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 couple of 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).

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

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. It's worth mentioning that the config value should be the same in every broker in the Kafka cluster for it to work as expected. This is also the case for a number of existing Kafka broker configs and since Kafka doesn't support cluster configs at this point, it seems acceptable.

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).

There are a couple more interfaces that need to be updated slightly to support protocol labels. The first is the broker registration data stored in ZooKeeper. Protocol labels would be used instead of security protocols in version 4 of the format:

{
  "version":4,
  "jmx_port":9999,
  "timestamp":2233345666,
  "host":"localhost",
  “port”:9092,
  "rack":"rack1",
  "endpoints": [
    "CLIENT://cluster1.foo.com:9092",
    “REPLICATION://broker1.replication.local:9093”,
    “INTERNAL_PLAINTEXT://broker1.local:9094”,
	"INTERNAL_SASL://broker1.local:9095"
  ]    
}

The second and final interface change is to the UpdateMetadataRequest protocol type. Version 2 would have a protocol_label field instead of security_protocol_type:

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 [end_points] 
    id => INT32
    end_points => port host protocol_label (instead of security_protocol_type)
      port => INT32
      host => STRING
      protocol_label => String (instead of security_protocol_type => INT16)

Note that 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

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

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, if a client makes a metadata request to cluster1.foo.com:9092, it will be mapped to the 192.1.1.8:9092 interface. The broker will return SASL_PLAINTEXT://cluster1.foo.com:9092 in the metadata response. It returned the entry in advertised.listeners with the same protocol label, but it translated the protocol label to the associated security protocol before returning it to the client.

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:

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

As such, the change is backwards compatible.

For users upgrading, they should only use protocol labels once all the brokers have been upgraded.

Future work

1. Different security protocol settings per listener. For example, one may want to configure SSL differently for internal versus external traffic.

Rejected Alternatives

1. 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.