Mailing-List: contact issues-help@hbase.apache.org; run by ezmlm
Precedence: bulk
Date: Wed, 22 Apr 2015 18:03:00 +0000 (UTC)
From: "Andrew Purtell (JIRA)" <jira@apache.org>
To: issues@hbase.apache.org
Message-ID: <JIRA.12819065.1428439433000.78315.1429725780080@Atlassian.JIRA>
In-Reply-To: <JIRA.12819065.1428439433000@Atlassian.JIRA>
References: <JIRA.12819065.1428439433000@Atlassian.JIRA>
 <JIRA.12819065.1428439433448@arcas>
Subject: [jira] [Comment Edited] (HBASE-13420)
 RegionEnvironment.offerExecutionLatency Blocks Threads under Heavy Load
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/HBASE-13420?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14507560#comment-14507560 ] 

Andrew Purtell edited comment on HBASE-13420 at 4/22/15 6:02 PM:
-----------------------------------------------------------------

bq. Seems not worth it to me, unless John Leach has a more specific use case where we'd see a larger difference.

I think you misunderstand Lars. Here are the benchmarks that matter: https://issues.apache.org/jira/browse/HBASE-13420?focusedCommentId=14486865&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14486865 . Sampled timings for offer() at percentiles < 1.0 for BoundedConcurrentLinkedQueue are much better than that for ArrayBlockingQueue. This is on top of John's original report indicating a problem with the BlockingQueues holding up threads in his test case.

You then asked for more information. I ran two macro-benchmarks to see if there's any perf hit. Here https://issues.apache.org/jira/browse/HBASE-13420?focusedCommentId=14504156&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14504156 and here https://issues.apache.org/jira/browse/HBASE-13420?focusedCommentId=14506064&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14506064 . I'm not seeing any significant degradation. The purpose of these last two tests were not to show some kind of improvement. 


was (Author: apurtell):
bq. Seems not worth it to me, unless John Leach has a more specific use case where we'd see a larger difference.

I think you misunderstand Lars. Here are the benchmarks that matter: https://issues.apache.org/jira/browse/HBASE-13420?focusedCommentId=14486865&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14486865 . Sampled timings for offer() at percentiles < 1.0 for BoundedConcurrentLinkedQueue are much better than that for ArrayBlockingQueue. This is on top of John's original report indicating a problem with the BlockingQueues holding up threads in his test case.

You then asked for more information. I ran two macro-benchmarks to see if there's any perf hit. Here https://issues.apache.org/jira/browse/HBASE-13420?focusedCommentId=14504156&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14504156 and here https://issues.apache.org/jira/browse/HBASE-13420?focusedCommentId=14506064&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14506064 . I'm not seeing a significant degradation. The purpose of these last two tests were not to show some kind of improvement. 

> RegionEnvironment.offerExecutionLatency Blocks Threads under Heavy Load
> -----------------------------------------------------------------------
>
>                 Key: HBASE-13420
>                 URL: https://issues.apache.org/jira/browse/HBASE-13420
>             Project: HBase
>          Issue Type: Improvement
>            Reporter: John Leach
>            Assignee: Andrew Purtell
>             Fix For: 2.0.0, 1.1.0, 0.98.13, 1.0.2
>
>         Attachments: 1M-0.98.12.svg, 1M-0.98.13-SNAPSHOT.svg, HBASE-13420.patch, HBASE-13420.txt, hbase-13420.tar.gz, offerExecutionLatency.tiff
>
>   Original Estimate: 3h
>  Remaining Estimate: 3h
>
> The ArrayBlockingQueue blocks threads for 20s during a performance run focusing on creating numerous small scans.  
> I see a buffer size of (100)
>     private final BlockingQueue<Long> coprocessorTimeNanos = new ArrayBlockingQueue<Long>(
>         LATENCY_BUFFER_SIZE);
> and then I see a drain coming from
>          MetricsRegionWrapperImpl with 45 second executor
>          HRegionMetricsWrapperRunable
>          RegionCoprocessorHost#getCoprocessorExecutionStatistics()   
>          RegionCoprocessorHost#getExecutionLatenciesNanos()
> Am I missing something?


--
This message was sent by Atlassian JIRA
(v6.3.4#6332)