I guess this gets into the point of whether "children" or whatever field is used for child documents actually needs to be in the schema. Schemaless mode creates one, but that's not a defining factor. Because if it needs to be in the schema, then the code should reflect its cardinality. But if it does not, then all bets are off.
Regards, Alex. P.s. I added this question to SOLR-12298, as I don't think I know enough about this part of Solr to judge. On Mon, 24 Aug 2020 at 02:28, Munendra S N <sn.munendr...@gmail.com> wrote: > > > > > Interestingly, I was forced to add children as an array even when the > > child was alone and the field was already marked multivalued. It seems > > the code does not do conversation to multi-value type, which means the > > query code has to be a lot more careful about checking field return > > type and having multi-path handling. That's not what Solr does for > > string class (tested). Is that a known issue? > > > > https://github.com/arafalov/SolrJTest/blob/master/src/com/solrstart/solrj/Main.java#L88-L89 > > Not sure about this. Maybe we might need to check in Dev list or Slack > > If I switch commented/uncommented lines around, the retrieval will fail > > part way through, because one 'children' field is returned as array, but > > not the other one: > > This might be because of these checks > https://github.com/apache/lucene-solr/blob/e1392c74400d74366982ccb796063ffdcef08047/solr/core/src/java/org/apache/solr/response/transform/ChildDocTransformer.java#L201-L209 > but > not sure > > Regards, > Munendra S N > > > > On Sun, Aug 23, 2020 at 7:53 PM Alexandre Rafalovitch <arafa...@gmail.com> > wrote: > > > Thank you Nunedra, > > > > That was very helpful. I am looking forward to that documentation Jira > > to be merged into the next release. > > > > I was able to get the example working by switching away from anonymous > > children to the field approach. Which means hasChildren() call also > > did not work. It seems the addChildren/hasChildren will need a > > different schema, without _nest_path_ defined. I did not test. > > > > Interestingly, I was forced to add children as an array even when the > > child was alone and the field was already marked multivalued. It seems > > the code does not do conversation to multi-value type, which means the > > query code has to be a lot more careful about checking field return > > type and having multi-path handling. That's not what Solr does for > > string class (tested). Is that a known issue? > > > > https://github.com/arafalov/SolrJTest/blob/master/src/com/solrstart/solrj/Main.java#L88-L89 > > > > If I switch commented/uncommented lines around, the retrieval will > > fail part way through, because one 'children' field is returned as > > array, but not the other one: > > > > {responseHeader={status=0,QTime=0,params={q=id:p1,fl=*,[child],wt=javabin,version=2}},response={numFound=1,numFoundExact=true,start=0,docs=[SolrDocument{id=p1, > > name=[parent1], class=[foo.bar.parent1.1, foo.bar.parent1.2], > > _version_=1675826293154775040, children=[SolrDocument{id=c1, > > name=[child1], class=[foo.bar.child1], _version_=1675826293154775040, > > children=SolrDocument{id=gc1, name=[grandChild1], > > class=[foo.bar.grandchild1], _version_=1675826293154775040}}, > > SolrDocument{id=c2, name=[child2], class=[foo.bar.child2], > > _version_=1675826293154775040}]}]}} > > > > Regards, > > Alex. > > > > On Sun, 23 Aug 2020 at 01:38, Munendra S N <sn.munendr...@gmail.com> > > wrote: > > > > > > Hi Alex, > > > > > > Currently, Fixing the documentation for nested docs is under progress. > > More > > > context is available in this JIRA - > > > https://issues.apache.org/jira/browse/SOLR-14383. > > > > > > > > https://github.com/arafalov/SolrJTest/blob/master/src/com/solrstart/solrj/Main.java > > > > > > The child doc transformer needs to be specified as part of the fl > > parameter > > > like fl=*,[child] so that the descendants are returned for each matching > > > doc. As the query q=* matches all the documents, they are returned. If > > only > > > parent doc needs to be returned with descendants then, we should either > > use > > > block join query or query clause which matches only parent doc. > > > > > > Another thing I noticed in the code is that the child docs are indexed as > > > anonymous docs (similar to old syntax) instead of indexing them in the > > new > > > syntax. With this, the nested block will be indexed but since the schema > > > has _nested_path_ defined [child] doc transformer won't return any docs. > > > Anonymous child docs need parentFilter but specifying parentFilter