Uber工程团队发布技术文章，介绍了一套错误所有权机制用于防止重试风暴。该机制通过区分错误来源——判断错误是源自服务本身还是从下游服务传递而来——在共享基础设施中实现上下文感知的重试决策，从而避免级联失败。1
2025年11月18日，Uber Core Entity服务发生故障并导致高错误率。1该错误所有权机制在这次事件中被验证了其有效性，阻止了9500万次虚假请求。1相比之下，传统重试配置会导致已降级服务的流量增加46%-135%。1具体来看，最大重试风暴半径从25降至3，平均值从20降至2。1该机制目前已在Uber全生产环境部署并运行。1
Uber's engineering team has developed an error ownership mechanism designed to prevent retry storms in its microservices infrastructure.1 The system distinguishes between errors originating from a service itself and those propagated from downstream dependencies, enabling context-aware retry decisions across shared infrastructure while avoiding cascading failures.1
The effectiveness of this mechanism was demonstrated during a Core Entity service outage on November 18, 2025.1 When the service experienced elevated error rates, the error ownership system blocked 9.5 million erroneous requests that would have otherwise triggered retry storms.1 Traditional retry configurations in such scenarios would have increased traffic to the degraded service by 46% to 135%.1 The mechanism reduced the maximum retry storm radius from 25 to 3, while bringing the average radius down from 20 to 2.1 The solution is now deployed across Uber's entire production environment.1
评论
还没有评论，欢迎留下第一条。