Existing architectures

Question 1

CouchBase

Answer 1

• Cluster Mgr on each node (No single master)
• Everything goes to cache first. One Disk persistence queue. Another Replication Queue
• Immediately consistent on all nodes (Replication queue)
• Async to disk queue
• Replicas co-exist on other nodes in cluster
• Joins on json documents supported

Question 2

HDFS

Answer 2

• Client + Name node (master) + Data nodes (slaves)
• No hot standby for NameNode, rather cold standby (Secondary name node exists and pulls data occasionally)
• Replication factor used to replicate both on same rack as well as another rack (to sustain rack failure). 3 min recommended.
• Each block of a file replicated individually
• Writes are serially replicated one block at a time, but success is only if all of replication succeeds
• First block on first rack, 2nd and 3rd on second rack