we_are_coded.by CODE · The world, decoded
БГ

Gitea · Tools

Gitea Runner 6.0.0: two breaking changes

Gitea shipped Runner 6.0.0 on 11 November with two breaking changes: builtin:checkout is removed and the runner now requires Gitea 1.26.2 or newer.

Source
Gitea
Published
The event
Checked on
Responsible editor
Cvetelin Ivanov
How we work
MethodCorrections

Two things break with this release. One of them is your pipeline.

Fact

The facts: on 11 November 2026 Gitea announced Runner 6.0.0. The release has two breaking changes: builtin:checkout is removed and the runner now requires Gitea 1.26.2 or newer. Steps with uses: builtin:checkout now fail; use actions/checkout instead, and if your job image has no Node, taiki-e/checkout-action. An older Gitea leaves most job logs in the database's dbfs_data table instead of archiving them. The shared tool cache (runner.tool_cache_mode: shared) is now safe at any capacity: until now it was only safe at runner.capacity: 1, because concurrent jobs installed into the same cache and could corrupt it. Each capacity slot now has its own tool cache (#1401); the first slot keeps the existing volume or host directory, and each further slot downloads its own tools, so disk use grows with capacity. The runner now prefers IPv4 and detects the address again for each job, unless cache.host sets it (#1398). Newly registered runners get an ubuntu-26.04 label (#1396). Windows is a supported platform since 5.4.0. The release is 6.0.0, dated 11 November 2026.

The address bug is not theory. Gitea describes it as something they ran into themselves while setting up a macOS runner.

The shared cache was safe only at one slot. Now it is safe at any, but each further slot pays with its own disk space.

The practical part is in the upgrade notes and it is three lines. Upgrade Gitea to 1.26.2 or newer before you touch the runner. Replace uses: builtin:checkout with actions/checkout, or with taiki-e/checkout-action if your job image has no Node. Replace gitea/runner:5.x with gitea/runner:6.0.0 in Docker Compose, Kubernetes manifests and service units.

There are two quiet things too. If you upgrade from 5.2.x or older, caches start cold once, because entries are now scoped by ref. If you upgrade from 5.4.0 or older and rely on the runner removing leftover containers from its Docker daemon, set container.sweep: true; it is opt-in since 5.4.1.

Upgrade Gitea first. The runner after. The order is not a wish, it is a condition written in the notes.

The visual is generated code-art, no third-party images.