Grafana Dashboard Extractor

A GENERIC "point at any Grafana instance" Apify actor — the dashboard-inventory sibling of the netdata / gatus / uptime-kuma observability runners, over a different, far larger platform population. Grafana (grafana/grafana, ~66k★) is THE self-hosted dashboarding / observability platform, on virtually every observability stack, and a large fraction of instances leaves the HTTP API anonymous-readable — so ONE actor spans every instance, no per-host scraper and no host enumeration.

Input

Field	Mode	Meaning
sources (required)	all	One or more Grafana base URLs — https://play.grafana.org, https://grafana.wikimedia.org, or any self-hosted instance. The /api/... path is appended for you.
mode	—	search (default), health, or discovery.
query	search	Case-insensitive substring matched against dashboard/folder titles.
type	search	dash-db (dashboards) or dash-folder (folders). Empty = both.
tag	search	One or more tags (AND-ed by Grafana).
maxRecords	all	Global cap across every source (one row = one billable event).
bearer	all	Optional API key / service-account token (Authorization: Bearer). Never logged.
extraHeaders	all	Optional extra request headers (JSON).

Modes

• search (default) — paginated /api/search → one flat row per dashboard/folder: id, uid, title, kind (dashboard/folder/other), type, url, tags, isStarred, folderId/folderUid/folderTitle, plus the instance _version / _grafanaBuildCommit (best-effort from a cheap health probe) and a lossless _raw.
• health — /api/health (+ best-effort /api/frontend/settings) → one row per instance: database, version, commit, edition, and the non-secret auth fingerprint (disableLoginForm, samlEnabled, ldapEnabled, oauthProviders [names only], featureToggleCount).
• discovery — health + settings + a bounded search count → one summary row per instance: version, edition, dashboardCount, and authDisabled (did anonymous /api/search return 200?). A cheap way to fingerprint many instances at once.

Behaviour

• Every field is read by key presence (dict.get) — Grafana's /api/search + /api/health shapes drift across the 8→13 major versions and instances populate different optional fields, so an entry omitting folderTitle / tags / sortMeta yields null / [] for that column, never a crash. Unknown fields survive in _raw.
• Many instances gate /api/search (401/403) but leave /api/health open. In search mode, such a source falls back to a discovery row (the instance fingerprint) instead of yielding nothing.
• A source behind an anti-bot interstitial (a non-JSON HTTP 200), unreachable (Cloudflare 52x), or returning a non-Grafana body is a clean skip-with-warning — the runner continues with the other sources. The run only fails fast (exit 91) when every source was skipped, so nothing broken ships and the cloud smoke gate stays honest.
• A valid instance with 0 matching dashboards → 0 records and a clean exit 0. Unpopular is fine; the actor never fabricates rows.

Pricing

Pay-per-event: one record charge per emitted dataset row.

Each row is charged as it is saved, within the maximum cost you set for the run. The Actor stops before it requests a search page or an instance your budget cannot pay for, so you never receive rows you did not pay for. If an instance fails after some rows were delivered, those rows stay in your dataset and the run is marked failed, so a partial export never looks complete.

Development

python3.12 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
python tests_logic.py          # pure-logic tests (no network)
apify validate-schema
apify run                       # local run (set storage/key_value_stores/default/INPUT.json)