[MAPREDUCE-6323] Support pass-by-reference for ChainMapper/ChainReducer

• Type: Improvement
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: 2.4.1
• Component/s: mrv2, performance
• Assignee: Unassigned
• Reporter: Laurent Goujon
• Created: 21/Apr/15 15:44
• Updated: 21/Apr/15 15:44

Description

ChainMapper and ChainReducer using the new mapreduce API always copy the (key,value) pair between chain elements. It actually looks the copy is done twice, once when queueing, and once when dequeuing!

When transferred objects are immutable (or not reused), this is a waste of resource (especially when passing big records)