Question on SOLR join query

[Jaimin Patel <re...@gmail.com>]

I am facing a performance problem and could narrow it down to a join query
that we are using. The join is on a unique field.

We have a person profile stored in RDB in a relational way. Like person
name table , address table etc. SOLR indexes are build using this RDB
data,Each children is stored as separate document with parent's unique id.
At query time , unique id of parent is joined with same in child
documents({!join
to=par_id from=par_id }) to allow search with AND condition for search
terms involving children data

I am reading similar issues and it says "initial join implementation is
O(nterms)".. What does this mean ? I could not find any reference
explaining meaning of 0 num_terms_in_field.

 Regards,
Jai

Re: Question on SOLR join query

[Erick Erickson <er...@gmail.com>]

First of all, Solr is a _search_ engine, it wasn't built to be an
RDBMS. Whenever I see this question (paraphrasing) "I've indexed my
tables and want to use Solr just like a DB" I cringe.

The join performance goes up with the number of unique values for the
join field. High-cardinality fields are the poorer performing type,
which is what "initial join implementation is O(nterms)" is
expressing.

Have you considered either denormalizing the data or using block joins?

Best,
Erick

On Mon, Sep 25, 2017 at 6:59 PM, Jaimin Patel <re...@gmail.com> wrote:
> I am facing a performance problem and could narrow it down to a join query
> that we are using. The join is on a unique field.
>
> We have a person profile stored in RDB in a relational way. Like person
> name table , address table etc. SOLR indexes are build using this RDB
> data,Each children is stored as separate document with parent's unique id.
> At query time , unique id of parent is joined with same in child
> documents({!join
> to=par_id from=par_id }) to allow search with AND condition for search
> terms involving children data
>
> I am reading similar issues and it says "initial join implementation is
> O(nterms)".. What does this mean ? I could not find any reference
> explaining meaning of 0 num_terms_in_field.
>
>  Regards,
> Jai

Re: Question on SOLR join query

[Mikhail Khludnev <mk...@apache.org>]

> I am reading similar issues and it says "initial join implementation is
O(nterms)".. What does this mean?

It enumerates all par_id terms every time. As an alternative for some of
field types you can add {!join ... score=none ...}.. to trigger Lucene's
join algorithm with O(fromDocs) ie if subordinate query returns no result
it gets back rapidly.

On Tue, Sep 26, 2017 at 4:59 AM, Jaimin Patel <re...@gmail.com> wrote:

> I am facing a performance problem and could narrow it down to a join query
> that we are using. The join is on a unique field.
>
> We have a person profile stored in RDB in a relational way. Like person
> name table , address table etc. SOLR indexes are build using this RDB
> data,Each children is stored as separate document with parent's unique id.
> At query time , unique id of parent is joined with same in child
> documents({!join
> to=par_id from=par_id }) to allow search with AND condition for search
> terms involving children data
>
> I am reading similar issues and it says "initial join implementation is
> O(nterms)".. What does this mean ? I could not find any reference
> explaining meaning of 0 num_terms_in_field.
>
>  Regards,
> Jai
>



-- 
Sincerely yours
Mikhail Khludnev