[
https://issues.apache.org/jira/browse/CASSANDRA-6638?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13886570#comment-13886570
]
Ignace Desimpel commented on CASSANDRA-6638:
--------------------------------------------
Thanks for the simple patch!
Related to efficiency :
Suppose we have sstable data in every other range.
Then the first range gets data matching the range. The second does not, and a
'file' seek is
done to the first key greater than the left of the this second range, thus to
the first sample key in the third range.
And then a loop is started over all the ifile entries until end of ifile file
(there is no upper boundary check)!
That would repeat itself over and over again if we happen to have such a data
and range arrangement and depending on the number of vnodes.
Correct? That means a lot of work for nothing?
> SSTableScanner can Skip Rows with vnodes
> ----------------------------------------
>
> Key: CASSANDRA-6638
> URL: https://issues.apache.org/jira/browse/CASSANDRA-6638
> Project: Cassandra
> Issue Type: Bug
> Components: Core
> Reporter: Tyler Hobbs
> Assignee: Sylvain Lebresne
> Priority: Blocker
> Fix For: 2.0.5
>
> Attachments: 6638-repro-test.txt, 6638.txt
>
>
> CASSANDRA-2524 added multiple range support to SSTableScanner, but it looks
> like there is at least one case where keys can be skipped. This can result
> in cleanup removing legitimate keys.
> See the attached patch that adds a unit test to reproduce the case.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
