Deploy the server build to the path crate-api expects

[erikhfj]

Deployment gap raised by rachelks on #42: the CI workflow builds server/ and restarts crate-api.service, but nothing puts the built server where the service actually runs it.

The gap

crate-api.service on the VM has:

WorkingDirectory=/home/erikhfj/T19-Project-2/server
ExecStart=/home/erikhfj/.nvm/versions/node/v24.21.0/bin/node dist/index.js

The CI deploy job builds in the runner's own workspace, ~/actions-runner-p2/_work/T19-Project-2/T19-Project-2, and publishes only client/dist to /var/www/html/project2. The server build in that workspace never reached ~/T19-Project-2/server/dist, so restarting crate-api brought back whatever process was last installed by hand. That was an early prototype copy from 23 September, which is why search errors against the newly migrated database.

Progress

• ~/T19-Project-2 is a git clone of main, not a hand copy. The old hand copy is kept at ~/T19-Project-2-prototype-backup.
• The VM can fetch from git.ntnu.no. It had no SSH key, so ~/.ssh/id_ed25519_git was generated there and added to the account as key 3852.
• .env survives a deploy. It is gitignored, and git checkout --force was verified to leave it untouched (same md5 before and after).
• The deploy step checks out $GITHUB_SHA in ~/T19-Project-2, installs, builds server/, restarts crate-api, polls for 200, and runs a smoke suite through Apache. In feat: #44 deploy the server build to where crate-api runs it #45.
• pnpm build output for server/ lands where crate-api.service runs it. Verified by a dry run of the deploy step against the VM on 15 October.
• push to main deploys the server and restarts crate-api without a manual step. The commands are verified; the job invoking them on a real push is what the merge of feat: #44 deploy the server build to where crate-api runs it #45 will show.
• the smoke suite runs on the VM against port 3001 after deploy. Verified by the same dry run, plus the fuller suite in docs/VM-setup.md.
• a reboot of the VM leaves a working API. systemctl is-enabled crate-api is enabled, so it should. Not yet confirmed by an actual reboot.

#42 merged on 13 October and the skip guards became hard failures in #45, so the deploy no longer skips.

Things to get right

• EnvironmentFile=/home/erikhfj/T19-Project-2/.env must survive whatever happens to that directory. It holds DATABASE_URL.
• Poll, do not sleep. Type=simple reports active before Node binds the port. Measured at about 1850 ms; a fixed sleep 2 has already caused a 503 through Apache.
• The sudoers rule covers restart, is-active and status on crate-api. Anything else needs adding to /etc/sudoers.d/crate-deploy first.
• sudo needs NOPASSWD, because the runner has no TTY.

Related

• GraphQL Yoga server with searchAlbums #6 adds server/ and is what makes this worth automating
• Deploy the API on the group VM #8 recorded the gap in AGENTS.md
• feat: #6 GraphQL Yoga server with searchAlbums #42 is where rachelks spotted it
• feat: #44 deploy the server build to where crate-api runs it #45 is the deploy step