[
https://issues.apache.org/jira/browse/HBASE-21021?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16581799#comment-16581799
]
Andrew Purtell commented on HBASE-21021:
----------------------------------------
bq. The test passes on master after considering the change and asserting for
returned result isEmpty(). Hence, the ordering issue does not exist in master.
Ok, but we still need a patch for master, please, just the unit test, just to
let us know the issue isn't there now and to catch it as a regression going
forward. Can't have coverage in branch-1 that doesn't extend forward for an
expectation that should also carry forward. Once done +1 for branch-1 for me.
I think this is a bug. Increment results are ordered (in effect) so this
violates the principle of least surprise. Let's make sure Javadoc for append
specifies the results are returned in order. Therefore I would like to further
apply this to branch-1.2 and branch-1.3 even though there is a behavioral
change. I guess ordering of the return results have not been an issue to date,
because this issue has not been filed previously to my knowledge, so a further
reordering won't matter, but going forward semantics will better match
expectations. Let me know if you have a concern [~busbey] [~toffer]
> Result returned by Append operation should be ordered
> -----------------------------------------------------
>
> Key: HBASE-21021
> URL: https://issues.apache.org/jira/browse/HBASE-21021
> Project: HBase
> Issue Type: Bug
> Affects Versions: 1.3.0, 1.5.0
> Reporter: Nihal Jain
> Assignee: Nihal Jain
> Priority: Major
> Fix For: 1.5.0
>
> Attachments: HBASE-21021.branch-1.001.patch
>
>
> *Problem:*
> The result returned by the append operation should be ordered. Currently, it
> returns an unordered list, which may cause problems like if the user tries to
> perform Result.getValue(byte[] family, byte[] qualifier), even if the
> returned result has a value corresponding to (family, qualifier), the method
> may return null as it performs a binary search over the unsorted result
> (which should have been sorted actually).
>
> The result is enumerated by iterating over each entry of tempMemstore hashmap
> (which will never be ordered) and adding the values (see
> [HRegion.java#L7882|https://github.com/apache/hbase/blob/1b50fe53724aa62a242b74444f64adf7845048df/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/HRegion.java#L7882]).
>
> *Actual:* The returned result is unordered
> *Expected:* Similar to increment op, the returned result should be ordered.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
