[
https://issues.apache.org/jira/browse/HBASE-17355?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15766302#comment-15766302
]
ramkrishna.s.vasudevan commented on HBASE-17355:
------------------------------------------------
Note that in the attached screen shot, the PE tool was run with 50 cols per row.
> Create a simplifed version of flush scanner
> -------------------------------------------
>
> Key: HBASE-17355
> URL: https://issues.apache.org/jira/browse/HBASE-17355
> Project: HBase
> Issue Type: Sub-task
> Affects Versions: 2.0.0
> Reporter: ramkrishna.s.vasudevan
> Assignee: ramkrishna.s.vasudevan
> Fix For: 2.0.0
>
> Attachments: HBASE-17354.patch, after patch.png, before patch.png
>
>
> Currently we use StoreScanner for performing the flushes which actuallly goes
> row by row. Probably that is not needed and we could always go ahead with a
> simple loop in collecting the cells and writing them to the file. Inside
> write path we have the required sanity check so it is not needed that the
> store scanner does a sanity check.
> Also the limit that could be retrieved in one next() call could be equivalent
> to the block size configured as we do for compaction.
> Are there any filters that we want to do (i mean any version check or
> deletion) that we need to check in flush? If so then this simplified version
> will not work. I may be missing something but if so we need to see what are
> those and add it here.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
