+Added flow control capacity utilization gauges. `llm_d_epp_flow_control_capacity_utilization_requests` and `llm_d_epp_flow_control_capacity_utilization_bytes` report occupancy as a fraction of a priority band's effective capacity (0.0-1.0), labelled by `priority` and `inference_pool`. `llm_d_epp_flow_control_global_capacity_utilization_requests` and `llm_d_epp_flow_control_global_capacity_utilization_bytes` report the all-bands rollup, emitted only when a global capacity is configured. Operators can alert on "band at N% of its limit" without joining config values into PromQL.
0 commit comments