[HBASE-20526] multithreads bulkload performance

• Type: Improvement
• Status: Patch Available
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: 1.2.5, 1.3.2
• Component/s: mapreduce, Zookeeper
• Labels: performance

Environment

hbase-server-1.2.0-cdh5.12.1 

spark version 1.6

• Assignee: Key Hutu
• Reporter: Key Hutu
• Created: 03/May/18 10:01
• Updated: 09/Aug/19 01:21

Description

When doing bulkload , some interactive with zookeeper to getting region key range may be cost more time.

In multithreads enviorment, the duration maybe cost 5 minute or more.

From the executor log, like 'Reading reply sessionid:0x262fb37f4a07080 , packet:: clientPath:null server ...' contents appear many times.

 

It likely to provide new method for bulkload, caching the key range outside