fix(cloudflare): add backend network to compose templates and improve tunnel diagnostics

- Add 'backend' external network to all compose templates so cloudflared can reach tool containers - Add better error handling and logging to create_tunnel() with specific error messages for auth failures - Add check_cloudflare_config() diagnostic function - Add /health/cloudflare endpoint to verify Cloudflare configuration - Import Any type for type hints
2026-05-20 15:55:20 +02:00
parent e7c42c17b9
commit e6f64c39f3
3 changed files with 187 additions and 68 deletions
@@ -147,3 +147,21 @@ async def health_check_db() -> dict[str, Any]:
            status="unhealthy",
            response_time_ms=0.0,
        ).model_dump()
+
+
+@router.get(
+    "/health/cloudflare",
+    summary="Cloudflare tunnel health check",
+    description="Returns Cloudflare tunnel configuration status and diagnostics.",
+    tags=["Health"],
+)
+async def health_check_cloudflare() -> dict[str, Any]:
+    """Check Cloudflare tunnel configuration.
+
+    Returns:
+        Dict with Cloudflare configuration status.
+    """
+    from src.services.cloudflare_tunnel import check_cloudflare_config
+
+    result = await check_cloudflare_config()
+    return result
@@ -122,7 +122,13 @@ services:
      - {{REPO_PATH}}:/config/workspace
    ports:
      - "8443:8443"
-    restart: unless-stopped""",
+    networks:
+      - backend
+    restart: unless-stopped
+
+networks:
+  backend:
+    external: true""",
                "default_port": 8443,
                "required_variables": ["REPO_PATH", "TOOL_NAME"],
            },
@@ -144,7 +150,13 @@ services:
      - {{REPO_PATH}}:/home/jovyan/work
    ports:
      - "8888:8888"
-    restart: unless-stopped""",
+    networks:
+      - backend
+    restart: unless-stopped
+
+networks:
+  backend:
+    external: true""",
                "required_variables": ["REPO_PATH", "TOOL_NAME"],
            },
            {
@@ -170,10 +182,16 @@ services:
             tail -f /dev/null"
    stdin_open: true
    tty: true
+    networks:
+      - backend
    restart: unless-stopped

 volumes:
-  opencode_home:""",
+  opencode_home:
+
+networks:
+  backend:
+    external: true""",
                "required_variables": ["REPO_PATH", "TOOL_NAME"],
            },
        ]
@@ -5,6 +5,7 @@ import logging
 import os
 import uuid
 from pathlib import Path
+from typing import Any

 import httpx
 from src.config import Settings
@@ -53,78 +54,98 @@ async def create_tunnel(
    headers = _get_headers(settings)
    account_id = settings.cloudflare_account_id

-    # Create tunnel
-    async with httpx.AsyncClient() as client:
-        logger.info("Step 1: Creating tunnel via Cloudflare API...")
-        response = await client.post(
-            f"{CLOUDFLARE_API_BASE}/accounts/{account_id}/cfd_tunnel",
-            headers=headers,
-            json={
-                "name": f"headquarter-{instance_name}",
-                "config_src": "cloudflare",
-            },
-        )
-        logger.info("Tunnel creation response: status=%d, body=%s", response.status_code, response.text[:500])
-        response.raise_for_status()
-        data = response.json()
+    try:
+        # Create tunnel
+        async with httpx.AsyncClient() as client:
+            logger.info("Step 1: Creating tunnel via Cloudflare API...")
+            response = await client.post(
+                f"{CLOUDFLARE_API_BASE}/accounts/{account_id}/cfd_tunnel",
+                headers=headers,
+                json={
+                    "name": f"headquarter-{instance_name}",
+                    "config_src": "cloudflare",
+                },
+            )
+            logger.info("Tunnel creation response: status=%d, body=%s", response.status_code, response.text[:500])
+            
+            if response.status_code == 403:
+                raise ValueError(f"Cloudflare API authentication failed. Check your API token permissions. Body: {response.text}")
+            if response.status_code == 400:
+                raise ValueError(f"Cloudflare API bad request: {response.text}")
+            
+            response.raise_for_status()
+            data = response.json()

