[
https://issues.apache.org/jira/browse/SOLR-9668?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Mikhail Khludnev updated SOLR-9668:
-----------------------------------
Description:
SolrEntityProcessor paginates using the start and rows parameters which can be
very inefficient at large offsets. In fact, the current implementation is
impracticable to import large amounts of data (10M+ documents) because the data
import rate degrades from 1000docs/second to 10docs/second and the import gets
stuck.
This patch introduces support for cursor paging which offers more or less
predictable performance. In my tests the time to fetch the 1st and 1000th pages
was about the same and the data import rate was stable throughout the entire
import.
To enable cursor paging a user needs to:
* add {{cursorMark='true'}} (!) attribute in the entity configuration;
* "sort" attribute in the entity configuration see note about sort at
https://cwiki.apache.org/confluence/display/solr/Pagination+of+Results ;
* remove {{timeout}} attribute.
{code}
<?xml version="1.0" encoding="UTF-8" ?>
<dataConfig>
<document>
<entity name="se" processor="SolrEntityProcessor"
query="*:*"
rows="1000"
cursorMark='true'
sort="id asc"
url="http://localhost:8983/solr/collection1";>
</entity>
</document>
</dataConfig>
{code}
If the "sort" attribute is missing then the default start/rows pagination is
used.
was:
SolrEntityProcessor paginates using the start and rows parameters which can be
very inefficient at large offsets. In fact, the current implementation is
impracticable to import large amounts of data (10M+ documents) because the data
import rate degrades from 1000docs/second to 10docs/second and the import gets
stuck.
This patch introduces support for cursor paging which offers more or less
predictable performance. In my tests the time to fetch the 1st and 1000th pages
was about the same and the data import rate was stable throughout the entire
import.
To enable cursor paging a user needs to add a "sort" attribute in the entity
configuration:
{code}
<?xml version="1.0" encoding="UTF-8" ?>
<dataConfig>
<document>
<entity name="se" processor="SolrEntityProcessor"
query="*:*"
rows="1000"
sort="id asc" <!-- turns on cursor paging. Must be a uniqueKey field tie
breaker -->
url="http://localhost:8983/solr/collection1";>
</entity>
</document>
</dataConfig>
{code}
If the "sort" attribute is missing then the default start/rows pagination is
used.
> Support cursor paging in SolrEntityProcessor
> --------------------------------------------
>
> Key: SOLR-9668
> URL: https://issues.apache.org/jira/browse/SOLR-9668
> Project: Solr
> Issue Type: Improvement
> Security Level: Public(Default Security Level. Issues are Public)
> Components: contrib - DataImportHandler
> Reporter: Yegor Kozlov
> Assignee: Mikhail Khludnev
> Priority: Minor
> Labels: dataimportHandler
> Fix For: master (7.0)
>
> Attachments: SOLR-9668.patch, SOLR-9668.patch
>
>
> SolrEntityProcessor paginates using the start and rows parameters which can
> be very inefficient at large offsets. In fact, the current implementation is
> impracticable to import large amounts of data (10M+ documents) because the
> data import rate degrades from 1000docs/second to 10docs/second and the
> import gets stuck.
> This patch introduces support for cursor paging which offers more or less
> predictable performance. In my tests the time to fetch the 1st and 1000th
> pages was about the same and the data import rate was stable throughout the
> entire import.
> To enable cursor paging a user needs to:
> * add {{cursorMark='true'}} (!) attribute in the entity configuration;
> * "sort" attribute in the entity configuration see note about sort at
> https://cwiki.apache.org/confluence/display/solr/Pagination+of+Results ;
> * remove {{timeout}} attribute.
> {code}
> <?xml version="1.0" encoding="UTF-8" ?>
> <dataConfig>
> <document>
> <entity name="se" processor="SolrEntityProcessor"
> query="*:*"
> rows="1000"
> cursorMark='true'
> sort="id asc"
> url="http://localhost:8983/solr/collection1";>
> </entity>
> </document>
> </dataConfig>
> {code}
> If the "sort" attribute is missing then the default start/rows pagination is
> used.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
