[ 
https://issues.apache.org/jira/browse/CASSANDRA-6922?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13946466#comment-13946466
 ] 

Sylvain Lebresne commented on CASSANDRA-6922:
---------------------------------------------

What's the expected benefits? The cons are pretty simple to see: there is 
almost no way we will be able to support live upgrade between partitioner (an 
not only will that be a non-live upgrade, but it will be a very costly one), 
and even if we were, provided you were using BOP/OPP for a good reason, there 
just isn't any simple upgrade path.

I'm less sure about the benefits. Surely BOP/OPP hasn't require too much 
maintenance by themselves. I get that assuming they doesn't exist might make 
some tickets like CASSANDRA-6696 easier, but it doesn't sound like a huge deal, 
is it?

Note that I'm fine with (and would even favor some of):
# rewriting unit tests to use M3P since that's our default.
# adding some warning in the log when you use BOP/OPP clearly stating that you 
better know what you're doing if you use it.
# use the simplest possible solution that work when confronted with code paths 
that needs special casing for BOP/OPP (like CASSANDRA-6696), and even why not 
say that some specific features are not available with those partitioners.

But a priori, removing their support completely feels really really hard on 
those that uses it. If the claim is that "nobody uses it anyway", then I'd like 
to have at some some form of data that suggests that it's true (for instance 
sending an email on the mailing list to gauge if it's really true).

> Investigate if we can drop ByteOrderedPartitioner and 
> OrderPreservingPartitioner in 3.0
> ---------------------------------------------------------------------------------------
>
>                 Key: CASSANDRA-6922
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-6922
>             Project: Cassandra
>          Issue Type: Improvement
>            Reporter: Marcus Eriksson
>             Fix For: 3.0
>
>
> We would need to add deprecation warnings in 2.1, rewrite a lot of unit 
> tests, and perhaps provide tools/guidelines to migrate an existing data set 
> to Murmur3Partitioner.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Reply via email to