This section discusses the recovery procedures to use if data loss or problems occur as a result of alter table changes.
Executor thread problems require no extra recovery. Data may be discarded when there is a normalization error in an executor thread.
When datatype conversion has completed, and if the table has been altered such that the executor thread cannot normalize the data, the data row may be discarded. Use the “Migration procedure” to avoid data loss.
If data in the inbound queue is incompatible with the column datatype in a replication definition, the distributor thread may shut down. The resume distributor command has been extended to allow you to skip one transaction:
resume distributor ds.db skip transaction
When there is bad data in the outbound queue, use resume connection skip [n] transaction to skip the bad data. In the case of replicate (not standby) Data Server Interface (DSI) threads, you may be able to alter the replication definition and resume the DSI to recover the data.
