v2.77.1

Binaries and the merged-pull-request list are on the GitHub releases page.

This release fixes four problems that appear when a lane runs several jobs at once on one runner. Plans without lanes, and lanes of one job, were not affected by the last three.

A re-run job's success is seen

gh run rerun --failed resumes one execution: orun run --lane … --retry re-opens the lane's failed jobs. A job re-opened that way and then passed was still read as failed, so a lane in a later environment waiting on it kept waiting until its 60-minute limit. orun now reads the re-open (state.job.retried) the way the platform does.

A completed job is not "taken over"

A job's heartbeat kept running after the job reported its result, while it ran post-steps and printed its output. The platform answers every heartbeat for a finished job with lease_lost, and orun printed ↻ lease lost on <job> — another runner has taken it over and cancelled the job's remaining work. The heartbeat now stops when the job reports its result.

Tool installers no longer race

Jobs on one runner share the tool cache. Two actions/setup-node steps installing the same Node version at once could leave one job with a half-written install (ENOENT … bin/npm) or running a node binary still being written (text file busy). Actions whose name says setup (actions/setup-node, pnpm/action-setup, hashicorp/setup-terraform, …) now run one at a time per tool cache; the second finds the first's install and reuses it.

A failed job's log is printed

With several jobs in flight, a lane prints each job's step groups when the job ends. A job that failed a step printed nothing of its own, its failing step included. It now prints its steps, the error annotation and its footer.