Suppose there are five nodes:
When A is down,we failover it.
The question is:
After failover all requests that were supposed to go to A now go to the remaining nodes or just go to one node?
Replicas are (nearly) evenly split among remaining nodes. So load will be evenly distributed among them.
There's limit of number of 'slave' nodes per node. Currently 10. So in larger clusters after failing over node load will be distributed slightly unevenly. But in larger clusters (say 50 nodes) loss of single node is small extra load.
© 2013 COUCHBASE All rights reserved.