lucene-solr-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jack Krupansky" <j...@basetechnology.com>
Subject Re: [custom data structure] aligned dynamic fields
Date Wed, 22 May 2013 15:32:31 GMT
Although we are entering the era of "Big Data", that does not mean there are 
no limits or restrictions on what a given technology can do.

Maybe you need to consider either a smaller scope for your project, or more 
limited features, or some other form of simplification.

Solr can do "billions" of documents - for a heavily sharded cluster, but you 
will have to work really hard to make that work well.

So, I can confirm, that maybe in this case, there is no free lunch - unless 
you are willing to strip down the project. Or, maybe we just need a deeper 
feel for what your data model is really trying to achieve.

Suggestion: Think about your data model again, and then try rephrasing it 
for this group. You have violated one cardinal rule of this group: you 
focused on a proposed solution rather than focusing our attention on the 
original problem you are trying to solve. That short-circuited our focus on 
really solving your problem.

-- Jack Krupansky

-----Original Message----- 
From: Dmitry Kan
Sent: Wednesday, May 22, 2013 6:50 AM
To: solr-user@lucene.apache.org
Subject: Re: [custom data structure] aligned dynamic fields

Jack,

Thanks for your response.

1. Flattening could be an option, although our scale and required
functionality (runtime non DocValues backed facets) is beyond what solr3
can handle (billions of docs). We have flattened the meta data at the
expense of "over"-generating solr documents. But to solve the problem I
have described via flattening would make big impact on the scalability and
price.

2. We have quite the opposite of what you have described about the dynamic
fields: there will be very few per document. I agree, that caution should
be taken here, as we have "suffered" (or should I say "experienced") having
multivalued fields (the good thing is we never had to facet on them).

Any other options? Maybe someone can share their experience with dynamic
fields and discourage from pursuing this path?

Dmitry


On Mon, May 20, 2013 at 4:23 PM, Jack Krupansky 
<jack@basetechnology.com>wrote:

> Before you dive off the deep end and "go crazy" with dynamic fields, try a
> clean, simple, Solr-oriented static design. Yes, you CAN do an
> over-complicated design with dynamic fields, but that doesn't mean you
> should.
>
> In a single phrase, denormalize and flatten your design. Sure, that will
> lead to a lot of rows, but Solr and Lucene are designed to do well in that
> scenario.
>
> If you are still linking in terms of "C Struct", go for a long walk or do
> SOMETHING else until you can get that idea out of your head. It is a
> sub-optimal approach for exploiting the power of Lucene and Solr.
>
> Stay with a static schema design until you hit... just stay with a static
> schema, period.
>
> Dynamic fields and multi-valued fields do have value, but only when used
> in moderation - small numbers. If you start down a design path and find
> that you are heavily dependent on dynamic fields and/or multi-valued 
> fields
> with large numbers of values per document, that is feedback that your
> design needs to be denormalized and flattened further.
>
> -- Jack Krupansky
>
> -----Original Message----- From: Dmitry Kan
> Sent: Monday, May 20, 2013 7:06 AM
> To: solr-user@lucene.apache.org
> Subject: [custom data structure] aligned dynamic fields
>
>
> Hi all,
>
> Our current project requirement suggests that we should start storing
> custom data structures in solr index. The custom data structure would be 
> an
> equivalent of C struct.
>
> The task is as follows.
>
> Suppose we have two types of fields, one is FieldName1 and the other
> FieldName2.
>
> Suppose also that we can have multiple pairs of these two fields on a
> document in Solr.
>
> That is, in notation of dynamic fields:
>
> doc1
> FieldName1_id1
> FieldName2_id1
>
> FieldName1_id2
> FieldName2_id2
>
> doc2
> FieldName1_id3
> FieldName2_id3
>
> FieldName1_id4
> FieldName2_id4
>
> FieldName1_id5
> FieldName2_id5
>
> etc
>
> What we would like to have is a value for the Field1_(some_unique_id) and 
> a
> value for Field2_(some_unique_id) as input for search. That is we wouldn't
> care about the some_unique_id in some search scenarios. And the search
> would automatically iterate the pairs of dynamic fields and respect the
> pairings.
>
> I know it used to be so, that with dynamic fields a client must provide 
> the
> dynamically generated field names coupled with their values up front when
> searching.
>
> What data structure / solution could be used as an alternative approach to
> help such a "structured search"?
>
> Thanks,
>
> Dmitry
> 


Mime
View raw message