[
https://issues.apache.org/jira/browse/SOLR-18034?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18046059#comment-18046059
]
Chris M. Hostetter edited comment on SOLR-18034 at 12/18/25 12:42 AM:
----------------------------------------------------------------------
I'm creating this issue based on these comments [~abenedetti] made in
SOLR-17974...
{quote}Nested vectors -> build nested documents automatically (as this is the
route we've decided in Lucene after many discussions)
...
I'll also open a draft Pull request soon with some ideas, at least for the HNSW
use case (pretty much an alternative syntax to indexing nested vectors)
{quote}
...I haven't been following the lucene level discussions about supporting
multi-valued HNSW docs (or that a decision had been made that nested documents
was *the* solution) but I _DO_ have existing code that makes it easy to build
parent->(many)child docs where each child contains a single vector field value
(used for dealing with chunked vectors at my day job, which I was given
permission to share)
Attaching this for consideration.
The goal of this plugin (as i currently use it) is to make external clients as
agnostic as possible to the internal parent/child structures:
* External clients *ONLY* deal with indexing "parent" documents
* The internal plugins ensure that:
** One child documents is automatically created for each (individual) vector
in the "parent" document
** Any DBQs against parent documents automatically wrap all children of those
documents. (which we can do safely/automatically because our only use of
hierarchical documents are these vector based children; and clients should
never be sending us any DBQs _intentionally_ matching vector-child documents)
was (Author: hossman):
I'm creating this issue based on these comments [~abenedetti] made in
SOLR-17974...
{quote}Nested vectors -> build nested documents automatically (as this is the
route we've decided in Lucene after many discussions)
{quote}
...I haven't been following the lucene level discussions about supporting
multi-valued HNSW docs (or that a decision had been made that nested documents
was *the* solution) but I _DO_ have existing code that makes it easy to build
parent->(many)child docs where each child contains a single vector field value
(used for dealing with chunked vectors at my day job, which I was given
permission to share)
Attaching this for consideration.
The goal of this plugin (as i currently use it) is to make external clients as
agnostic as possible to the internal parent/child structures:
* External clients *ONLY* deal with indexing "parent" documents
* The internal plugins ensure that:
** One child documents is automatically created for each (individual) vector
in the "parent" document
** Any DBQs against parent documents automatically wrap all children of those
documents. (which we can do safely/automatically because our only use of
hierarchical documents are these vector based children; and clients should
never be sending us any DBQs _intentionally_ matching vector-child documents)
> Indexing helper logic for nested vector documents
> -------------------------------------------------
>
> Key: SOLR-18034
> URL: https://issues.apache.org/jira/browse/SOLR-18034
> Project: Solr
> Issue Type: New Feature
> Components: vector-search
> Reporter: Chris M. Hostetter
> Priority: Major
> Attachments: SOLR-18034.patch
>
>
> We can provide (server side) index time assistance for making it easy to
> create parent/child document structures to model nested vector documents.
> ie: the index side bits of SOLR-17736
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
