[
https://issues.apache.org/jira/browse/CASSANDRA-7787?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14102066#comment-14102066
]
Sergio Bossa commented on CASSANDRA-7787:
-----------------------------------------
[~slebresne], we don't actually *index* the row marker, we rather need it among
the indexed columns (hence indexes() needs to return true for us) in order to
detect pk-only rows.
> 2i index indexing the cql3 row marker throws NPE
> ------------------------------------------------
>
> Key: CASSANDRA-7787
> URL: https://issues.apache.org/jira/browse/CASSANDRA-7787
> Project: Cassandra
> Issue Type: Bug
> Components: Core
> Reporter: Berenguer Blasi
> Priority: Minor
> Fix For: 2.0.10
>
> Attachments: cassandra-2.0-7787.txt
>
>
> If you have a secondary index implementation that 'indexes()' the cql3 row
> marker you get a NPE in SecondaryIndexManager/deleteFromIndexes() as the call
> to getColumnDefinitionFromColumnName() returns null which is not checked for.
> This has been detected in the context of inserting PK only rows, where the
> row marker is expected to be present. When 'indexes()' returned 'false', the
> row would mistakenly get deleted as the row marker didn't go through.
> If 'indexes()' returns 'true' the row marker goes through but you get a NPE.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
