Deprecation

OpenStack projects should adhere to the guidelines outlined here regarding deprecation of features and functionality. This process aims to ensure that operators and users of OpenStack are able to gracefully upgrade between releases, as well as migrate away from dependence on functional aspects that are removed over time. Care must be taken to avoid breaking compatibility between adjacent releases, as well as between tick releases as defined in the TC resolution on release cadence.

Justification for careful deprecation

End users of a given service need to know if a feature or an API they are using and rely on will still be supported by the software tomorrow. Operators and deployers of a given service want to be able to roll out code and configuration changes asynchronously, and therefore rely on new code working correctly with the existing config files.

At the early stages of development it’s important to be agile, experiment, and fail fast. At that point it’s not reasonable to commit to support those early mistakes forever. But as the project matures and gets more users that rely on existing features, knowing under which conditions the project can remove features, APIs or alter configuration options in the future becomes important. It can be a factor in deciding if the project is stable and mature enough for a specific use case.

Guidelines

Project teams should follow the following process for end-user-visible or operator-visible features deprecation:

  1. Features, APIs or configuration options are marked deprecated in the code. Appropriate warnings will be sent to the end user, operator or library user. Code will be frozen and only receive minimal maintenance (just so that it continues to work as-is).

  2. A migration path will be documented for current users of the feature. An email thread will be started on openstack-discuss to determine how many people are using the deprecated API or feature, and how costly the migration plan is to implement. A migration path may be “stop using that feature”: the cost is then very related to the number of people using the feature and how dependent they are to that feature.

  3. If the deliverable is part of an Interop Working Group Guideline, the project will check if the deprecated feature is part of the exposed capabilities. If it is, the obsolescence date (see below) additionally needs to take into account Interop WG capabilities deprecation schedule.

  4. Based on that data, an obsolescence date will be set. At the very minimum the feature (or API, or configuration option) should be marked deprecated (and still be supported) in at least one “tick” release branch, and for at least 12 months of releases. Consider the following examples:

    1. A feature deprecated in the 2023 tick release may be removed in the 2023 tock release. Operators deploying only tick releases receive one notice and have 12 months before the next tick where the feature is removed. Operators deploying tick-tock releases receive one notice and have 6 months before the release where the feature is removed, as was the case before the tick-tock model.

    2. A feature deprecated in the 2023 tock release must also be present and supported the 2024 tick release, and must also be called out as deprecated (so that operators only deploying tick releases receive that warning). The feature may be removed in the 2024 tock release. Operators deploying only tick releases receive one notice and have 12 months before the next tick where the feature is (long-since) removed. Operators deploying tick-tock receive two notices and still have 12 months before the release where the feature is removed.

Note that this delay is a required minimum. For significant features, it is recommended that the deprecated feature appears at least in the next two tick release branches. Ideally, features are deprecated and removed only in tick releases to keep the accounting to a minimum, but there may be cases where it is advantageous to be more flexible.

Another scenario to consider is experimental features that are introduced in a tock release, which “don’t survive” and need to be removed quickly. In that case the earliest possible sequence would be deprecation in the following tick (effectively introducing the feature as deprecated to tick-only deployers) and then removal in the following tock release.

In addition, projects should:

  • Use automated testing to verify that configuration files and database schema are forward-compatible from release to release and that this policy is not accidentally broken (for example, a gating grenade test).

  • Use automated testing to verify the above compatibility also remains stable between tick releases so that operators are able to skip tock releases.

  • No existing config options will have their meaning changed in such a way that it would alter the software behavior or otherwise render an existing config file broken.

Note: these guidelines apply to both service and library feature deprecation policies and procedures.