b728696020
Removes the rsync --delete / manual-migration / no-rollback footguns from the deploy path. Validated: clean deploy passes the health gate; a forced health failure correctly rolls back from the .prev snapshot and recovers. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
143 lines
5.3 KiB
Markdown
143 lines
5.3 KiB
Markdown
# Deploy notes — Void 2.0
|
||
|
||
## DB role posture (CT 310 — `void2-db`, alpha-9+)
|
||
- The `void` DB role is **NOSUPERUSER** (least privilege). It owns the `void` + `void_test`
|
||
databases and the `public` schema, so it can run all migrations and the test-harness
|
||
`resetDb` without superuser.
|
||
- The `vector` (pgvector) extension was marked **trusted** so the non-superuser `void` role
|
||
can `CREATE EXTENSION vector` (needed by `tests/helpers/db.js` on each reset):
|
||
```
|
||
echo 'trusted = true' >> /usr/share/postgresql/16/extension/vector.control
|
||
```
|
||
**⚠ Re-apply this after any pgvector package upgrade** (the package may overwrite the
|
||
control file). `pgcrypto` ships trusted already.
|
||
- Revert (emergency): as `postgres` on CT 310, `ALTER ROLE void SUPERUSER;`.
|
||
|
||
## App deploy (CT 311 — `void2-app`)
|
||
|
||
One-time setup on the target host:
|
||
|
||
```bash
|
||
# Node 22 (from nodesource if Debian's default is older)
|
||
curl -fsSL https://deb.nodesource.com/setup_22.x | bash -
|
||
apt install -y nodejs
|
||
|
||
# Service user + working dir
|
||
useradd -r -m -d /opt/void-server void
|
||
mkdir -p /opt/void-server
|
||
chown void: /opt/void-server
|
||
|
||
# systemd
|
||
install -m 644 void-server.service /etc/systemd/system/void-server.service
|
||
systemctl daemon-reload
|
||
systemctl enable void-server
|
||
|
||
# Secrets — /opt/void-server/.env must contain:
|
||
# DATABASE_URL=postgres://void:<password>@<db-host>:5432/void
|
||
# OWNER_TOKEN=<32+ char secret>
|
||
# PORT=3000
|
||
# NODE_ENV=production
|
||
chmod 600 /opt/void-server/.env
|
||
chown void: /opt/void-server/.env
|
||
```
|
||
|
||
Then from the dev box:
|
||
|
||
```bash
|
||
cd /project/src/void-v2
|
||
./deploy/push.sh
|
||
```
|
||
|
||
## Maintenance
|
||
|
||
```bash
|
||
journalctl -u void-server -f # follow logs
|
||
systemctl status void-server # check status
|
||
systemctl restart void-server # cycle
|
||
|
||
# Run migrations on the deployed copy:
|
||
ssh root@void2-app 'cd /opt/void-server && npm run migrate'
|
||
```
|
||
|
||
## Notes
|
||
|
||
- `.env` is excluded from the rsync to avoid clobbering production secrets with dev values.
|
||
- The push script uses `--omit=dev` to skip test deps on the target.
|
||
- `tests/` is excluded — they're for the dev environment only.
|
||
|
||
## Workers (Python void-workers — Plan 4+)
|
||
|
||
Runs alongside void-server as a second systemd unit.
|
||
|
||
One-time setup on CT 311:
|
||
|
||
```bash
|
||
apt install -y python3.12 python3.12-venv python3-pip \
|
||
ffmpeg tesseract-ocr tesseract-ocr-eng poppler-utils
|
||
|
||
useradd -r -m -d /opt/void-workers -s /bin/bash voidworkers
|
||
mkdir -p /opt/void-workers /var/lib/void/whisper-models
|
||
chown voidworkers: /opt/void-workers
|
||
chown -R voidworkers: /var/lib/void/whisper-models
|
||
|
||
# voidworkers needs to read the shared blob store
|
||
usermod -aG void voidworkers
|
||
chmod -R g+rX /var/lib/void/blobs
|
||
|
||
install -m 644 deploy/void-workers.service /etc/systemd/system/
|
||
systemctl daemon-reload
|
||
systemctl enable void-workers
|
||
```
|
||
|
||
`/opt/void-workers/.env` (mode 600, owned by voidworkers):
|
||
|
||
```
|
||
DATABASE_URL=postgres://void:<pw>@192.168.1.215:5432/void
|
||
BLOB_ROOT=/var/lib/void/blobs
|
||
WHISPER_MODEL=small.en
|
||
WHISPER_CACHE=/var/lib/void/whisper-models
|
||
```
|
||
|
||
Deploy after edits:
|
||
|
||
```bash
|
||
cd /project/src/void-v2
|
||
./deploy/push-workers.sh
|
||
```
|
||
|
||
## SQL_ASCII cluster note
|
||
|
||
`void2-db` was initialized as SQL_ASCII (not UTF-8). The data is already
|
||
UTF-8 in practice but Python's psycopg refuses to decode without an
|
||
explicit `client_encoding=UTF8` parameter. Workers set this on every
|
||
connection (`lib/db/pool.py` equivalent in `workers/void_workers/`).
|
||
Node's `pg` lib is more lenient and doesn't need this. If you ever
|
||
re-initdb the cluster, use `--encoding=UTF8 --locale=C.UTF-8`.
|
||
|
||
## Plan 6 (alpha-8)
|
||
|
||
- **Migrations 012–014** (dashboard_layout, speedtest_results, service_status) are applied by the standard `npm run migrate` — no manual steps needed.
|
||
- **`speedtest-cli` on CT 311** — the hourly speedtest job requires it:
|
||
```bash
|
||
pip install --break-system-packages speedtest-cli
|
||
```
|
||
Until installed, speedtest jobs will fail but the Sacred Valley speedtest card still renders any existing history without error.
|
||
- **Icon cache** — the server writes cached service icons to `ICON_CACHE` (default `/var/lib/void/icons`) and auto-creates the directory on first use. You can pre-create and own it for clarity:
|
||
```bash
|
||
mkdir -p /var/lib/void/icons
|
||
chown void: /var/lib/void/icons
|
||
```
|
||
- **Service registry** — edit `config/services.json` to the real homelab service URLs and CT numbers. The committed seed values are best-guess placeholders and should be updated before the health band is meaningful.
|
||
|
||
## Deploy safety (push.sh, hardened)
|
||
`./deploy/push.sh` now does an atomic-ish, self-verifying deploy:
|
||
1. **Snapshots** the current remote code (excl `node_modules`/`.env`) to `/opt/void-server.prev` for rollback.
|
||
2. rsyncs the new code (`--delete`; preserves `node_modules` + `.env`).
|
||
3. Runs **`npm install --omit=dev` + `npm run migrate`** as part of the deploy (no more separate manual migrate step).
|
||
4. Restarts `void-server`.
|
||
5. **Health-gates**: polls `/health` until it reports the expected `package.json` version + `db_ok` (≈25s).
|
||
6. **Auto-rolls-back** on any failure: restores the `.prev` snapshot, reinstalls, restarts.
|
||
|
||
Override the health endpoint with `HEALTH_URL=…` if the target IP differs.
|
||
Caveat: forward-only migrations are not auto-reverted on rollback (they're additive by convention, so a code rollback against the new schema is safe; a destructive migration needs manual care).
|