[jira] [Comment Edited] (CASSANDRA-12605) Timestamp-order searching of sstables does not handle non-frozen UDTs, frozen collections correctly

["Tyler Hobbs (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/CASSANDRA-12605?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15510844#comment-15510844 ] 

Tyler Hobbs edited comment on CASSANDRA-12605 at 9/21/16 7:02 PM:
------------------------------------------------------------------

Thank you.  The new runs look better.  The failures in the trunk dtest appear to be from a new dtest that was just added to test a change in Cassandra, so it should be fine to ignore that.


was (Author: thobbs):
Thank you.  The new runs look better.

> Timestamp-order searching of sstables does not handle non-frozen UDTs, frozen collections correctly
> ---------------------------------------------------------------------------------------------------
>
>                 Key: CASSANDRA-12605
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-12605
>             Project: Cassandra
>          Issue Type: Bug
>            Reporter: Tyler Hobbs
>            Assignee: Tyler Hobbs
>
> {{SinglePartitionReadCommand.queryNeitherCountersNorCollections()}} is used to determine whether we can search sstables in timestamp order.  We cannot use this optimization when there are multicell values (such as unfrozen collections or UDTs).  However, this method only checks {{column.type.isCollection() || column.type.isCounter()}}.  Instead, it should check {{column.type.isMulticell() || column.type.isCounter()}}.
> This has two implications:
> * We are using timestamp-order searching when querying non-frozen UDTs, which can lead to incorrect/stale results being returned.
> * We are not taking advantage of this optimization when querying frozen collections.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)