Status

Current state: Draft

Discussion thread: here

...

When the group is Stable (i.e. when there are active consumers in the group), the offsets associated with the group are returned as part of the info returned for each consumer in the group in terms of a ConsumerSummary record. But there is no ConsumerSummary record if there is no consumer in the group, and therefore, in that scenario a different method is required for extracting those offsets.

Public Interfaces

This is the current schema for DescribeGroups (version 0).

...

DescribeGroups Request (Version: 1) => [group_ids] 
  group_ids => STRING
 
DescribeGroups Response (Version: 1) => [groups] 
  groups => error_code group_id state protocol_type protocol [members] [offsets]
    error_code => INT16
    group_id => STRING
    state => STRING
    protocol_type => STRING
    protocol => STRING
    members => member_id client_id client_host member_metadata member_assignment 
      member_id => STRING
      client_id => STRING
      client_host => STRING
      member_metadata => BYTES
      member_assignment => BYTES
    offsets => topic partition offset metadata
      topic => STRING
      partition => INT32
      offset => INT64
      metadata => STRING

...

Proposed Changes

The proposal is to include a list offsets of tuples in the response to DescribeGroups request. Each of these tuples contains the following fields:

• topic: topic from which a previous member of the group consumed.
• partition: partition of the above topic for which offset information associated with the group exists
• offset: associated committed offset of the above topic partition within the group
• metadata: metadata associated with the above offset

Compatibility, Deprecation, and Migration Plan

There should be no impact to existing users of this protocol. The Request schema remains unchanged, and the Response schema is only expanded. There is no change in the existing Response structure, and, therefore, users who use version 0 of the schema should not be affected. The code behind kafka-consumer-groups.sh will also be updated (as requested in the JIRA) to return more helpful information. In case it is necessary to provide the old output as well, a new parameter can be defined for the tool to report the new information; while the old way of running the tool still prints out the old output. 

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.