We want to have 1 database on 3 servers.
Each one should have 1by1 data.
So when we write value to server1 - we want to read this value from server2 and from server3.
And if after this - server1 will be shut down - we want to have working server2 and server3. And want to write and read data from server 2 and 3.
Well, you would use replication factor 3 which will provide 1 copy on each of the 3 nodes in your cluster. By default, clients will only read from the node owning the master copy of the partition the record belongs to (but this can be changed in the policy configured on the client library side).