Introduction to Modern High Availability Web Routing

In the contemporary enterprise landscape, system downtime directly translates to financial loss and reputational damage. As applications migrate toward decentralized architecture, the edge routing layer—the entry point for all client requests—must be resilient, performant, and horizontally scalable. While traditional solutions like Nginx or HAProxy are widely deployed, Caddy Server has rapidly emerged as a powerful alternative due to its memory-safe Go footprint, native HTTP/3 support, and revolutionary automatic SSL/TLS lifecycle management.

However, when scaling Caddy across multiple Virtual Private Servers (VPS) behind a layer 4 load balancer, a critical architectural challenge emerges: how do ephemeral, independent server instances share and coordinate SSL certificates without hitting Let's Encrypt rate limits or causing race conditions during ACME challenges?

The definitive solution lies in a High Availability (HA) Caddy Architecture backed by a centralized Redis storage provider. This technical blueprint explores how to decouple certificate storage from the local filesystem, enabling seamless cluster-wide synchronization, automated renewal coordination, and true horizontal scalability across multiple VPS environments.

The Core Challenge: ACME and Horizontal Scaling

Caddy’s standout feature is its ability to automatically provision, validate, renew, and staple TLS certificates using ACME providers like Let's Encrypt and ZeroSSL. By default, Caddy writes these cryptographic assets to the local disk. In a multi-server setup, this default behavior introduces severe architectural bottlenecks:

Rate Limiting Thresholds: If three independent VPS instances behind a round-robin load balancer attempt to serve the same domain, each will trigger individual ACME challenges. This rapidly exhausts the strict weekly rate limits imposed by public Certificate Authorities (CAs).
Validation Race Conditions: During HTTP-01 or TLS-ALPN-01 challenges, the CA sends a token to verification endpoints. If the load balancer routes the CA's request to VPS-B while VPS-A holds the generated validation token on its local disk, the challenge fails, breaking automatic SSL provisioning.
State Inconsistency: A client connecting to VPS-A might receive a freshly renewed certificate, while a subsequent connection routed to VPS-B encounters an old or mismatched session ticket, compromising the user experience and security compliance.

To overcome this, we must transition from local file storage to a distributed, shared-state storage mechanism using Redis.

Architectural Overview: Caddy, Redis, and Layer 4 Load Balancing

A resilient HA Caddy topology relies on three distinct layers, ensuring that traffic distribution, SSL management, and state storage are completely decoupled.

"True high availability is achieved only when state is externalized from the computing node. By treating edge routers as stateless workers, we can scale fluidly based on incoming traffic demands."

The layout consists of the following components:

Layer 4 Load Balancer (TCP Mode): Positioned at the edge, this load balancer routes raw TCP packets on ports 80 and 443 directly to the backend VPS cluster. It does not terminate TLS; it merely acts as a high-throughput traffic distributor.
Stateless Caddy Cluster (Multiple VPS): A minimum of two or three VPS nodes running identical Caddy binaries compiled with the caddy-dns or caddy-storage-redis plugin. These nodes ingest the inbound TCP traffic and terminate the TLS connections.
Centralized Redis Layer: A secure, high-performance Redis instance (or a Redis Sentinel cluster) that serves as the single source of truth for all ACME certificates, private keys, and operational lock files.

Step-by-Step Implementation Guide

Step 1: Compiling Caddy with the Redis Storage Module

Standard distributions of Caddy do not include third-party storage backends. To support Redis, we must build a custom Caddy binary using xcaddy, Caddy's official command-line build tool. Execute the following commands on your development or deployment environment:

# Install xcaddy via package manager or download the binary
sudo apt install -y debian-keyring debian-archive-keyring apt-transport-https
curl -1sLf 'https://dl.cloudsmith.io/public/caddy/xcaddy/cfg/gpg/gpg.1046445B6D443CB1.key' | sudo gpg --dearmor -o /usr/share/keyrings/caddy-xcaddy-archive-keyring.gpg
curl -1sLf 'https://dl.cloudsmith.io/public/caddy/xcaddy/cfg/setup/config.deb.txt?arch=amd64' | sudo tee /etc/apt/sources.list.y/caddy-xcaddy.list
sudo apt update
sudo apt install xcaddy

# Compile Caddy with the Redis storage plugin
xcaddy build --with github.com/gamalan/caddy-tlsredis

Once the compilation completes, distribute this custom binary to the /usr/bin/caddy directory across all target VPS instances in your cluster.

