Refresh a table from its source now
Start an import immediately. Where the provider supports it and the table has synced before, only records changed since the last successful run are fetched; otherwise everything is re-listed.
full_relist re-fetches every record without deleting anything. delete_missing_rows additionally removes rows whose records the source no longer returns — a separate, destructive ask, refused while the source has a row limit.
Refused while an import is already running: two concurrent imports would race the same rows.
Authorizations
Bearer authentication header of the form Bearer <token>, where <token> is your auth token.
Path Parameters
Body
What kind of refresh to run.
Two independent asks, not one dial. Re-fetching every record is safe and routine; deleting local rows whose records are gone is destructive and deliberate. Both default to off, so a plain refresh stays the cheap incremental one.
Delete imported rows whose records the source no longer returns. Implies a full re-list, and is refused while the source has a row limit — a capped import cannot tell a deleted record from one it never fetched.
Re-fetch every record instead of only those changed since the last run. Updates rows; never deletes any.
Response
Successful Response
One import run's progress and outcome.
Whether this run asked the provider for everything.
Records returned without the source's identity field, and so dropped.
Where one import run has got to.
The terminal states are absorbing: a run never leaves them. There is no PENDING here — unlike a CSV upload, which sits parsed and uncommitted until the user says go, a source run is only ever created at the moment something decides to run it.
CANCELLING is a request, not an outcome. Stopping an import is a database status flip the workflow observes between pages, never a Temporal terminate: a terminate could land mid-page and leave rows half-written with nothing to settle the run, which is exactly the stuck state that blocks every future refresh of the table.
running, cancelling, cancelled, succeeded, failed What asked for a run.
initial, manual, scheduled Classified cause of the failure: unauthorized, rate_limited, server_unavailable, network_error, not_found, pagination_error, bad_request. NULL when the run did not fail.
Whether fixing this means reconnecting the integration. Only true for a credentials failure, so the prompt keeps its meaning.