[
https://issues.apache.org/jira/browse/SOLR-8349?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15144017#comment-15144017
]
David Smiley commented on SOLR-8349:
------------------------------------
1. the classLoader separation is simply because there are libs and conf/
resources per-core, as cores might be for entirely separate purposes with
different dependencies that might not even be compatible if it were all
together.
2. RE memory-leak: I propose the weakValues() feature of Guava's cache. It's
a nice cache :-) No reference counting needed; we're in a VM that GCs.
3. you're right; this isn't just about analysis components. But it would
probably be incompatible with anything that implements CoreAware or SchemaAware
as both those things are tied to a SolrCore.
4. RE "what's left to be loaded in the core level loader": I think this re-use
feature should be explicitly opt-in. Most stuff will cotinue as-is -- core
level loader. Docs for HunspellFactory on the ref guide could be augmented to
suggest flagging global-reuse because we know it uses a lot of memory.
5. RE keys: I think just the W3C DOM Node.java instance is probably fine. We
might want to double-check it's equals() method makes sense and doesn't refer
to the parent, or if it does find a way to clone/detach it. Actually I think
it does; so we'd call cloneNode(true) and use the result of that. I'm not
concerned on the efficiency of the keys; this cache isn't going to be hit hard.
Any way, lets continue this discussion if need be on a follow-on issue for
actually _using_ the cache that this issue will create. Cool?
> Allow sharing of large in memory data structures across cores
> -------------------------------------------------------------
>
> Key: SOLR-8349
> URL: https://issues.apache.org/jira/browse/SOLR-8349
> Project: Solr
> Issue Type: Improvement
> Components: Server
> Affects Versions: 5.3
> Reporter: Gus Heck
> Attachments: SOLR-8349.patch
>
>
> In some cases search components or analysis classes may utilize a large
> dictionary or other in-memory structure. When multiple cores are loaded with
> identical configurations utilizing this large in memory structure, each core
> holds it's own copy in memory. This has been noted in the past and a specific
> case reported in SOLR-3443. This patch provides a generalized capability, and
> if accepted, this capability will then be used to fix SOLR-3443.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
