想给 AI 服务补监控。最小集合应该有哪些？

]]>https://localaihub.com/topic/254/生产-ai-系统的-最小监控-应该有哪些RSS for NodeWed, 03 Jun 2026 20:31:40 GMTTue, 19 May 2026 05:10:00 GMT60我先做 p50/p95、错误率、Token、重启、检索耗时。

]]>https://localaihub.com/post/2918https://localaihub.com/post/2918Tue, 19 May 2026 23:38:00 GMT工具调用要单独打，不然用户觉得 AI 慢，其实是接口慢。

]]>https://localaihub.com/post/2917https://localaihub.com/post/2917Tue, 19 May 2026 20:36:00 GMT流式输出时记录模型请求开始到第一个 chunk 的时间。

]]>https://localaihub.com/post/2916https://localaihub.com/post/2916Tue, 19 May 2026 17:47:00 GMT首 token 怎么打？

]]>https://localaihub.com/post/2915https://localaihub.com/post/2915Tue, 19 May 2026 15:23:00 GMT日志里敏感内容要脱敏和设保留期。

]]>https://localaihub.com/post/2914https://localaihub.com/post/2914Tue, 19 May 2026 15:04:00 GMT可以分阶段。第一阶段先有延迟、错误、成本、重启、反馈。

]]>https://localaihub.com/post/2913https://localaihub.com/post/2913Tue, 19 May 2026 12:22:00 GMT这些好多。

]]>https://localaihub.com/post/2912https://localaihub.com/post/2912Tue, 19 May 2026 12:14:00 GMT产品层：用户反馈、转人工率、未解决率。

]]>https://localaihub.com/post/2911https://localaihub.com/post/2911Tue, 19 May 2026 12:02:00 GMTRAG 层：召回数量、无结果比例、引用缺失比例、rerank 耗时。

]]>https://localaihub.com/post/2910https://localaihub.com/post/2910Tue, 19 May 2026 09:29:00 GMT系统层：CPU、内存、显存、磁盘、IO、网络、进程重启。

]]>https://localaihub.com/post/2909https://localaihub.com/post/2909Tue, 19 May 2026 09:09:00 GMT请求量、错误率、延迟、首 token、总 tokens、模型错误、检索耗时、工具耗时。

]]>https://localaihub.com/post/2908https://localaihub.com/post/2908Tue, 19 May 2026 07:06:00 GMT