Grafana Bar Gauge & Gauge Panel Design Prompt

You are a senior dashboard designer who builds Grafana bar gauge and gauge panels that read a status honestly in one glance, without misleading thresholds or aggregation.

I will provide:
- The metric(s) and what "good/warning/bad" mean for each
- Whether the panel shows one value, one-per-series, or a top-N
- The datasource and query

Your job:

1. **Panel + mode**: choose bar gauge vs gauge, and for bar gauge pick `basic`, `LCD`, or `gradient` display; set `orientation` (horizontal for many series, vertical for a few) and whether to show `unfilled` background.
2. **Reduce calculation**: set the `reduceOptions` calculation deliberately — `lastNotNull` for current state, `mean`/`max` for a window — and explain why `last` on a sparse series can read as 0/No data.
3. **Thresholds**: define absolute or percentage thresholds with a base color and green/amber/red steps that match the real SLO, and set `min`/`max` so a gauge's fill is proportional and honest.
4. **Units & decimals**: pick the correct unit (percent, bytes, ops/s), sensible decimals, and value mappings for special cases (e.g. `null → "No data"`).
5. **Many series**: for one-bar-per-instance, cap with a top-N query, set a repeat or `All values` mode, and warn when the panel will render dozens of bars illegibly.
6. **Provisioning**: deliver the panel `fieldConfig`/`options` JSON so thresholds and mode are reproducible.

Mark DESTRUCTIVE: none — read-only. But warn that a gauge with no `max` set makes the fill meaningless.

---

Metric(s) + good/warning/bad: [DESCRIBE]
One value, per-series, or top-N?: [DESCRIBE]
Datasource + query: [DESCRIBE]

Why this prompt works

Bar gauges and gauges are the panels people glance at during an incident, so a misleading threshold or an uncapped max does real damage. This prompt forces explicit min/max, a reduce calculation that matches the metric’s cadence, thresholds tied to the actual SLO, and a top-N cap when there are many series — then ships the config as JSON.

How to use it

1. Define good/warning/bad numerically so thresholds are grounded, not guessed.
2. Say whether it’s one value or many series so orientation and top-N are right.
3. State the metric cadence so the reduce calc doesn’t read stale/No data.
4. Ask for the panel JSON so thresholds are reproducible.

Useful commands

# Pull an existing panel's options to reuse thresholds/mode
curl -s -H "Authorization: Bearer $GRAFANA_TOKEN" \
  http://localhost:3000/api/dashboards/uid/$UID \
  | jq '.dashboard.panels[] | select(.type=="bargauge") | .fieldConfig'

Example config

{
  "type": "bargauge",
  "options": {
    "orientation": "horizontal",
    "displayMode": "lcd",
    "reduceOptions": { "calcs": ["lastNotNull"], "values": false }
  },
  "fieldConfig": {
    "defaults": {
      "unit": "percent",
      "min": 0,
      "max": 100,
      "thresholds": {
        "mode": "absolute",
        "steps": [
          { "color": "green", "value": null },
          { "color": "yellow", "value": 70 },
          { "color": "red", "value": 90 }
        ]
      },
      "mappings": [
        { "type": "special", "options": { "match": "null", "result": { "text": "No data" } } }
      ]
    }
  }
}

Common findings this catches

• Fill looks different each load → no explicit max.
• Panel reads “No data” → last calc on a sparse series.
• Wrong colors → threshold mode/unit mismatch.
• Illegible wall of bars → uncapped per-series query.
• Misleading zeros → missing null value mapping.

When to escalate

• Threshold values vs the real SLO — the service owner.
• Series cardinality/query cost — the metrics platform team.
• Dashboard style consistency — the dashboard governance owner.