Clustering in MongoDB - MongoDB High Availability - page 32

Secondary nodes will download and apply primary's oplog , that is, secondaries that
have the same dataset. As already mentioned, each replica set has one and only one
primary node that accepts all read/write operations. If the primary node is unavail-
able, another secondary node will be the primary node that uses an election process.
A secondary node will be elected to a primary node if there is a clear majority vote
for a node.
Tip
By default, clients read from the primary node; however, clients can specify a read
preference to send read operations to secondaries at http://docs.mongodb.org/manual/core/
replication-introduction/ .
To update the nodes' availability, each node sends a ping or heartbeat packets to all other
nodes. The following diagram illustrates this process:
To facilitate the election process, you can add an extra mongod node as an arbiter
node. Arbiter nodes don't have any dataset. Arbiter nodes exist to vote in the elec-
tion process. They come in handy when you have an even number of nodes, so these
nodes can help you to choose the correct secondary node for the primary node.
Note
What is an arbiter?
Arbiter is a mongod node that doesn't have any dataset and can't become a primary
node. Arbiter nodes exist to vote in the election process of a primary node. They
don't need dedicated hardware.