Open hanfei1991 opened 6 years ago
This paper introduces a way to reorder join tables during execution by using the information of bloom filter. It builds multiple hash tables(and bloom filters) for each table concurrently, then probed by one biggest table.
Should this one also be in executor component too? @SunRunAway
I suppose to close this issue, planner and executor should both be involved.
@SunRunAway Thanks for the paper, I'll read it.
I suppose to close this issue, planner and executor should both be involved.
You can use this issue to trace all the progress. For example: proposal -> develop tasks -> tests -> user documents
For Multi-Way Hash Join: Build a single hash table, probed by multiple other tables. The performance can be obtained to avoid building hash table multiple times.
For Multi-Way Merge Join: compare join keys belonging to different tables at the same time. All the join keys are only compared once. Saved many compare operations performed on the current two-way Merge Join
For Multi-Wat Index Join: We can read from a single table, transform join key into index keys for different tables to read multiple tables according to the index at the same time. Which can saves a lot of index key construction operations on the current two-way Index Join.