High Availability & Multi-Server Clustering

Scale your application across multiple servers with rolling zero-downtime updates, cross-host load balancing, and private network routing.

---

Multi-Server Architecture

When deploying to multiple hosts, epd can either route traffic locally on each server or establish a distributed ingress mesh across all nodes.

                  DNS (Round Robin or Cloudflare)
                   ┌──────────────┴──────────────┐
                   │                             │
                   ▼                             ▼
         ┌───────────────────┐         ┌───────────────────┐
         │ Server 1 (Host A) │         │ Server 2 (Host B) │
         │ Traefik Proxy     │◄───────►│ Traefik Proxy     │
         │ Replicas (Slot G) │ Private │ Replicas (Slot G) │
         └───────────────────┘ Network └───────────────────┘

---

1. Cross-Host Routing (proxy.cross_host: true)

By default, Traefik on Server A balances requests only among containers running on Server A.

With proxy.cross_host: true:

• Each server's Traefik instance includes upstream replica targets across the entire cluster.
• If Server B experiences hardware failure or is undergoing maintenance, requests arriving at Server B's IP address can be transparently forwarded to healthy containers on Server A.

name: my-app
mode: docker
image: ghcr.io/my-org/my-app

proxy:
  cross_host: true
  # Bind internal service traffic to private VPC network IPs
  private_ips:
    203.0.113.10: 10.0.0.10
    203.0.113.11: 10.0.0.11
    203.0.113.12: 10.0.0.12

servers:
  web:
    hosts:
      - 203.0.113.10
      - 203.0.113.11
      - 203.0.113.12
    replicas: 2
    port: 3000
    domains: [app.example.com]

---

2. Deployment Strategies: Rolling vs. Parallel

When deploying updates to a multi-server cluster, epd controls the rollout strategy:

Rolling Deployment (Default)

strategy: rolling

In a rolling deployment:

1. epd deploys to Host 1 and waits until health probes pass.
2. Once Host 1 is live and serving new traffic, it moves to Host 2.
3. This sequence repeats until the entire fleet is updated.
4. If a failure occurs on Host 2, Host 1 is already healthy, and the deployment stops safely before breaking the remaining hosts.

Parallel Deployment

strategy: parallel

Deploys to all configured hosts concurrently. Recommended for staging environments or non-critical worker clusters where deploy speed is prioritized.

---

Next Steps

• Set up Automated Multi-Provider DNS Synchronization
• Configure Advanced Routing Rules
• Check out the Complete Configuration Schema