working on caddy sidecar

.env.sample

@@ -1,6 +1,6 @@
 TYPE=bluesky-pds
 
-DOMAIN=pds.example.com
+DOMAIN=bluesky-pds.example.com
 LETS_ENCRYPT_ENV=production
 
 # Blob upload limit in bytes (default: 100MB)

Caddyfile

@@ -0,0 +1,12 @@
+{
+	on_demand_tls {
+		ask http://app:3000/tls-check
+	}
+}
+
+{$DOMAIN}, *.{$DOMAIN} {
+	tls {
+		on_demand
+	}
+	reverse_proxy app:3000
+}

README.md

@@ -77,17 +77,42 @@ abra app run YOURAPPDOMAIN app -- \
 
 User handles on a PDS can work in two ways:
 
-1. **Subdomain handles** (e.g. `user.pds.example.com`): Requires a wildcard DNS
+1. **Subdomain handles** (e.g. `user.pds.example.com`): The default. Requires a
-   record (`*.pds.example.com`) pointing to your server, and wildcard TLS
+   wildcard DNS record (`*.pds.example.com`) pointing to your server. TLS is
-   certificates (which require DNS challenge configuration in Traefik).
+   handled automatically by the Caddy sidecar (see below).
 
 2. **Domain handles** (e.g. `user.com`): Users can use their own domain as a
    handle by adding a DNS TXT record at `_atproto.user.com` with the value
    `did=did:plc:<their-did>`. This works without any additional server
    configuration.
 
-Domain handles are recommended for most deployments as they don't require
+## TLS architecture (Caddy sidecar)
-wildcard TLS configuration.
+
+This recipe uses a **Caddy sidecar** for TLS instead of letting Traefik terminate
+TLS directly. This is needed because Bluesky subdomain handles require TLS
+certificates for each `user.pds.example.com` subdomain, and Traefik cannot issue
+on-demand per-subdomain certificates.
+
+The architecture:
+
+1. **Traefik** receives TLS connections on port 443 and does **TCP passthrough**
+   (no TLS termination) for traffic matching `DOMAIN` and `*.DOMAIN`, forwarding
+   the raw TLS stream to Caddy.
+2. **Caddy** terminates TLS using **on-demand certificates** — it automatically
+   obtains a Let's Encrypt certificate for each subdomain the first time a
+   connection arrives, using the TLS-ALPN-01 challenge.
+3. **Caddy** reverse proxies the decrypted HTTP traffic to the PDS on port 3000.
+
+This matches how the [upstream PDS](https://github.com/bluesky-social/pds) is
+designed to work (it ships with Caddy), adapted for Co-op Cloud's Traefik-based
+routing. The PDS exposes a `/tls-check` endpoint that Caddy consults before
+issuing a certificate, preventing abuse.
+
+**Note:** The first request to a new subdomain handle may take 10-30 seconds while
+Caddy obtains the TLS certificate from Let's Encrypt. Subsequent requests are instant.
+
+No changes to the Traefik recipe are needed — the TCP passthrough is configured
+entirely via deploy labels on the Caddy service in this recipe's `compose.yml`.
 
 ## DNS setup
 

abra.sh

@@ -1 +1,2 @@
 export ENTRYPOINT_VERSION=v1
+export CADDYFILE_VERSION=v1

compose.yml

@@ -5,7 +5,7 @@ services:
   app:
     image: ghcr.io/bluesky-social/pds:0.4
     networks:
-      - proxy
+      - internal
     environment:
       - PDS_HOSTNAME=${DOMAIN}
       - PDS_DATA_DIRECTORY=/pds
@@ -39,15 +39,11 @@ services:
     deploy:
       restart_policy:
         condition: on-failure
+        max_attempts: 5
       labels:
-        - "traefik.enable=true"
-        - "traefik.docker.network=proxy"
-        - "traefik.http.services.${STACK_NAME}.loadbalancer.server.port=3000"
-        - "traefik.http.routers.${STACK_NAME}.rule=Host(`${DOMAIN}`)"
-        - "traefik.http.routers.${STACK_NAME}.entrypoints=web-secure"
-        - "traefik.http.routers.${STACK_NAME}.tls.certresolver=${LETS_ENCRYPT_ENV}"
         - "coop-cloud.${STACK_NAME}.timeout=${TIMEOUT:-120}"
-        - "coop-cloud.${STACK_NAME}.version=0.1.0+0.4"
+        - "coop-cloud.${STACK_NAME}.version=0.1.0+v0.4"
+        - "backupbot.backup=true"
     healthcheck:
       test: ["CMD", "wget", "-q", "--spider", "http://localhost:3000/xrpc/_health"]
       interval: 30s
@@ -55,12 +51,39 @@ services:
       retries: 5
       start_period: 30s
 
+  caddy:
+    image: caddy:2
+    networks:
+      - proxy
+      - internal
+    environment:
+      - DOMAIN=${DOMAIN}
+    configs:
+      - source: caddyfile
+        target: /etc/caddy/Caddyfile
+    volumes:
+      - caddy_data:/data
+    deploy:
+      restart_policy:
+        condition: on-failure
+        max_attempts: 5
+      labels:
+        - "traefik.enable=true"
+        - "traefik.swarm.network=proxy"
+        - "traefik.tcp.routers.${STACK_NAME}.rule=HostSNI(`${DOMAIN}`) || HostSNIRegexp(`^.+\\.${DOMAIN}$$`)"
+        - "traefik.tcp.routers.${STACK_NAME}.ruleSyntax=v3"
+        - "traefik.tcp.routers.${STACK_NAME}.entrypoints=web-secure"
+        - "traefik.tcp.routers.${STACK_NAME}.tls.passthrough=true"
+        - "traefik.tcp.services.${STACK_NAME}.loadbalancer.server.port=443"
+
 networks:
   proxy:
     external: true
+  internal:
 
 volumes:
   pds_data:
+  caddy_data:
 
 secrets:
   pds_jwt_secret:
@@ -78,3 +101,6 @@ configs:
     name: ${STACK_NAME}_entrypoint_${ENTRYPOINT_VERSION}
     file: entrypoint.sh.tmpl
     template_driver: golang
+  caddyfile:
+    name: ${STACK_NAME}_caddyfile_${CADDYFILE_VERSION}
+    file: Caddyfile