[
https://issues.apache.org/jira/browse/SOLR-12630?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16573404#comment-16573404
]
David Smiley commented on SOLR-12630:
-------------------------------------
Although not a strict alternative, I've been interested in ways of exploring
how to make updates faster, and thus make choices like the above more favorable
to simply sticking with block join if only updates cold be cheaper. Admittedly
these ideas are wildly out of scope with this issue (and broadly beneficial to
many use-cases unrelated to nested docs!). But what the heck; I'll share them
here any way. Some ideas:
* For full-text fields with an expensive analysis configuration, the update
process could be made to leverage term vectors to avoid re-analysis. Although
given the weight of them; I'm unsure how fruitful it is. I have a pretty good
idea of what's involved internally though, and have always wanted to try a
POC/experiment.
* We have numeric in-place updates but not yet String in-place updates for
docValue only fields. I recall there was WIP for that by Shai but for whatever
reason it stalled.
* Lucene's ParallelLeafReader is a hidden gem, allowing you to partition
updates of fields to different internal indexes (and other use-cases). PLR is
itself actually rather simple; it would take a lot of other (complex) code to
completely implement the strategy. If you only need to update certain fields,
typically small ones (while leaving lots of other data alone), then this is an
interesting possibility. Implementation of this would likely result in a new
attribute in the schema on fields called "fieldGroup" or some-such to segment
certain fields aside that are able to be updated faster alone, since it
wouldn't involve re-indexing fields in other field groups.
> Query children without using block join
> ---------------------------------------
>
> Key: SOLR-12630
> URL: https://issues.apache.org/jira/browse/SOLR-12630
> Project: Solr
> Issue Type: Improvement
> Security Level: Public(Default Security Level. Issues are Public)
> Reporter: mosh
> Priority: Major
>
> Currently work is being put into Solr to support nested document hierarchies,
> keeping the original relation between parent and child docs
> ([SOLR-12298|https://issues.apache.org/jira/browse/SOLR-12298],
> [SOLR-12441|https://issues.apache.org/jira/browse/SOLR-12441]). Currently the
> only way to query child documents is using BlockJoin. I was thinking perhaps
> we could introduce a way to query documents that are not indexed in the same
> block with a different transformer, or perhaps add this feature to the
> ChildDocTransformer.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
