I think that’s because the Spark in the maven repos is tied to hadoop 2 and the 
default in the master is 1.2.1

Sounds like you are the closest to the build machines. Can you try 
https://github.com/pferrel/mahout/tree/hadoop-client

This is a merge of Gokhan’s patch with master. It should default to hadoop 2 
and theoretically should have all artifacts in alignment.

On Oct 30, 2014, at 8:11 PM, Andrew Palumbo <[email protected]> wrote:



> Subject: Re: Jenkins build became unstable: mahout-nightly » Mahout Spark 

I cleaned out my mvn repo, unset SPARK_HOME, and ran 

 $ mvn clean install 

from the latest master. now am getting the failure you're talking about:

- ddsvd - naive - q=1 *** FAILED ***
 org.apache.spark.SparkException: Job aborted due to stage failure: Task 9 in 
stage 28.0 failed 1 times, most recent failure: Lost task 9.0 in stage 28.0 
(TID 81, localhost): java.io.IOException: PARSING_ERROR(2)
       org.xerial.snappy.SnappyNative.throw_error(SnappyNative.java:78)





bindings #1728
> From: [email protected]
> Date: Thu, 30 Oct 2014 19:10:19 -0700
> To: [email protected]
> 
> I took Gokhan’s PR and merged the master with it and compiling with 
> 
> mvn clean install package -Dhadoop.version=1.2.1
> 
> I get the same build error as the nightly.
> 
> Changing back to the master it builds fine. The default hadoop version is 
> 1.2.1 in master so I don’t need a profile or CLI options to build for my 
> environment.
> 
> This seems like more than cosmic rays as Dmitriy guessed.
> 
> On Oct 30, 2014, at 12:41 PM, Dmitriy Lyubimov <[email protected]> wrote:
> 
> more likely spark thing .
> 
> the error is while using torrent broadcast. AFAIK that was not default
> choice until recently.
> 
> On Thu, Oct 30, 2014 at 10:27 AM, Suneel Marthi <[email protected]> wrote:
> 
>> The nightly builds often due to running on an old machine and the failure
>> is also a function of the number of concurrent jobs that are running.  If u
>> look at the logs from the failure, it most likely would have failed due to
>> a JVM crash (or something similar).  Its the daily builds that we need to
>> ensure are not failing.
>> 
>> 
>> On Thu, Oct 30, 2014 at 1:21 PM, Andrew Palumbo <[email protected]>
>> wrote:
>> 
>>> I just built and tested with no problems.  Probably just Jenkins acting
>> up.
>>> 
>>>> Subject: Re: Jenkins build became unstable:  mahout-nightly » Mahout
>>> Spark bindings #1728
>>>> From: [email protected]
>>>> Date: Thu, 30 Oct 2014 09:26:45 -0700
>>>> To: [email protected]
>>>> 
>>>> At first blush this looks unrelated to the stuff I pushed to move to
>>> Spark 1.1.0
>>>> 
>>>> The error is in snappy parsing during some R-like ops
>>>> 
>>>> I don’t use native snappy myself, is anyone else seeing this or is it
>>> just  cosmic rays?
>>>> 
>>>> 
>>>> On Oct 29, 2014, at 4:43 PM, Apache Jenkins Server <
>>> [email protected]> wrote:
>>>> 
>>>> See <
>>> 
>> https://builds.apache.org/job/mahout-nightly/org.apache.mahout$mahout-spark_2.10/1728/
>>>> 
>>>> 
>>>> 
>>> 
>> 
> 
                                          

Reply via email to