ADK Java Error Recovery — Handling Agent Failures

Why it matters

Unhandled errors turn agents into bad UX. Understanding recovery patterns builds reliable apps.

The architecture

Retry: transient errors → exponential backoff.

Fallback: alternative path when primary fails.

Circuit breaker: stop calling failing service.

How it works end to end

Retry policy: max attempts + exponential backoff + jitter.

Fallback: cached response or simpler answer.

Graceful degradation: agent tells user 'I couldn't do X' instead of failing.

Idempotency: ensure retries safe.