wenjin272 commented on issue #1065: URL: https://github.com/apache/flink-agents/issues/1065#issuecomment-5616733143
Thanks for raising this and for putting together the initial proposal. I agree that budget enforcement is valuable, especially for continuously running streaming jobs where an unnoticed loop or retry could cause significant token consumption. I have a few initial thoughts on the design. We may need token limits at different scopes, such as per model request, per ReAct loop, per time window, and per subtask. A job-wide per-agent budget may be difficult to enforce correctly because the agent runs in a distributed environment. The current proposal also needs to clarify its error-handling behavior and persistence and recovery semantics. Following the community’s current practice, I suggest opening a GitHub Discussion with a detailed design covering these semantics and trade-offs before moving to implementation. Please note that the community is currently focused on developing the planned features for the 0.4 release, so we may not have enough bandwidth to discuss the proposal in depth before the release. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
