systemd – deploy-and.run

Boot the Release Against a Copy of Production Before You Switch

November 15, 2026November 12, 2026 by halimat

Terminal sketch of a deploy preflight: the new release starts on a spare port against a VACUUM INTO copy of the database; a 200 from /healthz allows the switch, a crash aborts it.

A health check after the switch only tells you how long the outage was. Starting the new release on a spare port against a VACUUM INTO copy of the live SQLite database catches broken dependencies, syntax errors and failing migrations while the old version keeps serving — and a deliberately broken release proved it with HTTP 200 the whole way through.

rsync -a Stole the setgid Bit, and systemd Blocked the Rollback

November 15, 2026November 11, 2026 by halimat

Terminal sketch: a release directory loses its setgid bit and shared group after rsync -a, node fails with EACCES on node_modules, and systemctl refuses the restart with 'start request repeated too quickly'.

The first automated deploy of a small Node.js app went red, and the rollback failed too. Two independent mechanisms, each harmless alone: rsync copying permissions it had no business copying, and a systemd start limit that outlived the crash loop it was meant to contain.