AI Inference
AI 服务 504 频发背后：大模型公司的运维短板正在暴露
An AI inference P2 outage from thin concurrency and missing circuit breakers exposes the ops wall LLM firms hit moving from benchmarks to production.
Back to home
Postmortem
2 articles tagged with this topic
AI Service Stability
AI Service Crashed 12% of Users for 28 Minutes — Engineering, Not the Model
AI inference outage postmortem: 12% of user requests timed out for 28 minutes. Root cause: missing queue length limits, not the model.