Kafka server configuration - listeners vs. advertised.listeners

From this link: https://cwiki.apache.org/confluence/display/KAFKA/KIP-103%3A+Separation+of+Internal+and+External+traffic

During the 0.9.0.0 release cycle, support for multiple listeners per broker was introduced. Each listener is associated with a security protocol, ip/host and port. When combined with the advertised listeners mechanism, there is a fair amount of flexibility with one limitation: at most one listener per security protocol in each of the two configs (listeners and advertised.listeners).

In some environments, one may want to differentiate between external clients, internal clients and replication traffic independently of the security protocol for cost, performance and security reasons. A few examples that illustrate this:

• Replication traffic is assigned to a separate network interface so that it does not interfere with client traffic.
• External traffic goes through a proxy/load-balancer (security, flexibility) while internal traffic hits the brokers directly (performance, cost).
• Different security settings for external versus internal traffic even though the security protocol is the same (e.g. different set of enabled SASL mechanisms, authentication servers, different keystores, etc.)

As such, we propose that Kafka brokers should be able to define multiple listeners for the same security protocol for binding (i.e. listeners) and sharing (i.e. advertised.listeners) so that internal, external and replication traffic can be separated if required.

So,

listeners - Comma-separated list of URIs we will listen on and their protocols. Specify hostname as 0.0.0.0 to bind to all interfaces. Leave hostname empty to bind to default interface. Examples of legal listener lists:

• PLAINTEXT://myhost:9092,TRACE://:9091
• PLAINTEXT://0.0.0.0:9092, TRACE://localhost:9093

advertised.listeners - Listeners to publish to ZooKeeper for clients to use, if different than the listeners above. In IaaS environments, this may need to be different from the interface to which the broker binds. If this is not set, the value for listeners will be used.

Since I cannot comment yet I will post this as an "answer", adding on to M.Situations answer.

Within the same document he links there is this blurb about which listener is used by a KAFKA client (https://cwiki.apache.org/confluence/display/KAFKA/KIP-103%3A+Separation+of+Internal+and+External+traffic):

As stated previously, clients never see listener names and will make metadata requests exactly as before. The difference is that the list of endpoints they get back is restricted to the listener name of the endpoint where they made the request.

This is important as depending on what URL you use in your bootstrap.servers config that will be the URL* that the client will get back if it is mapped in advertised.listeners (do not know what the behavior is if the listener does not exist).

Also note this:

The exception is ZooKeeper-based consumers. These consumers retrieve the broker registration information directly from ZooKeeper and will choose the first listener with PLAINTEXT as the security protocol (the only security protocol they support).

As an example broker config (for all brokers in cluster):

advertised.listeners=EXTERNAL://XXXXX.compute-1.amazonaws.com:9990,INTERNAL://ip-XXXXX.ec2.internal:9993

inter.broker.listener.name=INTERNAL

listener.security.protocol.map=EXTERNAL:SSL,INTERNAL:PLAINTEXT

If the client uses XXXXX.compute-1.amazonaws.com:9990 to connect, the metadata fetch will go to that broker. However, the returning URL to use with the Group Coordinator or Leader could be 123.compute-1.amazonaws.com:9990* (a different machine!). This means that the match is done on the listener name as advertised by KIP-103 irrespective of the actual URL (node).

Since the protocol map for EXTERNAL is SSL this would force you to use an SSL keystore to connect.

If on the other hand you are within AWS lets say, you can then issue ip-XXXXX.ec2.internal:9993 and the corresponding connection would be plaintext as per the protocol map.

This is especially needed in IaaS where in my case brokers and consumers live on AWS, whereas my producer lives on a client site, thus needing different security protocols and listeners.

EDIT: Also adding Inbound Rules is much easier now that you have different ports for different clients (brokers, producers, consumers).

EDIT2: This article is a great in depth guide if the above is still not clear: https://rmoff.net/2018/08/02/kafka-listeners-explained/

listeners is what the broker will use to create server sockets.

advertised.listeners is what clients will use to connect to the brokers.

The two settings can be different if you have a "complex" network setup (with things like public and private subnets and routing in between).

There's so much confusion or little information in answers provided here for the question. So posting my elaborate answer for clarity.

1. listeners - Used by the embedded jetty web server in kafka to bind to. This jetty web server is used to provide REST API that provides the control plane for Kafka Connect workers. The hostname in this setting can be left empty if you want kafka to bind to localhost (it does by calling InetAddress.getLocalHost().getCanonicalHostName() java api)
2. advertised.listeners: This address is published to zookeeper by every kafka broker. If this setting is not set, then value of listeners will be used here and published to zookeeper. That's the only purpose of this setting for notifying others. Kafka Clients use the 'advertised.listeners' setting published to zookeeper (as /brokers/ids/<id>/ # endpoints) to talk to Kafka broker.

Now the question is why to have two setting? Why not a single setting? Let's say your kafka broker is sitting behind a proxy. And all the kafka clients have to talk to the proxy to reach the broker. In this case, we want kafka's embedded jetty server to bind to localhost and local port, but we can't publish this to zookeeper as clients can't use it. So kafka admin can set the setting advertised.listeners to the proxy host and port.

Also, in some of our production hosts, InetAddress.getLocalHost().getCanonicalHostName() returns empty and so listeners setting's hostname was empty which was fine for jetty to bind. But advertised.listeners was published to zookeeper as NULL:9092 since it took the same value as listeners by default. Now all the brokers tried to publish in this way to zookeeper and so brokers got the error java.lang.IllegalArgumentException: requirement failed: Configured end points null:14092 as advertised.listeners as NULL:9092 is already registered by broker 101. The fix was to change the advertised.listeners setting to have hostname in it.