Author SHA1 Message Date
ahmza 94c561581a docs: show dynamic cloudflare trusted_proxies as primary example
caddy-cf now bundles WeidiDeng/caddy-cloudflare-ip, so document
'trusted_proxies cloudflare { interval 12h; timeout 15s }' as the
primary (auto-refreshing) approach, with the static-list method noted
as the fallback for builds without the plugin. Also explains
client_ip_headers (Cf-Connecting-Ip preferred, X-Forwarded-For fallback).
2026-07-25 05:03:12 +01:00
ahmza ce7a72728d Use dynamic cloudflare trusted_proxies IP source
caddy-cf now includes WeidiDeng/caddy-cloudflare-ip, so replace the
hardcoded 22-range static list with 'trusted_proxies cloudflare' which
auto-fetches and refreshes Cloudflare's IPv4/IPv6 ranges (interval 12h,
timeout 15s). No static list to keep in sync when Cloudflare updates.

Kept client_ip_headers Cf-Connecting-Ip X-Forwarded-For (prefer CF's
authoritative end-user IP, fall back to XFF) and documented why.

Validated + fmt-clean and tested end-to-end against the updated caddy-cf
image: starts cleanly, serves HTTP 200 with gzip + security headers.
2026-07-25 05:01:49 +01:00
ahmza c24ed01e0a docs: document Cloudflare-fronted HTTP-only Caddy topology
The live production setup is Cloudflare -> Caddy (HTTP :80) -> SWS, which
differed from the DNS-01 topology the doc previously assumed. Add a
'Deployment topologies' overview and a dedicated Cloudflare-fronted
section pointing at the committed root Caddyfile, covering reverse_proxy,
encode, security headers, and Cloudflare trusted_proxies. Label the
existing Compose/K8s examples as the DNS-01 alternative.
2026-07-25 04:29:33 +01:00
ahmza 7e1728a037 Restrict Caddy trusted_proxies to Cloudflare IP ranges
Replace 'private_ranges' with Cloudflare's official published edge ranges
(15 IPv4 + 7 IPv6 from cloudflare.com/ips-v4 and /ips-v6). This ensures
Cf-Connecting-Ip / X-Forwarded-For are only honoured for requests that
genuinely originate from Cloudflare, preventing client-IP header spoofing
from any other source.

Validated + fmt-clean and tested at runtime against the caddy-cf image:
Caddy starts cleanly with all 22 ranges, serves HTTP 200 with gzip and
security headers.
2026-07-25 04:25:25 +01:00
ahmza 30ef38b143 Add HTTP-only Caddyfile for Cloudflare-fronted deployment
The live Caddy was returning a blank page (NOP, status 0) because its
site block had no responder \u2014 no reverse_proxy to the blog. This adds a
known-good HTTP-only Caddyfile matching the actual topology
(Cloudflare terminates TLS -> Caddy :80 -> SWS blog:80):

- reverse_proxy blog:80  (the missing responder that caused the blank page)
- encode zstd gzip        (fixes: origin was serving everything uncompressed)
- security headers        (fixes: HSTS/X-Frame-Options/etc were absent)
- trusted_proxies + client_ip_headers for correct Cloudflare visitor IPs
- http:// scheme so Caddy does not try to provision TLS (Cloudflare does)

Validated + fmt-clean against the caddy-cf image; tested end-to-end
locally (Caddy -> SWS) returning HTTP 200 with gzip and headers.
2026-07-25 04:19:34 +01:00
ahmza b7145e1323 Remove temporary Netbird connectivity debug step
Build Hugo / build (push) Successful in 1m32s
Build Hugo / deploy (push) Successful in 39s
The deploy pipeline works end-to-end now (Gimli deploys successfully),
so the diagnostic step used to trace the WireGuard/TUN issue is no
longer needed. RPIC continue-on-error is kept until the cluster returns.
2026-07-25 03:07:00 +01:00
ahmza 87867b9f30 Point compose.yml at Gitea registry image with :latest tag
Build Hugo / build (push) Successful in 1m27s
Build Hugo / deploy (push) Successful in 40s
CI pushes to tea.ahmza.com/ahmza/blog-ahmza-com, not ghcr.io, and a
bare reference could resolve to the non-runnable :buildcache tag. Pin
the image to tea.ahmza.com/ahmza/blog-ahmza-com:latest so docker compose
pulls the actual runnable image.
