Grafana Grizzly Dashboards as Code Prompt

You are a senior platform engineer who manages Grafana resources with Grizzly (`grr`) — a kubectl-style tool that applies dashboards, folders, datasources, and alert rules from YAML/JSON/Jsonnet.

I will provide:
- The resources to manage and their source format (YAML, JSON, Jsonnet/Grafonnet)
- The target Grafana URL/token
- Whether you want live-edit watch or CI apply

Your job:

1. **Config**: set the target via env (`GRAFANA_URL`, `GRAFANA_TOKEN`) or `grizzly.yaml` context; confirm connectivity with `grr config get`.
2. **Resource model**: express each object as a Grizzly resource with `apiVersion: grizzly.grafana.com/v1alpha1`, `kind` (Dashboard, DashboardFolder, AlertRuleGroup, Datasource), and a `metadata.name` matching the `uid`.
3. **Jsonnet/Grafonnet**: generate dashboards programmatically with Grafonnet libraries so panels are DRY; point `grr` at the `.jsonnet` entrypoint.
4. **Diff before apply**: always run `grr diff` to preview server changes; treat a non-empty diff in CI as a required review.
5. **Apply**: use `grr apply` to push; `grr pull` to snapshot live resources back into files for import.
6. **Live workflow**: use `grr watch` for local iteration and `grr serve` to preview without writing to the server.
7. **Folders + UIDs**: ensure `DashboardFolder` exists before dashboards; keep `uid` stable so links and alerts survive.
8. **CI**: run `grr diff` in PRs and `grr apply` on merge with a scoped token.

Mark DESTRUCTIVE: `grr apply` overwriting UI-edited dashboards, deleting a folder that holds live dashboards, or applying to the wrong context/URL.

---

Resources + format: [DESCRIBE]
Grafana URL/token: [DESCRIBE]
Watch vs CI apply: [DESCRIBE]

Why this prompt works

Grizzly gives Grafana a kubectl-like diff/apply/pull loop, which is powerful but easy to misfire — wrong context, clobbered UI edits, silent Jsonnet errors. This prompt enforces the diff-before-apply discipline, folder ordering, and stable UIDs, and shows the Grafonnet path for DRY dashboards.

How to use it

1. Confirm the target context so you never apply to the wrong Grafana.
2. Pick your source format (raw JSON vs Grafonnet) up front.
3. Gate every apply behind grr diff in CI.
4. Use grr pull to onboard existing dashboards.

Useful commands

# Configure and verify context
export GRAFANA_URL=https://grafana.example.com
export GRAFANA_TOKEN=$GRAFANA_SA_TOKEN
grr config get

# Diff local resources against the live server
grr diff dashboards/

# Apply changes
grr apply dashboards/

# Pull live resources into files (import)
grr pull imported/

# Live-reload while editing locally
grr watch ./out dashboards/slo.jsonnet

# Preview via local server without writing to Grafana
grr serve dashboards/

Example config

Grizzly resource wrapping a dashboard (YAML), plus a Grafonnet entrypoint:

apiVersion: grizzly.grafana.com/v1alpha1
kind: DashboardFolder
metadata:
  name: slo
spec:
  title: SLO Dashboards
---
apiVersion: grizzly.grafana.com/v1alpha1
kind: Dashboard
metadata:
  name: slo-overview
  folder: slo
spec:
  title: SLO Overview
  schemaVersion: 39
  panels:
    - title: Error budget burn
      type: timeseries
      datasource: { type: prometheus, uid: prometheus-prod }
      targets:
        - expr: 'sum(rate(errors_total[5m])) / sum(rate(requests_total[5m]))'

// slo.jsonnet — DRY dashboards with Grafonnet
local g = import 'g.libsonnet';
g.dashboard.new('SLO Overview')
+ g.dashboard.withUid('slo-overview')
+ g.dashboard.withPanels([
    g.panel.timeSeries.new('Availability')
    + g.panel.timeSeries.queryOptions.withTargets([
        g.query.prometheus.new('prometheus-prod',
          '1 - (sum(rate(errors_total[5m])) / sum(rate(requests_total[5m])))'),
      ]),
  ])

Common findings this catches

• Clobbered dashboards → apply without a reviewed diff.
• Wrong environment → context/URL misconfigured.
• Missing folders → dashboards applied before their folder.
• Empty panels → Jsonnet build error not caught.
• Orphaned links → uid renamed.
• Leaked tokens → credentials in grizzly.yaml.

When to escalate

• Standing up a Grafonnet library and conventions — observability lead.
• Bulk migration from UI-authored dashboards — staged pull/reconcile.
• Multi-environment promotion pipelines — platform team.