On Thu, Jun 26, 2014 at 11:41 PM, Elliot Ickovic <elliot.icko...@gmail.com>
wrote:
> Hi Mikhail, Thank you for the quick response!
>
> If I instead of:
>
> add() - fakeparent000 : [single001]
>
> I do :
>
> add() - single000 : [fakeChild001]
>
> will this prevent the index from appearing corrupted? This way I can
> retain my logical top level docs.
>
> What is the reason I need add a fake doc?
Because BlockJoinQueries are straightforward and relies on regular block
structure: single parent doc follows contiguous children and marked by the
bitset. The same works in case of block nesting.
> If I add nested documents N
> levels deep, creating a tree, do I need to ensure fake documents are
> created to tree max depth? Or is it sufficient only for the top level?
>
tree should be regular. if in a branch you need just single children doc or
parent, you need to accomplish them by the necessary counterpart. It's
mandatory for all levels.
>
>
> On Thu, Jun 26, 2014 at 2:12 PM, Mikhail Khludnev <
> mkhlud...@griddynamics.com> wrote:
>
> > Hello Elliot,
> >
> > Parent doc is mandatory, you can't omit it.
> > Thus instead of:
> > add() - single001
> >
> > you have to
> >
> > add() - fakeparent000 : [single001]
> >
> > there was no plans to support any sort of flexibility there...
> >
> >
> >
> > On Thu, Jun 26, 2014 at 9:52 PM, Elliot Ickovic <
> elliot.icko...@gmail.com>
> > wrote:
> >
> > > Using Solr 4.8.1.
> > >
> > > I am creating an index containing Solr documents both with and without
> > > nested documents. When Indexing documents from a single SolrJ client
> on
> > a
> > > single thread if I do not call commit() after each document add() I see
> > > some erroneous documents returned from my "child of" or "parent which"
> > > block queries. Here is an example:
> > >
> > > I have a required field 'doc-type' used for the block join. For
> example
> > I
> > > have 3 types {'parent', 'child', 'single'} where parent has 1 or more
> > > nested child docs and single never has a nested doc.
> > >
> > > CASE 1:
> > > ================================
> > > I add / commit docs in this order
> > >
> > > add() - single001
> > > add() - parent001 : [child001, child002]
> > > commit()
> > >
> > > then query all child docs of every parent doc
> > >
> > > ..... {!child of='doc-type:parent'}doc-type:parent
> > >
> > > response contains *single001*, child001, child002 -- INCORRECT
> *single001
> > > *is
> > > not a child.
> > > ================================
> > >
> > > CASE 2:
> > > ================================
> > > I add / commit docs in this order
> > >
> > > add() - single001
> > > *commit() added commit here*
> > >
> > > add() - parent001 : [child001, child002]
> > > commit()
> > >
> > > then query all child docs of every parent doc
> > >
> > > ..... {!child of='doc-type:parent'}doc-type:parent
> > >
> > > response contains only {child001, child002} -- CORRECT
> > > ================================
> > >
> > > CASE 2 will only work when a single solr client is adding docs. If
> > spawn
> > > a bunch of threads that do adds I will not be able to guarantee that
> > commit
> > > is called between adds.
> > > Additionally, I am not 100% sure calling commit() is the correct
> solution
> > > it just seems to work where I have tested.
> > >
> > > I'd greatly appreciate if someone can shed some light on this problem
> or
> > > help me better understand the limitations of the block indexing and
> block
> > > join.
> > >
> >
> >
> >
> > --
> > Sincerely yours
> > Mikhail Khludnev
> > Principal Engineer,
> > Grid Dynamics
> >
> > <http://www.griddynamics.com>
> > <mkhlud...@griddynamics.com>
> >
>
--
Sincerely yours
Mikhail Khludnev
Principal Engineer,
Grid Dynamics
<http://www.griddynamics.com>
<mkhlud...@griddynamics.com>