-        if not data.get("success"):
-            raise ValueError(f"Failed to create tunnel: {data.get('errors')}")
+            if not data.get("success"):
+                errors = data.get('errors', [])
+                raise ValueError(f"Failed to create tunnel: {errors}")

-        tunnel = data["result"]
-        tunnel_id = tunnel["id"]
-        logger.info("Step 1 complete: tunnel_id=%s", tunnel_id)
+            tunnel = data["result"]
+            tunnel_id = tunnel["id"]
+            logger.info("Step 1 complete: tunnel_id=%s", tunnel_id)

-        # Get tunnel token
-        logger.info("Step 2: Getting tunnel token...")
-        token_response = await client.get(
-            f"{CLOUDFLARE_API_BASE}/accounts/{account_id}/cfd_tunnel/{tunnel_id}/token",
-            headers=headers,
-        )
-        logger.info("Token response: status=%d", token_response.status_code)
-        token_response.raise_for_status()
-        token_data = token_response.json()
-        tunnel_token = token_data["result"]
-        logger.info("Step 2 complete: got tunnel token")
+            # Get tunnel token
+            logger.info("Step 2: Getting tunnel token...")
+            token_response = await client.get(
+                f"{CLOUDFLARE_API_BASE}/accounts/{account_id}/cfd_tunnel/{tunnel_id}/token",
+                headers=headers,
+            )
+            logger.info("Token response: status=%d", token_response.status_code)
+            
+            if token_response.status_code != 200:
+                raise ValueError(f"Failed to get tunnel token: {token_response.status_code} - {token_response.text}")
+            
+            token_data = token_response.json()
+            tunnel_token = token_data["result"]
+            logger.info("Step 2 complete: got tunnel token")

-        # Create DNS record for the tunnel
-        subdomain = f"instance-{instance_id[:8]}"
-        hostname = f"{subdomain}.{settings.cloudflare_base_domain}"
-        logger.info("Step 3: Creating DNS record for subdomain=%s, hostname=%s", subdomain, hostname)
+            # Create DNS record for the tunnel
+            subdomain = f"instance-{instance_id[:8]}"
+            hostname = f"{subdomain}.{settings.cloudflare_base_domain}"
+            logger.info("Step 3: Creating DNS record for subdomain=%s, hostname=%s", subdomain, hostname)

-        dns_response = await client.post(
-            f"{CLOUDFLARE_API_BASE}/zones/{settings.cloudflare_zone_id}/dns_records",
-            headers=headers,
-            json={
-                "type": "CNAME",
-                "name": subdomain,
-                "content": f"{tunnel_id}.cfargotunnel.com",
-                "ttl": 1,
-                "proxied": True,
-            },
-        )
-        logger.info("DNS response: status=%d, body=%s", dns_response.status_code, dns_response.text[:500])
-        dns_response.raise_for_status()
-        logger.info("Step 3 complete: DNS record created")
+            dns_response = await client.post(
+                f"{CLOUDFLARE_API_BASE}/zones/{settings.cloudflare_zone_id}/dns_records",
+                headers=headers,
+                json={
+                    "type": "CNAME",
+                    "name": subdomain,
+                    "content": f"{tunnel_id}.cfargotunnel.com",
+                    "ttl": 1,
+                    "proxied": True,
+                },
+            )
+            logger.info("DNS response: status=%d, body=%s", dns_response.status_code, dns_response.text[:500])
+            
+            if dns_response.status_code == 403:
+                raise ValueError(f"Cloudflare DNS API authentication failed. Check token has Zone:Edit permission.")
+            if dns_response.status_code == 400:
+                raise ValueError(f"Cloudflare DNS bad request: {dns_response.text}")
+            
+            dns_response.raise_for_status()
+            logger.info("Step 3 complete: DNS record created")

