[
https://issues.apache.org/jira/browse/HBASE-12431?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14201615#comment-14201615
]
Jingcheng Du commented on HBASE-12431:
--------------------------------------
The test results are listed following.
Before applying the patch
loadValue(): 260087188618
getValue(): 280990331150
After applying the patch
loadValue(): 262340481358
getValue(): 281435842873
In loadValue(), it shows 0.87% performance degrade.
In getValue(), it shows 0.16% performance degrade.
> Use of getColumnLatestCell(byte[], int, int, byte[], int, int) is Not Thread
> Safe
> ---------------------------------------------------------------------------------
>
> Key: HBASE-12431
> URL: https://issues.apache.org/jira/browse/HBASE-12431
> Project: HBase
> Issue Type: Bug
> Components: Client
> Affects Versions: 0.98.1
> Reporter: Jonathan Jarvis
> Assignee: Jingcheng Du
> Fix For: 2.0.0, 0.98.8, 0.99.2
>
> Attachments: HBASE-12431-V2.diff, HBASE-12431-V3.diff,
> HBASE-12431.diff
>
>
> Result declares that it is NOT THREAD SAFE at the top of the source code, but
> one would assume that refers to many different threads accessing the same
> Result object. I've run into an issue when I have several different threads
> accessing their own Result object that runs into an issue because of use of
> common static member variable.
> I noticed the problem when I switched from:
> getColumnLatestCell(byte[], byte[]) to
> getColumnLatestCell(byte[], int, int, byte[], int, int)
> These methods call different binarySearch methods, the latter invoking:
> protected int binarySearch(final Cell [] kvs,
> 309 final byte [] family, final int foffset, final int flength,
> 310 final byte [] qualifier, final int qoffset, final int qlength) {
> This method utilizes a private static member variable called "buffer"
> If more than one thread is utilizing "buffer" you'll see unpredictable
> behavior unless you synchronize(Result.class) {}.
> If buffer is to remain a static variable, I would recommend changing it to a
> ThreadLocal<byte[]> instead.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
