A Kafka broker receives messages from producers and stores them on a disc with a unique offset.
A Kafka broker enables consumers to retrieve messages by topic, partition, and offset. 

If a Kafka broker goes offline for an extended time, it will likely return online but be far behind in logs.
Therefore, it will need to catch up on logs from the remaining brokers before being reintroduced into the In-Sync Replica (ISR).
During normal operation, replicas communicate with one another.
However, if one broker is significantly behind, we adjust settings to enable Kafka to spend more resources on keeping replicas up to date.
Because all partition replicas on the broker become unavailable during a broker outage, the availability of the affected partitions is determined by the existence and status of their other replicas.
If a partition has no additional replicas, it becomes inaccessible.
The settings on which they depend are:

• Min.isync.replicas
• Default.replication.factor
• Unclean.leader.election.enable
  

When the broker goes down, however, Kafka does not create a new replica.
You will have to wait for the broker to become available, but if you need a replica, you can use the reassignment tool kafka-reassign-partitions.sh to move partitions to available brokers.
In addition, when you restart the broker, it will attempt to reconnect.
Once completed, whether it remains a follower or becomes the leader is determined by whether it is the appropriate replica.

Reference: https://jaceklaskowski.gitbooks.io/apache-kafka/content/kafka-brokers.html