fix(断点续传): 外观/结果作业/去重三处中断语义补齐,重启不再静默丢行
- 外观专利行标改两态(in-flight 15min / processed 24h):只有 CAS 写回 payload 成功 才转「已处理」,异常、覆盖率不足、CAS 打满等未落库出口一律撤销标记 —— 修「发 LLM 请求前就盖章,进程被杀的那一批永不再补检」 - 结果文件作业开机自回收:按 owner_instance_id 只收本机名下 RUNNING 且心跳超 90s 的行; 卡死阈值 30 → 5 分钟,重启接回从最长半小时压到 5 分钟 - 外观结果表缺检行显式写「未取得检测结果」并判失败,不再留空靠 0.5 覆盖率阈值蒙过 - 去重进度改 NodeSharedStore 跨节点共享,runId 用已落库的 task_no:轮询打到另一节点或 发版重启后仍能读到真实进度、已完成文件仍可下载;共享态缺失时按 task_no 从库重建; 逐文件心跳防 StaleTaskRepair 把长任务误判死;开机把本机残留 RUNNING 判失败并给提示
This commit is contained in:
@@ -81,7 +81,7 @@ AIIMAGE_RESULT_FILE_JOB_LOCAL_DISPATCH_POOL_SIZE=2
|
||||
AIIMAGE_RESULT_FILE_JOB_LOCAL_DISPATCH_QUEUE_CAPACITY=200
|
||||
AIIMAGE_RESULT_FILE_JOB_LOCAL_WORKER_ENABLED=false
|
||||
AIIMAGE_RESULT_FILE_JOB_LOCAL_WORKER_DELAY_MS=10000
|
||||
AIIMAGE_RESULT_FILE_JOB_STUCK_TIMEOUT_MINUTES=30
|
||||
AIIMAGE_RESULT_FILE_JOB_STUCK_TIMEOUT_MINUTES=5
|
||||
|
||||
AIIMAGE_ZINIAO_ENABLED=false
|
||||
AIIMAGE_ZINIAO_BASE_URL=https://sbappstoreapi.ziniao.com/openapi-router
|
||||
|
||||
@@ -275,7 +275,11 @@ aiimage:
|
||||
local-worker-enabled: ${AIIMAGE_RESULT_FILE_JOB_LOCAL_WORKER_ENABLED:true}
|
||||
local-worker-delay-ms: ${AIIMAGE_RESULT_FILE_JOB_LOCAL_WORKER_DELAY_MS:10000}
|
||||
stuck-scan-delay-ms: ${AIIMAGE_RESULT_FILE_JOB_STUCK_SCAN_DELAY_MS:60000}
|
||||
stuck-timeout-minutes: ${AIIMAGE_RESULT_FILE_JOB_STUCK_TIMEOUT_MINUTES:30}
|
||||
# 作业运行期间每 60s touch 一次 updated_at,所以"5 分钟没心跳"就等于持有者已经没了;
|
||||
# 30 分钟是发版/容器重建后结果文件作业被白白冻结的时长。
|
||||
stuck-timeout-minutes: ${AIIMAGE_RESULT_FILE_JOB_STUCK_TIMEOUT_MINUTES:5}
|
||||
# 开机自回收的"已停止心跳"判据(≈1.5 个心跳周期),只回收 owner=本机 的 RUNNING 作业。
|
||||
boot-reclaim-stale-seconds: ${AIIMAGE_RESULT_FILE_JOB_BOOT_RECLAIM_STALE_SECONDS:90}
|
||||
heartbeat-interval-ms: ${AIIMAGE_RESULT_FILE_JOB_HEARTBEAT_INTERVAL_MS:60000}
|
||||
batch-size: ${AIIMAGE_RESULT_FILE_JOB_BATCH_SIZE:20}
|
||||
# 客户端兜底拉取:页面没推送成功(只解析没点启动 / 推送前关页面 / 纯浏览器打开)、
|
||||
|
||||
Reference in New Issue
Block a user