On Tue, Apr 5, 2011 at 9:50 AM, Jonathan Colby <jonathan.co...@gmail.com> wrote: > Hi experts, > > We have a cluster of 4 nodes across 2 data centers and we are using > OldNetworkTopology to balance the replicas (RF=3). > > Question: Can we afford to lose 1 node in a data center and still have > Quorum work?
Insomuch as it works already, yes. (See below.) > My understanding is that OldNetworkTopology writes 2 to the "local" data > center and "1" to the remote data center. Sort of. It writes two to the DC where the key consistently hashes to, and one to the other. NOT "local" as the client sees it. So using quorum + ONTS you have no control over when you have to wait for cross-DC latency. This is why we introduced NTS and LOCAL_QUORUM CL. Eric wrote a good explanation here: http://www.datastax.com/dev/blog/deploying-cassandra-across-multiple-data-centers -- Jonathan Ellis Project Chair, Apache Cassandra co-founder of DataStax, the source for professional Cassandra support http://www.datastax.com
