Replicat abending with database error 1403() during the initial startup during a compressed update.
This error can be seen very frequently on a production environment when implementing Goldengate for the first time.
During the instantiation of the target database, if there is any open transaction which is yet to be committed, the extract is started with “begin now” and then the export/rman backup is started. This open transaction could be missed out by either of the processes and when this open transaction is committed at a later point of time and an update/delete is fired against this transaction which is captured by the extract, guess what happens?
Replicat would not be able to find the row for doing the latter transaction.
There is a way to avoid this situation by altering the extract to the oldest long-pending transaction from gv$transaction/v$transaction.
Use the below to find:
SQL> select min(START_TIME) from gv$transaction;
Alter the extract with the begin timestamp to process the oldest open transaction:
# delete the existing extract and pump trails
# create a new trail sequence for extract and point the pump to read the new trail sequence
# ggsci> alter EXTRACT BEGIN 2016-10-25 21:13:04
# ggsci> alter EXTRACT BEGIN 2016-10-25 21:13:04, threads
# alter the replicat to the new trail sequence and start the processes.
Of course remember to use HANDLECOLLISIONS, until you pass by the time your ORIGINAL TRAIL was processed until.
And for NEXT TIME make sure you run your GV$TRANSACTION sql above. And create the extract process to start capturing from the time you get from GV$TRANSACTION.