Space Management

The undo subsystem relies on the backend recycle thread to recycle free space. It recycles the space of the undo module on the primary node. As for the standby node, it recycles the space by replaying the Xlog. The recycle thread traverses the undo zones in use. The txn pages in the undo zone are scanned in the ascending order of XIDs. The transactions that have been committed or rolled back are also recycled. The commit time of transactions must be earlier than $(current_time – undo_retention_time). For a transaction that needs to be rolled back during a traversal, the recycle thread adds an asynchronous rollback task for the transaction.

When the database has transactions that run for a long time and contain a large amount of modified data, or it takes a long time to enable flashback, the undo space may continuously expand. When the undo space is close to the value specified by undo_space_limit_size, forcible recycling is triggered. As long as a transaction has been committed or rolled back, the transaction may be recycled even if it is committed later than $(current_time – undo_retention_time).