[
https://issues.apache.org/jira/browse/SOLR-1787?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12837424#action_12837424
]
Michael Henson commented on SOLR-1787:
--------------------------------------
There isn't any new code to make a pre-warming call. The design relies on
filling the cache on the first call to getAllNonCachedRows().
I'm also relying on the synchronize block of
ThreadedEntityProcessorWrapper.nextRow() to block other threads while the cache
filling query runs. It looks like I'm missing some aspect of multi-threading
because I get SQLException's with "could not execute query" when I have
threads="3" on the root entity.
> CachedSqlEntityProcessor pre-warmed cache use case
> --------------------------------------------------
>
> Key: SOLR-1787
> URL: https://issues.apache.org/jira/browse/SOLR-1787
> Project: Solr
> Issue Type: Improvement
> Components: contrib - DataImportHandler
> Affects Versions: 1.5
> Environment: jdk 1.6.x, windows xp, tomcat 6.x
> Reporter: Michael Henson
> Assignee: Noble Paul
> Priority: Minor
> Fix For: 1.5
>
> Attachments: solr-1787.patch
>
>
> The CachedSqlEntityProcessor currently builds a cache of rows it sees as it
> goes, so later requests for that same key can be served from data that has
> already been fetched. The primary query could be written to fetch all
> possible rows, which would then be set into the cache on the first request
> for a row. In that case the database would only receive another query when
> there is a cache miss. However, the query it would execute is the one that
> pulls all rows, negating any performance gain.
> This patch adds the ability to configure behavior on cache miss with the
> "onCacheMiss" attribute on an "entity" tag in the data-config.xml file. The
> current behavior is the default, corresponding to the setting
> onCacheMiss="fill". Any other value explicitly given for onCacheMiss will
> cause cache misses to be ignored - no query will be made to the db to fulfill
> them.
> I've encountered two cases where this capability is useful:
> 1. Relatively small datasets, such as category id -> category name mappings,
> which will not change during the course of indexing.
> 2. Queries which are heavy on db resources per-query, particularly if the
> query for an individual record is slow, and can't be fixed easily on the db
> side for whatever reason.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
