With 9 partitions spread across 3 brokers, each broker typically hosts 3 leaders (assuming even distribution). When Broker 0 fails, the partitions for which it was leader will elect new leaders on brokers 1 or 2 if enough in-sync replicas (ISRs) remain. But since min.insync.replicas=3 and only 2 brokers are up, no partition can meet the minimum in-sync replica requirement, so producers with acks=all will fail to write. However, for partitions where Broker 0 is not the leader, consumers can still read committed messages. Given that only 3 partitions likely had Broker 0 as leader, six partitions remain accessible for reads, but not writes.