知识卡片
退避抖动与熔断控重试
过载请求失败后立即重试,会与新请求叠加并放大流量,多层重试尤甚。指数退避延长间隔,随机抖动避免客户端同步重试,熔断在持续错误时暂缓请求;恢复后逐步放量,防止冷系统被全量流量再次击倒。它们减少自激负载,但仍要控制总重试预算与业务截止时间。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
原书出处 · 1 处
- Chapter 16. Common Failure Patterns · 查看原书依据
Exponential backoff is good, but it is imperfect; in addition to exponential backoff, it is also very useful to add jitter. Jitter is a small amount of randomness which spreads load across time. Without jitter, a situation can occur where everyone fails at the same time, waits
参考来源
- 位置:《Designing Distributed Systems, 2nd Edition》第16章《Chapter 16. Common Failure Patterns》(来源项目 book-sources;源文件:extracted/Designing Distributed Systems, 2nd Edition/_epub-src/OEBPS/ch16.html)
- 结论依据:原文由过载翻倍推导退避、jitter、circuit breaker与渐进恢复。
- 原始内容:Exponential backoff is good, but it is imperfect; in addition to exponential backoff, it is also very useful to add jitter. Jitter is a small amount of randomness which spreads load across time. Without jitter, a situation can occur where everyone fails at the same time, waits