[
https://issues.apache.org/jira/browse/SOLR-8251?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14997531#comment-14997531
]
Hoss Man commented on SOLR-8251:
--------------------------------
bq. The mailing-list discussion referenced above can be found here ...
specifically:
https://mail-archives.apache.org/mod_mbox/lucene-solr-user/201511.mbox/%3CCAAJUMYrO32eXRiUvULJxRdNe0p9xmXyH5RawVAcDNZHh%2BffoZw%40mail.gmail.com%3E
bq. Of course executing a MatchAllDocsQuery just to Filter the results is
stupid. As Filters are now plain Queries in 5.x, why not execute all Filters in
Solr as a plain query? MatchAllDocsQuery is only useful for pure negative
queries.
This isn't as weird as it may seem when discussed abstractly ... practically
speaking it can be common for people to build up a faceting/navigation system
based on adding fqs to a "base query" as the user drills down. If the user
starts their navigation with a search then they have a non-trivial "q" for
their base query -- but if the user didn't start with some explicit search,
they may just be drilling down from a starting view of "all documents"
something like dismax/edismax with a {{q.alt=*:*}} is even designed around this
usage pattern.
----
Perhaps the "if (null==filter) \{ wrap main + filter in a BooleanQuery \}"
logic yonik mentioned LUCENE-6583 putting in SolrIndexSearcher should be
tweaked to include a similar optimization to what use to be in FilteredQuery ?
(ie: "if (null==main) \{ don't wrap, just use main=filter \}")
https://mail-archives.apache.org/mod_mbox/lucene-solr-user/201511.mbox/%3CCAB_8Yd8ForGSSXRC=hq+i6mkxhy5xy1kehvh1iml4pdx_ey...@mail.gmail.com%3E
> MatchAllDocsQuery is much slower in solr5.3.1 compare to solr4.7
> ----------------------------------------------------------------
>
> Key: SOLR-8251
> URL: https://issues.apache.org/jira/browse/SOLR-8251
> Project: Solr
> Issue Type: Bug
> Components: search
> Affects Versions: 5.3
> Reporter: wei shen
>
> I am trying to upgrade our production solr instance from 4.7 to 5.3.1.
> Unfortunately when I do load testing I find the MatchAllDocsQuery is much
> slower in solr 5.3.1 compare to 4.7. (solr 5.3.1 is faster in load test with
> queries other than MatchAllDocsQuery). I asked solr-user and discussed with
> Yonik Seeley. He confirmed that he can see the problem too comparing solr
> 5.3.1 and 4.10.
> here is the query I use:
> {code}
> q={!cache=false}*:*&fq=+categoryIdsPath:1001&fl=id&start=0&rows=2&debug=true
> {code}
> for me the query is consistently about 60-70% slower on solr5 than solr4.
> Yonik mentioned in his email "For me, 5.3.1
> is about 5x slower than 4.10 for this particular query."
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
