lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Yonik Seeley <yo...@lucidworks.com>
Subject Re: Is this a SolrCloud bug? Or expected behavior?
Date Tue, 18 Sep 2012 16:21:12 GMT
On Tue, Sep 18, 2012 at 11:49 AM, jimtronic <jimtronic@gmail.com> wrote:
> I've got a set up as follows:
>
> - 13 cores
> - 2 servers
> - running Solr 4.0 Beta with numShards=1 and an external zookeeper.
>
> I'm trying to figure out why some complex queries are running so slowly in
> this setup versus quickly in a standalone mode.
>
> Given a query like: /select?q=(some complex query)
>
> It runs fast and gets faster (caches) when only running one server:
>
> 1. ?fl=*&q=(complex query)&wt=json&rows=24 (QTime 3)
>
> When, I issue the same query to the cluster and watch the logs, it looks
> like it's actually performing the query 3 times like so:
>
> 1. ?q=(complex
> query)&distrib=false&wt=javabin&rows=24&version=2&NOW=1347911018556&shard.url=(server1)|(server2)&fl=id,score&df=text&start=0&isShard=true&fsv=true
> (QTime 2)
>
> 2. ?ids=(ids from query
> 1)&distrib=false&wt=javabin&rows=24&version=2&df=text&fl=*&shard.url=(server1)|(server2)&NOW=1347911018556&start=0&q=(complex
> query)&isShard=true (QTime 4)
>
> 3.  ?fl=*&q=(complex query)&wt=json&rows=24 (QTime 459)


#1 and #2 are sub-requests (different phases of the top level request).
#3 is the log for the top level original request that made #1 and #2
and encompasses the time for both of them.


> Why is it performing #3? It already has everything it needs in #2 and #3
> seems to be really slow even when warmed and cached.
>
> As stated above, this query is fast when running on a single server that is
> warmed and cached.
>
> Since my query is complex, I could understand some slowness if I was
> attempting this across multiple shards, but since there's only one shard,
> shouldn't it just pick one server and query it?

A distributed search that happens to only hit a single shard is
something we haven't optimized for yet.

-Yonik
http://lucidworks.com

> I can "fix" this by adding "distrib=false" to my original queries, but then
> that sort of makes the whole cluster meaningless.
>
> Ideally, I'd just spin up a new server that attaches itself to zookeeper and
> add it to my load balancer and forget about it.
>
> Thanks!
> Jim

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org


Mime
View raw message