You are viewing a plain text version of this content. The canonical link for it is here.

Posted to solr-user@lucene.apache.org by Jun Ohtani <jo...@gmail.com> on 2013/08/29 10:22:03 UTC

Too slow UIMA with Solr

Dear all,

My co-worker use UIMA with Solr 4.4.0.
But too slow UIMA with Solr.

I read source code solr/contirb/uima/.
Solr UIMA integration source code call repeatedly AEProvider.getAE().
Maybe it's create AnalysisEngine instance every request and every processText().

But Lucene code (o.a.l.analysis.uima.BaseUIMATokenizer) reuse AnalysisEngine instance.

Why does Solr UIMA integration call AEProvider.getAE() every update request ?

Regards,

------------
Jun Ohtani

Re: Too slow UIMA with Solr

Posted by Jun Ohtani <jo...@gmail.com>.

Hi Tommaso,

Thanks!!

------------
Jun Ohtani




On 2013/08/29, at 17:56, Tommaso Teofili <to...@gmail.com> wrote:

> p.s.
> see https://issues.apache.org/jira/browse/SOLR-5201
> 
> 
> 2013/8/29 Tommaso Teofili <to...@gmail.com>
> 
>> Hi Jun,
>> 
>> I agree the AE (instead of the AEProvider) should be cached on the
>> UpdateRequestProcessor.
>> In previous revisions [1] it was cached directly by the BasicAEProvider so
>> there wasn't need of that in the UIMAUpdateRequestProcessor but, since that
>> has changed, I agree that should be done there as well, I'll open an issue.
>> 
>> Thanks,
>> Tommaso
>> 
>> 
>> [1] :
>> http://svn.apache.org/viewvc/lucene/dev/branches/branch_4x/lucene/analysis/uima/src/java/org/apache/lucene/analysis/uima/ae/BasicAEProvider.java?revision=1244236&view=markup
>> 
>> 
>> 
>> 2013/8/29 Jun Ohtani <jo...@gmail.com>
>> 
>>> Dear all,
>>> 
>>> My co-worker use UIMA with Solr 4.4.0.
>>> But too slow UIMA with Solr.
>>> 
>>> I read source code solr/contirb/uima/.
>>> Solr UIMA integration source code call repeatedly AEProvider.getAE().
>>> Maybe it's create AnalysisEngine instance every request and every
>>> processText().
>>> 
>>> But Lucene code (o.a.l.analysis.uima.BaseUIMATokenizer) reuse
>>> AnalysisEngine instance.
>>> 
>>> Why does Solr UIMA integration call AEProvider.getAE() every update
>>> request ?
>>> 
>>> Regards,
>>> 
>>> ------------
>>> Jun Ohtani
>>> 
>>> 
>>> 
>>> 
>>

Re: Too slow UIMA with Solr

Posted by Tommaso Teofili <to...@gmail.com>.

p.s.
see https://issues.apache.org/jira/browse/SOLR-5201


2013/8/29 Tommaso Teofili <to...@gmail.com>

> Hi Jun,
>
> I agree the AE (instead of the AEProvider) should be cached on the
> UpdateRequestProcessor.
> In previous revisions [1] it was cached directly by the BasicAEProvider so
> there wasn't need of that in the UIMAUpdateRequestProcessor but, since that
> has changed, I agree that should be done there as well, I'll open an issue.
>
> Thanks,
> Tommaso
>
>
> [1] :
> http://svn.apache.org/viewvc/lucene/dev/branches/branch_4x/lucene/analysis/uima/src/java/org/apache/lucene/analysis/uima/ae/BasicAEProvider.java?revision=1244236&view=markup
>
>
>
> 2013/8/29 Jun Ohtani <jo...@gmail.com>
>
>> Dear all,
>>
>> My co-worker use UIMA with Solr 4.4.0.
>> But too slow UIMA with Solr.
>>
>> I read source code solr/contirb/uima/.
>> Solr UIMA integration source code call repeatedly AEProvider.getAE().
>> Maybe it's create AnalysisEngine instance every request and every
>> processText().
>>
>> But Lucene code (o.a.l.analysis.uima.BaseUIMATokenizer) reuse
>> AnalysisEngine instance.
>>
>> Why does Solr UIMA integration call AEProvider.getAE() every update
>> request ?
>>
>> Regards,
>>
>> ------------
>> Jun Ohtani
>>
>>
>>
>>
>

Re: Too slow UIMA with Solr

Posted by Tommaso Teofili <to...@gmail.com>.

Hi Jun,

I agree the AE (instead of the AEProvider) should be cached on the
UpdateRequestProcessor.
In previous revisions [1] it was cached directly by the BasicAEProvider so
there wasn't need of that in the UIMAUpdateRequestProcessor but, since that
has changed, I agree that should be done there as well, I'll open an issue.

Thanks,
Tommaso


[1] :
http://svn.apache.org/viewvc/lucene/dev/branches/branch_4x/lucene/analysis/uima/src/java/org/apache/lucene/analysis/uima/ae/BasicAEProvider.java?revision=1244236&view=markup



2013/8/29 Jun Ohtani <jo...@gmail.com>

> Dear all,
>
> My co-worker use UIMA with Solr 4.4.0.
> But too slow UIMA with Solr.
>
> I read source code solr/contirb/uima/.
> Solr UIMA integration source code call repeatedly AEProvider.getAE().
> Maybe it's create AnalysisEngine instance every request and every
> processText().
>
> But Lucene code (o.a.l.analysis.uima.BaseUIMATokenizer) reuse
> AnalysisEngine instance.
>
> Why does Solr UIMA integration call AEProvider.getAE() every update
> request ?
>
> Regards,
>
> ------------
> Jun Ohtani
>
>
>
>