Skip to content

v0.4.1

Latest

Choose a tag to compare

@github-actions github-actions released this 20 Oct 09:15
· 65 commits to main since this release
a9ff940

中文 | English

0.4.1 补丁版本,汇总了自 v0.4.0 以来的主要变更。

问题修复

  • 更新锁释放时由持有者删除 .dsh-ops-update-*.lock 边车:先重新打开名字并比较 (device, inode),只有绑定成功的持有者才能发布,并且只在仍持锁且名字仍指向自己的 inode 时 unlink;成功写入、拿锁后被拒的请求与整目录 replace 都不再留下文件,被硬杀留下的边车会被下一次同名写入复用并清除。 @T-Auto
  • 恢复 ops_bash 的后台任务:run_in_background 不再必然报 EXECUTOR_UNAVAILABLE,改为经 ctx.inject(['jobs'], …) 绑定一次宿主 jobs 注册表并传入工具定义,前台超时也能提升为后台。 @T-Auto
  • 面板采集恢复时先计入已有追加日志或新会话头部的字节,再接纳排队事件;恢复已封顶日志不再追加事件或重复的封顶标记,日志初始化失败即停止该会话采集,JSON 解析丢失记录时标记采集降级与用量不完整并保留有效记录。 @ccch1mneyyy
  • PTC 批量审计不再为审计记录创建前的已确认校验失败、或工作进程创建前的同步失败返回不存在的引用,同时保留 worker 返回的引用、已有 resume 标识与执行中断或状态不明时的恢复标识,且不增加 worker 调用。 @ccch1mneyyy

The 0.4.1 patch release summarizes the main changes since v0.4.0.

Bug Fixes

  • The .dsh-ops-update-*.lock sidecar is now removed on release by the writer that holds it: the holder reopens the name and compares the (device, inode) identity, only a bound holder publishes, and it unlinks only while still holding the lock on the inode the name carries. Published writes, requests refused after the lock and directory-wide replace calls leave nothing behind, and a sidecar left by a killed writer is reused and removed by the next write of that name. by @T-Auto
  • Restore ops_bash background jobs: run_in_background no longer always fails with EXECUTOR_UNAVAILABLE. The host jobs registry is bound once through ctx.inject(['jobs'], …) and passed into the tool definition, and a foreground timeout can now promote the command to the background. by @T-Auto
  • Panel capture recovery now counts the bytes already written to the append log or the new-session header before admitting queued events. Recovering a capped log no longer appends events or duplicate cap markers, a failed log open stops that capture, and JSON parse losses now mark the capture degraded and usage incomplete while preserving valid records. by @ccch1mneyyy
  • PTC batch audits no longer return references for confirmed validation failures that occur before the audit record is created, or for synchronous failures before the worker starts. Worker-provided references, existing resume identities and recovery ids for interrupted or ambiguous execution are preserved, and no additional worker calls are made. by @ccch1mneyyy

Full Changelog: v0.4.0...v0.4.1