grenadeandClaude Opus 5
f724d8f256
deploy-web failed on run 27 after every health check had passed. The
service was live, the frontend was serving, the public vhost was
answering and /metrics was closed -- and then the job went red because
`tail` could not read an nginx log.
5244585 dropped sudo here on the premise that "nginx's logs are
world-readable and /var/log/nginx is traversable". Half right:
/var/log/nginx is drwx--x--x, so traversable but not listable, and
logrotate creates the files 0640 nginx:root. The deploy account cannot
read them as itself at all, so the step could only ever have failed.
Try sudo, say so plainly when it is not permitted, and mark the step
continue-on-error. A step that exists to print context after a deploy
should never be able to fail one.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PqNtYNhov3fukx46KS9R7L