[
https://issues.apache.org/jira/browse/HBASE-8420?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13656370#comment-13656370
]
Jean-Daniel Cryans commented on HBASE-8420:
-------------------------------------------
On the latest patch:
- I think you have a leak if the client doesn't come back to get the
prefetched data, at least you should decrement globalPrefetchedResultSize in
RegionScannerHolder.closeScanner().
- 512MB seems big, I'd be more comfortable with 128MB.
- Is it possible to unit test this?
- Have you run a concurrent workload against capped pre-fetching?
- Should we report on the state of globalPrefetchedResultSize? I'm wondering
what could be helpful to see when debugging scan slowness WRT this rate
limiting. Maybe a TRACE every time we skip a prefetch and output
totalPrefetchedResultSize?
> Port HBASE-6874 Implement prefetching for scanners from 0.89-fb
> -----------------------------------------------------------------
>
> Key: HBASE-8420
> URL: https://issues.apache.org/jira/browse/HBASE-8420
> Project: HBase
> Issue Type: Improvement
> Reporter: Jimmy Xiang
> Assignee: Jimmy Xiang
> Attachments: 0.94-8420_v1.patch, 0.94-8420_v2.patch,
> trunk-8420_v1.patch, trunk-8420_v2.patch, trunk-8420_v3.patch
>
>
> This should help scanner performance. We should have it in trunk.
--
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
