hadoop-common-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Harsh J <ha...@cloudera.com>
Subject Re: Modify hadoop settings without restart
Date Thu, 14 Jun 2012 13:19:36 GMT
Matthias,

Good to know! Remember the caveat that the Generic Options (but don't
use that Class directly, its API is not marked for public consumption,
use Tool and ToolRunner instead) need to be the preset at the
beginning, followed by any app-specific arguments. Otherwise, they may
not work. The reasoning for this has once been explained at
https://issues.apache.org/jira/browse/HADOOP-7995

On Thu, Jun 14, 2012 at 1:17 PM, Matthias Zengler
<matthias.zengler@googlemail.com> wrote:
> Thanks Harsh!
>
> That did the trick after some other search. This link was helpful:
> http://hadoop.apache.org/common/docs/r1.0.3/api/org/apache/hadoop/util/GenericOptionsParser.html
>
> The right way to do it was the following:
> hadoop jar some.jar some.class.to.execute
> -Dmapred.map.child.java.opts="-Xmx2048M"
> -Dmapred.reduce.child.java.opts="-Xmx2048M" param1 param2
>
> Just in case, someone else has the same problem.
>
> Greetings,
> Mat
>
> 2012/6/13 Harsh J <harsh@cloudera.com>
>
>> That is a per job property and you can raise it when submitting a job
>> itself.
>>
>> You can pass it via -D args (See
>>
>> http://hadoop.apache.org/common/docs/current/api/org/apache/hadoop/util/Tool.html
>> )
>> or via a conf.set(name, value) from inside the code. Mostly, any prop
>> that is not of the format mapred(uce).{jobtracker,tasktracker}.*
>> aren't going to be requiring a restart of services.
>>
>> On Wed, Jun 13, 2012 at 6:24 PM, Matthias Zengler
>> <matthias.zengler@googlemail.com> wrote:
>> > I want to change the mapreduce.(map|reduce).java.opts for a bigger heap
>> > space of my map and reduce parts.
>> >
>> > 2012/6/13 Harsh J <harsh@cloudera.com>
>> >
>> >> Matthias,
>> >>
>> >> It depends on what config you're trying to change. There are several
>> >> per-job configs that do not require changes on JT/TT mapred-site.xml
>> >> and can be passed via -D parameters to a job's configuration directly
>> >> from CLI (if you use the Tool and ToolRunner methods of writing a
>> >> driver), or manually via the JobConf/Job.getConfiguration() objects
>> >> inside the code.
>> >>
>> >> So question to you: Which set of configs are you wishing to change
>> >> every job run? Cause only config changes that apply to JT or TTs need
>> >> those services to be restarted, the rest can be applied on a per job
>> >> basis, without requiring a restart of anything.
>> >>
>> >> On Wed, Jun 13, 2012 at 5:50 PM, Matthias Zengler
>> >> <matthias.zengler@googlemail.com> wrote:
>> >> > Hi,
>> >> >
>> >> > I've got a question regarding hadoop configuration. Is it possible
to
>> >> pass
>> >> > configuration parameters on job start up?
>> >> > Something like that:
>> >> >
>> >> > hadoop -HADOOP_HEAPSIZE=4G jar some.jar some.class.to.execute param1
>> >> param2
>> >> >
>> >> > Or do I have to restart the hadoop cluster every time I want to change
>> >> > something even if it is just for a specific job or workflow?
>> >> > We have some jobs running which needs a lot of time and we want to
>> start
>> >> > another one with a slightly different configuration because it needs
>> more
>> >> > memory to finish.
>> >> > We are using CDH3.
>> >> >
>> >> > Greetings,
>> >> > Mat
>> >>
>> >>
>> >>
>> >> --
>> >> Harsh J
>> >>
>>
>>
>>
>> --
>> Harsh J
>>



-- 
Harsh J

Mime
View raw message