Mailing-List: contact mapreduce-user-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Reply-To: mapreduce-user@hadoop.apache.org
Received-SPF: pass (nike.apache.org: local policy)
From: Marek Miglinski <mmiglinski@seven.com>
To: "mapreduce-user@hadoop.apache.org" <mapreduce-user@hadoop.apache.org>
Subject: RE: Capacity Scheduler problem
Thread-Topic: Capacity Scheduler problem
Thread-Index: AczPtRM0hOzE6nfTSjWzNAXtuQu0gwACS3qAACNtIyM=
Date: Wed, 11 Jan 2012 10:31:27 +0000
Message-ID: <03AF7A47D43C954E92AC1F7CD5C124020C2153B0@TLN-MBX1.corp.seven.com>
References: 
 <03AF7A47D43C954E92AC1F7CD5C124020C21514E@TLN-MBX1.corp.seven.com>,<CAJVBNiHMNnPwVmvdkmzEyO1ssHxAp4LUW0MmDkttEZCzxvyZmQ@mail.gmail.com>
In-Reply-To: 
 <CAJVBNiHMNnPwVmvdkmzEyO1ssHxAp4LUW0MmDkttEZCzxvyZmQ@mail.gmail.com>
Accept-Language: en-US
Content-Language: en-US
Content-Type: text/plain; charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0


That's not the case, I've even removed hadoop-fairscheduler-0.20.2-cdh3u2.j=
ar from hadoop lib folder and fair-scheduler.xml from hadoop conf folder an=
d it didn't help... Any ideas?

________________________________
From: Bai Shen [baishen.lists@gmail.com]
Sent: Tuesday, January 10, 2012 9:35 PM
To: mapreduce-user@hadoop.apache.org
Subject: Re: Capacity Scheduler problem

Turn off the fairscheduler multiple task assign setting.  I just had the sa=
me problem with my cluster.

On Tue, Jan 10, 2012 at 11:34 AM, Marek Miglinski <mmiglinski@seven.com<mai=
lto:mmiglinski@seven.com>> wrote:
Hello guys,

1. I have concern with my 3 node cluster, I run capacity scheduler with 4 q=
ueues and one has 30% of cluster resources, the problem is that when I sche=
dule a job, all tasks are assigned to one single node which takes all of it=
's mappers and works quite slow. Are there any settings for mapred/capacity=
 scheduler to assign mappers on all nodes by even number of tasks?

2. I've set my capacity-scheduler.xml settings "mapred.capacity-scheduler.q=
ueue.job1.capacity" for all queues as wanted but jobtracker doesn't realloc=
ate resources from free queues if only one is working, why?


Thanks,
Marek M.