Provider fallback logic for errors midstream

[fhalde]

Hi,

Envoy AI Gateway supports provider fallback: https://aigateway.envoyproxy.io/docs/0.4/capabilities/traffic/provider-fallback/

My question is: if I have an ongoing LLM stream and the provider fails mid-stream, does the gateway automatically make a call to another provider, or is fallback only triggered before streaming begins?

Do we provide any hint back to the caller?