2026-07-25 03:02:39 +01:00
ahmza 8a7833e8f5 Make RPIC deploy non-blocking while cluster is out of service
Build Hugo / build (push) Successful in 1m26s
Build Hugo / deploy (push) Failing after 38s
The Raspberry Pi cluster is temporarily out of service. Mark its deploy
step continue-on-error so its failure no longer skips the Gimli deploy
or fails the whole job. Revert once the cluster is back online.
2026-07-25 02:42:15 +01:00
ahmza 4426668518 Use registry build cache instead of gha cache
Build Hugo / build (push) Successful in 3m23s
Build Hugo / deploy (push) Failing after 39s
The GitHub-hosted 'gha' cache backend is unreachable from the
self-hosted Gitea runner (dial tcp 172.18.0.2:44181 i/o timeout), which
failed the build step even though the image pushed successfully. Switch
cache-from/cache-to to a registry-backed cache stored as a :buildcache
tag in the same Gitea registry, which works on self-hosted runners.
2026-07-25 02:21:38 +01:00
ahmza eacf71d676 Debug: check runner WireGuard capability instead of UDP ports
Build Hugo / build (push) Failing after 13m45s
Build Hugo / deploy (push) Skipped
Other Netbird peers connect fine, so the server and 51820/udp are OK --
the hang is specific to the Gitea runner. Replace the UDP port probes
with runner-side checks: /dev/net/tun presence, effective capabilities
(NET_ADMIN), WireGuard tooling/module, and a definitive 'ip tuntap add'
test. A runner container lacking TUN/NET_ADMIN cannot create the
WireGuard interface, which explains 'netbird up' timing out after the
control plane connects.
2026-07-25 01:53:09 +01:00
ahmza b05814980f Update Netbird debug to probe correct data-plane ports
Build Hugo / build (push) Successful in 1m47s
Build Hugo / deploy (push) Failing after 1m15s
The Netbird server multiplexes management + signal + dashboard behind
Traefik on :443 (no separate :10000 signal port). The tunnel data plane
is WireGuard on 51820/udp and STUN on 3478/udp. Update the debug step to
probe those UDP ports plus a general outbound-UDP check (public STUN) to
determine whether the runner network blocks UDP egress -- the likely
cause of 'netbird up' timing out after the control plane connects fine.
2026-07-25 01:46:28 +01:00
ahmza eb67d6361e Add Netbird connectivity debug step to deploy job
Build Hugo / build (push) Successful in 1m44s
Build Hugo / deploy (push) Failing after 1m8s
Temporary diagnostic step before 'Netbird Connect' that checks secret
presence (masked), DNS resolution, HTTPS reachability to the management
API, and the signal server TCP port from the runner. Helps distinguish
DNS/connectivity issues from auth/registration failures behind the
recurring 'netbird up' DeadlineExceeded error. Marked continue-on-error
so it never blocks the build.
2026-07-25 01:39:09 +01:00
ahmza 8062b9d208 Read Netbird management URL from secret
Build Hugo / build (push) Successful in 1m46s
Build Hugo / deploy (push) Failing after 58s
Switch the deploy job's Netbird Connect step to read management-url from
the NETBIRD_URL secret instead of a hardcoded value, so the URL can be
managed in the Gitea UI alongside the setup key.
2026-07-25 01:33:47 +01:00
ahmza aa01def998 Merge pull request 'Optimizations: switch to static-web-server + Caddy reverse proxy docs' (#1) from optimizations into main
Build Hugo / build (push) Successful in 1m51s
Build Hugo / deploy (push) Failing after 57s
Reviewed-on: #1
2026-07-25 00:19:53 +00:00
ahmza f23fe01f3c Fix GitLab/Gitea inconsistency in README
Build Hugo / deploy (push) Canceled after 0s
Build Hugo / build (push) Canceled after 48s
Update the README to reflect the actual infrastructure: CI/CD runs on
Gitea Actions (not GitLab), images live in the Gitea container registry,
and the clone URL points to tea.ahmza.com. Also clarify the Blowfish
theme is a Hugo module, so no submodules are needed.
