Overview

Ustore is an in-place update storage engine launched by GaussDB. The biggest difference between Ustore and Astore lies in that, the latest data and historical data (excluding indexes) are stored separately.

Ustore Advantages

The latest data and historical data are stored separately. Compared with Astore, Ustore has a smaller scanning scope. The HOT chain of Astore is removed. Non-index columns, index columns, and heaps can be updated in-place without change to row IDs. Historical versions can be recycled in batches, and space bloat is controllable.
The transaction information is added to B-tree indexes to perform MVCC independently. This increases the proportion of IndexOnlyScan and greatly reduces the number of times that TABLE ACCESS BY INDEX ROWID is executed.
VACUUM is not the only way to clear historical data. Spaces are recycled independently. Indexes are decoupled from heap tables and can be cleared independently. The I/O stability is better.
In the scenario where a large number of concurrent updates are performed on the same row, row ID offset may occur in an Astore table. The in-place update mechanism of Ustore ensures the stability of tuple row IDs and update latency.
The flashback function is supported.

When modifying data pages, Ustore DML operations also trigger undo logs. Therefore, the update cost is higher. In addition, the scanning overhead of a single tuple is high because of replication (Astore returns pointers).