[
https://issues.apache.org/jira/browse/CASSANDRA-5292?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13587294#comment-13587294
]
Jonathan Ellis commented on CASSANDRA-5292:
-------------------------------------------
IMO there are two principles we should be following here:
# schema options should be "just a Map." This is the most straightforward and
trying to be clever here will surprise people more than it will help. Nor do I
want to maintain "almost-Map" structures in the code.
# CQL should not "helpfully" lowercase Map keys.
My understanding is that we observe #2 elsewhere, but schema options are
not-quite-a-Map in this respect. In that case I think we should suck it up and
take the backwards incompatibility to fix this.
If we violate #2 in general, that is a bitter pill to swallow, but it's still a
bug that should be fixed.
> CQL3 shouldn't lowercase DC names for NTS
> -----------------------------------------
>
> Key: CASSANDRA-5292
> URL: https://issues.apache.org/jira/browse/CASSANDRA-5292
> Project: Cassandra
> Issue Type: Bug
> Affects Versions: 1.2.0
> Reporter: Sylvain Lebresne
> Assignee: Sylvain Lebresne
> Fix For: 1.2.3
>
> Attachments: 5292-option3.txt
>
>
> In CREATE and ALTER statements, when a property map is given (replication,
> compaction and compression options), CQL3 lowercase the map keys to provide
> case insensitivity. The goal is to allow things like:
> {noformat}
> replication = { 'Class' : 'SimpleStrategy', 'Replication_factor' : '1' }
> {noformat}
> However, this messes up with NTS, as it ends up lower-casing the datacenter
> names. As a consequence,
> {noformat}
> replication = { 'class' : 'NetworkTopologyStrategy', 'DC1' : '1' }
> {noformat}
> will currently create a DC that is really called 'dc1', which is a problem
> because DC names are interpreted case sensitively otherwise (at least in
> PropertyFileSnitch).
> That's the problem. Now I'm kind of hesitant on what is the right fix. I see
> the following possibilities:
> # Remove the CQL3 lower-casing completely. I'll admit that providing case
> insensitivy for property map keys may not have been such a good idea in the
> first place. After all, those map keys are string literal, which rather
> suggest case sensitivity. However, making that change would be a break
> strictly speaking.
> # Make DC name case insensitive. As much as I think DC names ought to be case
> insensitive, I'm not sure that's very doable in practice because that would
> imply storing DC names lower-cased internally, but DC names are exchanged
> over gossip and whatnot, so that would probably break all hell loose.
> # Keep CQL3 case insensitivity for property map keys in general but special
> case internally for NTS. The problems I see with that is that 1) this will be
> ugly and 2) if we special case too much, we might break potential custom
> strategies inspired by NTS. I also had the idea of changing strategy options
> internally from Map<String, String> to some custom object that would be
> essentially a case insensitive string map (general case), but that would also
> hold the original case of keys so NTS (and any other likely-minded strategy)
> can do its stuff. This happens to not be a small patch however (I'm attaching
> the patch for reference because I wrote it, but I'm seriously wondering if
> it's not too overkill).
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
