Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

Table of Contents

This page is meant as a template for writing a KIP. To create a KIP choose Tools->Copy on this page and modify with your content and replace the heading with the next KIP number and a description of your issue. Replace anything in italics with your own description.

Status

Current state: "Under Discussion"

Discussion thread: here [Change the link from the KIP proposal email archive to your own email thread]

JIRA:here [Change the link from KAFKA-1 to your own ticket] KAFKA-15922

PR: 1486014984

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

Motivation

The metadata version MetadataVersion (MV) is used to define the on disk and over the wire record format for messages stored in the metadata log. A new metadata version is introduced each time a record format is changed or a new record is introduced to support a new feature. Examples of this are the support of SCRAM in MV IBP_3_5_IV2, support of Delegation Tokens in MV IBP_3_6_IV2 and the support of JBOD in MV IBP_3_7_IV2. All brokers and controllers must support a given MV before the cluster can start using that MV.  Once the brokers and controllers agree to use a MV then the record format for that MV and all earlier MV is supported thus also enabling all earlier features, thus if you want to use Delegation Tokens you get the record format changes necessary to support SCRAM also.

The development of a new feature that requires a new MV often requires multiple commits before the feature is complete, sufficiently tested, and ready for production. Sometimes there is a need for multiple features to be developed at the same time and in some cases not all of these features with be complete and ready for production for a given release. In these scenarios we need to reorder MetadataVersions such that production ready features have their MV before those that are not ready so that customers can specify to use only those features which are stable.

Proposed Changes

The idea is to define a production-ready MetadataVersion defined in MetadataVersion.java as as LATEST_PRODUCTION such that all MetadataVersions less than or equal to this value are stable and ready for production. Any MetadataVersion greater than this value is unstable and requires a special property to use in a clusterBy default only those features with a MV less than or equal to the LATEST_PRODUCTION MV will be available to use in a cluster. A cluster can set "unstable.metadata.versions.enable" to true in the broker and controller properties file to enable the use of a MV greater than LATEST_PRODUCTION in a cluster. Unit tests will default to use all MetadataVersions defined not just those less than or equal to LATEST_PRODUCTION so that developers can add unit tests for new features easily. Any unstable feature which requires a new MV can have that MV change , as in a later release of Apache Kafka to a higher MV. This will occur when other features with a higher MV become stable and ready for production. A MV defined as unstable will never have its MV moved to a lower MV.

Public Interfaces

Briefly list any new interfaces that will be introduced as part of this proposal or any existing interfaces that will be removed or changed. The purpose of this section is to concisely call out the public contract that will come along with this feature.

A public interface is any change to the following:

...

Binary log format

...

The network protocol and api behavior

...

Any class in the public packages under clientsConfiguration, especially client configuration

  • org/apache/kafka/common/serialization

  • org/apache/kafka/common

  • org/apache/kafka/common/errors

  • org/apache/kafka/clients/producer

  • org/apache/kafka/clients/consumer (eventually, once stable)

...

Monitoring

...

Command line tools and arguments

...

A new property "unstable.metadata.versions.enable" will be created to allow for developers to create clusters and use features with an unstable MV. Using this feature has risks and may result in the cluster with this feature being unable to upgrade to a future release or even a more recent build from trunk.

Compatibility, Deprecation, and Migration Plan

  • What impact (if any) will there be on existing users?
  • If we are changing behavior how will we phase out the older behavior?
  • If we need special migration tools, describe them here.
  • When will we remove the existing behavior?

Test Plan

Describe in few sentences how the KIP will be tested. We are mostly interested in system tests (since unit-tests are specific to implementation details). How will we know that the implementation works as expected? How will we know nothing broke?

Rejected Alternatives

  • Features with a stable MV will never have that MV change!
  • Users will only be able to use features with a stable MV unless they explicitly configure their clusters with the new property.
  • Features with an unstable MV may have that MV increased if a feature with a higher MV becomes stable and ready for production.
  • Using an unstable MV may result in the cluster being unable to upgrade to a future AK release or even a more recent build from trunk.

Test Plan

We will validate that the StorageTool will only format the metadata log to a stable MV unless "unstable.metadata.versions.enable" is set to true in the properties file.

We will validate that only stable features are enabled in the defaultFeatureMap unless the enableUnstable is set to true.

We will validate that stable features are enabled in a their specific MV. (TODO)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.