OK. Given the correction of my unfortunate partitioner error, you can, and probably should, upgrade in place to 1.2, but with num_tokens=1 so it will initially behave like 1.1 non vnodes would. Then you can do a rolling conversion to more than one vnode per node, and once complete, shuffle your vnodes.
http://www.datastax.com/dev/blog/upgrading-an-existing-cluster-to-vnodes-2 There should be no time where your cluster is unbalanced. -Tupshin On Dec 30, 2013 9:46 AM, "Katriel Traum" <katr...@google.com> wrote: > Hello list, > > I have a 2 DC set up with DC1:3, DC2:3 replication factor. DC1 has 6 > nodes, DC2 has 3. This whole setup runs on AWS, running cassandra 1.1. > Here's my nodetool ring: > 1.1.1.1 eu-west 1a Up Normal 55.07 GB 50.00% > 0 > 2.2.2.1 us-east 1b Up Normal 107.82 GB 100.00% > 1 > 1.1.1.2 eu-west 1b Up Normal 53.98 GB 50.00% > 28356863910078205288614550619314017622 > 1.1.1.3 eu-west 1c Up Normal 54.85 GB 50.00% > 56713727820156410577229101238628035242 > 2.2.2.2 us-east 1d Up Normal 107.25 GB 100.00% > 56713727820156410577229101238628035243 > 1.1.1.4 eu-west 1a Up Normal 54.99 GB 50.00% > 85070591730234615865843651857942052863 > 1.1.1.5 eu-west 1b Up Normal 55.1 GB 50.00% > 113427455640312821154458202477256070484 > 2.2.2.3 us-east 1e Up Normal 106.78 GB 100.00% > 113427455640312821154458202477256070485 > 1.1.1.6 eu-west 1c Up Normal 55.01 GB 50.00% > 141784319550391026443072753096570088105 > > > I am going to upgrade my machine type, upgrade to 1.2 and change the > 6-node to 3 nodes. I will have to do it on the live system. > I'd appreciate any comments about my plan. > 1. Decommission a 1.1 node. > 2. Bootstrap a new one in-place, cassandra 1.2, vnodes enabled (I am > trying to avoid a re-balance later on). > 3. When done, decommission nodes 4-6 at DC1 > > Issues i've spotted: > 1. I'm guessing I will have an unbalanced cluster for the time period > where I have 1.2+vnodes and 1.1 mixed. > 2. Rollback is cumbersome, snapshots won't help here. > > Any feedback appreciated > > Katriel > >