Aerospike cluster behaviour during network partition


#1

I want to understand Aerospike behaviour during network partition. This is without strong consistency feature. Assume that there is a 4 node cluster (n1 to n4). Because of network issue, cluster gets divided into n1,n2 and n3,n4. Consider a partition whose master copy is there in n1 and replica is there in n4. I am looking for answers for the following questions -

  1. Would any one node from these 2 sub clusters become master for the given partition and remaining node as replica?
  2. What would be the view of client? (let’s assume that client is able to connect to all the 4 nodes). How would regular db operations of read/write take place?
  3. What would be the behaviour when network partition gets resolved? – I suppose answer to this question is data merging based on updation time?

#2

The architecture overview has a section on consistency that you should read. The community edition is in available mode.

Enterprise edition supports strong consistency. See Aerospike 4.0, Strong Consistency, and Jepsen.