It sounds like you need a full copy of the data so you can discard all rows except the one you want to update. This will eliminate all of the state changes to your current datawindow, and should provide the shortest code path (although copying large quantities of data into the blob & back out to datastore multiple times might be slower).

Conceptually I think this should work. Additional error checking is probably required.

blob lblob_fullstate
long ll_rows, ll_ctr, ll_b4rows, ll_afterrows
datastore ds

ds = create datastore
ds.settransobject(sqlca)

ll_rows = dw.getfullstate(lblob_fullstate)

for ll_ctr = 1 to dw.rowcount()
	// check row modified status, if it's not changed, ignore and go to next
	if dw.getitemstatus(ll_ctr, 0, primary!) = NotModified! then
		continue
	end if
	ds.SetFullState(lblob_fullstate)
	ll_b4rows = ll_ctr - 1
	ll_afterrows = ll_counter + 1

	if ll_afterrows < dw.rowcount() then
		ds.rowsdiscard(ll_afterrows, dw.rowcount(), Primary!)
	end if

	if ll_b4rows > 0 then
		ds.rowsdiscard(1, ll_b4rows, primary!)
	end if

	ds.update()
	// do external call
	if external success then
		dw.SetItemStatus(ll_ctr, 0, NotModified!)
		commit using sqlca;
	else
		rollback using sqlca;
	end if
end if