When the key is null the default partitioner will be used. This means that, as you noted, the message will be sent to one of the available partitions at random. A round-robin algorithm will be used in order to balance the messages among the partitions.

After Kafka 2.4, the round-robin algorithm in the default partitioner is sticky - this means it will fill a batch of messages for a single partition before going onto the next one.

Of course, you can specify a valid partition when producing the message and it will be respected.

Ordering will not differ - messages will get appended to the log in the same order by their arrival time regardless if they have a key or not.