The cause: 

This is the reason for heartbeat failure for group because it’s rebalancing. The fundamental way to determine whether all consumers are still functional is to listen for heartbeats. If you see a heartbeat failure when the group rebalances, it means that your consumer instance delayed too long to submit the next heartbeat, at which point the group deemed it dead and initiated a rebalance.

Solution:

To avoid this, you may either lengthen the timeout (session.timeout.ms) or make sure your customer transmits heartbeats more frequently (heartbeat.interval.ms). Because heartbeats are essentially built into poll(), you must call poll often enough. This is often accomplished by limiting the number of records that a single poll returns using the max.poll.records setting (to reduce the time needed to process all of the fetched data).