When the consumer enters the group and gets no assignment (for ex. there is not enough partitions in the topic),
librdkafka waits for the rebalancing sequence to be finished by calling assign with the empty list of partitions
(just as was passed by librdkafka to rebalance callback).
But cppkafka pass nullptr instead (which means unassign). And consumer stuck forever in that
state, not being able to pick the partition during the next rebalance (failover), because the previous rebalance sequence
was not finished.
When the consumer enters the group and gets no assignment (for ex. there is not enough partitions in the topic), librdkafka waits for the rebalancing sequence to be finished by calling assign with the empty list of partitions (just as was passed by librdkafka to rebalance callback).
But cppkafka pass nullptr instead (which means unassign). And consumer stuck forever in that state, not being able to pick the partition during the next rebalance (failover), because the previous rebalance sequence was not finished.
Fixes https://github.com/mfontanini/cppkafka/issues/273 , https://github.com/ClickHouse/ClickHouse/issues/21118 , etc.