The goals behind the command line shell are fundamentally to provide a centralized management for Kafka operations.

Status

Current state: Accepted

Discussion thread: here

JIRA:  Unable to render Jira issues macro, execution error.

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

Motivation

Users of Kafka have created dozens of different systems to work with Kafka. Providing a wire protocol that allows the brokers to execute administrative code and public api/client has many benefits including:

• Allows clients in any language to administrate Kafka
  • Wire protocol is supported by any language
• Provides public client for performing admin operations
  • Ensures integration test code in other projects and clients maintains compatibility
  • Prevents users from needing to use the Command classes and work around standard output and system exits
• Removing the need for admin scripts (kafka-topics.sh, kafka-acls.sh, etc) to talk directly to Zookeeper. 
  • Allows ZNodes to be completely locked down via ACLs
  • Further hides the Zookeeper details of Kafka

Public Interfaces

• Changes to Wire Protocol:
  • Adds the following new Request/Response messages:
    • CreateTopics
    • AlterTopics
    • DeleteTopics
    • ListAcls
    • AlterAcls
    • DescribeConfig (moved to KIP-133: Describe and Alter Configs Admin APIs)
    • AlterConfig (moved to KIP-133: Describe and Alter Configs Admin APIs)
  • Modifies Metadata Request/Response to allowing polling for in-progress or complete admin operations. Added fields include:
    • Add the ability to request no topics with a null topics list
    • Boolean indicating if a topic is marked for deletion
    • Boolean indicating if a topic is an internal topic
    • Rack information (if not added by KIP-36 Rack aware replica assignment)
    • Boolean indicating if a broker is the controller

Proposed Changes

Proposed changes include 2 parts:

1. Wire protocol additions and changes
2. Server-side message handlers and authorization

Follow Up Changes

Changes that should be considered shortly after or are enabled by this KIP included:

• General

  • New Java AdminClient implementation (KIP-117)
  • Refactor admin scripts and code to use new client where appropriate
  • Support forwarding requests to the required broker (coordinator, group leader, partition leader) (KAFKA-1912)
    • See Request Forwarding below

• Metadata Schema

  • Consider supporting regex topic filters in the request 
  • Filter internal topics using the returned metadata ( Unable to render Jira issues macro, execution error. )

• Topic Admin Schema

  • Improve the broker side delete topic implementation
    • Delete is likely to get used more once creation/deletion is made easier with the client. The broker side implementation has had many jiras. 
    • Currently can't delete unhealthy topics. 
  • Support cluster consistent blocking to wait until all relevant brokers have the required metadata
    • This may require significant re-work of the controller to do correctly
    • See Cluster Consistent Blocking below
  • Implement auto-topic creation client side (KAFKA-2410)
  • Add topic creation to the MirrorMaker client?
  • Support renaming topics (KAFKA-2333)
    • This might required unique ids for topics instead of using the name (this would improve delete too)
  • Improve reliability and speed of topic deletion
    • Support force deleting unhealthy topics
    • Support marking for deletion and async data cleanup
      • This would required unique ids for topics instead of using the name (this is needed for rename too)
      • The topic can then be marked as deleted instead of requiring all data to be removed immediately and in the mean time a new topic with the same name can be created. 

• ACL Admin Schema

  • Review privileges for listing and altering ACLs to be more fine grained.
  • Provide an Authorizer interface using the new Java classes used by the ACL requests/responses (KAFKA-3509)
    • Deprecate the old one to encourage transition
  • Define standard Exceptions that can be thrown by the Authorizer in the interface (KAFKA-3266)
    • Otherwise all exceptions are unknown server exception to the client
  • Consider building a sync call into the Authorizer to ensure changes are propagated

Details

1. Wire Protocol Extensions

Schema

TopicExistsException

InvalidPartitionsException

InvalidReplicationFactorException

InvalidReplicaAssignmentException

InvalidConfigurationException

NotControllerException

Metadata Schema (Voted and Adopted in 0.10.0.0)

Metadata Request (version 1)

MetadataRequest => [topics] 

Metadata Response (version 1)

MetadataResponse => [brokers] controllerId [topic_metadata]   
  brokers => node_id host port rack 
    node_id => INT32
    host => STRING
    port => INT32
    rack => NULLABLE_STRING
  controllerId => INT32
  topic_metadata => topic_error_code topic is_internal [partition_metadata] 
    topic_error_code => INT16
    topic => STRING
    is_internal => BOOLEAN
    partition_metadata => partition_error_code partition_id leader [replicas] [isr] 
      partition_error_code => INT16
      partition_id => INT32
      leader => INT32
      replicas => INT32
      isr => INT32

Topic Admin Schema

Create Topics Request (KAFKA-2945): (Voted and Committed for 0.10.1.0)

CreateTopics Request (Version: 0) => [create_topic_requests] timeout 
  create_topic_requests => topic num_partitions replication_factor [replica_assignment] [configs] 
    topic => STRING
    num_partitions => INT32
    replication_factor => INT16
    replica_assignment => partition_id [replicas] 
      partition_id => INT32
      replicas => INT32
    configs => config_key config_value 
      config_key => STRING
      config_value => STRING
  timeout => INT32

Create Topics Response

CreateTopics Response (Version: 0) => [topic_error_codes] 
  topic_error_codes => topic error_code 
    topic => STRING
    error_code => INT16

DeleteTopics Request (Version: 0) => [topics] timeout 
  topics => STRING
  timeout => INT32

DeleteTopics Response (Version: 0) => [topic_error_codes] 
  topic_error_codes => topic error_code 
    topic => STRING
    error_code => INT16

ListAcls Request (Version: 0) => principal resource 
  principal => NULLABLE_STRING  
  resource => resource_type resource_name 
    resource_type => INT8
    resource_name => STRING