2026-07-25 01:19:29 +01:00
5 changed files with 176 additions and 19 deletions
+10 -3
View File
@@ -46,8 +46,11 @@ jobs:
tags: |
${{ env.REGISTRY }}/${{ gitea.repository_owner }}/blog-ahmza-com:latest
${{ env.REGISTRY }}/${{ gitea.repository_owner }}/blog-ahmza-com:${{ gitea.event.head_commit.id }}
cache-from: type=gha
cache-to: type=gha,mode=max
# Registry-backed build cache. The GitHub-hosted 'gha' cache is not
# reachable from self-hosted Gitea runners, so store the cache as a
# dedicated tag in the same registry instead.
cache-from: type=registry,ref=${{ env.REGISTRY }}/${{ gitea.repository_owner }}/blog-ahmza-com:buildcache
cache-to: type=registry,ref=${{ env.REGISTRY }}/${{ gitea.repository_owner }}/blog-ahmza-com:buildcache,mode=max
deploy:
runs-on: ubuntu-latest
needs: build
@@ -58,9 +61,13 @@ jobs:
with:
setup-key: ${{ secrets.NETBIRD_SETUP_KEY }}
hostname: 'gitea-actions-blog'
management-url: 'https://netbird.ahmza.com'
management-url: ${{ secrets.NETBIRD_URL }}
- name: Deploy to RPIC-Cluster
# RPIC cluster is temporarily out of service. Don't let its failure
# block the Gimli deploy or fail the whole job. Remove this line once
# the cluster is back.
continue-on-error: true
uses: appleboy/ssh-action@v1
with:
host: ${{ secrets.RPIC_HOST }}
+59
View File
@@ -0,0 +1,59 @@
# Caddyfile for the blog behind Cloudflare.
#
# Topology: Cloudflare (terminates TLS) -> Caddy (HTTP :80) -> SWS (blog:80)
#
# Cloudflare handles the public HTTPS cert, so Caddy serves plain HTTP on :80
# and does NOT need the Cloudflare DNS-01 / automatic-HTTPS setup. Set the
# Cloudflare SSL/TLS mode to "Full" so CF->origin is encrypted at the network
# layer (e.g. over your tunnel/Netbird), even though this listener is HTTP.
#
# Copy this file to /etc/caddy/Caddyfile on the caddy_cf host and reload:
# docker exec caddy_cf caddy reload --config /etc/caddy/Caddyfile
# or restart the container.
{
# Trust only Cloudflare's edge IPs so real visitor IPs (Cf-Connecting-Ip /
# X-Forwarded-For) are honoured only when the request actually comes from
# Cloudflare. This prevents header spoofing from any other source.
#
# The `cloudflare` IP source (WeidiDeng/caddy-cloudflare-ip, baked into the
# caddy-cf image) auto-fetches the ranges from cloudflare.com/ips-v4 and
# /ips-v6 and refreshes them periodically, so there's no static list to
# keep in sync.
servers {
trusted_proxies cloudflare {
interval 12h
timeout 15s
}
# Read the real visitor IP from these headers (only honoured for
# requests from trusted_proxies above). Cf-Connecting-Ip is
# Cloudflare's single authoritative end-user IP; X-Forwarded-For is
# the fallback if a request ever arrives without the CF header.
client_ip_headers Cf-Connecting-Ip X-Forwarded-For
}
}
# Bind to the hostname on the HTTP port only. Because the address has an
# explicit ":80" scheme/port, Caddy will NOT try to provision a TLS cert.
http://blog.ahmza.com {
# THE responder: proxy every request to the blog (SWS) container.
# This is what was missing before (requests fell through to a NOP -> blank page).
reverse_proxy blog:80
# Compression happens here (SWS has it disabled on purpose).
encode zstd gzip
# Security headers (single source of truth; SWS has security-headers off).
header {
Strict-Transport-Security "max-age=31536000; includeSubDomains; preload"
X-Content-Type-Options "nosniff"
X-Frame-Options "SAMEORIGIN"
Referrer-Policy "strict-origin-when-cross-origin"
# Drop Caddy's default Server header.
-Server
}
log {
output stdout
format console
}
}
+8 -7
View File
@@ -4,10 +4,10 @@
This repository contains the source code for my personal blog hosted at [blog.ahmza.com](https://blog.ahmza.com).
