SGLang waiting-queue gauge

SGLang main, python/sglang/srt/observability/metrics_collector.py, read 2026-10-04. NOT ON BY DEFAULT: nothing is exported unless --enable-metrics (ServerArgs default False), which is also the only thing that mounts /metrics on the server port. Label keys are model_name, engine_type, tp_rank, pp_rank, moe_ep_rank (plus dp_rank under data parallelism, priority under priority scheduling, and anything in extra_metric_labels). Bucket ladders are overridable per histogram via get_histogram_conf_from_env. 'The number of requests in the waiting queue.' This is the waiting-queue depth; the running count is sglang:num_running_reqs. There is no by-reason split of the waiting set - the vLLM-only capability - but disaggregated deployments get separate queue gauges per stage instead.