[
https://issues.apache.org/jira/browse/SOLR-564?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12595960#action_12595960
]
Jason Rutherglen commented on SOLR-564:
---------------------------------------
The patches are a way to see if integration with Ocean is possible.
Some issues that have been cropping up include the anywhere direct field cache
access is performed. This includes SimpleFacets, ValueSource, and other
places. These are areas I do not have an immediate solution to and so the
first version of OceanCore will not implement.
Because Ocean makes use field caching per segment, and new segments are being
created all the time, something like MultiDocSet would need to be created to
fit Ocean into the existing Lucene field cache framework. However this is
impossible given Lucene field caching uses direct array access whereas DocSet
is an interface. Ocean has a flexible field caching framework that includes a
primitive based StringIndex like set of classes which save ram and allow fast
range queries.
Performing work inside the segments and then merging the results later is how
the Ocean is designed. I'm not sure how much work it would take to have Solr
do the same. The concept is similar to the distributed indexing in Solr.
> Realtime search in Solr
> -----------------------
>
> Key: SOLR-564
> URL: https://issues.apache.org/jira/browse/SOLR-564
> Project: Solr
> Issue Type: New Feature
> Components: replication, search
> Affects Versions: 1.3
> Reporter: Jason Rutherglen
>
> Before when I looked at this, the changes required to make Solr realtime
> would seem to break the rest of Solr. Is this still the case? In project
> Ocean http://code.google.com/p/oceansearch/ there is a realtime core however
> integrating into Solr has looked like a redesign of the guts of Solr.
> - Support for replication per update to transaction log
> - Custom realtime index creation
> - Filter and facet merging
> - Custom IndexSearcher that ties into realtime subsystem
> - Custom SolrCore that ties into realtime subsystem
> Is there a way to plug into these low level Solr functions without a massive
> redesign? A key area of concern is the doclist caching which is not used in
> realtime search because after every update the doclists are no longer valid.
> The doclist caching and handling is default in SolrCore. Ocean relies on a
> custom threaded MultiSearcher rather than a single IndexSearcher is a
> difficulty. DirectUpdateHandler2 works directly on IndexWriter is
> problematic.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
