You are viewing a plain text version of this content. The canonical link for it is here.
Posted to solr-dev@lucene.apache.org by "Yonik Seeley (JIRA)" <ji...@apache.org> on 2007/02/03 00:13:05 UTC

[jira] Created: (SOLR-138) queries can stack up

queries can stack up
--------------------

                 Key: SOLR-138
                 URL: https://issues.apache.org/jira/browse/SOLR-138
             Project: Solr
          Issue Type: Bug
            Reporter: Yonik Seeley


A burst of "expensive" queries can cause contention within lucene, greatly reducing effective throughput and causing more and more queries to stack up.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Commented: (SOLR-138) queries can stack up

Posted by "Lance Norskog (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/SOLR-138?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12579094#action_12579094 ] 

Lance Norskog commented on SOLR-138:
------------------------------------

Our use case is the classic battle between interactive use and running reports: reports use long queries that shut out interactive use.

One option is to have two search servlets instead of one, where each can have a separate queueing depth. One servlet is reserved for long queries and may only have one or two queries active at one time, while the other servlet is for smaller interactive queries and might allow 5 concurrent queries.

> queries can stack up
> --------------------
>
>                 Key: SOLR-138
>                 URL: https://issues.apache.org/jira/browse/SOLR-138
>             Project: Solr
>          Issue Type: Bug
>            Reporter: Yonik Seeley
>
> A burst of "expensive" queries can cause contention within lucene, greatly reducing effective throughput and causing more and more queries to stack up.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Issue Comment Edited: (SOLR-138) queries can stack up

Posted by "Ian Holsman (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/SOLR-138?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12579165#action_12579165 ] 

iholsman edited comment on SOLR-138 at 3/15/08 10:16 PM:
------------------------------------------------------------

the 2nd option is to use a query timeout like the one in #SOLR-502 so that "interactive" queries can have a fixed end time which kill the badly behaved queries

      was (Author: iholsman):
    the 2nd option is to use a query timeout like the one in #502 so that "interactive" queries can have a fixed end time which kill the badly behaved queries
  
> queries can stack up
> --------------------
>
>                 Key: SOLR-138
>                 URL: https://issues.apache.org/jira/browse/SOLR-138
>             Project: Solr
>          Issue Type: Bug
>            Reporter: Yonik Seeley
>
> A burst of "expensive" queries can cause contention within lucene, greatly reducing effective throughput and causing more and more queries to stack up.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Commented: (SOLR-138) queries can stack up

Posted by "Ian Holsman (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/SOLR-138?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12579165#action_12579165 ] 

Ian Holsman commented on SOLR-138:
----------------------------------

the 2nd option is to use a query timeout like the one in #502 so that "interactive" queries can have a fixed end time which kill the badly behaved queries

> queries can stack up
> --------------------
>
>                 Key: SOLR-138
>                 URL: https://issues.apache.org/jira/browse/SOLR-138
>             Project: Solr
>          Issue Type: Bug
>            Reporter: Yonik Seeley
>
> A burst of "expensive" queries can cause contention within lucene, greatly reducing effective throughput and causing more and more queries to stack up.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Re: [jira] Commented: (SOLR-138) queries can stack up

Posted by Erik Hatcher <er...@ehatchersolutions.com>.
On Feb 2, 2007, at 6:32 PM, Yonik Seeley (JIRA) wrote:

>
>     [ https://issues.apache.org/jira/browse/SOLR-138? 
> page=com.atlassian.jira.plugin.system.issuetabpanels:comment- 
> tabpanel#action_12469907 ]
>
> Yonik Seeley commented on SOLR-138:
> -----------------------------------
>
> I'm starting to like this idea...
> Before, if some queries were taking a long time, you couldn't tell  
> which ones because they don't get logged until after completion.
>
> But if you are queuing in a manner that lets you keep track of  
> currently executing requests, you could export that info to the  
> admin page, along with how long each query has been running.

+1

I presume this is related, but I've been bummed when making queries  
to Solr that cause an exception (yeah, my own request handlers  
probably) and I don't see in the jetty console the request logged  
*before* the exception.  That would be mighty handy!  Especially with  
admin reporting.  We're getting closer to needing that JMX interface,  
aren't we?  :)

	Erik


[jira] Commented: (SOLR-138) queries can stack up

Posted by "Yonik Seeley (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/SOLR-138?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12469907 ] 

Yonik Seeley commented on SOLR-138:
-----------------------------------

I'm starting to like this idea...
Before, if some queries were taking a long time, you couldn't tell which ones because they don't get logged until after completion.

But if you are queuing in a manner that lets you keep track of currently executing requests, you could export that info to the admin page, along with how long each query has been running.



> queries can stack up
> --------------------
>
>                 Key: SOLR-138
>                 URL: https://issues.apache.org/jira/browse/SOLR-138
>             Project: Solr
>          Issue Type: Bug
>            Reporter: Yonik Seeley
>
> A burst of "expensive" queries can cause contention within lucene, greatly reducing effective throughput and causing more and more queries to stack up.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Commented: (SOLR-138) queries can stack up

Posted by "Yonik Seeley (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/SOLR-138?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12469905 ] 

Yonik Seeley commented on SOLR-138:
-----------------------------------

One method to "fix" this would be to limit the number of queries that may be executing at any one time (queueing).
http://www.nabble.com/TermInfosReader-lazy-term-index-reading-tf3162657.html#a8776274

One could do this in the servlet container by limiting the number of threads in the thread pool, but would that lock out the admin pages?

A more flexible way might be to put the queueing in the SolrCore.exec() method.
Admin pages could bypass the exec() method by calling a handler directly, or could set a priority on the request that avoided queueing.


> queries can stack up
> --------------------
>
>                 Key: SOLR-138
>                 URL: https://issues.apache.org/jira/browse/SOLR-138
>             Project: Solr
>          Issue Type: Bug
>            Reporter: Yonik Seeley
>
> A burst of "expensive" queries can cause contention within lucene, greatly reducing effective throughput and causing more and more queries to stack up.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.