[
https://issues.apache.org/jira/browse/LUCENE-1483?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12655976#action_12655976
]
Michael McCandless commented on LUCENE-1483:
--------------------------------------------
Duh: I just realized that when we switched back to a single pqueue for
gathering results across the N subreaders, we lost the original
intended "benefit" for this issue. Hard to keep the forrest in mind
when looking at all the trees....
Ie, we are now (again) creating a single FieldSortedHitQueue, which
pulls the FieldCache for the entire MultiReader, not per-segment. So
warming time is still slow, when sorting by fields.
Really we've "stumbled" on 2 rather different optimizations:
# Run Scorer at the "sub reader" level: this gains performance
because you save the cost of going through a MultiReader. This
requires the new DocCollector class, so we can setDocBase(...).
# Do collection (sort comparison w/ pqueue) at the "sub reader"
level: this gains warming performance because we only ask for
FieldCache for each subreader. But, it seems to hurt search
performance (pqueue comparison & insertion cost went up), so it's
no longer a no-brainer tradeoff (by default at least).
Given that #1 has emerged as a tentatively fairly compelling gain, I
now think we should decouple it from #2. Even though #2 was the
original intent here, let's now morph this issue into addressing #1
(since that's what current patch does), and I'll open a new issue for
#2?
> Change IndexSearcher to use MultiSearcher semantics for multiple subreaders
> ---------------------------------------------------------------------------
>
> Key: LUCENE-1483
> URL: https://issues.apache.org/jira/browse/LUCENE-1483
> Project: Lucene - Java
> Issue Type: Improvement
> Affects Versions: 2.9
> Reporter: Mark Miller
> Priority: Minor
> Attachments: LUCENE-1483.patch, LUCENE-1483.patch, LUCENE-1483.patch,
> LUCENE-1483.patch, LUCENE-1483.patch, LUCENE-1483.patch, LUCENE-1483.patch,
> LUCENE-1483.patch, LUCENE-1483.patch, LUCENE-1483.patch
>
>
> FieldCache and Filters are forced down to a single segment reader, allowing
> for individual segment reloading on reopen.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
To unsubscribe, e-mail: java-dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: java-dev-h...@lucene.apache.org