Step 2: Provisioning the Centralized Redis Instance

For optimal reliability, your Redis node should be accessible over a secure, low-latency private network interface common to all VPS nodes. Ensure your Redis configuration (/etc/redis/redis.conf) enforces strong authentication and limits access to trusted internal IP addresses:

# Bind to localhost and the private network IP
bind 127.0.0.1 10.130.0.5

# Require complex authentication tokens
requirepass YourExtremelySecureAndComplexPasswordHere

# Optimize for persistence and high-availability consistency
appendonly yes
appendfsync everysec

Step 3: Configuring the Distributed Caddyfile

With the customized binaries and central database operational, we configure the uniform Caddyfile block. This layout must be mirrored exactly across every active VPS worker node. Open /etc/caddy/Caddyfile and define the global option block:

{
    # Initialize distributed Redis storage backend
    storage redis {
        host     "10.130.0.5"
        port     6379
        password "YourExtremelySecureAndComplexPasswordHere"
        db       0
        timeout  5
        key_prefix "caddy_ssl_cluster:"
    }

    # Optimal global options for cloud environments
    email admin@yourdomain.com
    acme_ca https://acme-v02.api.letsencrypt.org/directory
}

# Define Application Routing Block
app.yourdomain.com {
    # Reverse proxy upstream application payload
    reverse_proxy 10.130.0.50:8080 {
        health_uri /healthz
        health_interval 10s
        health_timeout 5s
    }

    # Enable compression and standard security profiling
    encode gzip zstd
    header {
        Strict-Transport-Security "max-age=63072000; includeSubDomains; preload"
        X-XSS-Protection "1; mode=block"
        X-Content-Type-Options "nosniff"
    }
}

Behind the Scenes: Distributed Locking and Consensus

Understanding how Caddy coordinates renewals over Redis illustrates the elegance of this architecture. When an ACME certificate approaches its expiration date, Caddy nodes query the centralized Redis storage block using a deterministic key structure prefixed by caddy_ssl_cluster:.

Instead of multiple nodes simultaneously requesting a certificate, the primary node attempting the operation acquires a distributed lock inside Redis with a strict Time-To-Live (TTL). If VPS-A successfully sets the lock key, it proceeds with the ACME validation process. Meanwhile, VPS-B and VPS-C observe the active lock, step back, and monitor the storage cluster. Once VPS-A finishes the verification, it uploads the generated certificate and key array to Redis and releases the lock. The remaining cluster nodes instantly fetch the updated cryptographic assets directly from memory, maintaining seamless sync without a single duplicated external API call.

Monitoring, Maintenance, and Security Hardening

To guarantee enterprise-grade uptime, implementation is only half the battle; continuous observation is imperative. Consider these production optimization vectors:

Prometheus Metrics Extraction: Enable Caddy’s native telemetry metrics endpoints globally. Scraping internal processing latency, connection limits, and certificates status profiles allows DevOps teams to map real-time performance inside Grafana dashboards.
Failover Testing (Chaos Engineering): Periodically drop connections to a single VPS worker. The layer 4 load balancer should instantly exclude the node, while remaining nodes process ongoing connections without prompting TLS negotiation errors.
Redis Clustering: For high-density microservices handling thousands of concurrent domain validations, upgrade from standalone Redis instances to a fully managed Redis Sentinel deployment to eliminate the storage engine as a single point of failure.

Conclusion

Decoupling Caddy’s certificate storage from individual server nodes and centralizing it within an optimized Redis instance unlocks elite horizontal scalability. This High Availability layout effectively solves ACME validation limits, accelerates SSL termination across diverse regions, and creates a highly resilient routing tier capable of handling unpredictable traffic surges with zero manual intervention. Transitioning away from local filesystems toward distributed, in-memory state synchronization provides an engineering foundation that keeps applications fast, secure, and permanently online.

High Availability Architecture for Caddy Server: Implementing Cluster-Wide SSL Synchronization via Redis Across Multiple VPS Instances