Grafana Table Panel Transformations Prompt

You are a senior observability engineer who shapes Grafana table panels with the transformation pipeline so multiple queries become one clean, sortable table.

I will provide:
- The queries/frames feeding the panel
- The desired columns and per-row grouping
- Any joins across data sources or metrics

Your job:

1. **Order matters**: transformations run top-to-bottom; sequence them (reduce/group first, then join, then organize) — reordering changes the output.
2. **Reduce time series**: apply `Reduce` (calc: last, max, mean) or `Group by` to collapse time series into one value per series before display.
3. **Join**: use `Merge` to stack frames, or `Join by field` (outer/inner) to combine metrics on a shared key (e.g. `pod`) into one row.
4. **Group by + calc**: `Group by` on a label with aggregations (count, sum, mean) per numeric field to build summaries.
5. **Organize fields**: use `Organize fields` to rename, reorder, and hide columns — this is where you make labels human-readable.
6. **Filter/derive**: `Filter data by values`, `Filter by name`, and `Add field from calculation` (binary ops, e.g. errors/total) to compute derived columns like error rate.
7. **Display**: set cell display mode (colored background/gauge), per-column units/thresholds via field overrides, and enable column filters + pagination.
8. **Labels to fields**: use `Labels to fields` so Prometheus labels become columns; combine with `Extract fields` for JSON blobs.

Mark DESTRUCTIVE: none — transformations are client-side and non-mutating. Flag `Filter` steps that silently drop rows a viewer expects to see.

---

Queries/frames: [DESCRIBE]
Desired columns/grouping: [DESCRIBE]
Joins needed: [DESCRIBE]

Why this prompt works

The table panel’s power is its transformation pipeline, but transformations are order-dependent and easy to misuse — an inner join silently drops rows, a filter hides data, a reduce masks spikes. This prompt sequences the pipeline correctly (reduce/group, then join, then organize) and calls out the silent data-loss steps so the table stays trustworthy.

How to use it

1. List the frames each query returns.
2. Name the final columns and the row key.
3. Sequence transformations deliberately, top-to-bottom.
4. Verify row counts after any join or filter.

Useful commands

# Inspect a panel's transformation pipeline
curl -s -H "Authorization: Bearer $TOKEN" \
  http://localhost:3000/api/dashboards/uid/inventory \
  | jq '.dashboard.panels[] | select(.type=="table") | .transformations'

Example config

Table panel joining two metrics per pod and computing error rate:

{
  "type": "table",
  "title": "Pods: traffic and error rate",
  "targets": [
    { "refId": "A", "expr": "sum by (pod) (rate(requests_total[5m]))", "format": "table", "instant": true },
    { "refId": "B", "expr": "sum by (pod) (rate(errors_total[5m]))",   "format": "table", "instant": true }
  ],
  "transformations": [
    { "id": "labelsToFields", "options": {} },
    { "id": "joinByField", "options": { "byField": "pod", "mode": "outer" } },
    { "id": "calculateField", "options": {
        "mode": "binary", "alias": "error_rate",
        "binary": { "left": "Value #B", "operator": "/", "right": "Value #A" } } },
    { "id": "organize", "options": {
        "renameByName": { "Value #A": "req/s", "Value #B": "err/s" },
        "indexByName": { "pod": 0, "req/s": 1, "err/s": 2, "error_rate": 3 },
        "excludeByName": { "Time": true } } }
  ],
  "fieldConfig": {
    "overrides": [
      { "matcher": { "id": "byName", "options": "error_rate" },
        "properties": [
          { "id": "unit", "value": "percentunit" },
          { "id": "custom.cellOptions", "value": { "type": "color-background" } },
          { "id": "thresholds", "value": { "mode": "absolute", "steps": [
            { "color": "green", "value": null }, { "color": "red", "value": 0.05 } ] } }
        ] }
    ]
  }
}

Common findings this catches

• Missing rows → inner join or value filter dropped them.
• Wrong aggregates → Group by used sum where mean was needed.
• Unreadable columns → labels never mapped via Organize/Labels-to-fields.
• Masked spikes → reduced to last instead of max.
• Order bugs → join placed before reduce.
• Falsely hidden data → sensitive column hidden, not excluded from the frame.

When to escalate

• Cross-datasource joins that strain the browser — move logic to the backend.
• Complex derived tables reused widely — library panel + documentation.
• Correctness disputes on aggregations — data/service owner.