Infra Overview dashboard

The repository contains one dashboard JSON: infra-overview.json.

This file retains the original Grafana 13 dashboard V2 JSON model from dev / main, with its existing elements, layout, annotations, timeSettings, and other top-level fields. It preserves the original 42 panel elements (including the eight NVIDIA P4000 panels) and their queries unchanged, and adds 13 Lenovo host/CI runner panels under Lenovo CI (lenovo-lap). There is no Classic-format conversion and no second dashboard JSON or deployment script.

The preexisting dashboard is managed inside Grafana, not file-provisioned. If editing the existing dashboard's JSON Model, use this spec-only JSON in the same editor/workflow that successfully handled the original exported JSON. Back up any live-only edits before replacing the model; a merge may be necessary if the live database diverged from the repo. The PR does not deploy Grafana changes by itself.

Local structural check

python3 -m json.tool docs/services/dashboards/infra-overview.json >/dev/null
python3 - <<'PY'
import json
from pathlib import Path
d = json.loads(Path("docs/services/dashboards/infra-overview.json").read_text())
assert "elements" in d and "layout" in d and "panels" not in d
assert len(d["elements"]) == 55
assert len(d["layout"]["spec"]["rows"]) == 5
for row in d["layout"]["spec"]["rows"]:
    for item in row["spec"]["layout"]["spec"]["items"]:
        assert item["spec"]["element"]["name"] in d["elements"]
ids = [p["spec"]["id"] for p in d["elements"].values()]
assert len(ids) == len(set(ids))
assert d["layout"]["spec"]["rows"][-1]["spec"]["title"] == "Lenovo CI (lenovo-lap)"
assert len(d["layout"]["spec"]["rows"][-1]["spec"]["layout"]["spec"]["items"]) == 13
print("Grafana V2 structural checks passed")
PY

These structural checks are not Grafana server-side tests. On 2026-11-03 the operator separately confirmed the live Grafana 13 V2 dashboard renders correctly, including Caddy/GPU panels, Lenovo host metrics, runner CPU/RAM host percentages, and job transitions. No exported post-validation live dashboard JSON was provided for comparison with this repository file.

Lenovo CI row (deployed and operator-reviewed 2026-11-03)

The fifth row groups 13 panels: host up, CPU, memory, root disk; runner exporter health; backend and infrastructure container running indicators; current log-inferred job names; and each container's CPU and RAM percentage of host capacity. All metrics originate from the three live Prometheus scrape jobs on monitor-01: lenovo-lap-node, lenovo-lap-containers, and lenovo-lap-runners.

For future changes: export/backup the live Grafana dashboard model before applying a revised repository version. Its structure is the Grafana 13 V2 / Scenes model, not an importable classic panels JSON. If live-only changes have been made since the last committed export, reconcile them instead of overwriting them. After updating in the Grafana editor, inspect the fifth row; verify both runner gauges, and test a live job without restarting any runner. Changes to GitHub Actions registration are not required.

The current-job panels show None running when the log observer is healthy and has observed an idle state. They show no data if the Prometheus scrape itself fails. They are not authoritative GitHub online/busy telemetry. Container-CPU and RAM panels use the dedicated Lenovo exporter’s gha_runner_container_cpu_host_percent and gha_runner_container_memory_host_percent gauges, not cAdvisor's container-name labels. Both require deployment of the updated exporter and gha_runner_container_stats_up=1 for the corresponding runner. Container status panels use or on() vector(0) to produce a single 0/1 series, rather than displaying simultaneous UP and DOWN values.