with > > > _nested_path_ will lead to error > > > It is due to this check - > > > > > https://github.com/apache/lucene-solr/blob/1c8f4c988a07b08f83d85e27e59b43eed5e2ca2a/solr/core/src/java/org/apache/solr/response/transform/ChildDocTransformerFactory.java#L104 > > > > > > Instead of indexing the docs this way, > > > > > > > SolrInputDocument parent1 = new SolrInputDocument(); > > > > parent1.addField("id", "p1"); > > > > parent1.addField("name", "parent1"); > > > > parent1.addField("class", "foo.bar.parent1"); > > > > > > > > SolrInputDocument child1 = new SolrInputDocument(); > > > > > > > > parent1.addChildDocument(child1); > > > > child1.addField("id", "c1"); > > > > child1.addField("name", "child1"); > > > > child1.addField("class", "foo.bar.child1"); > > > > > > > > > > > modify it to indexing > > > > > > > SolrInputDocument parent1 = new SolrInputDocument(); > > > > parent1.addField("id", "p1"); > > > > parent1.addField("name", "parent1"); > > > > parent1.addField("class", "foo.bar.parent1"); > > > > > > > > SolrInputDocument child1 = new SolrInputDocument(); > > > > > > > > parent1.addField("sometag", Arrays.asList(child1)); > > > > child1.addField("id", "c1"); > > > > child1.addField("name", "child1"); > > > > child1.addField("class", "foo.bar.child1"); > > > > > > > > I think, once the documentation fixes get merged to master, indexing > > and > > > searching with the nested documents will become much clearer. > > > > > > Regards, > > > Munendra S N > > > > > > > > > > > > On Sun, Aug 23, 2020 at 5:18 AM Alexandre Rafalovitch < > > arafa...@gmail.com> > > > wrote: > > > > > > > Hello, > > > > > > > > I am trying to get up to date with both SolrJ and Nested Document > > > > implementation and not sure where I am failing with a basic test > > > > ( > > > > > > https://github.com/arafalov/SolrJTest/blob/master/src/com/solrstart/solrj/Main.java > > > > ). > > > > > > > > I am using Solr 8.6.1 with a core created with bin/solr create -c > > > > solrj (schemaless is still on). > > > > > > > > I then index a nested parent/child/grandchild document in and then > > > > query it back. Looking at debug it seems to go out fine as a nested > > > > doc but come back as a 3 individual ones. > > > > > > > > Output is: > > > > SolrInputDocument(fields: [id=p1, name=parent1, > > > > class=foo.bar.parent1], children: [SolrInputDocument(fields: [id=c1, > > > > name=child1, class=foo.bar.child1], children: > > > > [SolrInputDocument(fields: [id=gc1, name=grandChild1, > > > > class=foo.bar.grandchild1])])]) > > > > > > > > > > {responseHeader={status=0,QTime=1,params={q=*,wt=javabin,version=2}},response={numFound=3,numFoundExact=true,start=0,docs=[SolrDocument{id=gc1, > > > > name=[grandChild1], class=[foo.bar.grandchild1], > > > > _version_=1675769219435724800}, SolrDocument{id=c1, name=[child1], > > > > class=[foo.bar.child1], _version_=1675769219435724800}, > > > > SolrDocument{id=p1, name=[parent1], class=[foo.bar.parent1], > > > > _version_=1675769219435724800}]}} > > > > Found 3 documents > > > > > > > > Field: 'id' => 'gc1' > > > > Field: 'name' => '[grandChild1]' > > > > Field: 'class' => '[foo.bar.grandchild1]' > > > > Field: '_version_' => '1675769219435724800' > > > > Children: false > > > > > > > > Field: 'id' => 'c1' > > > > Field: 'name' => '[child1]' > > > > Field: 'class' => '[foo.bar.child1]' > > > > Field: '_version_' => '1675769219435724800' > > > > Children: false > > > > > > > > Field: 'id' => 'p1' > > > > Field: 'name' => '[parent1]' > > > > Field: 'class' => '[foo.bar.parent1]' > > > > Field: '_version_' => '1675769219435724800' > > > > Children: false > > > > > > > > Looking in Admin UI: > > > > * _root_ element is there and has 3 instances of 'p1' value > > > > * _nest_path_ (of type _nest_path_ !?!) is also there but is not > > populated > > > > * _nest_parent_ is not there > > > > > > > > I am not quite sure what that means and what other scheme modification > > > > (to the _default_) I need to do to get it to work. > > > > > > > > I also tried to reproduce the example in the documentation (e.g. > > > > > > https://lucene.apache.org/solr/guide/8_6/indexing-nested-documents.html > > > > and > > > > > > https://lucene.apache.org/solr/guide/8_6/searching-nested-documents.html#searching-nested-documents > > > > ) > > > > but both seem to also want some undiscussed schema (e.g. with ID field > > > > instead of id) and fail to execute against default schema. > > > > > > > > I am kind of stuck. Anybody has a working SolrJ/Nested example or > > > > ideas of what I missed. > > > > > > > > Regards, > > > > Alex. > > > > > >
