Merge pull request #157 from openqj/feat-final-audit-detail

fix(worker): 终态审计带上 error,复发卡死可定位
This commit is contained in:
openqj authored and GitHub committed 2026-10-30 01:53:39 +08:00
commit aa5bf7da72
2 files changed
+36 -1

No files matched your search

+11 -1
View File
@@ -5167,12 +5167,22 @@ class Worker:
if final_written:
# 审计名须与终态一致:interrupted 任务记 worker_failed
# 会把“可续跑的中断”误读成失败,误导后续排查。
# 终态审计细节(2026-10-29 任务 A 教训):interrupted /
# failed 时只记 task_id,事后完全无法区分是取消、
# 配额、host 返回 interrupted 还是其它失败——且自动
# 续跑会把 error_text 覆盖掉,第一轮的原因就此丢失。
# 因此非完成终态的审计 detail 带上 error(截 500 字、
# 去换行,不含 prompt/key),完成态仍只记 task_id。
_final_detail = task_id
if final in ("interrupted", "failed") and error:
_final_detail = "%s: %s" % (
task_id, str(error)[:500].replace("\n", " "))
self._audit(task["project_id"],
"worker_quota_stop" if quota_stop else
("worker_completed" if final == "completed"
else "worker_interrupted"
if final == "interrupted"
else "worker_failed"), task_id)
else "worker_failed"), _final_detail)
log_event(self.logger,
"worker_quota_stop" if quota_stop else
("worker_completed" if final == "completed"