RFC-042: Tail Latency Budget & Retry Policy
To prevent retry storms across upstream services, each gateway worker enforces an adaptive token bucket.
1. Latency SLO Formula
For a target availability of $99.95%$, bounded tail latency across $k$ serial hops satisfies:
$$L_{\text{total}} = \sum_{i=1}^{k} \mu_i + z_{0.995} \sqrt{\sum_{i=1}^{k} \sigma_i^2}$$
2. Service Budgets
| Hop | Timeout | Max Retries | Backoff |
|---|---|---|---|
| Edge Router | 1500ms |
1 |
Exponential + Jitter |
| Auth Verifier | 120ms |
2 |
25ms base |
| Primary Store | 400ms |
0 |
Circuit Breaker |
curl -fsSL https://supanote.app/health | jq .status