No Majority Node Partition

Suppose the shard is divided into multiple partitions, and no partition contains a majority of the Replication Nodes in the shard. In this case, the shard's partitions can service read requests, so long as the consistency policy in use for the read supports it. If the read requires tight consistency with the master, and the master is not available to ensure the consistency can be met, then the read will fail.

The partition containing the master can service write requests only if you are using the weakest available durability policy, in which no acknowledgements from replicas are required. If acknowledgements are required, then there will not be enough replicas to satisfy the durability policy and no write operations can occur.

Once the network partition is resolved, the shard will elect a new master, synchronize all replicas with it, and continue operations as normal.