[IGNITE-12609] SQL: GridReduceQueryExecutor refactoring.

• Type: Task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.9, 2.8.1
• Component/s: sql
• Labels: refactoring
• Assignee: Andrey Mashenkov
• Reporter: Andrey Mashenkov
• Created: 30/Jan/20 13:38
• Updated: 21/Apr/20 09:38
• Resolved: 12/Feb/20 09:39

Description

For now we have few issues that can be resolved.

1. We create fake H2 tables\indices for reduce stage even if there is no need to do so (skipMergeTable=true.
Let's decouple reduce logic from H2Index adapter code.

2. Partition mapping code look to complicated and non-optimal.
Let's use cached affinity mapping and avoid collections copying when possible.

3. Also there is no sense to pass RequestID to mapping code just for logging.
We'll never be able to match any request as no query really exists at a time when error with RequestID is logged.

4. Replicated only flag value semantic (calculation and usage) is not clear.

5. GridReduceQueryExecutor.reduce() method is too long (over 400 lines).

Issue Links

• links to: GitHub Pull Request #7361