mirror of
https://github.com/jmcorgan/fips.git
synced 2026-10-05 19:18:25 +00:00
Johnathan Corgan
856055db76
Build and container-start output went to /dev/null, so a failing scenario printed that it had failed and nothing about why. The ubuntu:26.04 e2e failure has been undiagnosable since July for exactly that reason. Output is now captured and emitted only when the command fails, naming which command it was, so a passing run stays as quiet as before. The systemd readiness wait, which is the check that actually times out, dumps the container state, the failed units and the journal when it gives up. This also fixes a defect the change uncovered. The script runs under pipefail, and systemctl is-system-running exits non-zero when the system is degraded, so the pipeline testing for 'running|degraded' failed even when grep matched. The degraded branch was dead, and since systemd inside a container always settles at degraded, every scenario burned the full 30-second boot timeout and emitted a spurious warning about a container that had booted correctly. The readiness poll now matches on the captured state instead of piping into grep. The original ubuntu:26.04 failure does not reproduce on this host, so that half stays open. What changes is that the next occurrence will say why.