---
title: "Discussion on \"Cloudflare Tunnel on Multiple VPSes: HA, Scaling, and Pitfalls\" | Hashnode"
description: "Discussion on \"Cloudflare Tunnel on Multiple VPSes: HA, Scaling, and Pitfalls\". Can Cloudflare Tunnel provide high availability or load balancing across multiple VPSes? Learn how connectors work, common deployment mistakes, and the best production architecture for self-hosted applications."
url: "https://hashnode.com/posts/cloudflare-tunnel-on-multiple-vpses-ha-scaling-and-pitfalls/6a2edb10014c010ccb62b0fd"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Cloudflare Tunnel on Multiple VPSes: HA, Scaling, and Pitfalls](https://rahulr.cc/blogs/cloudflare-tunnel-on-multiple-vpses-ha-scaling-and-pitfalls?utm_source=hashnode&utm_medium=feed)

Can Cloudflare Tunnel provide high availability or load balancing across multiple VPSes? Learn how connectors work, common deployment mistakes, and the best production architecture for self-hosted applications.

Jun 14 · 6 min read · rahulr.cc  
Read the article

-   [#cloudflare](https://hashnode.com/tag/cloudflare)
-   [#networking](https://hashnode.com/tag/networking)
-   [#devlife](https://hashnode.com/tag/devlife)
-   [#docker](https://hashnode.com/tag/docker)
-   [#selfhosting](https://hashnode.com/tag/selfhosting)
-   [#cloudflare-tunnel](https://hashnode.com/tag/cloudflare-tunnel)
-   [#system-design](https://hashnode.com/tag/system-design)

## Written by

[Rahul Ravindran](https://hashnode.com/@imrhlrvndrn)

@imrhlrvndrn · Mumbai, India

  
[View profile](https://hashnode.com/@imrhlrvndrn)

## Responses 1

Markdown supported · ⌘↵ to post

[Edge Ops](https://hashnode.com/@edgecacheops)

[Nov 5](https://hashnode.com/posts/cloudflare-tunnel-on-multiple-vpses-ha-scaling-and-pitfalls/6a2edb10014c010ccb62b0fd/comment/6aec18cbea7f40d33d835a03)

Running the same Cloudflare Tunnel identity from more than one VPS only helps when every connector can reach every ingress upstream it might serve. Prefer hostname-shaped routes to Compose or host service names on a private network both connectors share (or replicate the backend beside each connector), end with a catch-all that fails closed, and keep admin UIs behind Access so a failover path never exposes a login that was only firewalled on the primary box. TLS can terminate at the Cloudflare edge for the public names while the tunnel hop stays private; do not mix connectors that still point at host-localhost with ones that resolve Docker DNS — failover then becomes a portability bug.

After adding or replacing a connector I check that each member shows healthy, curl each upstream by internal name from inside that connector’s network, then hit the same hostnames through the public edge twice across a brief connector stop. A 502 with at least one green connector is almost always an upstream the standby cannot see, a renamed service still listed in ingress, or leftover path rules that only worked on the first VPS. When you scale past two members, treat identical config and identical backend reachability as one unit — asymmetric networks look like flaky HA when they are really split-brain routing.
