Description
As of now, the size of every table in cluster is hardcoded to 10k rows, which prevents cost-based planner from providing best plan.
In IGNITE-22619 method returning estimated size of a particular table has been introduced. Sql engine may take an advantage of this method in order to improve quality of returned plans.
Let's incorporate proper table size estimation in query optimization pipeline. Note: call to org.apache.ignite.internal.table.InternalTable#estimatedSize results in distributed call to primary replicas of all partitions of the table, those it's better to throttle it. Also, table size estimation must return the same number of rows within a single optimization session.
Attachments
Issue Links
- fixes
-
IGNITE-19558 Sql. Use statistics snapshot per sql request.
-
- Resolved
-
- links to
