Lightning Talk: Let's talk about: Retries
summary
Says simple retry limits plus exponential backoff still create huge amplification when services fail — one example shows 1,000 requests → 4,000 → 16,000 → 64,000 under nested retries — and backoff often fails in open systems. Recommends switching to retry budgets (leaky-token style) to cap extra load, cites an existing tower budget trait and implementation, and notes a request crate retry module that defaults to budgets so the approach ships and works.
AI-written summary · about
related talks
59mLightning Talks
26mThe tools and techniques behind resilient systems
46mLightning Talks
11mLightning Talk: Beyond Billions: Operating Thanos, Prometheus & OpenTelemetry...
52mPerformance & Stability Anti patterns
10mLightning Talk: A Culture of Resiliency through Chaos
05mLightning Talk: Scrape Trolley Dilemma
15mWhy I Prefer Good Testing Over Excellent Testing