When using the consumer group feature to subscribe to all partitions in a topic, I got some very strange behaviour where some runs of the code did consume messages and some runs were just stuck and did not get any messages back.

After a lot of debugging I found that k6 runs the init code (code outside of methods) multiple times: https://community.k6.io/t/why-init-called-4-times/973

This caused multiple consumer group instances to be created, which in turn meant that partitions were distributed between them. Thus, sometimes the consumer group created during the last init contained the partitions where my data was, and sometimes it did not contain the right partitions as they were assigned to consumers created in a previous run of init.

I solved this by using a UUID when naming my consumer group, thereby guaranteeing that my consumer group object was assigned to all partitions in a topic.

Although this is not per-se a bug in xk6-kafka, when I followed the documentation here there was no mention of this and the examples all used consumers directly in the init code.
I would suggest to either make this clear in the documentation of somehow alter the code to avoid this problem.