What happens if request_fnc of all agents rejects a request? Is the request retried after some time or dropped?

Thanks for reaching out, akshat. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

How to retry these requests in that case?

You're asking how to retry requests when every agent's

request_fnc

rejects them. By default those requests are dropped and not retried. You can implement retries yourself — two common patterns: Option 1 — Requeue at the application level (recommended) • When a request is rejected by all agents, have your orchestrator re-enqueue the job into an external queue (Redis, SQS, RabbitMQ). • Apply retry/backoff policies on the queue (fixed delay or exponential backoff) and a retry limit to avoid infinite retries. • Consumer flow: pull job from queue → attempt dispatch to LiveKit agents → on full rejection requeue with delay. Option 2 — Handle inside

request_fnc

or

on_request

• Instead of rejecting immediately, have

request_fnc

defer/hold and retry acceptance a few times (with short sleeps) before finally rejecting. • Or, on rejection, push the job into an internal retry mechanism (in-memory scheduler or DB) for later re-dispatch. Practical tips • Add metadata to the job (retry_count, next_retry_at) so you can limit retries and track progress. • Use exponential backoff and a max attempts value to prevent overload. • Log rejection reasons so you can decide whether to retry (transient capacity vs. permanently unsupported job). Which approach fits you best — external queue orchestration or in-agent retry logic? Tell me your current setup (language, orchestrator/queue if any) and I’ll give a concise implementation example. Sources: Agent server options | LiveKit docs | Server-side | livekit/livekit_composite | DeepWiki | server-sdk-go | livekit/livekit_composite | DeepWiki | examples/survey | livekit/livekit_composite | DeepWiki | Agents Framework | livekit/livekit_composite | DeepWiki