recall

← recall

request amplification pattern

One incoming request fans out into many internal calls. The user makes 1 request; your service makes 50. When traffic doubles, internal load multiplies — and the slowest dependency dominates p99.

One incoming request fans out into many internal calls. The user makes 1 request; your service makes 50. When traffic doubles, internal load multiplies — and the slowest dependency dominates p99.

symptoms

  • internal RPS far higher than external RPS
  • p99 worse than slowest dependency
  • small traffic spikes cause large internal spikes
  • dashboards on dependent services light up first

causes

  • chatty service decomposition
  • N+1 queries
  • denormalized reads requiring multiple lookups
  • unbatched parallel calls

fixes

  • batch / coalesce calls
  • cache fan-out responses
  • denormalize at write time
  • service-level SLOs that surface amplification

you might say

  • this endpoint fans out to like 30 services
  • we have a really high amplification factor
  • one request lights up the whole graph

related

aliases: fan-out amplification, read amplification, amplification factor

topics: failure-modes, distributed-systems, performance

references: