Reliability & Fault Tolerance

Reliability at a glance

Most incidents are caused by upstreams (RPCs, vendors, networks). Acceso is built to keep behavior predictable under partial failure.

Where failures usually happen

• RPC provider degradation.

• Protocol/API vendor downtime.

• Network partitions and latency spikes.

• Hot keys and burst traffic.

Platform reliability layers

1. Gateway protections (edge safety)

   • Auth + key validation, quotas, and rate limits.

   • Parameter validation and request shaping.

2. Service isolation (fault containment)

   • Domain services isolate protocol failures and hot spots.

   • One failing venue should not degrade unrelated endpoints.

3. Upstream resilience

   • Timeouts, bounded retries, circuit breakers, fallbacks.

4. Caching as resilience

   • Cache-first reads when safe.

   • Derived views for expensive analytics.

5. Observability

   • Request identifiers for tracing.

   • Metrics/logs to spot upstream degradation quickly.

Client defaults that work well

Do not infinite-retry on the client. Use bounded retries with backoff, or switch to streams/webhooks.

• Set client-side timeouts.

• Retry only on transient failures (5xx, timeouts).

• Back off hard on 429 (and batch requests when possible).

• Prefer WebSockets or webhooks for high-frequency updates.

Deep dive

Read the full failure model and degraded-mode behavior in Reliability, Fault Tolerance & Resilience.