Hi Marshall,
Thanks for reply.
> In most application scenarios, it is intended that a UIMA pipeline be
> instantiated once, and then reused.
I wanted to know about that.
I will try to post Solr ML.
Thanks!
------------
Jun Ohtani
On 2013/08/29, at 6:23, Marshall Schor <msa@schor.com> wrote:
> This may be a better question for the Solr list. These issues seem to come from
> the particular integration of UIMA into SOLR, where it seems to be repeatedly
> instantiating new UIMA pipelines, over and over.
>
> In most application scenarios, it is intended that a UIMA pipeline be
> instantiated once, and then reused.
>
> -Marshall
>
> On 8/26/2013 4:20 AM, Jun Ohtani wrote:
>> Dear all,
>>
>> My co-worker use UIMA with Solr.
>> We have two problems increasing heap size and cpu usage.
>>
>> We have two question.
>>
>> 1. o.a.u.jcas.impl.JCasImpl.loadJCasClasses() call every requests.
>> Using UIMA with Solr, we send to request using UIMAUpdateRequestProcessor , Solr(Lucene)
create AnalysisEngine instance every requests.
>> Maybe AnalysisEngine call loadJCasClasses() methods every request, and load CasType
from classloader.
>> Why does Solr(Lucene) not cache an Analysis Engine instance?
>>
>> 2. o.a.u.analysis_emgine.impl.AnalysisEngineManagementImpl.usedRootNames increase
entry.
>> The unique MBean name create classname and suffix number.
>> AnalysisEngineManagementImpl.setName() use HashSet to check unique MBean name.
>> Is it better using HashMap(classname, counter) Instead of HashSet ?
>>
>>
>> Regards
>>
>> ------------
>> Jun Ohtani
>
|