## Features
Using GitLab provides several advantages:
- **Highly Customizable CI/CD Pipelines**: Automated processes for building and deploying the blog.
- **GitLab Container Image Repository**: Efficient management and storage of container images.
- **Custom GitLab Runners**: Flexibility in managing the CI/CD environment.
Using Gitea provides several advantages:
- **Highly Customizable CI/CD Pipelines**: Automated processes for building and deploying the blog via Gitea Actions.
- **Gitea Container Image Repository**: Efficient management and storage of container images.
- **Custom Gitea Runners**: Flexibility in managing the CI/CD environment.
The CI/CD pipeline is configured to build the blog for both amd64 and arm64
architectures, allowing me to host the site on my [Pi cluster](https://blog.ahmza.com/posts/pi-docker-swarm/)
@@ -33,7 +33,7 @@ sacrificing performance.
## Implementation
- **Hugo**: Converts Markdown files into static HTML/CSS/JS files.
- **GitLab CI/CD**: Automates the build process and creates Docker images.
- **Gitea Actions**: Automates the build process and creates Docker images.
- **Kubernetes Cluster**: Serves the static files to users.
## To-Do List
@@ -47,7 +47,8 @@ sacrificing performance.
## Notes
### Cloning the Repository
To clone this repository along with its submodules, use the following command:
The Blowfish theme is managed as a Hugo module, so no submodules are required.
Clone the repository with:
```bash
git clone git@gitlab.com:ahmza/blog.git
git clone ssh://gitea@tea.ahmza.com:2222/ahmza/blog-ahmza-com.git
```
+1 -1
View File
@@ -1,6 +1,6 @@
services:
blog:
image: ghcr.io/ahmzaa/blog-ahmza-com
image: tea.ahmza.com/ahmza/blog-ahmza-com:latest
ports:
- 80:80
restart: unless-stopped
+98 -8
View File
@@ -1,15 +1,44 @@
# Caddy as a Reverse Proxy
This document describes how to run [Caddy](https://caddyserver.com/) as a
reverse proxy in front of the blog. It covers both deployment targets used by
this project: the Kubernetes (Raspberry Pi) cluster and the Docker Compose host
(Gimli).
reverse proxy in front of the blog.
It uses the custom [`caddy-cf`](https://github.com/ahmzaa/caddy-cf) image
(`ghcr.io/ahmzaa/caddy-cf:latest`) — a Caddy build with the **Cloudflare DNS
module** baked in. The stock Caddy image cannot solve the ACME **DNS-01**
challenge for Cloudflare; this custom build can, which is what lets us obtain
certificates without exposing ports `80`/`443` to the public internet.
module** baked in.
## Deployment topologies
There are two supported ways to run Caddy in front of the blog. **Pick one.**
### 1. Cloudflare in front (current production) — HTTP only
```
Cloudflare (terminates TLS) -> Caddy (HTTP :80) -> SWS (blog:80)
```
Cloudflare proxies the domain and provides the public HTTPS certificate, so
Caddy only needs to serve plain **HTTP on :80** and reverse-proxy to the blog.
Caddy does **not** provision a TLS cert in this mode. This is the live setup and
its config is the committed [`Caddyfile`](../Caddyfile) at the repo root —
see [Cloudflare-fronted setup](#cloudflare-fronted-setup-http-only) below.
Set the Cloudflare **SSL/TLS mode to "Full"** so the Cloudflare→origin hop is
still encrypted at the network layer (e.g. over Netbird), even though Caddy's
listener is HTTP.
### 2. Caddy terminates TLS via Cloudflare DNS-01
```
Client -> Caddy (HTTPS :443, ACME DNS-01) -> SWS (blog:80)
```
Caddy obtains/renews the certificate itself using the ACME **DNS-01** challenge
through the Cloudflare API — no inbound `80`/`443` from the public internet is
required, which suits a Netbird-only setup with no Cloudflare proxy in front.
The stock Caddy image cannot do this; the `caddy-cf` build bundles the
Cloudflare DNS provider plugin. See the Docker Compose / Kubernetes examples
later in this document.
