From 3d8a00b9c096757af61515edfcb903af7d94f580 Mon Sep 17 00:00:00 2001 From: aas008 Date: Fri, 2 Oct 2026 08:16:21 -0400 Subject: [PATCH] feat(dashboard): add Grafana link routing for Athena and Janus clusters Adds _is_athena_run() and _is_janus_run() routing functions and corresponding GRAFANA_DASHBOARDS entries for H200_ATHENA and H200_JANUS. Both point to the combined multi-cluster dashboard (df28i54z7okcgf) with the cluster pre-filtered via var-cluster_name, so runs from these clusters get a direct Grafana link like any other H200 cluster. SGLang runs from Athena/Janus route to the existing SGLang dashboard. Co-Authored-By: Claude Sonnet 4.6 (1M context) --- dashboard.py | 32 +++++++++++++++++++++++++++++++- 1 file changed, 31 insertions(+), 1 deletion(-) diff --git a/dashboard.py b/dashboard.py index 479532a..6c36778 100644 --- a/dashboard.py +++ b/dashboard.py @@ -10408,6 +10408,16 @@ def convert_epoch_to_date(row): "dashboard_id": "d35f19c8f56250", "dashboard_name": "vllm-2b-dcgm-metrics-psap-zeus-syd", }, + "H200_ATHENA": { + "dashboard_id": "df28i54z7okcgf", + "dashboard_name": "vllm-2b-dcgm-complete-metrics-combined", + "extra_params": "&var-cluster_name=psap-fire-athena", + }, + "H200_JANUS": { + "dashboard_id": "df28i54z7okcgf", + "dashboard_name": "vllm-2b-dcgm-complete-metrics-combined", + "extra_params": "&var-cluster_name=psap-de-h200-cluster", + }, "B200_PIRATE": { "dashboard_id": "b200-pirate-vllm-dcgm", "dashboard_name": "vllm-2b-dcgm-metrics-b200-pirate", @@ -10423,6 +10433,8 @@ def convert_epoch_to_date(row): "H200": _SGLANG_H200_DASHBOARD, "H200_HERA2": _SGLANG_H200_DASHBOARD, "H200_ZEUS2": _SGLANG_H200_DASHBOARD, + "H200_ATHENA": _SGLANG_H200_DASHBOARD, + "H200_JANUS": _SGLANG_H200_DASHBOARD, } # Jan 1, 2026 00:00:00 UTC in milliseconds @@ -10449,6 +10461,20 @@ def _is_zeus2_run(run_name): upper = run_name.upper() return upper.startswith("H200-ZEUS2-") or upper.startswith("H200_ZEUS2-") + def _is_athena_run(run_name): + """Check if a run belongs to the Athena cluster.""" + if not isinstance(run_name, str): + return False + upper = run_name.upper() + return upper.startswith("ATHENA-") or upper.startswith("ATHENA_") + + def _is_janus_run(run_name): + """Check if a run belongs to the Janus cluster.""" + if not isinstance(run_name, str): + return False + upper = run_name.upper() + return upper.startswith("JANUS-") or upper.startswith("JANUS_") + def create_grafana_link(row): """Create Grafana dashboard link if timestamps are available.""" start_time = row.get("guidellm_start_time_ms") @@ -10472,7 +10498,11 @@ def create_grafana_link(row): end_ms = int(end_time) # Determine which dashboard to use based on accelerator, cluster, and date - if _is_hera2_run(run_name): + if _is_athena_run(run_name): + dashboard_key = "H200_ATHENA" + elif _is_janus_run(run_name): + dashboard_key = "H200_JANUS" + elif _is_hera2_run(run_name): dashboard_key = "H200_HERA2" elif _is_hera_run(run_name): dashboard_key = "H200_HERA"