Deploy dccd on a server (run it unattended)¶

Run dccd start (scheduler + streams + web UI) 24/7 on a server, surviving crashes and reboots. Two blessed paths are below; systemd is recommended for a long-lived home/VPS box (no container overhead, journald integration, clean restart semantics). Use Docker for a containerised or ephemeral host.

Both were verified end to end on a real Ubuntu 24.04 server.

Note

Old CPUs without AVX2 (common on recycled home servers): the default polars wheel crashes the daemon at import with SIGILL. Check with grep -o -m1 avx2 /proc/cpuinfo — if it prints nothing, use the polars-lts-cpu variant (shown in both paths below).

Recommended: systemd¶

Run from a virtualenv as a dedicated dccd user. deploy/dccd.service ships with the repo.

# 1. Service user + venv
sudo useradd --system --no-create-home --shell /usr/sbin/nologin dccd
sudo python3 -m venv /opt/dccd/venv
sudo /opt/dccd/venv/bin/pip install "dccd[daemon]"
# CPU without AVX2 only:
#   sudo /opt/dccd/venv/bin/pip uninstall -y polars && \
#   sudo /opt/dccd/venv/bin/pip install polars-lts-cpu

# 2. Config at /etc/dccd/config.yml (XDG_CONFIG_HOME=/etc), readable by the
#    service group only. data_path must be under /var/lib/dccd.
sudo install -d -o root -g dccd -m 750 /etc/dccd
sudo install -o root -g dccd -m 640 config.yml /etc/dccd/config.yml

# 3. Install + enable the unit
sudo cp deploy/dccd.service /etc/systemd/system/dccd.service
sudo systemctl daemon-reload
sudo systemctl enable --now dccd

The unit runs ExecStart=/opt/dccd/venv/bin/dccd start as User=dccd with StateDirectory=dccd (systemd creates and owns /var/lib/dccd) and a hardened sandbox (ProtectSystem=strict, ProtectHome, PrivateTmp, NoNewPrivileges). Restart=on-failure brings it back after a crash, and because it is enabled it auto-starts after a reboot — verified by a real systemctl reboot: the daemon came back, the stream reconnected and the interval backfills re-armed with no data gap.

Check it:

systemctl status dccd
curl -fsS http://127.0.0.1:8080/health        # {"status":"ok"}
journalctl -u dccd -f                          # live logs

Operate it¶

Health: GET /health returns 200 when up. The systemd unit restarts on failure; the Docker image ships a HEALTHCHECK so an orchestrator sees healthy/unhealthy.
Logs & rotation: dccd writes no log files — logs go to the journal (journalctl -u dccd). Rotation is journald’s job (/etc/systemd/journald.conf: SystemMaxUse=, MaxRetentionSec=). Under Docker, the container engine owns rotation (--log-opt max-size=…).
Alerts: set alerts.webhook_url (and alerts.max_consecutive_errors) to POST a webhook when a job fails repeatedly. Verified: a failing job past the threshold delivers the alert.
Resource limits: the unit ships commented MemoryMax/CPUQuota/TasksMax — uncomment and tune for your host (too-low values can OOM-kill a busy daemon).
Secrets: keep the token out of the image/VCS — see Protect the web UI with a token.
Off-box backups: mirror the store with rclone — see Sync data to a remote (S3, GCS, …).

Alternative: Docker¶

# Build (add the build-arg only for a no-AVX2 host):
docker build -t dccd .
#   docker build --build-arg POLARS_VARIANT=polars-lts-cpu -t dccd .

# Run: mount the config (kept out of the image) + a named data volume.
docker run -d --name dccd --restart unless-stopped -p 8080:8080 \
    -v "$PWD/config.yml:/etc/dccd/config.yml:ro" \
    -v dccd-data:/data \
    dccd

The image is built on a digest-pinned python:3.12-slim, sets XDG_CONFIG_HOME=/etc (config at /etc/dccd/config.yml), exposes 8080, and declares a HEALTHCHECK against /health. Set settings.data_path: /data and settings.ui_host: 0.0.0.0 in the mounted config. Check docker inspect --format '{{.State.Health.Status}}' dccd → healthy.

Reaching the UI from another machine¶

The default bind is 127.0.0.1. To reach the UI remotely, set ui_host: 0.0.0.0 and a token (Protect the web UI with a token), and put it behind a private network (e.g. Tailscale) or a TLS reverse proxy — never expose the API plaintext on the public internet.