[jira] [Commented] (SOLR-4903) Solr sends all doc ids to all shards in the query counting facets

["Cassandra Targett (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SOLR-4903?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17303736#comment-17303736 ] 

Cassandra Targett commented on SOLR-4903:
-----------------------------------------

Someone asked me the other day when they might expect to hit this particular problem, and I said that I'd never seen this happen as far as I'm aware. A little offline discussion about it led a couple of us to believe that maybe it's no longer a problem? If it is still a problem, could we get a little more info about how or when it happens so we could investigate?

> Solr sends all doc ids to all shards in the query counting facets
> -----------------------------------------------------------------
>
>                 Key: SOLR-4903
>                 URL: https://issues.apache.org/jira/browse/SOLR-4903
>             Project: Solr
>          Issue Type: Improvement
>          Components: search
>    Affects Versions: 3.4, 4.3, 4.3.1
>            Reporter: Dmitry Kan
>            Priority: Major
>
> Setup: front end solr and shards.
> Summary: solr frontend sends all doc ids received from QueryComponent to all shards which causes POST request buffer size overflow.
> Symptoms:
> The query is: http://pastebin.com/0DndK1Cs
> I have omitted the shards parameter.
> The router log: http://pastebin.com/FTVH1WF3
> Notice the port of a shard, that is affected. That port changes all the time, even for the same request
> The log entry is prepended with lines:
> SEVERE: org.apache.solr.common.SolrException: Internal Server Error
> Internal Server Error
> (they are not in the pastebin link)
> The shard log: http://pastebin.com/exwCx3LX
> Suggestion: change the data structure in FacetComponent to send only doc ids that belong to a shard and not a concatenation of all doc ids.
> Why is this important: for scaling. Adding more shards will result in overflowing the POST request buffer at some point anyway.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)