[
https://issues.apache.org/jira/browse/CASSANDRA-7791?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14102408#comment-14102408
]
Sylvain Lebresne commented on CASSANDRA-7791:
---------------------------------------------
bq. It might be possible that a UDT from one keyspace is not available for a
table in another keyspace (different DC replication settings for different
keyspaces)
Schema tables are not subject to replication settings, there are replicate
everywhere no matter what.
> Consider re-allowing to refer to UDT outside of their keyspace
> --------------------------------------------------------------
>
> Key: CASSANDRA-7791
> URL: https://issues.apache.org/jira/browse/CASSANDRA-7791
> Project: Cassandra
> Issue Type: Improvement
> Reporter: Sylvain Lebresne
> Labels: cql
> Fix For: 3.0
>
>
> In CASSANDRA-6643 we decided to make UDT inaccessible outside of their
> keyspace of definition. Doing so mainly has the advantage that when we drop a
> keyspace, we don't have to worry its UDT being used in another keyspace.
> However, this directly conflict with functions (UDF) being global: we can't
> have functions working on UDT if functions are global and we don't allow UDT
> access outside their keyspace. Which, I believe, leave us with the following
> possible options:
> # we make UDT accessible anywhere (though their fully qualified name).
> # we don't support functions on UDT at all.
> # we make functions keyspace-scoped, either always, or only if they apply to
> UDT.
> # we revert CASSANDRA-6438 and make UDT global.
> In a perfect world I would lean towards 4: the arguments to make UDT
> keyspace-scoped where not wrong per-se but weak in hindsight given the other
> options here. It is however basically too late: changing it would be a
> breaking change so we can't reasonably change this post-2.1.0, and while it's
> not released yet, it's not a change we can make without substantially
> delaying the final.
> Option 2 feels rather lame in my book.
> Option 3 feels pretty messy. Having 2 types of UDF, some keyspace-scoped and
> some that are not would be super confusing. Saying that ll UDF are
> keyspace-scoped feels limiting, and we would still be somewhat inconsistent
> with our existing hard-coded functions that are global.
> Which leaves option 1 which might be the most pragmatic. Having to check that
> UDTs are not used before allowing keyspace drop don't sound like a huge deal
> to me.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
