Hi Everyone, We're trying the deploy the Kafka behind the network balancer and we have created the port map for each Kafka brokers under that network balancer--we only have one public IP and the Kafka clients are in other system and thus cannot access the brokers via internal IP directly.
So for example, we have the public IP 1.2.3.4, we map 1.2.3.4:9092 for broker1 and 1.2.3.4 : 9093 for broker2, etc. And in the server.properties, the advertised host and port will be 1.2.3.4:9092 and 1.2.3.4:9093 for broker 1 and broker 2 respectively. It works well at beginning. But then after several days with load, the replication between brokers fails due to connection timeout---it happens intermediately. But the outside connection to these Brokers are still working fine consistently. By looking at the tcpdump at the destination server, we find out that when timeout happens when the broker does not send TCP ACK back to the public IP 1.2.3.4. As you can see the Source IP here is the public IP, and the destination is the broker. The broker server does not send TCP ACK to the source. [image: Inline image 1] Did you see the similar problem before? Very appreciate for any kind of information. Thanks, Tony
