[Core feature] Add Round Robin Support for Agent Server Deployments

[Future-Outlier]

Motivation: Why do you think this is important?

It can improve performance when we have more than 1 agent deployments and support same task types.

Goal: What should the final outcome look like, ideally?

For example, 2 endpoint (localhost:8000 and localhost:8001) task type a 1st time send request to localhost:8000, and second time send request to localhost:8001.

Describe alternatives you've considered

Stay the same behavior.

Propose: Link/Inline OR Additional context

No response

Are you sure this issue hasn't been raised already?

• [X] Yes

Have you read the Code of Conduct?

• [X] Yes

[runllm[bot]]

Flyte already supports round-robin load balancing for agent server deployments. According to the Agent Metadata Servicer PR, you can configure multiple agents with round-robin load balancing by specifying the defaultServiceConfig in the agent configuration.

Example configuration:

plugins:
  agent-service:
    defaultAgent:
      endpoint: "dns:///localhost:8000"
      insecure: true
      timeouts:
        GetTask: 100s
      defaultTimeout: 100s
    agents:
      custom_agent:
        endpoint: "dns:///localhost:8001"
        insecure: true
        defaultServiceConfig: '{"loadBalancingConfig": [{"round_robin":{}}]}'
        timeouts:
          DoTask: 300s
          GetTask: 100s
        defaultTimeout: 300s
    agentForTaskTypes:
      - custom_task: custom_agent 
      - api_task: custom_agent

Would you like more details on how to set this up?

Sources:

• Agent Metadata Servicer

---

This is an AI-generated response and your feedback is appreciated! Please leave a 👍 if this is helpful and 👎 if it is not.

Mention @runllm to ask follow-up questions. See runllm.com for more!

[RRap0so]

Hey @Future-Outlier. I'm not entirely sure if I'm following this. If one endpoint points to a DNS which solves to multiple Ips (if SRV even ports and everything) , won't they be loadbalacing between them?

Ideally I would like to not even think about the port or at least not have a LB based on port since multiple instances of my agent will be all the same port.

[kumare3]

I agree with @RRap0so I do not think we should implement a custom loadbalancer. this is a load balancers job.