During a source-based commit session, the Integration Service commits data to the target based on the number of rows from some active sources in a target load order group. These rows are referred to as source rows.

When the Integration Service runs a source-based commit session, it identifies commit source for each pipeline in the mapping. The Integration Service generates a commit row from these active sources at every commit interval. The Integration Service writes the name of the transformation used for source-based commit intervals into the session log:

The Integration Service might commit less rows to the target than the number of rows produced by the active source. For example, you have a source-based commit session that passes 10,000 rows through an active source, and 3,000 rows are dropped due to transformation logic. The Integration Service issues a commit to the target when the 7,000 remaining rows reach the target.

The number of rows held in the writer buffers does not affect the commit point for a source-based commit session. For example, you have a source-based commit session that passes 10,000 rows through an active source. When those 10,000 rows reach the targets, the Integration Service issues a commit. If the session completes successfully, the Integration Service issues commits after 10,000, 20,000, 30,000, and 40,000 source rows.

If the targets are in the same transaction control unit, the Integration Service commits data to the targets at the same time. If the session fails or aborts, the Integration Service rolls back all uncommitted data in a transaction control unit to the same source row.

If the targets are in different transaction control units, the Integration Service performs the commit when each target receives the commit row. If the session fails or aborts, the Integration Service rolls back each target to the last commit point. It might not roll back to the same source row for targets in separate transaction control units.