Retry

Exponential Backoff

점진적으로 시간 간격이 늘어나는 Exponential Backoff 전략

지수에 비례하여 Backoff 시간을 조절한다.

한계점: 어차피 동시에 요청이 몰린다면 똑같은 시간 간격으로 모든 재시도가 동일하게 몰릴 것

 

With Jitter

원래 Jitter는 데이터 통신 용어로 사용될 때는 패킷 지연이 일정하기 않고, 수시로 변하면서 그 간격이 일정하지 않는 현상을 의미한다

동일한 재시도 시간 간격에 무작위성을 추가하여 서로 요청하는 시간대의 동시성(?)을 분산

지수로 증가하는 Backoff 시간에 일정 범위 안의 랜덤 대기 시간을 추가적으로 더하는 것

 

On WebFlux

Normal Retry

• .retry()
• .retry(long numRetries)
• .retryWhen(Retry retrySpec)
  • .retryWhen(Retry.max(3)) == .retry(3)
  • .retryWhen(Retry.fixedDelay(3, Duration.ofMillis(1000)))

Exponential Backoff

• .retryWhen(Retry retrySpec)
  • .retryWhen(Retry.backoff(3, Duration.ofMillis(2000)))

Jitter

An additional benefit of the backoff strategy is that it adds randomness or jitter to the computed delay interval. Consequently, jitter can help to reduce retry-storms where multiple clients retry in lockstep.

By default, this value is set to 0.5, which corresponds to a jitter of at most 50% of the computed delay.

We should note that the possible range of values is between 0 (no jitter) and 1 (jitter of at most 100% of the computed delay).

• .retryWhen(Retry retrySpec)
  • .retryWhen(Retry.backoff(3, Duration.ofMillis(2000)).jitter(0.75))

 

 

 

Reference

https://javacan.tistory.com/entry/Reactor-Start-5-error-handling

https://velog.io/@jazz_avenue/%EC%A2%80-%EB%8D%94-%EC%9A%B0%EC%95%84%ED%95%9C-Retry-Expenential-Backoff-with-Jitter

https://aws.amazon.com/ko/blogs/architecture/exponential-backoff-and-jitter/

https://www.baeldung.com/spring-webflux-retry#3-retrying-with-jitter

https://www.baeldung.com/resilience4j-backoff-jitter