Give an unhealthy dependency space to recover#
Immediate retries preserve contention. Exponential backoff increases delay between attempts - conceptually 1, 2, 4, then 8 seconds, normally with a cap. Exact delays depend on the workload.
Backoff alone can synchronize 100,000 clients into another wave. Jitter randomizes retry time so clients do not return in lockstep. Different algorithms distribute delay differently; the architectural requirement is to avoid a synchronized surge.
Related practice problems
Use the manifest-backed prompts below to apply this lesson in a complete design.
- Timeouts, retries, and backoff with jitterAmazon Builders' Library
- Addressing cascading failuresGoogle SRE