On 12.05.2010, at 22:02, Randall Leeds wrote: > No. Nothing is totally guaranteed. Replication ensures that live nodes will > eventually contain all updates. If you write a document and the node that > receives it goes down, a subsequent view request will hit another replica > which may not yet have the update. > > This is true in any distributed database that does not commit changes to > multiple replicas. > > I've considered adding something like dynamo/riak/cassandra along the lines > of R and W quorum values for requests but right now a write will return as > soon as the first couch responds and lounge leaves it up to couchdb > replication to ensure the other nodes holding that shard eventually receive > that update.
I got it, thanks for clarification. This is a CAP, and it's possible to reach just 2 of them. Would be nice to be able to specify R and W just like in Riak and make a decision on application level which part of CAP is necessary.
