Introduce progressive fallback ramp-up logic when switching providers/models.
Example Scenario:
-
Primary provider fails.
-
Instead of sending 100% of requests to fallback, we progressively ramp:
- 10% → 25% → 50% → 100% over a configurable window.
-
If primary recovers, traffic is gradually moved back.
fallback_ramp:
strategy: progressive
intervals:
- percentage: 25
after: 30s
- percentage: 50
after: 1m
Why
- Avoids overloading fallback providers during large failover events.
- Ensures graceful degradation and stabilization during outages.
Introduce progressive fallback ramp-up logic when switching providers/models.
Example Scenario:
Primary provider fails.
Instead of sending 100% of requests to fallback, we progressively ramp:
If primary recovers, traffic is gradually moved back.
Why