signalk-grib-downloader

Signal K plugin that schedules GRIB downloads for signalk-grib-weather-provider — automatic after each model run, or on demand for intermittent connections — with a management webapp.

What it does

Each configured source (model + resolution) downloads into <root>/<model>-<resolution>/ through a one-shot grib-downloader container job run via the signalk-container plugin. Point signalk-grib-weather-provider's root at the same directory and every source is served as a weather provider automatically.

Supported models: GFS (0.25°/0.5°/1°, NOMADS server-side area subsetting), AROME (0.025°/0.01°), ARPEGE (0.1°/0.25°), ICON-EU.

Modes

• Auto — knows each model's publication schedule (cadence + delay) and downloads every run as soon as it is published. Run availability is probed upstream, so a not-yet-published run shows as awaiting publication, not an error.
• Manual — nothing downloads without an explicit trigger (webapp button or POST /plugins/signalk-grib-downloader/download[/<source>]). Made for metered satellite connections.

The webapp

Everything operational is managed in the webapp (Webapps → GRIB Downloader), not in the plugin config panel:

• Download area drawn on a map (draggable rectangle corners) — applied to GFS, the only model with server-side subsetting
• Sources: model, resolution, forecast duration on a slider showing the model maximum, past-runs archive count, enable/disable
• Volume estimation per source and per run, following the area and duration live
• Status: per-source badges (up to date / downloading / awaiting publication / settings changed / behind / error) with the last download log expandable for analysis
• Deleting a source optionally deletes its downloaded data

Settings are stored in the plugin's data directory (settings.json), written only through the webapp — the admin config panel only holds infrastructure (GRIB root directory, downloader image).

Data lifecycle

• A run is downloaded atomically and finished with a marker recording the fetch fingerprint (area, duration, groups, variables…). Changing any of those re-fetches the data; nothing goes stale silently.
• When a newer run completes, the previous one is deleted — or moved to <source>/archive/ if the source keeps archives (configurable count, no upper limit). Leftovers of interrupted runs are purged.
• Old data is never deleted while offline: cleanup only happens when a newer run has been fully downloaded.

Requirements

• Signal K server ≥ 2.x
• signalk-container with a working container runtime (declared via signalk.requires — the App Store installs it automatically)
• The ghcr.io/macjl/grib-downloader image (pulled automatically on first use)
• signalk-grib-weather-provider ≥ 0.2.0 to serve the downloaded data

License

Apache-2.0