run the dev server as a system-wide systemd unit (boots without login/linger)

This commit is contained in:
Martin Dvořák 2026-09-29 15:46:31 +00:00
commit bfa0840b0e
4 changed files with 119 additions and 20 deletions

View file

@ -114,9 +114,24 @@ site. To force the raw server instead, use `npm run dev:next` (don't point a tun
### Running it as a service (survives reboot) ### Running it as a service (survives reboot)
On this host the dev server runs as a **user** systemd unit, so it comes back after a On this host the dev server runs as a **system** systemd unit, so it starts at boot on
reboot without a login (linger is enabled). The unit lives in the repo at its own — no login or linger needed. The unit lives in the repo at
[`deploy/dvorakmortonxd-dev.service`](./deploy/dvorakmortonxd-dev.service): [`deploy/dvorakmortonxd-dev.system.service`](./deploy/dvorakmortonxd-dev.system.service):
```bash
# installs /etc/systemd/system/dvorakmortonxd-dev.service and enables it
sudo bash deploy/install-system-service.sh # or: doas bash deploy/install-system-service.sh
```
Manage it with `systemctl status|restart|stop dvorakmortonxd-dev` and follow logs with
`journalctl -u dvorakmortonxd-dev -f`. It shares `.next/dev/lock` with a manual
`npm run dev`, so stop the service before starting one by hand.
<details>
<summary>No root? Use the per-user unit instead</summary>
[`deploy/dvorakmortonxd-dev.service`](./deploy/dvorakmortonxd-dev.service) is an equivalent
**user** unit you can install without root:
```bash ```bash
mkdir -p ~/.config/systemd/user mkdir -p ~/.config/systemd/user
@ -126,9 +141,10 @@ systemctl --user enable --now dvorakmortonxd-dev.service
loginctl enable-linger "$USER" # once, so it starts at boot loginctl enable-linger "$USER" # once, so it starts at boot
``` ```
Manage it with `systemctl --user status|restart|stop dvorakmortonxd-dev` and follow logs with Drive it with `systemctl --user …` and `journalctl --user -u …`. Run only one of the two —
`journalctl --user -u dvorakmortonxd-dev -f`. It shares `.next/dev/lock` with a manual they contend for `:3000` and `.next/dev/lock`.
`npm run dev`, so stop the service before starting one by hand.
</details>
## 📡 API Routes ## 📡 API Routes
@ -172,7 +188,9 @@ src/app/
scripts/ scripts/
dev-server.mjs → custom dev server: live updates without the dev-only debug surface dev-server.mjs → custom dev server: live updates without the dev-only debug surface
deploy/ deploy/
dvorakmortonxd-dev.service → systemd user unit so the dev server starts at boot dvorakmortonxd-dev.system.service → systemd unit (system-wide) so the dev server starts at boot
dvorakmortonxd-dev.service → equivalent per-user unit (no root; needs linger)
install-system-service.sh → installs the system unit (run with sudo/doas)
public/ → fonts, wallpapers, icons, friend pages, /revival-inc game, llms.txt, robots.txt, sitemap.xml public/ → fonts, wallpapers, icons, friend pages, /revival-inc game, llms.txt, robots.txt, sitemap.xml
``` ```

View file

@ -249,15 +249,18 @@ list reordering, not free placement. Don't confuse the two.
`localhost:3000` keeps the full dev experience; the LAN/tunnel get a plain, `localhost:3000` keeps the full dev experience; the LAN/tunnel get a plain,
live-reloading site. Use `npm run dev:next` only if you explicitly want the live-reloading site. Use `npm run dev:next` only if you explicitly want the
raw, unprotected server (never point the tunnel at it). raw, unprotected server (never point the tunnel at it).
- The dev server normally runs on this host as a **user** systemd unit - The dev server runs on this host as a **system** systemd unit
(`~/.config/systemd/user/dvorakmortonxd-dev.service`, kept in-repo at (`/etc/systemd/system/dvorakmortonxd-dev.service`, kept in-repo at
`deploy/dvorakmortonxd-dev.service`). `Linger=yes` for uid 1000 means it starts `deploy/dvorakmortonxd-dev.system.service`, installed by
at boot without a login. `systemctl --user restart dvorakmortonxd-dev` reloads `deploy/install-system-service.sh`). It is enabled at `multi-user.target`, so
it; logs with `journalctl --user -u dvorakmortonxd-dev -f`. From a bare shell it starts at boot with no login/linger. `systemctl restart dvorakmortonxd-dev`
with no session bus, prefix those with reloads it; logs with `journalctl -u dvorakmortonxd-dev -f`. The unit pins
`XDG_RUNTIME_DIR=/run/user/1000 DBUS_SESSION_BUS_ADDRESS=unix:path=/run/user/1000/bus`. `HOSTNAME=0.0.0.0` (the script reads that env var) so the tunnel can reach
The unit pins `HOSTNAME=0.0.0.0` (the script reads that env var) so the tunnel :3000, clears a stale `.next/dev/lock` on start, and sets `HOME` because system
can reach :3000, and clears a stale `.next/dev/lock` on start. units don't inherit a login environment. A per-user twin
(`deploy/dvorakmortonxd-dev.service`) exists for rootless installs; it needs
`loginctl enable-linger` and is driven via `systemctl --user`. Only one of the
