[
https://issues.apache.org/jira/browse/PHOENIX-1313?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14169371#comment-14169371
]
rajeshbabu commented on PHOENIX-1313:
-------------------------------------
[~jamestaylor]
bq. You want the start/stop key to be set according to the range on the
original scan (i.e. the one calculated by WhereOptimizer). Otherwise you're
doing a full region scan when it's not necessary.
Yes James and also on StaleRegionBoundaryCacheException I want to limit the
scans to the daughter regions only by seeing the old scan start key and
EXPECTED_UPPER_REGION_KEY attribute value.
Other wise currently we are always preparing scans for all the regions and
there may be unnecessary duplicate scans can happen.
bq. Do the scan start/stop look correct to you?
After split the scan boundaries are becoming wrong because we are prefixing the
region start key once again which we have already done first time.
if EXPECTED_UPPER_REGION_KEY attribute set to scan that means we already
prefixed the start key to scan boundaries so we need to stripe out the start
key first and later any way we are prefixing proper start key.
I feel these changes are mandatory to make
LocalIndexIT.testLocalIndexScanAfterRegionSplit to pass.
> Investigate why LocalIndexIT.testLocalIndexScanAfterRegionSplit() is failing
> ----------------------------------------------------------------------------
>
> Key: PHOENIX-1313
> URL: https://issues.apache.org/jira/browse/PHOENIX-1313
> Project: Phoenix
> Issue Type: Bug
> Affects Versions: 4.2
> Reporter: James Taylor
> Assignee: rajeshbabu
> Fix For: 5.0.0, 4.2
>
> Attachments: PHOENIX-1313.patch, PHOENIX-1313_2.patch
>
>
> I checked in a fairly big change
> (https://git-wip-us.apache.org/repos/asf?p=phoenix.git;a=commit;h=d018cc1c6e01d9836de6e67af4f8b91de3269bfd)
> to improve encapsulation and fix a bunch of stats issues, and the following
> test started failing: LocalIndexIT.testLocalIndexScanAfterRegionSplit(). I
> tried to diagnose it, but that test is complicated. Would you mind taking a
> look [~rajesh23]?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
