Release Management#

This page describes DataFusion release branches and backports. For the maintainer release guide, including release candidate artifacts, voting, and publication, see the release process README in dev/release.

Overview#

DataFusion typically has a major release about once per month, including breaking API changes. Patch releases are made on an ad hoc basis, but we try to avoid them because major releases are frequent.

New development happens on the main branch. Releases are made from release branches named branch-NN, such as branch-50 for the 50.x.y release series.

In general:

  • New features land on [main]

  • Patch releases are cut from the corresponding branch-NN

  • Only targeted, low-risk fixes should be added to a release branch

Changes reach a release branch in one of two ways:

  • (Most common) Fix the issue on main and then backport the merged change to the release branch

  • Fix the issue on the release branch and then forward-port the change to main

Releases are coordinated using GitHub issues. Each planned release is listed in the DataFusion Releases tracking issue, and each release is coordinated in a dedicated issue, such as the release issue for 50.3.0. If you think a fix should be included in a patch release, discuss it on the relevant tracking issue or open a backport PR and link it there.

To prepare for a new release series, maintainers:

  • Create a new branch from main, such as branch-50, in the Apache repository

  • Continue merging new features to main

  • Prepare the release branch for release by updating versions, changelog content, and any additional release-specific fixes via the Backport Workflow

  • Create release candidate artifacts from the release branch

  • After approval, publish to crates.io, ASF distribution servers, and Git tags

Backport Criteria#

A release branch is a stabilization branch for an imminent or recent patch release. The bar for landing a change on a release branch is therefore higher than the bar for landing on main, not lower. These criteria define what is eligible for backport; the Backport Workflow below describes the mechanics.

DataFusion follows Cargo SemVer, with breaking changes allowed at major version boundaries — see the API health policy for the full framing of public Rust and SQL API stability. Patch releases (x.y.z, z ≥ 1) carry fixes only and never introduce new features or breaking changes.

Eligible for backport#

  • Security fixes. Fixes for known or reported security issues should be backported to every actively maintained release branch.

  • Correctness fixes. Fixes for queries that produce incorrect results, panics, data loss, or crashes. If the fix itself changes user-visible SQL semantics to make a wrong result right, follow Behavior changes below.

  • Stability and regression fixes. Fixes for regressions introduced in the current release line, hangs, deadlocks, memory leaks, or other availability issues.

  • Build, CI, and test fixes required to keep the branch buildable and releasable.

  • Documentation fixes for behavior already in the release. Documentation for behavior that exists only on main does not belong on a release branch.

Behavior changes#

A “behavior change” is any fix that alters user-visible results: SQL semantics (values, ordering, types, null handling), error messages that downstream users may rely on, plan output, or default configuration values.

Behavior-changing fixes need extra scrutiny on a release branch because users upgrading between patch versions do not expect their queries to start returning different results. When proposing one for backport, state on the release tracking issue why the change should ship in this patch release rather than wait for the next major. The previous and new behavior should already be documented on the original issue or PR — link to that rather than restating it.

If in doubt, default to “land on main, ship in the next major.”

Who decides#

The release manager for the active release line is the final reviewer of what goes into the patch release. They coordinate via the release tracking issue (for example, the release issue for 50.3.0). Anyone may propose a backport by opening a backport PR and linking it from the tracking issue; inclusion is the release manager’s call.

Active release branches#

DataFusion does not maintain Long-Term Support branches. In general only the most recent branch-NN is actively maintained for backports, but if you need fixes in older releases, we are open to discussion.

Security fixes are an exception: a maintainer may choose to backport a critical security fix to an older branch even after it would otherwise be closed. Discuss on the dev list or in a tracking issue before doing so.

Backport Workflow#

The usual workflow is:

  1. Fix on main first, and merge the fix via a normal PR workflow.

  2. Cherry-pick the merged commit onto the release branch.

  3. Open a backport PR targeting the release branch (examples below).

Inputs#

To backport a change, gather the following information:

  • Target branch, such as apache/branch-52

  • The release tracking issue URL, such as https://github.com/apache/datafusion/issues/19692

  • The original PR URL, such as https://github.com/apache/datafusion/pull/20192

  • Optional explicit commit SHA to backport

Apply the Backport#

Start from the target release branch, create a dedicated backport branch, and use git cherry-pick. For example, to backport PR #1234 to branch-52 when the commit SHA is abc123, run:

git checkout apache/branch-52
git checkout -b alamb/backport_1234
git cherry-pick abc123

Test#

Run tests as described in the testing documentation.

Open the PR#

Create a PR against the release branch, not main, and prefix it with [branch-NN] to show which release branch the backport targets. For example:

  • [branch-52] fix: validate inter-file ordering in eq_properties() (#20329)

Use a PR description that links the tracking issue, original PR, and target branch, for example:

- Part of <tracking-issue-url>
- Closes <backport-issue-url> on <branch-name>

This PR:

- Backports <original-pr-url> from @<author> to the <branch-name> line