[jira] Commented: (CASSANDRA-874) System tests against multiple nodes

["Jeremy Hanna (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/CASSANDRA-874?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12900963#action_12900963 ] 

Jeremy Hanna commented on CASSANDRA-874:
----------------------------------------

As another data point, Voldemort uses EC2 to perform multi-node system tests.  The code they use along with documentation is found here http://wiki.github.com/voldemort/voldemort/ec2-testing-infrastructure  Perhaps there are some general ideas there for what and how to test at that level.

> System tests against multiple nodes
> -----------------------------------
>
>                 Key: CASSANDRA-874
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-874
>             Project: Cassandra
>          Issue Type: Test
>          Components: Tools
>            Reporter: Jonathan Ellis
>
> We have a system test suite (see http://wiki.apache.org/cassandra/HowToContribute ) but it only runs against the local machine, which is suboptimal even if we didn't have a bunch of optimizations to make writes to a local node skip most of the messaging layer.
> We'd like to automate testing against a whole cluster.  Paul has some code using libcloud to do stress tests against an ephemeral rackspace cloud servers cluster at http://github.com/pquerna/ccb, which is probably useful as a starting point, but rather than running stress.py you'd want to put together a more comprehensive suite, based on the existing system tests but also explicitly covering things like range queries that span multiple nodes and failure conditions like Hinted Handoff (see http://wiki.apache.org/cassandra/HintedHandoff ) and repair (http://wiki.apache.org/cassandra/ArchitectureAntiEntropy ) resulting from deliberately disabling a node temporarily.
> Adding tests for ops tasks like http://wiki.apache.org/cassandra/Operations would be awesome too.
> Finally, CASSANDRA-561 might make something like this possible using a simulator instead of hardware-level virtualization, which would be both more convenient and less expensive.  But someone just dumped a huge zip in our laps w/o really trying to collaborate with us there, so if you want to try to resurrect that code or otherwise go that route, be more open about what you're doing. :)

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.