Refresh Dataset

Re-pull this dataset's data from the origin it was imported from.

One request, no source pointer, no layer selection. The dataset keeps serving its current data throughout: the worker loads into an attempt-scoped staging table and swaps only once the new data is complete, so a refresh that fails leaves the live table and its freshness exactly as they were.

Different origin kinds use different execution strategies. A dataset registered from an existing PostGIS table serves directly from its origin, so refresh re-measures the live relation — recounting features, recomputing the extent, rebuilding the column schema snapshot and statistics. A dataset imported from a STAC item points to an external COG, so its refresh re-reads the item document and follows the asset if the publisher moved it. Admission, the run row and the history they write are identical across all three.

Refuses with 409 dataset_busy while another refresh or re-upload is active for this dataset. A partial unique index prevents simultaneous requests from both being admitted.

accept_blocked_run_id accepts a blocked run once. A blocked service refresh is fetched again and publishes only if the result matches the run it accepts. A blocked service re-upload is fetched again from the dataset's source and publishes only if its review reasons and changes match; it answers 409 origin_changed once that source is not the one the re-upload fetched. A blocked file replacement (an upload run) is replaced again from the upload that run kept, and publishes only if its review reasons and changes match. That answers 422 upload_unavailable once the upload is gone, and 409 review_superseded once newer data has replaced what the run was compared with. A review fingerprint is an acknowledgement, not a secret: it keeps a replacement nobody reviewed from publishing, and any caller with write access can still publish deliberately.