Design efficient event-driven job monitoring for agents

Design an agent-facing job-monitoring interface that delivers meaningful scheduler state changes efficiently, remains responsive, and avoids recreating the control-plane overhead of intensive polling.

Background

Agents poll schedulers much more aggressively than human users, creating super-linear control-plane pressure as job submissions increase. Event-driven notifications could reduce unnecessary polling, but may omit context or fail to capture conditions requiring continuous observation. A lightweight aggregation and notification service is another possibility, yet the paper leaves unresolved how to provide responsive and sufficiently informative monitoring without reproducing polling overhead.

References

An open question is therefore who should be responsible for monitoring jobs and how status updates should be delivered efficiently.

— Towards Efficient HPC Systems for Agents: Challenges and Opportunities  (2609.38723 - Zheng et al., 30 Sep 2026) in Section Compute, subsection “Opportunities: Trial-Aware, Hardware-Aware, and Event-Driven Scheduling”