Observability & SLOs
- Metrics
- Numbers over time: QPS, error rate, p50/p99 latency, saturation.
- Logs
- Per-event records; structured (JSON) with a request ID.
- Traces
- One request's path across services — finds which hop is slow.
- SLI / SLO / SLA
- Indicator (what you measure) / Objective (internal target, e.g. 99.9% < 300ms) / Agreement (contract with penalties).
- Error budget
- 1 − SLO. Spend it on releases; when exhausted, freeze features and fix reliability.
- p99 latency
- Average hides pain; tail latency is what users at scale feel.
1. Your API calls 3 services in series, each 99.9%. Best-case availability?
2. A dependency is timing out and your service's threads are all stuck. Which pattern most directly helps?
Answered 0 / 2
Explain What happens when 10k clients all retry immediately against a service that just recovered, and how do you prevent it?