[
https://issues.apache.org/jira/browse/PHOENIX-1672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14330032#comment-14330032
]
Lars Hofhansl commented on PHOENIX-1672:
----------------------------------------
LGTM.
Note that until this is in, this can lead to arbitrarily incorrect results when
the locking assumption are actually used in HBase...
... As I am planning to do to avoid expensive locking and memory fencing at the
StoreScanner level, this could happen as early as 0.98 (I have a patch, but no
jira yet, and of course the HBasers have to agree, etc).
> RegionScanner.nextRaw contract not implemented correctly
> --------------------------------------------------------
>
> Key: PHOENIX-1672
> URL: https://issues.apache.org/jira/browse/PHOENIX-1672
> Project: Phoenix
> Issue Type: Bug
> Reporter: Lars Hofhansl
> Assignee: Andrew Purtell
> Fix For: 5.0.0
>
> Attachments: PHOENIX-1672.patch
>
>
> The nextRaw() contract (per Javadoc) requires the caller to hold a lock on
> the RegionScannerImpl object. Phoenix' *Aggregate observers do not do that as
> far as I can see.
> That might accidentally work but it not guaranteed to remain that way.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
