cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sylvain Lebresne (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-10758) Dropping an index does not invalidate prepared statements anymore
Date Fri, 27 Nov 2015 13:21:11 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-10758?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15029862#comment-15029862
] 

Sylvain Lebresne commented on CASSANDRA-10758:
----------------------------------------------

bq. Had you got something more in mind?

Not particularly. My sentence should really be read "I'm tired of looking at fixing stuff
that are not directly related to the issue" and I hadn't really looked at how to fix it at
all tbh. But the fix does is trivial so good and +1 on the patch, modulo clean CI (which doesn't
seem to have been started yet).

> Dropping an index does not invalidate prepared statements anymore
> -----------------------------------------------------------------
>
>                 Key: CASSANDRA-10758
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-10758
>             Project: Cassandra
>          Issue Type: Bug
>          Components: CQL
>            Reporter: Sylvain Lebresne
>            Assignee: Sam Tunnicliffe
>              Labels: secondary_index
>             Fix For: 3.0.1, 3.1
>
>
> When we drop an index, we should invalidate ({{SELECT}}) statements on the base table
as those may involve the index (and, for instance, validating an index exists is done a preparation
time in {{SELECT}}, so a {{SELECT}} does potentially become invalid if an index is dropped).
As far as I can tell, this is properly done in pre-3.0, because {{CFMetaData.apply}} will
return {{true}} which will trigger the statements invalidations. And that is because, before
3.0, index definitions are part of the {{ColumnDefinition}} of the base table and thus dropping
an index is (rightfullly in a way) considered by {{CFMetaData}} as a change to the table columns.
In 3.0 however, as indexes has been moved out of {{ColumnDefinition}}, {{CFMetaData.apply}}
returns {{false}} in that case and statements are not invalidated.
> I'll note that the changes to {{CQLTester}} done in CASSANDRA-10631 actually expose this
problem (but indirectly so we should create a purpose built tests for this).



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

Mime
View raw message