[ 
https://issues.apache.org/jira/browse/KUDU-3723?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18057703#comment-18057703
 ] 

ASF subversion and git services commented on KUDU-3723:
-------------------------------------------------------

Commit cad295f98a9eb2b390eb096d34e445fb27b8cca2 in kudu's branch 
refs/heads/master from Alexey Serbin
[ https://gitbox.apache.org/repos/asf?p=kudu.git;h=cad295f98 ]

KUDU-3723 fix interpretation of new range partition specs

In the context of KUDU-2671, a new notation for specifying hash schema
of a range partition has been added.  With the introduction of
range-specific hash schemas, the client side is free to send in
information on the newly added range partition as if it had a custom
hash schema, even if the new range's hash schema is the same as the
table-wide one.  However, the system catalog should have normalized
the provided information, and stored the information on ranges with
the table-wide and custom hash schemas differently.  The normalization
part was missing, and that was the reason for the reported issue.

This changelist addresses the problem and adds a new test scenario that
triggers an assertion without the fix.

Change-Id: Icceb138a919cd7afb572c6dd74695a3fcaaac99e
Reviewed-on: http://gerrit.cloudera.org:8080/23778
Tested-by: Alexey Serbin <[email protected]>
Reviewed-by: Alexey Serbin <[email protected]>


> Dropping and adding back same range partition right away isn't working for 
> certain table schemas
> ------------------------------------------------------------------------------------------------
>
>                 Key: KUDU-3723
>                 URL: https://issues.apache.org/jira/browse/KUDU-3723
>             Project: Kudu
>          Issue Type: Bug
>          Components: tserver
>    Affects Versions: 1.17.0, 1.18.0, 1.18.1
>            Reporter: Alexey Serbin
>            Assignee: Alexey Serbin
>            Priority: Major
>
> For a Kudu table partitioned by both range and hash, the following sequence 
> leads to the table being stuck in a limbo state where it cannot be opened or 
> operated by any other means, effectively rendering it inaccessible:
> * adding a new range with custom hash partition scheme
> * immediately drop the newly added range
> * immediately try adding the same range again
> The reproduction scenario is available at 
> http://gerrit.cloudera.org:8080/23778  It's 100% repeatable/reproducible.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to