[
https://issues.apache.org/jira/browse/HBASE-15935?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15366607#comment-15366607
]
Joseph commented on HBASE-15935:
--------------------------------
I guess coming back to this, we also send a non-zero return code on Generation
jobs that do fail this check along with the Counter's of the failed types. So
the Loop will fail the instant the first failed Generation phase returns. Also
it might be difficult to kill the Generation mapper job, because I'm not sure
if the error will be reproducible, so if we throw an error the task might
succeed on another retry? So I think it might not be necessary to fail the task
that soon, what are your thoughts?
> Have a separate Walker task running concurrently with Generator
> ------------------------------------------------------------------
>
> Key: HBASE-15935
> URL: https://issues.apache.org/jira/browse/HBASE-15935
> Project: HBase
> Issue Type: Sub-task
> Components: integration tests
> Reporter: Joseph
> Assignee: Joseph
> Priority: Minor
> Attachments: HBASE-15935.patch
>
>
> Keep track of which linked lists have been flushed in an HBase table, so that
> we can concurrently Walk these lists during the Generation phase. This will
> allow us to test:
> 1. HBase under concurrent read/writes
> 2. Availability of data immediately after flushes (as opposed to waiting till
> the Verification phase)
> The review diff can be found at:
> https://reviews.apache.org/r/48294/
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