ListAcls Response (Version: 0) => [responses] error_code 
  responses => resource [acls] 
    resource => resource_type resource_name 
      resource_type => INT8
      resource_name => STRING
    acls => acl_principal acl_permission_type acl_host acl_operation 
      acl_principal => STRING
      acl_permission_type => INT8
      acl_host => STRING
      acl_operation => INT8
  error_code => INT16

AlterAcls Request (Version: 0) => [requests] 
  requests => resource [actions]     
    resource => resource_type resource_name 
      resource_type => INT8
      resource_name => STRING
    actions => action acl 
      action => INT8
      acl => acl_principal acl_permission_type acl_host acl_operation 
        acl_principal => STRING
        acl_permission_type => INT8
        acl_host => STRING
        acl_operation => INT8

AlterAcls Response (Version: 0) => [responses] 
  responses => resource [results] 
    resource => resource_type resource_name 
      resource_type => INT8
      resource_name => STRING
    results => action acl error_code 
      action => INT8
      acl => acl_principal acl_permission_type acl_host acl_operation 
        acl_principal => STRING
        acl_permission_type => INT8
        acl_host => STRING
        acl_operation => INT8
      error_code => INT16

At the highest level, admin requests will be handled on the brokers the same way that all message types are. However, because admin messages modify cluster metadata they should be handled by the controller. This allows the controller to propagate the changes to the rest of the cluster.  However, because the messages need to be handled by the controller does not necessarily mean they need to be sent directly to the controller. A message forwarding mechanism can be used to forward the message from any broker to the correct broker for handling. 

Because supporting all of this is quite the undertaking I will describe the "ideal functionality" and then the "intermediate functionality" that gets us some basic administrative support quickly while working towards the optimal state. 

Ideal Functionality:

1. A client sends an admin request to any broker
2. The admin request is forwarded to the required broker (likely the controller)
3. The request is handled and the server blocks until a timeout is reached or the requested operation is completed (failure or success)
   1. An operation is considered complete/successful when all required nodes have the correct/current state. 
   2. Immediate follow up requests to any broker will succeed.
   3. Requests that timeout may still be completed after the timeout. The users would need to poll to check the state. 
4. The response is generated and forwarded back to the broker that received the request.
5. A response is sent back to the client. 

Intermediate Functionality:

1. A client sends an admin write requests to the controller broker. Read requests can still go to any broker. 
   1. As a follow up request forwarding can be added transparently. (see below)
2. The request is handled and the server blocks until a timeout is reached or the requested operation is completed (failure or success)
   1. An operation is considered complete/successful when the controller node has the correct/current state.
   2. Immediate follow up requests to the controller will succeed. Others (not to the controller) are likely to succeed or cause a retriable exception that would eventually succeed. 
   3. Requests that timeout may still be completed after the timeout. The users would need to poll to check the state. 
3. A response is sent back to the client. 

The ideal functionality has 2 features that are more challenging initially. For that reason those features will be removed from the initial changes, but will be tracked as follow up improvements. However, this intermediate solution should allow for a relatively transparent  transition to the ideal functionality. 

Request Forwarding: KAFKA-1912

Request forwarding is relevant to any message the needs to be sent to the "correct" broker (ex: partition leader, group coordinator, etc). Though at first it may seam simple it has many technicall challenges that need to be decided in regards to connections, failure, retries, etc. Today, we depend on the client to choose the correct broker and clients that want to utilize the cluster "optimally" would likely continue to do so. For those reasons it can be handled it can be handled generically as an independent feature. 

Cluster Consistent Blocking:

Blocking an admin request until the entire cluster is aware of the correct/current state is difficult based on Kafka's current approach for propagating metadata. This approach varies based on the the metadata changing.

• Topic metadata changes are propagated via UpdateMetadata and LeaderAndIsr requests
• Config changes are propagated via zookeeper and listeners
• ACL changes depend on the implementation of the Authorizer interface 
  • The default SimpleACLAuthorizer uses zookeeper and listeners

Though all of these mechanisms are different, they are all commonly "eventually consistent". None of the mechanisms, as currently implemented, will block until the metadata has been propagated successfully. Changing this behavior would require a large amount of change to the KafkaController, additional inter-broker messages, and potentially a change to the Authorizer interface. These are are all changes that should not block the implementation of KIP-4.

The intermediate changes in KIP-4 should allow an easy transition to "complete blocking" when the work can be done. This is supported by providing optional local blocking in the mean time. This local blocking only blocks until the local state on the controller is correct. We will still provide a polling mechanism for users that do not want to block at all. A polling mechanism is required in the optimal implementation too because users still need a way to check state after a timeout occurs because operations like "create topic" are not transactional. Local blocking has the added benefit of avoiding wasted poll requests to other brokers when its impossible for the request to be completed. If the controllers state is not correct, then the other brokers cant be either. Clients who don't want to validate the entire cluster state is correct can block on the controller and avoid polling all together with reasonable confidence that though they may get a retriable error on follow up requests, the requested change was successful and the cluster will be accurate eventually.

Because we already add a timeout field to the requests wire protocols, changing the behavior to block until the cluster is consistent in the future would not require a protocol change. Though the version could be bumped to indicate a behavior change. 

Rejected Alternatives

If there are alternative ways of accomplishing the same thing, what were they? The purpose of this section is to motivate why the design is the way it is and not some other way.

TopicMetadataRequest/Response: After some debate we decided not to evolve the TopicMetadataResponse to remove the ISR field (which currently can return incorrect information). There is a use-case for this in KAFKA-2225, so we will treat this a bug and fix it going forward. See KAFKA-1367 for more details