Mailing-List: contact core-user-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Reply-To: core-user@hadoop.apache.org
Received-SPF: neutral (athena.apache.org: local policy)
DomainKey-Signature: a=rsa-sha1; s=serpent; d=yahoo-inc.com; c=nofws; q=dns;
	h=message-id:date:from:user-agent:mime-version:to:subject:
	references:in-reply-to:content-type:content-transfer-encoding;
	b=S2GVu1tTf4uTen91rdBRPioK72pjVieMIiLbsHL1Dd0Sy7ccpNCg62ml5fCSNpsf
Message-ID: <486C8596.1010109@yahoo-inc.com>
Date: Thu, 03 Jul 2008 13:23:58 +0530
From: Amar Kamat <amarrk@yahoo-inc.com>
User-Agent: Thunderbird 2.0.0.14 (X11/20080421)
MIME-Version: 1.0
To: core-user@hadoop.apache.org
Subject: Re: scaling issue, please help
References: <560DF7D2-5A04-4308-8B9E-4C5F4319788C@apple.com>
 <486B0CD4.7020602@yahoo-inc.com>
 <7FC5A2B6-BDF8-47A7-AA4A-ADB8F1013844@apple.com>
In-Reply-To: <7FC5A2B6-BDF8-47A7-AA4A-ADB8F1013844@apple.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Mori Bellamy wrote:
> i discovered that some of my code was causing out of bounds 
> exceptions. i cleaned up that code and the map tasks seemed to work. 
> that confuses me -- i'm pretty sure hadoop is resilient to a few map 
> tasks failing (5 out of 13k). before this fix, my remaining 2% of 
> tasks were getting killed.
Mori, I am not sure what the confusion is. Hadoop is resilient to few 
task failures but not by default. The parameter that does it is 
mapred.max.map.failures.percent and mapred.max.reduce.failures.percent. 
Every task internally consists of attempts (internally, for the 
framework). Hadoop allows some attempt failures too. If the number of 
attempts that failed of a task exceeds the threshold 
(mapred.map.max.attempts/mapred.reduce.max.attempts : default is 4) then 
the task is considered failed. If the number of map/reduce task failures 
exceeds the threshold 
(mapred.max.map.failures.percent/mapred.max.reduce.failures.percent : 
default is 0) then the job is considered failed.
Amar
>
>
> On Jul 1, 2008, at 10:06 PM, Amar Kamat wrote:
>
>> Mori Bellamy wrote:
>>> hey all,
>>> i've got a mapreduce task that works on small (~1G) input. when i 
>>> try to run the same task on large (~100G) input, i get the following 
>>> error around when the map tasks are almost done (~98%)
>>>
>>> 2008-07-01 13:10:59,231 INFO org.apache.hadoop.mapred.ReduceTask: 
>>> task_200807011005_0005_r_000000_0: Got 0 new map-outputs & 0 
>>> obsolete map-outputs from tasktracker and 0 map-outputs from 
>>> previous failures
>>> 2008-07-01 13:10:59,232 INFO org.apache.hadoop.mapred.ReduceTask: 
>>> task_200807011005_0005_r_000000_0 Got 0 known map output 
>>> location(s); scheduling...
>>> 2008-07-01 13:10:59,232 INFO org.apache.hadoop.mapred.ReduceTask: 
>>> task_200807011005_0005_r_000000_0 Scheduled 0 of 0 known outputs (0 
>>> slow hosts and 0 dup hosts)
>>> 2008-07-01 13:10:59,232 INFO org.apache.hadoop.mapred.ReduceTask: 
>>> task_200807011005_0005_r_000000_0 Need 1 map output(s)
>> ...
>> ...
>> These are not error messages. The reducers are stuck as not all maps 
>> are completed. Mori, could you let us know what is happening to the 
>> other 2% maps. Are they getting executed? Are they still pending 
>> (waiting to run)? Were they killed/failed? Is there any lost tracker?
>>> I'm running the task on a cluster of 5 workers, one DFS master, and 
>>> one task tracker.
>> What do you mean by 5 workers and 1 task tracker?
>>> i'm chaining mapreduce tasks, so i'm using SequenceFileOutput and 
>>> SequenceFileInput. this error happens before the first link in the 
>>> chain sucessfully reduces.
>> Can you elaborate this a bit. Are you chaining MR jobs?
>> Amar
>>>
>>> does anyone have any insight? thanks!
>>
>