The nodes don't need to be up 100% of the time. Thanks to replication slots, the WAL on the other nodes will be kept until the node connects again and catches up. So it really depends on how much disk space you have on the other nodes.
Regarding the PK conflicts - I'm not sure I understand the question, but it's possible to use global sequences (which is one of the parts that did not make it into core yet). Otherwise it'll generate conflicts, and you'll have to resolve them somehow (e.g. it's possible to implement a custom conflict handler).
Regarding the PK conflicts - I'm not sure I understand the question, but it's possible to use global sequences (which is one of the parts that did not make it into core yet). Otherwise it'll generate conflicts, and you'll have to resolve them somehow (e.g. it's possible to implement a custom conflict handler).
See http://bdr-project.org/docs/stable/conflicts.html