On Wed, Aug 13, 2014 at 4:35 AM, DuyHai Doan <doanduy...@gmail.com> wrote:
> "add an additional integer column to the partition key (making it a > composite partition key if it isn't already). When inserting, randomly > pick a value between, say, 0 and 10 to use for this column" --> Due to the > low cardinality of bucket (only 10), there is no guarantee that the > partitions would be distributed evenly. But it's better than nothing. > It's important to think about it probablistically, i.e. "what is the probability that all ten partitions belong to the same node?" If you have a ten node cluster (assume RF=1 for simplicity), there's a 1/10^9 (one in a billion) chance that a single node is the owner for all partitions. So it's quite a bit better than nothing. If you want to improve your odds, bump the number up. But, keep in mind that it's a balance, because reads become more expensive. > > "Alternatively, instead of using a random number, you could hash the > other key components and use the lowest bits for the value. This has the > advantage of being deterministic" --> Does it work with VNodes, where > tokens are split in 256 ranges and shuffled in all nodes ? > Yes, it works perfectly fine with vnodes. -- Tyler Hobbs DataStax <http://datastax.com/>
