feat(心跳): 影子判官先跑读数——心跳断/进度停滞/排队/服务端处理分开判

现有判死线各看各的证据,假活(心跳线程活着、主线程卡死)与假死(长任务/排队被
判死)同时存在。新增只读影子扫描:每 5 分钟扫 RUNNING 文件任务,按「心跳中断 /
进度停滞 / 排队宽限 / 服务端结果任务 / 24h 硬顶」输出 would-finalize 日志,
不改任何任务状态。进度用「计数+文本+阶段」指纹,写影子独立快照键,不影响正式
判死线的未变化起点。规则纯函数单测 9 例 + 只读边界守卫 3 例。
This commit is contained in:
2026-10-20 12:30:09 +08:00
parent 4c1a2a9ffb
commit eb6f082674
7 changed files with 671 additions and 6 deletions
@@ -240,6 +240,18 @@ aiimage:
# 置 0 即整体关闭闸门、退回旧行为(只需改环境变量,不用重新构建)。
task-stale:
first-heartbeat-grace-minutes: ${AIIMAGE_TASK_STALE_FIRST_HEARTBEAT_GRACE_MINUTES:240}
# 心跳影子判官(2026-10-20):只读扫描 RUNNING 文件任务,按「心跳中断 / 进度停滞 / 排队宽限 /
# 服务端处理中 / 总量硬顶」输出 would-finalize 日志,不修改任何任务状态。先用真实读数校准阈值,
# 再决定是否接进判死线;enabled=false 即整体停用。
task-liveness-shadow:
enabled: ${AIIMAGE_TASK_LIVENESS_SHADOW_ENABLED:true}
scan-fixed-delay-ms: ${AIIMAGE_TASK_LIVENESS_SHADOW_SCAN_FIXED_DELAY_MS:300000}
scan-initial-delay-ms: ${AIIMAGE_TASK_LIVENESS_SHADOW_SCAN_INITIAL_DELAY_MS:120000}
scan-limit: ${AIIMAGE_TASK_LIVENESS_SHADOW_SCAN_LIMIT:500}
beat-lost-minutes: ${AIIMAGE_TASK_LIVENESS_SHADOW_BEAT_LOST_MINUTES:10}
no-progress-minutes: ${AIIMAGE_TASK_LIVENESS_SHADOW_NO_PROGRESS_MINUTES:30}
first-heartbeat-grace-minutes: ${AIIMAGE_TASK_LIVENESS_SHADOW_FIRST_HEARTBEAT_GRACE_MINUTES:240}
hard-cap-hours: ${AIIMAGE_TASK_LIVENESS_SHADOW_HARD_CAP_HOURS:24}
module-cleanup:
enabled: ${AIIMAGE_MODULE_CLEANUP_ENABLED:true}
cron: ${AIIMAGE_MODULE_CLEANUP_CRON:0 0 0 * * *}