Discover, assign and sync

Discovery is started from the cloud: open the source system's Data catalogue and press Discover catalogue (see On-prem tables). Tables appear within seconds, by name; the columns and key of a table are fetched when you Discover it, and Assign binds it to a dataset. From then on the dataset's Refresh now (Full or Incremental) is executed by the Gateway, and the Service tab lists the bound tables with their last watermark and detected strategy.

Change detection. Incremental syncs use SQL Server Change Tracking when the table has it, otherwise a watermark column that advances when a row changes (discovery guesses one from column names and types; ERP spellings such as CHANGEDDATE are recognised). A table without either is loaded as a snapshot every time. Manage source table integration on the catalogue row overrides the guess, forces snapshot, supplies key columns for a table whose driver reports no primary key, and switches on Read in windows for very large ODBC tables.

Deletes. A sync with Change Tracking carries deletions. Otherwise the Gateway periodically reads every key of the table and reports the ones that disappeared; how often is Setup → Runtime → On-prem delete check (default every 24 hours). A full load finds deletes on its own.

Loads that stay alive. The Gateway reports progress with every batch, so a long load keeps running as long as rows arrive. Batches are compressed and split when too large, and a cloud that is busy makes the batch wait rather than fail. A job that reports no progress for the On-prem job stall limit (default 30 minutes) is failed by the cloud so it does not block further refreshes.