Coolify v4 and Hetzner EX131: Scaling Self-hosted PaaS

TL;DR: The April 2026 deprecation of Docker Swarm in Coolify v4 marks a strategic pivot towards proprietary multi-server orchestration. Combined with Hetzner’s price-adjusted cloud and powerful new EX131 dedicated hardware, this creates a compelling, cost-optimised foundation for scaling high-availability, self-hosted PaaS clusters with enterprise-grade compliance.

Introduction: An Architectural Pivot Driven by Economics

The perennial challenge of scaling a self-hosted PaaS has been balancing orchestration complexity with infrastructure cost. Traditionally, platforms like Coolify relied on Docker Swarm for multi-server management—a pragmatic but increasingly limited solution. As of April 21, 2026, the release of Coolify v4.0.0-beta.474 officially marks Docker Swarm support as deprecated. This is not merely a feature removal; it signals a deliberate architectural shift towards a new proprietary orchestration engine, designed explicitly for the high-density, cost-conscious infrastructure landscape of 2026. The catalyst for this shift is clear: following Hetzner’s April 1 price adjustments, where cloud instances like the CX23 saw a 37% increase, the economic imperative to maximise resource utilisation on dedicated hardware has intensified. This convergence of software ambition and hardware economics redefines the blueprint for Coolify v4 clusters.

What is a Self-hosted PaaS in 2026?

A self-hosted Platform as a Service (PaaS) in 2026 is an infrastructure automation layer that abstracts server provisioning, application deployment, networking, and monitoring, while remaining under the operator’s direct control. Unlike managed cloud PaaS, it runs on infrastructure the user owns or leases, such as Hetzner servers. The core value proposition has evolved from simple convenience to strategic advantage: mitigating escalating egress costs, ensuring data sovereignty, and leveraging high-performance, dedicated hardware like the Hetzner EX131 to achieve density and performance unattainable in shared-cloud environments. Coolify v4 exemplifies this, acting as the control plane for such a privately-managed platform.

The Orchestration Engine: From Swarm to Proprietary Control

The deprecation of Docker Swarm in Coolify v4 is a foundational change. Swarm provided a known, albeit somewhat simplistic, clustering model. Coolify’s new engine, while not fully detailed in the beta release, is implied to offer deeper integration with Coolify’s core concepts—like its project and environment model—and direct control over proxy configuration and health monitoring.

A critical feature enabling this shift is the introduction of database-backed proxy configuration storage. Instead of ephemeral configuration files on a single manager node, Traefik or Caddy configurations are now stored, versioned, and managed within Coolify’s central database. This allows for automatic recovery and synchronisation across all nodes in a distributed cluster, removing a significant single point of failure and manual intervention point.

Pro Tip: When planning your migration from a Swarm-based Coolify v3 cluster, first ensure your proxy configurations are fully defined within Coolify’s UI. This will allow the v4 database to capture the state before you dismantle the Swarm overlay network.

This architectural move aligns with a broader industry trend towards bespoke orchestration for niche platforms, allowing for optimisations impossible with general-purpose tools.

Hardware Foundation: The Hetzner EX131 and ARM64 Optimisation

Hetzner’s April 2026 infrastructure refresh presents the ideal hardware counterpart to Coolify’s software evolution. The new EX131 Dedicated Server provides a high-performance foundation for Coolify build-servers, featuring DDR5 RAM and enhanced NVMe RAID10 configurations. For a self-hosted PaaS acting as both build engine and runtime host, this raw I/O and memory performance drastically reduces pipeline times and improves concurrent application performance.

Simultaneously, Hetzner’s CAX (Ampere Altra ARM64) cloud instances have emerged as a cost-performance leader. Benchmarking data shows they deliver 22% higher single-core Geekbench 6 scores compared to legacy shared-CPU Intel instances at the same price point. For Coolify clusters running numerous, often concurrent, containers, this translates to more efficient scheduling and better overall throughput per euro.

