omitting RTX 30 which remains an allowed TensorRT audience in product copy and detection.

Authority: this PR’s release-notes ask plus #456 SPEC-1; CONTRIBUTING.md alignment; docs/RELEASE_CHECKLIST.md §四 MUST (TensorRT optional for RTX 30 and earlier); README.md (“TensorRT 是 RTX 30 系及以下的可选方案”); NvidiaGpuDetector.recommend / NvidiaGpuDetectorTest still ALLOW TensorRT on RTX 3070 (and RTX 2080 / GTX 1660).

Impact: an RTX 30 user following this pre-release page would skip the optional offline splits; notes diverge from README, in-app Auto Setup, and the frozen checklist. Mechanical validate_release_notes.py still passes (structure/URLs only).

Acceptance: every language’s before-download bullet and TensorRT download-table label restore “RTX 30 series and earlier” (RTX 30 系及以下 / equivalent). GTX 16 may stay as an extra allowed example, not the exclusive audience.

The LoggingRuntimeTest awaitIdle(timeout) fix and production-untouched recovery contract match the originating flake ask; no separate findings there.

Follow-up ledger

Owned by existing tickets

No entries

Candidates — Fix

No entries

Candidates — Diagnose

No entries

Closed

No entries

Open IN_SCOPE: 1
Owned: 0
Fix candidates: 0
Diagnose candidates: 0
Closed: 0


- 现有 Windows 用户可选 core-update.zip,先退出软件,再覆盖原目录;更新前备份 user-data 和棋谱。
- 新安装或需要新 B11 的用户请下载完整包。B11 单次判断较强但搜索较慢;追求速度可在“一键设置 → 权重”选择 B10。
- RTX 30/40/50 优先 CUDA;TensorRT 是 RTX 20、GTX 16 的可选项,实际速度以本机测速为准。实验后端仍需对应硬件验证。

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

SPEC-1 (IN_SCOPE) — TensorRT optional audience still drops RTX 30 (carried from #456)

This bullet (and the matching TensorRT table label, plus the same pattern in all five other languages) limits TensorRT to RTX 20 / GTX 16. Product copy and detection still treat TensorRT as optional for RTX 30 and earlier (README.md, docs/RELEASE_CHECKLIST.md §四, NvidiaGpuDetector). .2 only retargeted tag/URLs from .1; the exclusive audience wording was not fixed.

Acceptance: restore “RTX 30 series and earlier” (RTX 30 系及以下 / equivalent) in every language’s before-download bullet and TensorRT download-table label. GTX 16 may remain as an extra example, not the exclusive audience.

@wimi321

wimi321 commented Sep 13, 2026

Copy link
Copy Markdown
Owner Author

补充成品验收:主线 CI 34729371705 已通过。Linux 发布构建 34729623885、Apple Silicon 发布构建 34729626603 均成功。已下载实际生成的 Apple Silicon DMG,在本机核对 SHA-256 e31b903d3e4c4ccb3d4e3d191e754f4465294857284680fd89e1cab85f5066ac,与 GitHub 一致;codesign --verify --deep --strict、Gatekeeper Notarized Developer ID、stapler validate 通过。内置 B11 SHA 73f6454eba62d2f6d099af8ce66d8c3fde6225e223c55817da0627590e98b0ae 正确;实际 Metal KataGo v1.18.1 对空棋盘和四手局面各完成一次分析,返回有效候选与胜率后正常退出。此项是成品引擎冒烟,不替代完整 UI/Windows GPU 验收。继续等待 Windows、Intel 与最终资产发布门禁。

@wimi321

wimi321 commented Sep 13, 2026

Copy link
Copy Markdown
Owner Author

已完成发布:https://github.com/wimi321/lizzieyzy-next/releases/tag/next-2026-09-13.2 。四个平台构建、主线 CI、发布门禁 34729371697 和测试频道 34732259227 全部成功。公开回读确认非 Draft、Pre-release=true,共 27 个资产,tag 精确指向 c1857c2;六语正文与仓库审核版本逐字一致,Windows core-update manifest 的文件名、大小、SHA-256、下载 URL 与实际发布资产一致。正式 Latest 仍为 next-2026-09-04.1。发布期间新增 #458 未混入冻结的本次发布,#449 继续保留草稿。

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.

2 participants