Linux runner image 0.3.0 is available as ghcr.io/tuist/tuist-linux-runner:0.3.0.
This version makes self-hosted CI builds more diagnosable and opens the fleet to pull requests from forks.
-
A resource vitals probe now runs while a job is executing, writing memory usage, OOM kill events, and CPU/memory pressure to the runner logs. When a Linux runner disappears mid-job, the last samples are preserved, so you can determine whether the workload was OOM-killed or starved of resources instead of guessing. Log collection also now runs on the runner nodes, so those diagnostics actually reach the logging backend.
-
The runner image uses a split-container design: a small initialization stage mints a single-job credential, and the workflow runs in a separate container that never holds the pool-scoped dispatch token. Because untrusted fork code no longer shares a container with a broad token, CI for pull requests from forks can run on Tuist’s own Linux runners. Android and macOS jobs remain on their existing runners for now.