A retry storm happens when concurrent retries pile onto a struggling service, multiplying load instead of restoring service. A two-second delay at a Payment API can trigger the problem: an Order Service mistakes the delay for a lost request and retries. section 12.8.3
When retries make things worse
With 1,000 users and a naive loop that makes three unspaced retries, the struggling Payment API receives 3,000 extra requests. The database can melt under this amplified traffic. section 12.8.3
Dampen the failure with two shock absorbers:
- Use a circuit breaker to stop retrying and fail fast when a downstream service is struggling. section 12.8.3
- Use exponential backoff with jitter to space retries and give the service time to recover. section 12.8.3
Figure 12.8 The Retry Storm. Without exponential backoff or a circuit breaker, a minor 2-second network delay causes naive loops to multiply traffic, accidentally launching a Distributed Denial of Service (DDoS) attack against your own database.

Shows how a minor delay and naive retry loops can multiply traffic into a self-inflicted denial-of-service attack. section 12.8.3

