Actions compatibility
Your workflows run as they do on GitHub-hosted runners. What differs is listed here.
Actions services
-
Cache service v2
actions/cache@v4, setup-* caches, restore keys, cross-branch scopes
-
Artifacts v4
actions/upload-artifact@v4 and download-artifact@v4, retention, merge
Also in the app, the API and the CLI.
-
Live logs and log archive
Streaming step logs, groups, masks, the downloadable archive
Searchable across runs.
-
Job summaries
GITHUB_STEP_SUMMARY Markdown on the run page
-
Annotations and problem matchers
::error, ::warning, ::notice with file and line
-
OIDC tokens
id-token: write for keyless sign-in to AWS, GCP, Azure, Vault
Issued by GitHub; claims are unchanged.
-
Artifact attestations
actions/attest-build-provenance, SLSA provenance
-
Tool cache
setup-node, setup-python, setup-go, setup-java, rustup
Popular versions are already in the image.
-
Packages and container registry
GITHUB_TOKEN sign-in to ghcr.io and GitHub Packages
Workflow syntax
-
Matrix, needs, if, strategy
Fan-out, fail-fast, max-parallel, include and exclude
-
Reusable workflows and composite actions
workflow_call, uses: ./.github/actions/...
-
Outputs and environment files
GITHUB_OUTPUT, GITHUB_ENV, GITHUB_PATH, GITHUB_STATE
-
Concurrency groups
concurrency, cancel-in-progress
-
Environments and approvals
Protected environments, required reviewers, secrets
-
Re-run failed jobs
Re-run all, re-run failed, re-run with debug logging
Also from the app and the CLI.
-
Timeouts and cancel
timeout-minutes, cancel from GitHub
-
Every trigger
push, pull_request, schedule, workflow_dispatch, merge_group, ...
Containers
-
Service containers
services: postgres, redis, ... with health checks and ports
-
Container jobs
container: image for the whole job
-
Docker, buildx and compose
docker build, buildx bake, docker compose up
-
Docker layer cache
Layers kept between runs
Add-on, billed per GB.
-
Sticky disks
A disk that follows a job between runs (node_modules, Bazel, Gradle)
Add-on, billed per GB.
Machines
-
Linux x86-64, Ubuntu 24.04
play-2vcpu to play-16vcpu, 4 GiB of memory per vCPU
-
Memory added as the job needs it
up to 4 GiB per vCPU, given back when freed
-
Linux arm64
ARM runners
-
macOS and Windows
macos-*, windows-* labels
Keep these jobs on GitHub-hosted runners.
-
GPU
GPU runners
Moving in
-
One-line change
runs-on: play-4vcpu
-
Pull request opened for you
The GitHub App finds every runs-on and opens one PR