feat: use service labels to identify service that needs to be annotated

gateway_services.yaml

@@ -1,4 +0,0 @@
-gateway_services:
-  - name: traefik-tcp
-    namespace: kube-system
-    label_pattern: app.kubernetes.io/name=traefik

node-external-ip-controller.Dockerfile

@@ -6,7 +6,6 @@ RUN pip install kubernetes kubernetes_asyncio PyYAML
 
 # Copy the controller script into the container
 COPY node_external_ip_controller_async.py /app/node_external_ip_controller.py
-COPY gateway_services.yaml /app/gateway_services.yaml
 
 # Set the working directory
 WORKDIR /app

node_external_ip_controller_async.py

@@ -3,32 +3,24 @@ from kubernetes_asyncio import client, config, watch
 import os
 
 # Configuration
-import yaml
-
-GATEWAY_SERVICES_FILE = os.getenv("GATEWAY_SERVICES_FILE", "gateway_services.yaml")
-with open(GATEWAY_SERVICES_FILE, "r") as f:
-    gateway_services = yaml.safe_load(f)
-
-services = gateway_services["gateway_services"]
 ANNOTATION_KEY = os.getenv("ANNOTATION_KEY", "kube-vip.io/loadbalancerIPs")
 ZERO_GATEWAY_IP = os.getenv("ZERO_GATEWAY_IP", "172.28.10.1")
 NODE_LABEL = os.getenv("NODE_LABEL", "svccontroller.k3s.cattle.io/enablelb=true")
+SERVICE_LABEL_KEY = os.getenv("SERVICE_LABEL_KEY", "zlanservice")
+SERVICE_LABEL_VALUE = os.getenv("SERVICE_LABEL_VALUE", "true")
 SERVICE_REQUEST_TIMEOUT = int(os.getenv("SERVICE_REQUEST_TIMEOUT", 300))
 NODE_REQUEST_TIMEOUT = int(os.getenv("NODE_REQUEST_TIMEOUT", 30))
 
 
 async def update_service_annotation(v1, service, external_ips):
     try:
-        # Get the current service object
+        service_name = service.metadata.name
-        service_name = service["name"]
+        namespace = service.metadata.namespace
-        namespace = service["namespace"]
         service_obj = await v1.read_namespaced_service(service_name, namespace)
 
-        # Check if the annotation needs to be updated
         current_annotation = service_obj.metadata.annotations.get(ANNOTATION_KEY)
         target_annotation = ",".join(external_ips) + "," + ZERO_GATEWAY_IP
         if current_annotation != target_annotation:
-            # Update the annotation
             body = {"metadata": {"annotations": {ANNOTATION_KEY: target_annotation}}}
             await v1.patch_namespaced_service(service_name, namespace, body)
             print(
@@ -40,97 +32,71 @@ async def update_service_annotation(v1, service, external_ips):
         print(f"API Exception in update_service_annotation: {e}", flush=True)
 
 
-async def watch_nodes():
+async def watch_nodes(v1, external_ips_update_queue):
-    config.load_incluster_config()
-    v1 = client.CoreV1Api()
     w = watch.Watch()
-
     while True:
         try:
-            # Check for external IP
-            external_ips, node_names = [], []
             async for event in w.stream(
                 v1.list_node,
                 label_selector=NODE_LABEL,
                 _request_timeout=NODE_REQUEST_TIMEOUT,
             ):
                 node = event["object"]
-                node_name = node.metadata.name
+                external_ips = [
-
+                    addr.address
-                for address in node.status.addresses:
+                    for addr in node.status.addresses
-                    if address.type == "ExternalIP":
+                    if addr.type == "ExternalIP"
-                        external_ips.append(address.address)
+                ]
-                        node_names.append(node_name)
+                if external_ips:
-            if len(external_ips) > 0:
+                    await external_ips_update_queue.put(external_ips)
-                print(
-                    f"Detected external IPs {str(external_ips)} for node {str(node_names)}",
-                    flush=True,
-                )
-                for service in services:
-                    await update_service_annotation(v1, service, external_ips)
 
         except client.exceptions.ApiException as e:
             print(f"API Exception in watch_nodes: {e}", flush=True)
             await asyncio.sleep(5)
-
         except asyncio.CancelledError:
             print("Watch task was cancelled.", flush=True)
             break
-
         except Exception as e:
             print(f"Unexpected error in watch_nodes: {e}", flush=True)
             await asyncio.sleep(5)
 
 
-async def watch_service(service):
+async def watch_services(v1, external_ips_update_queue):
-    config.load_incluster_config()
-    v1 = client.CoreV1Api()
     w = watch.Watch()
-    service_name = service["name"]
+    label_selector = f"{SERVICE_LABEL_KEY}={SERVICE_LABEL_VALUE}"
-    namespace = service["namespace"]
-    label_selector = service["label_pattern"]
 
     while True:
         try:
             async for event in w.stream(
-                v1.list_namespaced_service,
+                v1.list_service_for_all_namespaces,
-                namespace,
                 label_selector=label_selector,
                 _request_timeout=SERVICE_REQUEST_TIMEOUT,
             ):
-                service_obj = event["object"]
+                service = event["object"]
-                if (
+                if event["type"] in {"ADDED", "MODIFIED"}:
-                    event["type"] == "ADDED"
+                    external_ips = await external_ips_update_queue.get()
-                    and service_obj.metadata.name == service_name
-                ):
-                    print(f"New service detected: {service_name}", flush=True)
-
-                    # Fetch current external IP for nodes that match the label
-                    nodes = await v1.list_node(label_selector=NODE_LABEL)
-                    external_ips = []
-                    for node in nodes.items:
-                        for address in node.status.addresses:
-                            if address.type == "ExternalIP":
-                                external_ips.append(address.address)
                     await update_service_annotation(v1, service, external_ips)
 
         except client.exceptions.ApiException as e:
             print(f"API Exception in watch_services: {e}", flush=True)
             await asyncio.sleep(5)
-
         except asyncio.CancelledError:
             print("Watch task was cancelled.", flush=True)
             break
-
         except Exception as e:
             print(f"Unexpected error in watch_services: {e}", flush=True)
             await asyncio.sleep(5)
 
 
 async def main():
-    # Run watch_nodes and watch_services concurrently
+    config.load_incluster_config()
-    service_watchers = [watch_service(service) for service in services]
+    v1 = client.CoreV1Api()
-    await asyncio.gather(watch_nodes(), *service_watchers)
+
+    external_ips_update_queue = asyncio.Queue()
+    await asyncio.gather(
+        watch_nodes(v1, external_ips_update_queue),
+        watch_services(v1, external_ips_update_queue),
+    )
 
 
 if __name__ == "__main__":