a slave should be able to replicate from a repl pair
multiple servers per shard (WAS: use replica pair as shard)
replciation should be rack and data center aware
replica sets will address many (most) limitations of replica pairs.
Would it be possible to handle some sort of geographical location of the mongodb nodes within the replica set? This would allow us to, for example, have 2 mongodb nodes per datacenter, and when one node of the "main" datacenter fails, the other node within the same datacenter will be the preferred failover node.
Yes - we're thinking of different ways to handle that - but its definitely something we'll be adding support for.
Another question: will it be possible to include geographical topology of nodes with the replication acknowledgment via getlasterror? For example, being able to block until the data has been replicated to at least 1 node in a different datacenter.
yes - although ther'es a chance that may come in a point release shortly after first version
Is there an ETA for a production release of Sharding (other than just release 1.6)?
ETA for 1.6 is end of july
Is datacenter awareness actually a supported feature in 1.6.1 now? If so, how would one configure MongoDB to only sync data once from the main DC to the 2nd one and distribute the changes to the other RS members in the 2nd DC only locally?
@frank you cannot do that yet. Although you could have local replication to a second server in the primary data center, and one remote server elsewhere today.