[azlekov]

Every job on the control runners has failed since today's slots release. In Set up runner, the JOB_STARTED hook printed [pre-job] Previous job did not finish; sweeping… and exited 1.

Cause. The runner runs the hook under bash -e, whatever the script's shebang or its "not using set -e" comment say. When the bun cache has no @scope folder, "$bun_cache"/@*/ reaches find as a literal path, find exits 1 (its stderr is discarded), and the hook aborts.

Fix. Glob with nullglob and only when the cache exists, end each sweep command in || true, and end the script in exit 0. The header comment now says why.

Test. tests/pre-job.sh runs the hook under bash -e with no bun cache, with a cache that has no @scope folder, and with leftovers to sweep. It fails on the old hook (2 checks) and passes on the new one. shellcheck -x is clean.

Second commit: cleanup.sh, same root cause. This answers the open question. The JOB_COMPLETED hook also runs under bash -e:

• On a runner with no ~/.gradle/caches (every control runner, and any runner that never ran Gradle), find exits 1.
• cleanup.sh then aborts, and "Complete runner" fails the job even though every step passed.
• busy is left behind, so the next job takes the crash-recovery path.

Since the slots release this failed jobs on control and regular runners: smart-irrigation, nis-suite, kodni-mobile, infrasensing-senslogging.

The fix:

• busy is cleared first;
• every command ends in || true;
• the main path is cleanup || true.

tests/cleanup.sh runs the hook under bash -e with and without a Gradle cache. It fails on the old hook (4 checks) and passes on the new one.

🤖 Generated with Claude Code