## Why a reverse proxy?
@@ -21,7 +50,7 @@ the reverse proxy so that each layer owns exactly one job:
| Concern | Owned by | Notes |
| -------------------- | -------- | ------------------------------------------------- |
| TLS / HTTPS certs | Caddy | Automatic certs via Cloudflare **DNS-01** challenge |
| TLS / HTTPS certs | Cloudflare *or* Caddy | Cloudflare in topology 1; Caddy (DNS-01) in topology 2 |
| Compression (gzip/zstd) | Caddy | `encode` directive; SWS compression is off |
| Security headers | Caddy | Single source of truth, avoids duplicate headers |
| Serving static files | SWS | Reads from `/public`, sets `Cache-Control`/`ETag` |
@@ -50,11 +79,72 @@ Whatever the deployment, the SWS origin exposes:
- **Health endpoint:** `GET /health` → `200 OK`
- **Cache headers:** already set by SWS (Caddy should pass them through, not override)
Caddy only needs to terminate TLS (via Cloudflare DNS-01), add compression +
Caddy only needs to handle TLS (or let Cloudflare do it), add compression +
security headers, and forward requests to that origin.
---
## Cloudflare-fronted setup (HTTP only)
**This is the current production topology.** The canonical config is the
committed [`Caddyfile`](../Caddyfile) at the repo root; copy it to the
`caddy_cf` host and reload. Highlights:
- **`http://blog.ahmza.com`** — the explicit `http://` scheme binds Caddy to
`:80` only and stops it from trying to provision a TLS certificate (Cloudflare
already serves HTTPS to visitors).
- **`reverse_proxy blog:80`** — proxies to the blog (SWS) container by its
compose service name. Caddy and the blog container must share a Docker
network. Omitting this responder makes Caddy return an empty `NOP` response —
i.e. a **blank page**.
- **`encode zstd gzip`** — compression happens here (SWS has it disabled).
- **`header { ... }`** — security headers (SWS has them disabled).
- **`trusted_proxies cloudflare { ... }`** + **`client_ip_headers
Cf-Connecting-Ip X-Forwarded-For`** — so the real visitor IP is used for
logging/matchers, but only when the request genuinely comes from Cloudflare
(prevents header spoofing from any other source). The `cloudflare` IP source
([WeidiDeng/caddy-cloudflare-ip](https://github.com/WeidiDeng/caddy-cloudflare-ip),
baked into the `caddy-cf` image) auto-fetches the ranges from
<https://www.cloudflare.com/ips-v4/> and <https://www.cloudflare.com/ips-v6/>
and refreshes them on the configured `interval`, so there is no static list to
maintain:
```caddy
trusted_proxies cloudflare {
interval 12h
timeout 15s
}
```
`client_ip_headers` lists which header holds the real visitor IP:
`Cf-Connecting-Ip` is Cloudflare's single authoritative end-user IP, with
`X-Forwarded-For` as the fallback. It is only trusted for requests coming from
the `trusted_proxies` above.
> If you ever run without the `caddy-cloudflare-ip` plugin, you can instead
> pin the static ranges with `trusted_proxies static <ip/cidr ...>` copied
> from the Cloudflare IP pages, but then you must keep them in sync manually.
Deploy and reload:
```bash
# Put the Caddyfile in place on the caddy_cf host (e.g. into the ./conf mount),
# then reload without downtime:
docker exec caddy_cf caddy reload --config /etc/caddy/Caddyfile
# Verify from the host (expect HTTP 200, Content-Encoding: gzip, X-Frame-Options):
curl -sI -H "Accept-Encoding: gzip" http://localhost/ | grep -iE 'HTTP|content-encoding|x-frame'
```
> **Cloudflare SSL/TLS mode:** set it to **Full** so the Cloudflare→origin hop
> is encrypted at the network layer (e.g. over Netbird), even though Caddy's
> listener is plain HTTP.
---
The remaining sections describe **topology 2** (Caddy terminates TLS itself via
the Cloudflare DNS-01 challenge) for setups without a Cloudflare proxy in front.
## Docker Compose (Gimli)
Run Caddy and the blog container on the same Docker network. Caddy reaches the