Help Center/ GaussDB/ Distributed_3.x/ Storage Engine/ Ustore/ Storage Format/ Index/ Inserting, Deleting, Updating, and Scanning UB-Tree

Updated on 2024-07-04 GMT+08:00

View PDF

Inserting, Deleting, Updating, and Scanning UB-Tree

Insert: The insertion logic of UB-tree is basically not changed, except that you need to directly obtain the transaction information and fill in the xmin column during index insertion.
Delete: The index deletion process is added for UB-tree. The main procedure of index deletion is similar to that of index insertion. That is, obtain the transaction information, fill in the xmax column (The B-tree index does not maintain the version information so that the deletion operation is required.), and update active_tuple_count on pages. If active_tuple_count is reduced to 0, the system attempts to recycle the page.
Update: For Ustore, data update operations on UB-tree index columns are different from those on Astore. Data update includes index column update and non-index column update. The following figure shows the processing of UB-tree data update.

The preceding figure shows the differences between UB-tree updates in index columns and non-index columns.
1. When a non-index column is updated, the index does not change. The index tuple points to the data tuple inserted at the first time. The Uheap does not insert a new data tuple. Instead, the Uheap modifies the current data tuple and saves historical data to the undo segment.
2. When the index column is updated, a new index tuple is inserted into UB-tree and points to the same data linepointer and data tuple. To scan the data of historical versions, you need to read it from the undo segment.
Scan: When reading data, you can use index to speed up scanning. UB-tree supports multi-version management and visibility check of index data. The visibility check at the index layer improves the performance of index scanning and index-only scanning.
For index scanning:
1. If the index column contains all columns to be scanned (index-only scanning), binary search is performed on indexes based on the scanning conditions. If a tuple that meets the conditions is found, data is returned.
2. If the index column does not contain all columns to be scanned (index scanning), binary search is performed on indexes based on the scanning conditions to find TIDs of the tuples that meet the conditions, and then the corresponding data tuples are found in data tables based on the TIDs. See the following figure.