activemq-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "devantor (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (AMQ-3801) Job scheduler can lose data on forced kill
Date Mon, 06 May 2013 20:00:18 GMT

    [ https://issues.apache.org/jira/browse/AMQ-3801?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13650046#comment-13650046
] 

devantor edited comment on AMQ-3801 at 5/6/13 7:59 PM:
-------------------------------------------------------

I compared release 5.4.2 performance vs 5.6.0. The fix solve the corruption problem at performance
cost.


# Write batch disabled
ant producer -Durl=tcp://127.0.0.1:6503 -Ddurable=true -DtimeToLive=86400000 -Dsubject=DELAY_TEST
-Dtransacted=false -DparallelThreads=10 -DmessageSize=10000 -Dmax=1000
BUILD SUCCESSFUL
Total time: 36 minutes 17 seconds


# Batch enabled
     [java] All threads completed their work
BUILD SUCCESSFUL
Total time: 14 seconds
                
      was (Author: devantor):
    I compared release 5.4.2 performance vs 5.6.0. The fix solve the corruption problem at
performance cost.


##Benchmark
### write batch disabled
ant producer -Durl=tcp://127.0.0.1:6503 -Ddurable=true -DtimeToLive=86400000 -Dsubject=DELAY_TEST
-Dtransacted=false -DparallelThreads=10 -DmessageSize=10000 -Dmax=1000
BUILD SUCCESSFUL
Total time: 36 minutes 17 seconds


### Batch enabled
     [java] All threads completed their work
BUILD SUCCESSFUL
Total time: 14 seconds
                  
> Job scheduler can lose data on forced kill
> ------------------------------------------
>
>                 Key: AMQ-3801
>                 URL: https://issues.apache.org/jira/browse/AMQ-3801
>             Project: ActiveMQ
>          Issue Type: Bug
>    Affects Versions: 5.5.0
>            Reporter: Dejan Bosanac
>            Assignee: Dejan Bosanac
>             Fix For: 5.6.0
>
>
> The page file that keeps the index of jobs to be scheduled is using default write batch
size of 1000. That means that index could get into incorrect state when broker is killed (and
there's no recovery for this index). Turning off batching (as it's not needed for this low-volume
index) will ensure that changes are always persisted immediately. 

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message