What problem does this PR solve?

Issue Number: close #9875

Problem Summary:
This optimization can provide a significant performance boost when the data being read is not updated frequently.

What is changed and how it works?

Reduce redundant pack reads during bitmap filter building for delta merge case

Run CHBenchmark 1500: Overall 26.5% Improvement

mysql> select sum(DELTA_MEMTABLE_ROWS+DELTA_PERSISTED_ROWS) delta_rows, TIDB_TABLE from INFORMATION_SCHEMA.TIFLASH_SEGMENTS where TIDB_DATABASE='chbench_delta' group by TIDB_TABLE order by delta_rows;
+------------+------------+
| delta_rows | TIDB_TABLE |
+------------+------------+
|          0 | item       |
|         15 | region     |
|         75 | nation     |
|       4704 | warehouse  |
|      20000 | supplier   |
|      49790 | district   |
|     183924 | new_order  |
|    1022080 | history    |
|    1193288 | orders     |
|    1996224 | customer   |
|    7977425 | stock      |
|   41277141 | order_line |
+------------+------------+
12 rows in set (0.09 sec)

mysql> select sum(DELTA_MEMTABLE_ROWS+DELTA_PERSISTED_ROWS) delta_rows, sum(stable_rows) stable_rows, 100 * sum(DELTA_MEMTABLE_ROWS+DELTA_PERSISTED_ROWS) / sum(stable_rows) delta_percent from INFORMATION_SCHEMA.TIFLASH_SEGMENTS where TIDB_DATABASE='chbench_delta';
+------------+-------------+---------------+
| delta_rows | stable_rows | delta_percent |
+------------+-------------+---------------+
|   53724666 |  2283695006 |        2.3525 |
+------------+-------------+---------------+
1 row in set (0.09 sec)

The above table shows the delta rows for each table in chbench_delta database.

All Stable Performance Comparison

The following table presents the performance improvements for All Stable compared to both the original version (with Delta) and the optimized version (with Delta):

• After compacting the data, All Stable only shows a 5.86% improvement compared to the current optimized version with delta data.
• However, compared to the original version with delta data, All Stable shows a significant 33.93% improvement.

The chbenchmark data was inserted into another database. Then the compact table command is executed. As a result, the data might differ. So there are some differences in the query plans. For example, in Q20, the query plan uses hash partition join and includes late materialization for order_line table in the All Stable version, while original version uses broadcast join and no late materialization for order_line table. This causes Q20 to run slower in All Stable compared to the original version due to exchange overhead and low selectivity.

Check List

Tests

• Unit test
• Integration test
• Manual test (add detailed scripts or steps below)
• No code

Side effects

• Performance regression: Consumes more CPU
• Performance regression: Consumes more Memory
• Breaking backward compatibility

Documentation

• Affects user behaviors
• Contains syntax changes
• Contains variable changes
• Contains experimental features
• Changes MySQL compatibility

Release note

None