Re:RethinkDB 2.1 beta: automatic failover with Raft
by coffeemug on Thursday July 16, 2015 @03:56PM
(#9899530)
Great questions!
The simplest HA deployment needs three replicas on three nodes. If one of the nodes fails, one of the other two will automatically take over its responsibilities, either until the failed node comes back up, or the user reconfigures the table to not put data on the failed node anymore. So to sustain a one node failure, you need three nodes and three replicas.
To automatically sustain two node failures, you'd need five nodes.
In general, to sustain X failures without manual intervention, you'll need X*2+1 nodes (with the minimum of three nodes to get started).
The simplest HA deployment needs three replicas on three nodes. If one of the nodes fails, one of the other two will automatically take over its responsibilities, either until the failed node comes back up, or the user reconfigures the table to not put data on the failed node anymore. So to sustain a one node failure, you need three nodes and three replicas.
To automatically sustain two node failures, you'd need five nodes.
In general, to sustain X failures without manual intervention, you'll need X*2+1 nodes (with the minimum of three nodes to get started).