madlib-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Rahul Iyer <ri...@apache.org>
Subject Re: kmeanspp error on GPDB
Date Tue, 05 Dec 2017 21:12:14 GMT
Thanks for getting back, Anthony. Hopefully the problem is resolved with
GCC 4.x.

The gist of the GCC issue is that there is a memory allocation/deallocation
issue in MADlib for specific C++ in-built types. These types are used in
elastic_net and others but not in the primitive modules. We'll get to
investigating and fixing it soon after the 1.13 release.

Best,
Rahul

On Tue, Dec 5, 2017 at 12:38 PM, Anthony Thomas <ahthomas@eng.ucsd.edu>
wrote:

> Hi Nikhil and Luis,
>
> Thanks for the follow up. Nikhil: to your questions about my environment:
>
>    1. yes - I downloaded the source tarball for 1.12 from
>    http://apache.org/dyn/closer.cgi?filename=madlib/1.12/
>    apache-madlib-1.12-src.tar.gz&action=download
>    <http://apache.org/dyn/closer.cgi?filename=madlib/1.12/apache-madlib-1.12-src.tar.gz&action=download>
>    2. I compiled using GCC 5.4.0 so perhaps that's the source of the
>    issue. I'll try recompiling with GCC < 5 and see if that fixes the problem.
>    3. When running madlib install-check I see the same error in the
>    elastic_net test.
>    4. Thanks for the tip about CentOS - I'm using GPDB alongside several
>    other tools that recommend/need Ubuntu and so if at all possible I'd like
>    to try and get it to work on Ubuntu.
>
> Thanks for the tips - hopefully this is just the result of compiling using
> GCC 5+. It seems a little strange though that the primitive matrix/array
> ops seem to work fine as do some of the other built-in algorithms, but
> KMeans and Elastic Net seem to have trouble.
>
> Anthony
>
> On Tue, Dec 5, 2017 at 10:44 AM, Nikhil Kak <nkak@pivotal.io> wrote:
>
>> Hi Anthony,
>>
>> Thank you for reporting the issue. It does look like that the exception
>> is coming from greenplum.
>>
>> A few follow up questions:-
>> 1. Did you compile madlib from source? If yes, which branch?
>> 2. What version of gcc did you use to compile madlib? Please note that
>> madlib does not work with gcc5 and up.
>> 3. Did you run madlib install-check?
>> 4. As Luis mentioned, you may want to try using CentOS 6/7.
>>
>> Thanks,
>> Nikhil Kak
>>
>> On Mon, Dec 4, 2017 at 3:32 PM Luis Macedo <lmacedo@pivotal.io> wrote:
>>
>>> Yep... Greenplum had an issue with that job. Maybe you could also post
>>> on GPDB user list.
>>>
>>> Check greenplum.org for the instructions.
>>>
>>> Not sure about running on Ubuntu... Ubuntu support is new so you might
>>> want to try on CentOS if you can.
>>>
>>> Tks
>>>
>>> *Luis Macedo | Sr Platform Architect | **Pivotal Inc *
>>>
>>> *Call Me @ **+55 11 97616-6438 <+55%2011%2097616-6438>*
>>> *Take care of the customers and the rest takes care of itself*
>>> *---*
>>> *Pivotal POV Blog* <https://content.pivotal.io/blog>
>>>
>>> *Pivotal Mission: Help Developers Do What They Love
>>> <https://www.youtube.com/watch?v=xdw_9dADM-4&list=PLAdzTan_eSPQ1fuLSBhyB4eEZF7JQM0Mx>*
>>> *Big Data Problems? Go here!* <https://pivotal.io/pivotal-greenplum>
>>> *Latency Issues? No more! <https://pivotal.io/pivotal-gemfire>*
>>> <https://pivotal.io/pivotal-gemfire>
>>> ---
>>>
>>>
>>> 2017-12-04 15:07 GMT-08:00 Anthony Thomas <ahthomas@eng.ucsd.edu>:
>>>
>>>> Hi MADLib Users,
>>>>
>>>> I'm running MADLib on Greenplum 5.1.0 under Ubuntu 16.04. If I try to
>>>> run the example for KMeanspp given here
>>>> <http://madlib.apache.org/docs/latest/group__grp__kmeans.html>, I get
>>>> the following error message:
>>>>
>>>> NOTICE:  Releasing segworker groups to finish aborting the transaction.
>>>> ERROR:  could not connect to segment: initialization of segworker group
>>>> failed (cdbgang.c:236)
>>>> CONTEXT:  PL/Python function "internal_compute_kmeanspp_seeding"
>>>> SQL statement "SELECT  ( SELECT madlib.internal_compute_kmeanspp_seeding(
>>>> '_madlib_kmeanspp_args', '_madlib_kmeanspp_state', textin(regclassout( $1
>>>> )),  $2 ) )"
>>>> PL/pgSQL function "kmeanspp_seeding" line 79 at assignment
>>>> SQL statement "SELECT  madlib.kmeans(  $1 ,  $2 ,
>>>> madlib.kmeanspp_seeding( $1 ,  $2 ,  $3 ,  $4 , NULL,  $5 ),  $4 ,  $6 ,
>>>> $7 ,  $8 )"
>>>> PL/pgSQL function "kmeanspp" line 4 at assignment
>>>>
>>>> I've tried running a few other built-in ML algorithms and they all seem
>>>> to work fine. The error message seems to suggest its a problem with
>>>> Greenplum, but some preliminary Googling has not yielded useful results.
Has
>>>> anyone seen this error before or have any ideas about how to diagnose it?
>>>>
>>>> Thanks,
>>>>
>>>> Anthony
>>>>
>>>
>>>
>

Mime
View raw message