# Example Coolify v4 server configuration hinting at specialisation
server:
  role: build-server
  provider: hetzner-dedicated
  type: EX131
  priority: high

server:
  role: runtime-node
  provider: hetzner-cloud
  type: CAX21
  priority: standard

Despite the April 1 price increases, Hetzner cloud instances remain approximately 60% more cost-effective than comparable DigitalOcean Droplets. Furthermore, their consistent 20 TB monthly egress allowance—a 5x advantage over DigitalOcean’s 4 TB cap—is a crucial consideration for data-intensive applications, making them a sustainable choice for the self-hosted PaaS model where egress costs can otherwise dominate.

Why Does Compliance and Monitoring Matter for Enterprise Scaling?

The scalability of a platform is not just technical; it is also governed by compliance and operational visibility. Hetzner’s attainment of the BSI C5 Type 2 certification in March 2026 is a pivotal development. This German federal security certification allows Coolify-managed clusters hosted on Hetzner to support regulated enterprise workloads for the first time, opening new verticals for the self-hosted approach.

To manage these larger, more critical clusters, Coolify v4’s ‘Sentinel’ monitoring engine has been enhanced with native eBPF-based metrics. eBPF (extended Berkeley Packet Filter) allows for deep, real-time observation of container network activity, system calls, and performance without requiring instrumentation inside the container itself. The claimed overhead of “less than 1.5% CPU” on high-density servers like the EX131 is significant, as it enables pervasive monitoring without degrading the primary workload performance—a necessity for high-availability clusters.

Pro Tip: For regulated workloads, document your cluster design using Hetzner’s C5 attestation documents and ensure Coolify’s Sentinel monitoring logs are stored persistently and immutably for audit purposes.

This combination of certified infrastructure and low-overhead, deep monitoring fulfils the non-functional requirements essential for scaling beyond developer and startup use.

The 2026 Outlook: Specialisation and Vertical Integration

Looking ahead, the trajectory for self-hosted PaaS tools like Coolify is towards increased specialisation and vertical integration. The general-purpose cloud is becoming a cost-prohibitive baseline for many workloads. We predict a rise in: 1) Hardware-Specific Optimisations: Coolify and similar platforms will develop features tailored for hardware like the GEX131 GPU servers, which startups already use for self-hosted AI inference at 4x the FLOPS-per-dollar efficiency of managed services. 2) Compliance-as-Code: Built-in compliance frameworks for standards like BSI C5 will become features, not just external considerations. 3) Financial Orchestration: Tight integration with infrastructure billing APIs to provide real-time cost-per-deployment analytics will emerge as a core feature, directly addressing the cloud cost optimisation driver. The architecture will become less about abstracting any infrastructure and more about optimally exploiting specific, cost-effective infrastructure.

Key Takeaways

• Coolify v4’s move away from Docker Swarm is a strategic shift enabling tighter control and reliability for multi-server clusters, exemplified by database-backed proxy configuration.
• Hetzner’s EX131 dedicated server and CAX ARM64 cloud instances provide the high-density, performance-per-cost hardware foundation required to offset broader cloud price increases.
• Hetzner’s BSI C5 certification and generous egress allowances address the compliance and data cost barriers to enterprise adoption of self-hosted PaaS.
• The integration of eBPF-based monitoring in Coolify v4’s Sentinel engine provides the deep visibility needed for high-availability clusters with minimal performance penalty.
• The future of self-hosted PaaS lies in vertical optimisation for specific hardware (like GPU servers) and compliance frameworks, moving beyond generic infrastructure abstraction.

Conclusion

The April 2026 developments in both Coolify v4 and Hetzner’s infrastructure represent a synchronous evolution. The software is shedding generic orchestration for purposeful control, while the hardware offers certified, high-performance, and economically rational foundations. This creates a viable, scalable path for organisations seeking autonomy from managed cloud pricing and constraints. Engineering such clusters requires a nuanced understanding of both the new orchestration paradigms and the specific capabilities of underlying servers. At Zorinto, we help clients navigate these precise architectural decisions, implementing and optimising self-hosted platforms that turn infrastructure cost pressure into a competitive advantage.