cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Paulo Motta (JIRA)" <>
Subject [jira] [Commented] (CASSANDRA-9474) DC/Rack property changed on live system
Date Tue, 17 Nov 2015 12:55:11 GMT


Paulo Motta commented on CASSANDRA-9474:

[~molsson] Thanks for spotting this. It seems the flag was removed during the [up-merge|]
of CASSANDRA-10242 to 2.2, since there was also a minor change in the implementation, where
the check was moved from SystemKeyspace to CassandraDaemon (which is a more adequate place
for this check anyway, so you should also place the dc check there). 

I believe it was an oversight, so could you please add {{cassandra.ignore_rack}} flag back
(in addition to the new {{cassandra.ignore_dc}} flag)? since it's already documented on 2.1
NEWS.txt, we should still keep both flags. Also could you mention these flags in the error
messages? So if somebody Knows What Is Doing©, then it will find out about the flags.

> DC/Rack property changed on live system
> ---------------------------------------
>                 Key: CASSANDRA-9474
>                 URL:
>             Project: Cassandra
>          Issue Type: Improvement
>         Environment: Cassandra 2.1.5
>            Reporter: Marcus Olsson
>            Assignee: Marcus Olsson
>             Fix For: 2.1.x
>         Attachments: cassandra-2.1-9474.patch, cassandra-2.1-dc_rack_healthcheck.patch
> When using GossipingPropertyFileSnitch it is possible to change the data center and rack
of a live node by changing the file. Should this really be possible?
In the documentation at
it's stated that you should ??Choose the name carefully; renaming a data center is not possible??,
but with this functionality it doesn't seem impossible(maybe a bit hard with changing replication
> This functionality was introduced by CASSANDRA-5897 so I'm guessing there is some use
case for this?
> Personally I would want the DC/rack settings to be as restricted as the cluster name,
otherwise if a node could just join another data center without removing it's local information
couldn't it mess up the token ranges? And suddenly the old data center/rack would loose 1
replica of all the data that the node contains.

This message was sent by Atlassian JIRA

View raw message