[
https://issues.apache.org/jira/browse/PHOENIX-4950?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16638690#comment-16638690
]
James Taylor commented on PHOENIX-4950:
---------------------------------------
FWIW, BaseResultIterators.getParallelScans(byte[] startKey, byte[] stopKey)
already detects if there are regions without guideposts when it intersects the
guideposts with the region boundaries. In this case, the GuidePostsInfo
timestamp array for the position corresponding to the guidepost index will be
null (i.e. gps.getGuidePostTimestamps()[guideIndex] == null).
> SYSTEM.STATS table should include the stopKey of guide posts
> ------------------------------------------------------------
>
> Key: PHOENIX-4950
> URL: https://issues.apache.org/jira/browse/PHOENIX-4950
> Project: Phoenix
> Issue Type: Improvement
> Reporter: Karan Mehta
> Priority: Major
>
> SYSTEM.STATS table should include stop key for each guidepost along with
> start key. This helps in determining if there are any gaps/holes in the
> statistics. UPDATE STATISTICS creates separate scanners for each region and
> runs it in parallel. This can potentially fail if particular region is
> unavailable or SYSTEM.STATS is unable when the region tries to commit stats.
> This Jira will help increase overall resiliency of stats feature. This
> concept is similar to how hbck determines holes in the chain.
> Kudos to [~sukumaddineni] for the idea!
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
