When we append to an existing table (whether partitioned or not), we should make sure the append is an atomic operation. We either fail without appending any data to the destination table or we succeed and append all the new data.
If Dataflow does not offer any guarantee for atomic append, then we need to do it "manually": write output to a temporary BQ table and after successful run of VT append the new table to the exiting destination table.
When we append to an existing table (whether partitioned or not), we should make sure the append is an atomic operation. We either fail without appending any data to the destination table or we succeed and append all the new data.
If Dataflow does not offer any guarantee for atomic append, then we need to do it "manually": write output to a temporary BQ table and after successful run of VT append the new table to the exiting destination table.