[ 
https://issues.apache.org/jira/browse/SOLR-12572?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16567671#comment-16567671
 ] 

Varun Thacker commented on SOLR-12572:
--------------------------------------

I did some testing on my local machine. The collection was 1 shard 1 replica. 
The collection has 25M docs where {{number_i}} was 1 though 25M .
{code:java}
http://localhost:8983/solr/test_export/export?q=*:*&sort=number_i 
desc&fl=number_i{code}
Results from 2 runs shows a ~30% improvement.  

without patch : 886s , 866s
with patch : 664s , 674s

The worrying thing though is that 7.4 is faster than master. I suspect that has 
something to do with SOLR-11598 . I'm creating a new Jira to track that 
slowdown .

This patch looks good to me otherwise so I plan on committing it in the next 
few days

> Reuse fieldvalues computed while sorting at writing in ExportWriter
> -------------------------------------------------------------------
>
>                 Key: SOLR-12572
>                 URL: https://issues.apache.org/jira/browse/SOLR-12572
>             Project: Solr
>          Issue Type: Improvement
>      Security Level: Public(Default Security Level. Issues are Public) 
>          Components: streaming expressions
>            Reporter: Amrit Sarkar
>            Assignee: Varun Thacker
>            Priority: Minor
>         Attachments: SOLR-12572.patch, SOLR-12572.patch, SOLR-12572.patch, 
> SOLR-12572.patch, SOLR-12572.patch, SOLR-12572.patch
>
>
> While exporting result through "/export" handler,
> {code:java}
> http://localhost:8983/solr/core_name/export?q=my-query&sort=severity+desc,timestamp+desc&fl=severity,timestamp,msg
> {code}
> Doc-values are sought for all the {{sort}} fields defined (in this example 
> 'severity, 'timestamp'). When we stream out docs we again make doc-value 
> seeks against the {{fl}} fields ('severity','timestamp','msg') . 
> In most common use-cases we have {{fl = sort}} fields, or atleast the sort 
> fields are subset of {{fl}} fields, so if we can *pre-collect* the values 
> while sorting it, we can reduce the doc-value seeks potentially bringing 
> *speed improvement*.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to