The config servers have a crucial role over sharding since they hold the mapping information of all moving pieces of a shard cluster.
For several use cases the current behavior of config servers can be quite limiting due to 2 aspects:
- Strick number of members
- can impact multiple datacenter deployments
- specially on slow networks
- Failover of members
- needs to have all 3 always online otherwise the cluster may encounter problems booting up mongos's
Due to historic and mathematical reasons the config servers have the current hard number of 3 members.
Setting them as replicas with enforced Write Concerns for balancer locking and migration operations we can achieve a more flexible shard cluster.
Also the current hard setting creates some confusion on clients and generates a bit of unrest while configuring a sharding environment.