Saltar al contenido principal

Set Up a Multi-Node Cluster

CapRover uses Docker Swarm. The manager runs CapRover's control services; workers run app tasks.

Before adding a node: Prepare Docker, private networking, SSH, and a shared registry.

1. Prepare server​

On the new Ubuntu server, install Docker Engine:

sudo apt update
sudo apt install ca-certificates curl
sudo install -m 0755 -d /etc/apt/keyrings
sudo curl -fsSL https://download.docker.com/linux/ubuntu/gpg -o /etc/apt/keyrings/docker.asc
sudo chmod a+r /etc/apt/keyrings/docker.asc
sudo tee /etc/apt/sources.list.d/docker.sources <<EOF
Types: deb
URIs: https://download.docker.com/linux/ubuntu
Suites: $(. /etc/os-release && echo "${UBUNTU_CODENAME:-$VERSION_CODENAME}")
Components: stable
Architectures: $(dpkg --print-architecture)
Signed-By: /etc/apt/keyrings/docker.asc
EOF
sudo apt update
sudo apt install docker-ce docker-ce-cli containerd.io docker-buildx-plugin docker-compose-plugin
sudo docker version
sudo docker info --format '{{.Swarm.LocalNodeState}}'
  • Use a stable IP reachable from the manager.
  • Docker's Swarm state should be inactive.
  • Choose worker for ordinary app hosts. Manager roles require quorum planning.

2. Networking and SSH​

Between trusted nodes only:

  • 2377/tcp: Swarm management
  • 7946/tcp and 7946/udp: Node communication
  • 4789/udp: Overlay networking

SSH:

  1. Use a dedicated private key and a user with Docker access (usually root).
  2. Add the public key to the new server's authorized_keys.
  3. Test SSH from the manager's network to the new server.
  4. Give CapRover the private key during node setup.

Docker group membership gives root-equivalent access. Keep both SSH and Swarm ports restricted to trusted hosts. Review UFW rules CapRover adds during joining.

3. Registry​

  1. Configure a default push registry reachable by every node. CapRover requires it before adding nodes.
  2. Redeploy apps built before the registry was enabled to push their images.
  3. Verify the new node can pull a newly built image. Prebuilt private images also need pull access.

4. Add and verify node​

  1. In Cluster → Nodes, enter manager and worker IPs, SSH host/port/user, private key, and node role.
  2. Join the node. Verify it appears healthy and check docker node ls on the manager.
  3. Scale a stateless test app and check placement with docker service ps SERVICE_NAME.
  4. Test the app domain. If placement fails, inspect registry pulls, node health, and task errors.

5. Persistent apps​

  • Local volumes and bind mounts stay on their node.
  • CapRover pins persistent apps; joining a node does not copy data.
  • Back up and migrate data before changing placement.
  • See Persistent Applications in Clusters.

Manage Nodes​

Use the Cluster → Nodes view and docker node ls on a manager to check node availability and role. docker service ps SERVICE_NAME shows where an app's tasks run and why a task failed to start.

  • Before removing or draining a node, identify its running apps and any local volumes or bind mounts.

  • Stateless tasks can usually be rescheduled if images are available from the registry; persistent data does not follow them automatically.

  • Back up and move data deliberately before changing placement.

  • Keep Swarm management and overlay ports restricted to trusted node addresses.

  • If a node cannot join, verify Docker is installed, its Swarm is inactive, SSH credentials work from the manager, and the registry and inter-node routes are reachable.

Persistent Applications in Clusters​

  • CapRover normally limits a persistent app to one instance on a selected node.

  • A Docker named volume using the default local driver and a bind mount both store data on that node; a second node will not automatically see those files.

  • To move a persistent app, back up and quiesce writes, restore data on the destination, check permissions, update placement, then verify the app with one instance.

  • Keep the old copy until the new instance is proven.

  • A node failure can make a pinned app unavailable until its data is restored elsewhere.

  • Shared or external storage is an advanced design that must support the application's consistency and concurrent-write requirements.

  • Increasing instance count alone does not provide that storage.

  • See Shared / External Storage.

Multi-Node Networking​

  • Docker Swarm's overlay network lets services communicate across nodes.

  • Between trusted nodes, permit 2377/tcp for management, 7946/tcp and 7946/udp for node communication, and 4789/udp for overlay traffic.

  • Restrict these ports to node addresses; public web traffic still reaches CapRover's HTTP/HTTPS entry point.

  • A public app port in ingress mode can use the Swarm routing mesh.

  • In host mode it is published where the task runs, so placement affects which node's address answers.

  • For internal database traffic, use the destination app's service name and container port without a public mapping.

If a new node joins but tasks cannot communicate, check inter-node firewalls, Docker service state, and network membership before changing app settings. See the full port reference.