Hello cassandra-dev, This is an update on the previous discussion on virtual nodes - for anyone who missed that, an archive is here: http://www.mail-archive.com/dev@cassandra.apache.org/msg03837.html
Work on our proposed design is being tracked on JIRA: https://issues.apache.org/jira/browse/CASSANDRA-4119 We feel that our current implementation has sufficient functionality to be useful in its own right, but is still a fairly self-contained set of changes that can be incrementally improved - something which was expressed as desirable in the last discussion. This is a good point for a first round of review and to reach out to anyone interested in testing or contributing. == Obtaining the source == The easiest way to test these changes is to clone our github repo and switch to the topic branch representing the top patch: > git clone git://github.com/acunu/cassandra.git > cd cassandra > git checkout --track remotes/origin/p/4127/01_migration_path Now just build with ant. To create a cluster with virtual-nodes support just un-comment/edit the following parameter in cassandra.yaml: > num_tokens: 256 Make sure you have left initial_token blank. The node should automatically assign itself this many tokens. You can view the token assignment as usual with "nodetool ring" but this becomes fairly useless with any large number of hosts/tokens, which is why we have added "nodetool clusterinfo" which shows ring ownership without cluttering the output with the token assignment. If you want to test with a specific token assignment, initial_token now supports a comma-separated list of tokens which will override the num_tokens setting. This is just for convenience of testing - users should no longer need to manually specify tokens as balancing is automatic. == Patches == The tickets which are in scope for this round of review are: https://issues.apache.org/jira/browse/CASSANDRA-4121 https://issues.apache.org/jira/browse/CASSANDRA-4122 https://issues.apache.org/jira/browse/CASSANDRA-4125 https://issues.apache.org/jira/browse/CASSANDRA-4127 Each of these tickets has links to the corresponding patches. The order they should be applied is just increasing numerical order of ticket number. These patches are based off another patch (currently pending review) for CASSANDRA-3881, which was an existing issue blocking virtual nodes. https://issues.apache.org/jira/browse/CASSANDRA-3881 Links to the individual patches can also be found all in one place on the github wiki: https://github.com/acunu/cassandra/wiki/CASSANDRA-4119 The current patches in the series are as follows: * p/4121/01_support_multiple_tokens_per_host support multiple tokens per host in TokenMetadata (CASSANDRA-4121) * p/4122/01_bootstrap_decommission support bootstrapping a node into the ring with multiple tokens and minor changes for decommission (CASSANDRA-4122) * p/4122/02_remove_tokens minor changes to support remove-token with multiple tokens (CASSANDRA-4122) * p/4125/01_admin_tools nodetool support (addition of nodetool clusterinfo and changes to nodetool ring) (CASSANDRA-4125) * p/4127/01_migration_path support for migration from 1-token to multi-token per node (CASSANDRA-4127) If you wish to contribute or work with a clone from github then it would be advisable to familiarise yourself with TopGit, the tool we have been using for branch-based patch queue management. We've written up a tutorial here: https://github.com/acunu/cassandra/wiki/TopGit-Tutorial == What's left? == We haven't included any patches for tickets CASSANDRA-4123 and CASSANDRA-4124 which relate to the replication strategy and repair. Currently replication and repair "just work" with the current patches without any additional changes required. CASSANDRA-4126 relates to testing. We're running virtual nodes builds through our own test suites but we will also be writing new tests in addition. I look forward to your questions and comments! -- Sam Overton Acunu | http://www.acunu.com | @acunu