-        # Update cloudflared config
-        logger.info("Step 4: Updating cloudflared config...")
-        await update_cloudflared_config(
-            tunnel_id=tunnel_id,
-            tunnel_token=tunnel_token,
-            hostname=hostname,
-            instance_name=instance_name,
-            instance_port=instance_port,
-            settings=settings,
-        )
-        logger.info("Step 4 complete: cloudflared config updated")
+            # Update cloudflared config
+            logger.info("Step 4: Updating cloudflared config...")
+            await update_cloudflared_config(
+                tunnel_id=tunnel_id,
+                tunnel_token=tunnel_token,
+                hostname=hostname,
+                instance_name=instance_name,
+                instance_port=instance_port,
+                settings=settings,
+            )
+            logger.info("Step 4 complete: cloudflared config updated")

-        logger.info("Tunnel creation complete: tunnel_id=%s, public_url=https://%s", tunnel_id, hostname)
-        return {
-            "tunnel_id": tunnel_id,
-            "public_url": f"https://{hostname}",
-            "subdomain": subdomain,
-        }
+            logger.info("Tunnel creation complete: tunnel_id=%s, public_url=https://%s", tunnel_id, hostname)
+            return {
+                "tunnel_id": tunnel_id,
+                "public_url": f"https://{hostname}",
+                "subdomain": subdomain,
+            }
+    except Exception as e:
+        logger.error("Failed to create Cloudflare tunnel: %s", str(e), exc_info=True)
+        raise


 async def delete_tunnel(
@@ -287,3 +308,65 @@ async def remove_tunnel_from_config(
        credentials_file.unlink()

    logger.info("Removed tunnel %s from cloudflared config", tunnel_id)
+
+
+async def check_cloudflare_config(settings: Settings | None = None) -> dict[str, Any]:
+    """Check if Cloudflare configuration is valid and working.
+
+    Args:
+        settings: Optional settings override
+
+    Returns:
+        Dict with status and diagnostic information
+    """
+    if settings is None:
+        settings = Settings()
+
+    result: dict[str, Any] = {
+        "configured": False,
+        "api_token_set": bool(settings.cloudflare_api_token),
+        "account_id_set": bool(settings.cloudflare_account_id),
+        "zone_id_set": bool(settings.cloudflare_zone_id),
+        "base_domain_set": bool(settings.cloudflare_base_domain),
+        "api_test": None,
+        "errors": [],
+    }
+
+    if not all([
+        settings.cloudflare_api_token,
+        settings.cloudflare_account_id,
+        settings.cloudflare_zone_id,
+        settings.cloudflare_base_domain,
+    ]):
+        result["errors"].append("Missing required Cloudflare configuration")
+        return result
+
+    result["configured"] = True
+
+    # Test API connectivity
+    try:
+        headers = _get_headers(settings)
+        async with httpx.AsyncClient() as client:
+            # Test account access
+            resp = await client.get(
+                f"{CLOUDFLARE_API_BASE}/accounts/{settings.cloudflare_account_id}",
+                headers=headers,
+            )
+            if resp.status_code == 200:
+                result["api_test"] = "ok"
+            elif resp.status_code == 403:
+                result["api_test"] = "auth_failed"
+                result["errors"].append("API token authentication failed - check token permissions")
+            else:
+                result["api_test"] = f"error_{resp.status_code}"
+                result["errors"].append(f"API test failed: {resp.status_code}")
+    except Exception as e:
+        result["api_test"] = "exception"
+        result["errors"].append(f"API test exception: {str(e)}")
+
+    # Check config directory
+    config_dir = Path(settings.cloudflared_config_dir)
+    result["config_dir_exists"] = config_dir.exists()
+    result["config_dir_writable"] = os.access(config_dir, os.W_OK) if config_dir.exists() else False
+
+    return result