Quanamo — release

quanamo-release is the end-to-end release pipeline for the three Quanamo apps. Tests, tag, deploy, warmup, health check — all gated, fail-fast on any phase.

Single profile

Unlike Kraite, Quanamo has exactly one profile: odin web. No CWD detection. The release always runs against:

  • ~/Herd/admin.quanamo.test → /home/odin/admin.quanamo.com (PostgreSQL master, runs core migrations)
  • ~/Herd/feedz.quanamo.test → /home/odin/feedz.quanamo.com (shares admin's PostgreSQL DB; no migrations of its own)
  • ~/Herd/kanban.quanamo.test → /home/odin/kanban.quanamo.com (owns its own MySQL DB; own migrations)

Run from anywhere on disk — the dispatchers operate on the project folders by absolute path.

Phases

Strict order. AFTER the local phase that prepares its inputs.

#PhaseTouchesHard gate?
1Testslocalyes — any failure stops the pipeline
2Taglocal + GitHubrecurses into quanamo-tag per project
3Deployodinrecurses into quanamo-deploy — admin → feedz → kanban
4Warmupodinrecurses into quanamo-warmup
5Healthodinrecurses into quanamo-health
6Cleanuplocalcancel scheduled polls, report

Modifiers

ModifierEffect
skip-testsSkip Phase 1 — dangerous, only for pure config/docs changes
skip-ciAccepted but no-op — none of the Quanamo repos have a .github/workflows/ directory, so the CI gate is already vacuous

Test phase (Phase 1)

Detection cascade per project — vendor/bin/pest present → pest directly with explicit 512M memory; pest absent but phpunit.xml / tests/Pest.php / pest.xml present → php artisan test --compact; nothing → "no test suite — skipped".

Order: admin → feedz → kanban. Pipes through tail are BANNED — they clobber $? and a missing binary silently reports green.

If any suite fails → STOP. Do NOT tag, deploy, warmup, or health-check.

Why vendor/bin/pest direct, not the artisan wrapper

The Laravel test runner wrapper triggers Pest's browser-plugin shutdown hook, which can blow the default 128 MB memory limit on symfony/console/Input/Input.php:40 and surface as a fatal AFTER the suite passes — masking a green run as a failure. vendor/bin/pest -d memory_limit=512M bypasses that. PHPUnit-only projects (feedz) don't have the wrapper-shutdown problem, so php artisan test --compact is fine.

Deploy phase (Phase 3)

Three apps, in order: admin → feedz → kanban.

  • admin — PostgreSQL backup → migrations → composer install --no-dev → frontend build → cache rebuild
  • feedz — shares admin's DB, no migrations → composer install --no-dev → frontend build → cache rebuild
  • kanban — MySQL backup → migrations → composer install --no-dev → frontend build → cache rebuild

DB backups are HARD gates: any non-zero pg_dump / mysqldump exit, or a snapshot smaller than 1KB, aborts that app's deploy before migrations run. Backups stay forever in db-backups/.

dev-master packages are forbidden in production — deploy.sh aborts if composer.lock ships any.

Hard rules

  • One profile, one server. No CWD detection, no fleet logic.
  • Tests are the first hard gate. Detect-first then run unmasked — never swallow pest's exit code with || echo.
  • Tag BEFORE deploy. Server pulls by tag, not by branch.
  • admin deploys first. PostgreSQL migrations land there; feedz reads the same DB.
  • No dev-master in production.
  • DB backups are HARD gates per app.
  • Don't touch kanban-scheduler's active.flag by hand. Package-managed under step-dispatcher v1.12+.
  • Cancel all scheduled polls at the end. Don't leak watcher loops between releases.

Error recovery

FailureAction
Tests failFix, re-run /do quanamo-release
Deploy fails on one appOthers may still be fine. /do quanamo-deploy admin|feedz|kanban for just the failing one
Migrations failedRestore from the backup written immediately before — see quanamo-deploy
Health degradedSystem is live but degraded. Investigate; don't roll back tags blindly