[FLINK-11882] Introduce BytesHashMap to batch hash agg

• Type: New Feature
• Status: Closed
• Priority: Major
• Resolution: Implemented
• Fix Version/s: 1.9.0
• Component/s: Runtime / Task
• Labels: pull-request-available
• Assignee: Jingsong Lee
• Reporter: Jingsong Lee
• Created: 12/Mar/19 07:55
• Updated: 15/Mar/19 07:58
• Resolved: 15/Mar/19 07:58

Description

Introduce bytes based hash table.
It can be used for performing aggregations where the aggregated values are fixed-width.
Because the data is stored in continuous memory, AggBuffer of variable length cannot be applied to this HashMap. The KeyValue form in hash map is designed to reduce the cost of key fetching in lookup.

Add a test to do a complete hash agg. When HashMap has enough memory, pure hash AGG is performed; when memory is insufficient, it degenerates into sort agg.

Issue Links

• is a child of: FLINK-11490 Add an initial Blink SQL batch runtime
• links to: GitHub Pull Request #7961