Run with Docker2.x

This sets up a new 2.x install. Coming from 1.x? Follow Upgrade from 1.x instead, so your library and history carry over.

Compose file

services:
  infinidysk:
    image: ghcr.io/infinidysk/infinidysk-next:edge
    container_name: infinidysk
    restart: unless-stopped
    # pid: host # optional: name the processes that read the mount, see Logs and traces
    environment:
      PUID: 1000
      PGID: 1000
      TZ: Etc/UTC # your time zone, e.g. Europe/Oslo; the maintenance window follows it
    ports:
      - "3000:3000"
    volumes:
      - ./config:/config
      - /mnt:/mnt:rshared
      - /sys/devices/virtual/bdi:/sys/devices/virtual/bdi:rw
    devices:
      - /dev/fuse:/dev/fuse
    cap_add:
      - SYS_ADMIN
    security_opt:
      - apparmor:unconfined

edge follows the latest 2.x build. There is no latest tag until 2.0 ships.

First start

1. Run docker compose up -d.
2. Open http://<host>:3000 and create the admin account.
3. Add your providers. See Providers.

Caution

Create the admin account before you expose port 3000 to an untrusted network. Until an admin exists, anyone who reaches the page can create one.

Where things are

• Port 3000 serves the UI, the SABnzbd API and WebDAV.
• The library is mounted at /mnt/infinidysk by default. Change it under Settings > Platform > Mount, or with INFINIDYSK__MOUNT__DIR.
• The database is /config/infinidysk.sqlite. Set INFINIDYSK_DATABASE_URL to a postgres://… URL to use PostgreSQL instead.
• Logs go to /config/logs/infinidysk.log and the System page.

The full list is in Ports, paths and mount layout.

FUSE in Docker

• /dev/fuse, SYS_ADMIN and apparmor:unconfined let the container mount. Without /dev/fuse, 2.x logs a warning and serves WebDAV only.
• :rshared on the parent directory makes the mount visible on the host and in other containers. They bind it with :rslave; see Connect your media server.
• The /sys/devices/virtual/bdi volume lets the preset under Settings > Platform > Performance set the kernel’s readahead. Docker mounts /sys read-only, so without it readahead stays at the kernel’s 128 KB.

Health check

The image has a built-in health check, so the compose file needs no healthcheck: block. docker ps shows the container as starting, healthy or unhealthy.

Healthy means all of these hold:

• The server answers on its port.
• The database answers.
• The import worker runs.
• The mount is mounted and answers. Without /dev/fuse the mount is not checked.

A normal start turns healthy within seconds. The first start after an upgrade from 1.x stays starting until the import is done. If the import takes longer than the 10-minute start period, the container shows unhealthy until it ends.

The check runs inside the container, so it cannot see whether :rshared carries the mount to the host. Check that once from your media server; see Connect your media server.

Wait for the mount

Containers that read the mount can wait until infinidysk is healthy:

services:
  plex:
    depends_on:
      infinidysk:
        condition: service_healthy

docker compose up then starts the media server after the mount is up. If infinidysk turns unhealthy first, for example during a long 1.x import, Compose does not start the media server. Run docker compose up -d again once infinidysk is healthy.

Change the timing

The built-in check runs every 30 seconds and marks the container unhealthy after 3 failures in a row. Override it in the compose file:

services:
  infinidysk:
    healthcheck:
      interval: 15s
      retries: 5

What it asks

The check runs infinidysk healthcheck, which asks GET /healthz on the address in INFINIDYSK_LISTEN and exits 0 when healthy, 1 otherwise. It needs no curl and never reads the mount itself, so a stuck mount cannot hang it.

/healthz answers without sign-in and shows no settings or secrets. It answers 200 when healthy and 503 while starting or unhealthy:

{"status":"healthy","checks":{"database":"pass","mount":"pass","queue":"pass"}}

Each check is pass, fail, pending (not ready yet) or off (does not apply, for example mount without /dev/fuse). During the first start after an upgrade, the body has one import check instead. The reason a check fails is in the log, not in the answer.