[ 
https://issues.apache.org/jira/browse/MAPREDUCE-7017?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

jiayuhan-it updated MAPREDUCE-7017:
-----------------------------------
    Description:   MRAppMaster uses TaskAttemptImpl::resolveHosts to determine 
the dataLocalHosts for each task when the location of data split is IP, which 
will call a lot of times ( taskNum * dfsReplication) of function InetAddress:: 
getByName and most of the funcition calls are redundant.  When the job has a 
great number of tasks and the speed of DNS resolution is not fast enough, it 
will take a lot of time at this stage before the job running.  (was:     In 
MRAppMaster each TaskAttempt call resolveHosts function too many times for 
getting dataLocalHosts.  When a job has a lot of tasks or the machine 
configures unreasonab, it will waste a lot of time.)

> Too many times of meaningless invocation in TaskAttemptImpl#resolveHosts
> ------------------------------------------------------------------------
>
>                 Key: MAPREDUCE-7017
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-7017
>             Project: Hadoop Map/Reduce
>          Issue Type: Improvement
>          Components: mr-am
>    Affects Versions: 3.0.0-alpha4
>            Reporter: jiayuhan-it
>
>   MRAppMaster uses TaskAttemptImpl::resolveHosts to determine the 
> dataLocalHosts for each task when the location of data split is IP, which 
> will call a lot of times ( taskNum * dfsReplication) of function 
> InetAddress:: getByName and most of the funcition calls are redundant.  When 
> the job has a great number of tasks and the speed of DNS resolution is not 
> fast enough, it will take a lot of time at this stage before the job running.



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

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to