Status

Current state: Under Discussion

Discussion thread: here

JIRA: KAFKA-4565

Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).

Motivation

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:

1. Replication traffic is assigned to a separate network interface so that it does not interfere with client traffic.
2. External traffic goes through a proxy/load-balancer (security, flexibility) while internal traffic hits the brokers directly (performance, cost).
3. 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.

Public Interfaces

Configuration

A new broker config listener.security.protocol.map will be introduced so that we can map a protocol label to a security protocol. The config value should be in the CSV Map format that is currently used by max.connections.per.ip.overrides. The config value should follow map semantics: each key should only appear once, but values may appear multiple times. For example, the config could be defined in the following way to match the existing behaviour:

listener.security.protocol.map=PLAINTEXT:PLAINTEXT,SSL:SSL,SASL_PLAINTEXT:SASL_PLAINTEXT,SASL_SSL:SASL_SSL

To ensure compatibility with existing configs, we propose the above as the default value for the new config.

The next step is to change the validation of advertised.listeners and listeners so that the protocol label has to be one of the keys in listener.security.protocol.map (only security protocols are allowed currently). For example, the following would configure a broker with two different host:port pairs mapped to the same security protocol in two cases:

listener.security.protocol.map=CLIENT:SASL_PLAINTEXT,REPLICATION:PLAINTEXT,INTERNAL_PLAINTEXT:PLAINTEXT,INTERNAL_SASL:SASL_PLAINTEXT
advertised.listeners=CLIENT://cluster1.foo.com:9092,REPLICATION://broker1.replication.local:9093,INTERNAL_PLAINTEXT://broker1.local:9094,INTERNAL_SASL://broker1.local:9095
listeners=CLIENT://192.1.1.8:9092,REPLICATION://10.1.1.5:9093,INTERNAL_PLAINTEXT://10.1.1.5:9094,INTERNAL_SASL://10.1.1.5:9095

We then introduce a second broker config as an alternative to security.inter.broker.protocol:

inter.broker.protocol.label=REPLICATION

It is an error to set both security.inter.broker.protocol and inter.broker.protocol.label at the same time. inter.broker.protocol.label will be null by default, which means that PLAINTEXT will be used by default (as is currently the case).

Finally, we make it possible to provide different security (SSL and SASL) settings for each protocol label by using a prefix in the config name. For example, if we wanted to set a different keystore for the CLIENT protocol label, we'd use set a config with name protocol.label.client.ssl.keystore.location. If the config for the protocol label is not set, we will fallback to generic config (i.e. ssl.keystore.location) for compatibility and convenience.

ZooKeeper

Version 4 of the broker registration data stored in ZooKeeper will have protocol labels instead of security protocols in the elements of the endpoints array and an additional listener.security.protocol.map field. The latter is not strictly needed if we assume that all brokers have the same config, but it would make config updates trickier (e.g. two rolling bounces would be required to add a new mapping from protocol label to security protocol). Also, we add an additional field instead of changing the endpoints schema to allow for rolling upgrades.

{
	"version": 4,
	"jmx_port": 9999,
	"timestamp": 2233345666,
	"host": "localhost",
	"port": 9092,
	"rack": "rack1",
	"listener.security.protocol.map": {
		"PLAINTEXT": "PLAINTEXT",
		"SSL": "SSL",
		"SASL_PLAINTEXT": "SASL_PLAINTEXT",
		"SASL_SSL": "SASL_SSL"
	},
	"endpoints": [
		"CLIENT://cluster1.foo.com:9092",
		"REPLICATION: //broker1.replication.local:9093",
		"INTERNAL_PLAINTEXT: //broker1.local:9094",
		"INTERNAL_SASL://broker1.local:9095"
	]
}

Protocol

Version 2 of UpdateMetadataRequest will be introduced and the elements of the end_points array would also have a protocol_label field.

UpdateMetadata Request (Version: 2) => controller_id controller_epoch [partition_states] [live_brokers] 
  controller_id => INT32
  controller_epoch => INT32
  partition_states => topic partition controller_epoch leader leader_epoch [isr] zk_version [replicas] 
    topic => STRING
    partition => INT32
    controller_epoch => INT32
    leader => INT32
    leader_epoch => INT32
    isr => INT32
    zk_version => INT32
    replicas => INT32
  live_brokers => id [listener_security_protocol_map] [end_points]
    id => INT32
    end_points => port host protocol_label security_protocol_type
      port => INT32
      host => STRING
      protocol_label => String
      security_protocol_type => INT16

Client

Protocol labels only exist in the brokers, clients never see them.

Proposed Changes

We would have to change a number of places in the code that currently use SecurityProtocol as a key to use SecurityLabel instead. A few examples:

1. Acceptor thread
2. Metadata request handler
3. ReplicaManager
4. Broker class

The changes are mostly mechanical and don't affect public API.

We would also have to change the various authenticator classes to look for security configs for the relevant protocol label before falling back to the generic ones.

As stated previously, clients never see protocol labels and will make metadata requests exactly as before. The difference is that the list of endpoints they get back is restricted to the protocol label of the endpoint where they made the request. In the example above, let's assume that all brokers are configured similarly and that a client sends a metadata request to cluster1.foo.com:9092 and it reaches broker1's 192.1.1.8:9092 interface via a load balancer. The security protocol would be SASL_PLAINTEXT and the metadata response would contain host=cluster1.foo.com,port=9092 for each broker returned.

The exception is ZooKeeper-based consumers. These consumers retrieve the broker registration information from ZooKeeper directly and would have to be updated to map from protocol label to security protocol.

Compatibility, Deprecation, and Migration Plan

As mentioned previously, the default value of listener.security.protocol.map maps the existing security protocols to a label with the same name to maintain compatibility:

listener.security.protocol.map=PLAINTEXT:PLAINTEXT,SSL:SSL,SASL_PLAINTEXT:SASL_PLAINTEXT,SASL_SSL:SASL_SSL

For users upgrading, they should only use protocol labels once all the brokers and ZooKeeper-based consumers have been upgraded to a version that supports protocol labels.

Rejected Alternatives

1. Instead of adding the listener.security.protocol.map config, we could extend the protocol part of the listener definition to include both the protocol label and security protocol. For example, CLIENT+SASL_PLAINTEXT://192.1.1.8:9092. This is appealing from a clarity perspective (the listeners are fully defined in a single config value), but it may lead to duplication between listeners and advertised.listeners. A way to avoid that issue (at the cost of loss of symmetry) would be for advertised.listeners to only include the protocol label (we can infer the security protocol by looking at the listeners entry with the same protocol label).
2. Assume that listener.security.protocol.map is the same in every broker. The slight benefit in terms of smaller broker registration JSON is not worth the additional operational complexity when it comes to changing the config values in a running cluster (two rolling upgrades would be needed in some simple cases).
3. Using hard-coded listener domains for internal and replication traffic. The config format is simpler and there's less scope for hard to understand configs. The main disadvantage is that it's a bit too specific and may need to be extended again as more sophisticated use cases appear. The current proposal is more general and it seems like a natural evolution of the existing system.