We are simulating 2 networks

• ccloud where Kafka resides
• onpremise where are the clients

onpremise can't reach ccloud, and thus needs a proxy with can connect to both.

The problem relies in the medata protocol, and thus a simple haproxy would not suffice. We need to proxy at Level7 to change the ip announced by Kafka by the ip of the proxy (kafka-proxy approach), or trick the metadata sent to go through the proxy (ha-proxy approach).

Enter https://github.com/grepplabs/kafka-proxy.

The main command is the following:

    command: server --bootstrap-server-mapping=broker-1:9092,0.0.0.0:9092,kafka-proxy:9092 --bootstrap-server-mapping=broker-2:9092,0.0.0.0:9093,kafka-proxy:9092 --bootstrap-server-mapping=broker-3:9092,0.0.0.0:9094,kafka-proxy:9092

Let's decompose the first argument

--bootstrap-server-mapping=broker-1:9092,0.0.0.0:9092,kafka-proxy:9092

1. We are bootstraping the broker-1:9092 kafka server
2. We are asking to proxy broker-1:9092 via 0.0.0.0:9092
3. We are announcing this proxy to clients as kafka-proxy:9092

To test, run the stack with docker-compose up -d then write something into a topic

docker-compose exec kafka-client kafka-console-producer --broker-list kafka-proxy:9092 --topic example`

type a lovely hello world and a firm [Enter].

And then consume it back via

docker-compose exec kafka-client kafka-console-producer --bootstrap-server kafka-proxy:9092 --topic example --from-beginning

To test, run the stack with docker-compose up -d

in kafka-client enter docker

docker-compose exec kafka-client bash
echo `host haproxy-1 | awk '{print $4}'` broker-1 >> /etc/hosts
echo `host haproxy-2 | awk '{print $4}'` broker-2 >> /etc/hosts
echo `host haproxy-3 | awk '{print $4}'` broker-3 >> /etc/hosts
echo example | kafka-client kafka-console-producer --broker-list haproxy-1:9092 --topic example
kafka-client kafka-console-producer --bootstrap-server haproxy-1:9092 --topic example --from-beginning