two may run at a time — they contend for :3000 and `.next/dev/lock`.
- `allowedDevOrigins` (`next.config.ts`) still matters: the tunnel host and any - `allowedDevOrigins` (`next.config.ts`) still matters: the tunnel host and any
LAN origin must be listed or Next rejects the HMR WebSocket with a 403 (the LAN origin must be listed or Next rejects the HMR WebSocket with a 403 (the
client then never hydrates — SSR only, icons stacked at `(0,0)`, dead client then never hydrates — SSR only, icons stacked at `(0,0)`, dead
@ -309,10 +312,10 @@ list reordering, not free placement. Don't confuse the two.
- There is **no test suite**. Verification is `tsc --noEmit`, `eslint`, and - There is **no test suite**. Verification is `tsc --noEmit`, `eslint`, and
`next build`. `next build`.
- A dev server may already be running on port 3000 — on this host it is normally - A dev server may already be running on port 3000 — on this host it is normally
the `dvorakmortonxd-dev` user systemd service, so check the **system** `dvorakmortonxd-dev` service, so check
`systemctl --user status dvorakmortonxd-dev` first. Two dev servers on this `systemctl status dvorakmortonxd-dev` first. Two dev servers on this repo
repo collide on `.next/dev/lock`; stop the service (or the old process) collide on `.next/dev/lock`; stop the service (or the old process) before
before starting the custom server by hand. starting the custom server by hand.
- Dev-server guard check: `curl -s -o /dev/null -w '%{http_code}' \ - Dev-server guard check: `curl -s -o /dev/null -w '%{http_code}' \
http://localhost:3000/__nextjs_server_status` must be `200` (localhost keeps http://localhost:3000/__nextjs_server_status` must be `200` (localhost keeps
debug), while the same path through the LAN IP or `beta.dvorakmortonxd.dev` debug), while the same path through the LAN IP or `beta.dvorakmortonxd.dev`

View file

@ -0,0 +1,31 @@
[Unit]
Description=dvorakmortonxd.dev dev server (custom Next.js dev wrapper)
Documentation=file:///home/martin/dvorakmortonxd.dev/README.md
After=network-online.target
Wants=network-online.target
[Service]
Type=simple
User=martin
Group=martin
WorkingDirectory=/home/martin/dvorakmortonxd.dev
# A system service does not inherit a login environment, so set HOME/PATH
# explicitly (Next.js wants a writable HOME for its cache).
Environment=HOME=/home/martin
Environment=PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
# Bind all interfaces so the Cloudflare tunnel host can reach :3000. The dev
# server reads process.env.HOSTNAME, so pin it rather than inheriting it.
Environment=HOSTNAME=0.0.0.0
Environment=PORT=3000
# A reboot can leave a stale dev lock behind; clear it before starting.
ExecStartPre=/bin/rm -f /home/martin/dvorakmortonxd.dev/.next/dev/lock
ExecStart=/bin/node /home/martin/dvorakmortonxd.dev/scripts/dev-server.mjs
Restart=always
RestartSec=3
# Give Next.js a moment to shut down cleanly.
TimeoutStopSec=15
StandardOutput=journal
StandardError=journal
[Install]
WantedBy=multi-user.target

View file

@ -0,0 +1,47 @@
#!/usr/bin/env bash
# Install the dvorakmortonxd.dev dev server as a SYSTEM systemd unit so it
# starts at boot for the whole machine (no login / no linger needed).
#
# Run it with root, e.g.:
# doas bash deploy/install-system-service.sh
#
# It also retires the per-user unit, so the two don't fight over :3000 and
# .next/dev/lock.
set -euo pipefail
if [ "$(id -u)" -ne 0 ]; then
echo "root required — run: doas bash \"$0\"" >&2
exit 1
fi
here="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
unit_src="$here/dvorakmortonxd-dev.system.service"
unit_dst="/etc/systemd/system/dvorakmortonxd-dev.service"
user_unit="/home/martin/.config/systemd/user/dvorakmortonxd-dev.service"
install -m 0644 "$unit_src" "$unit_dst"
systemctl daemon-reload
# Retire the per-user copy if it exists (installed earlier as martin).
if [ -e "$user_unit" ] || [ -e "/home/martin/.config/systemd/user/default.target.wants/dvorakmortonxd-dev.service" ]; then
if command -v runuser >/dev/null 2>&1; then
runuser -u martin -- env XDG_RUNTIME_DIR=/run/user/1000 \
DBUS_SESSION_BUS_ADDRESS=unix:path=/run/user/1000/bus \
systemctl --user disable --now dvorakmortonxd-dev.service || true
else
# runuser is unavailable (minimal container). The caller should have run
# `systemctl --user stop dvorakmortonxd-dev` first; we still remove the files
# so the two units can't fight over :3000 on the next boot.
echo "note: runuser not found — stop the per-user unit manually if it is still running" >&2
fi
rm -f "$user_unit" \
/home/martin/.config/systemd/user/default.target.wants/dvorakmortonxd-dev.service
fi
systemctl enable --now dvorakmortonxd-dev.service
systemctl --no-pager --full status dvorakmortonxd-dev.service || true
echo
echo "Installed. Manage it with:"
echo " systemctl status|restart|stop dvorakmortonxd-dev"
echo " journalctl -u dvorakmortonxd-dev -f"