You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@hbase.apache.org by "Rich Howarth (JIRA)" <ji...@apache.org> on 2017/09/19 12:31:00 UTC

[jira] [Created] (HBASE-18847) Huge performance bottleneck in HFileWriterV2 in 1.2.6

Rich Howarth created HBASE-18847:
------------------------------------

             Summary: Huge performance bottleneck in HFileWriterV2 in 1.2.6
                 Key: HBASE-18847
                 URL: https://issues.apache.org/jira/browse/HBASE-18847
             Project: HBase
          Issue Type: Bug
    Affects Versions: 1.2.6
            Reporter: Rich Howarth


The below code block starts at line 277 of HFileWriterV2.java. Class-level synchronization in a heavily used code path has a demonstrably significant negative effect on performance. I tested forcing a major compaction with 18 compaction threads per node; removing the synchronization resulted in an order of magnitude performance increase, with the bottleneck then being at the disks (where I want it to be).



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)