RESILIENCEAI SRE

Timeout Budgets

A timeout budget divides the maximum acceptable user latency across retrieval, model calls, tools and validation.

WHY IT MATTERS

Independent long timeouts cause total request latency to grow far beyond the user SLO.

PRODUCTION EXAMPLE

A 10-second request budget allows 1.5s retrieval, 5s inference, 2s tool time and 1.5s reserve.

SRE DECISION

Does every downstream timeout fit inside the end-to-end SLO?

REMEMBERTimeouts should be designed as one budget, not chosen independently.
No uploads · No company data · No account required