[
https://issues.apache.org/jira/browse/HBASE-8188?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Lars Hofhansl updated HBASE-8188:
---------------------------------
Attachment: 8188-0.94-experimental.txt
So now that I said this, I have thought of a more radical/experimental change
that does away with SQM.setRow(...) altogether.
The first row KV is passed to SQM anyway, so it can avoid another comparison,
and also make the logic more self-contained.
Passes:
* TestQueryMatcher
* TestFromClientSide
* TestAtomicOperation
* My own local performance testing
I am still validating my earlier performance claim.
Please have a look, if you don't think I am crazy I'll make a 0.95/0.96 patch.
> Avoid unnecessary row compare in StoreScanner
> ---------------------------------------------
>
> Key: HBASE-8188
> URL: https://issues.apache.org/jira/browse/HBASE-8188
> Project: HBase
> Issue Type: Bug
> Reporter: Lars Hofhansl
> Assignee: Lars Hofhansl
> Attachments: 8188-0.94-experimental.txt, 8188-0.94.txt,
> 8188-0.96.txt, 8188-0.96.txt, 8188-0.96.txt, 8188-0.96.txt
>
>
> When scanner batching disable (which is the default) a row compare in
> StoreScanner avoided, because only entire rows of data are requested.
> This provides a slight performance gain, especially for tall tables (and when
> all data is in the cache)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
