指数退避重试架构

Architecture

指数退避重试架构图，展示完整的重试策略引擎，包括可重试与不可重试错误分类、带抖动的指数延迟计算、最大重试阈值和优雅的耗尽处理。该模板模拟处理分布式系统中瞬态故障的基本弹性模式，通过随机退避延迟防止惊群问题。对于云架构中任何服务间通信都是基础性的。

下载 .fz 文件

完整 FlowZap 代码

Caller { # Calling Service
n1: circle label:"Initiate Request"
n2: rectangle label:"Send HTTP Request"
n3: rectangle label:"Receive Response"
n4: circle label:"Request Complete"
n1.handle(right) -> n2.handle(left)
n2.handle(bottom) -> RetryPolicy.n5.handle(top) [label="Execute"]
n3.handle(right) -> n4.handle(left)
}
RetryPolicy { # Retry Policy Engine
n5: rectangle label:"Attempt Request"
n6: diamond label:"Response Status?"
n7: rectangle label:"Return Success"
n8: diamond label:"Retryable Error?"
n9: rectangle label:"Return Permanent Failure"
n10: rectangle label:"Calculate Backoff Delay"
n11: diamond label:"Max Retries Exceeded?"
n12: rectangle label:"Wait with Jitter"
n13: rectangle label:"Return Exhausted Error"
n5.handle(right) -> n6.handle(left)
n6.handle(right) -> n7.handle(left) [label="2xx"]
n6.handle(bottom) -> n8.handle(top) [label="Error"]
n7.handle(top) -> Caller.n3.handle(bottom) [label="Success"]
n8.handle(right) -> n10.handle(left) [label="429/503/Timeout"]
n8.handle(bottom) -> n9.handle(top) [label="400/401/404"]
n9.handle(top) -> Caller.n3.handle(bottom) [label="Non-Retryable"]
n10.handle(right) -> n11.handle(left) [label="2^n * base"]
n11.handle(bottom) -> n13.handle(top) [label="Yes"]
n11.handle(right) -> n12.handle(left) [label="No"]
n12.handle(top) -> n5.handle(bottom) [label="Retry"]
n13.handle(top) -> Caller.n3.handle(bottom) [label="Max Retries"]
}
Target { # Target Service
n14: rectangle label:"Process Request"
n15: diamond label:"Service Healthy?"
n16: rectangle label:"Return 200 OK"
n17: rectangle label:"Return 503 Unavailable"
n14.handle(right) -> n15.handle(left)
n15.handle(right) -> n16.handle(left) [label="Yes"]
n15.handle(bottom) -> n17.handle(top) [label="No"]
n16.handle(top) -> RetryPolicy.n6.handle(bottom) [label="Success"]
n17.handle(top) -> RetryPolicy.n6.handle(bottom) [label="Error"]
n5.handle(bottom) -> Target.n14.handle(top) [label="HTTP"]
}

为什么需要这个工作流？

Transient failures (network blips, temporary overloads) are inevitable in distributed systems. Naive retry strategies with fixed intervals cause thundering herd problems where all clients retry simultaneously. Exponential backoff with jitter spreads retries over time, giving the failing service time to recover.

工作原理

Step 1: The calling service sends a request to the target service.
Step 2: If the response indicates a transient error (429, 503, timeout), the retry policy activates.
Step 3: Non-retryable errors (400, 401, 404) are returned immediately without retry.
Step 4: The backoff delay is calculated as 2^n × base_delay with random jitter.
Step 5: If the maximum retry count is exceeded, an exhaustion error is returned to the caller.
Step 6: Successful responses reset the retry counter for subsequent requests.

替代方案

Fixed-interval retries cause thundering herds. No retries mean transient failures become permanent. Circuit breakers complement retries by stopping attempts to known-failing services. This template shows the complete retry policy engine.

Key Facts

Template Name	指数退避重试架构
Category	Architecture
Steps	6 workflow steps
Format	FlowZap Code (.fz file)