ARTESCA-11245 repair metalk8s monitoring salt module

CHANGELOG.md

@@ -8,13 +8,18 @@
   Workload Plane/Control Plane CIDRs.
   (PR[#4217](https://github.com/scality/metalk8s/pull/4217))
 
+### Bug fixes
+
+- Salt module metalk8s_monitoring fixed and improved
+  (PR[#4238](https://github.com/scality/metalk8s/pull/4238))
+
 ## Release 126.0.2
 
 ### Bug fixes
 
 - [2137](https://github.com/scality/metalk8s/issues/2137) - Fix a bug
   that prevents re-run of the bootstrap script if it fails at a specific point
-  (PR[4196](https://github.com/scality/metalk8s/pull/4196))
+  (PR[#4196](https://github.com/scality/metalk8s/pull/4196))
 
 - Bump the attempts to wait for the Operator and ClusterConfig
   to be Ready

salt/_modules/metalk8s_kubernetes_utils.py

@@ -150,7 +150,7 @@ def read_and_render_yaml_file(source, template, context=None, saltenv="base"):
         return salt.utils.yaml.safe_load(contents)
 
 
-def get_service_endpoints(service, namespace, kubeconfig):
+def get_service_endpoints(service, namespace, **kwargs):
     error_tpl = "Unable to get kubernetes endpoints for {} in namespace {}"
 
     try:
@@ -159,7 +159,7 @@ def get_service_endpoints(service, namespace, kubeconfig):
             kind="Endpoints",
             apiVersion="v1",
             namespace=namespace,
-            kubeconfig=kubeconfig,
+            **kwargs,
         )
         if not endpoint:
             raise CommandExecutionError("Endpoint not found")
@@ -184,3 +184,58 @@ def get_service_endpoints(service, namespace, kubeconfig):
         raise CommandExecutionError(error_tpl.format(service, namespace)) from exc
 
     return result
+
+
+def get_service_ips_and_ports(service, namespace, **kwargs):
+    try:
+        service_object = __salt__["metalk8s_kubernetes.get_object"](
+            name=service,
+            kind="Service",
+            apiVersion="v1",
+            namespace=namespace,
+            **kwargs,
+        )
+        if not service_object:
+            raise CommandExecutionError("Service not found")
+    except CommandExecutionError as exc:
+        raise CommandExecutionError(
+            f"Unable to get kubernetes service {service} in namespace {namespace}"
+        ) from exc
+
+    cip = service_object["spec"].get("clusterIP")
+    cips = [cip] if cip else []
+    cips.extend(
+        [ip for ip in service_object["spec"].get("clusterIPs", []) if ip != cip]
+    )
+
+    try:
+        unamed = 0
+        ports = {}
+        service_ports = service_object["spec"]["ports"]
+
+        for port in service_ports:
+            if port.get("name"):
+                ports[port["name"]] = port["port"]
+            else:
+                ports[f"unnamed-{unamed}"] = port["port"]
+                unamed += 1
+
+        if not ports:
+            raise CommandExecutionError("No ports")
+    except (
+        AttributeError,
+        IndexError,
+        KeyError,
+        TypeError,
+        CommandExecutionError,
+    ) as exc:
+        raise CommandExecutionError(
+            f"No ports for service {service} in namespace {namespace}",
+        ) from exc
+
+    result = {}
+    if cips:
+        result["ips"] = cips
+    result["ports"] = ports
+
+    return result

salt/_modules/metalk8s_monitoring.py

@@ -17,7 +17,7 @@
 
 def __virtual__():
     if MISSING_DEPS:
-        error_msg = "Missing dependencies: {}".format(", ".join(MISSING_DEPS))
+        error_msg = f"Missing dependencies: {', '.join(MISSING_DEPS)}"
         return False, error_msg
 
     return __virtualname__
@@ -33,7 +33,7 @@ def add_silence(
     time_format="%Y-%m-%dT%H:%M:%S",
     author="",
     comment="",
-    **kwargs
+    **kwargs,
 ):
     """Add a new silence in Alertmanager.
 
@@ -106,9 +106,7 @@ def delete_silence(silence_id, **kwargs):
         salt-call metalk8s_monitoring.delete_silence \
             64d84a9e-cc6e-41ce-83ff-e84771ff6872
     """
-    _requests_alertmanager_api(
-        "api/v1/silence/{}".format(silence_id), "DELETE", **kwargs
-    )
+    _requests_alertmanager_api(f"api/v1/silence/{silence_id}", "DELETE", **kwargs)
 
 
 def get_silences(state=None, **kwargs):
@@ -164,42 +162,46 @@ def get_alerts(state=None, **kwargs):
 
 
 def _requests_alertmanager_api(route, method="GET", **kwargs):
-    endpoint = __salt__["metalk8s_kubernetes.get_service_endpoints"](
+    endpoints = __salt__["metalk8s_kubernetes.get_service_ips_and_ports"](
         "prometheus-operator-alertmanager",
         "metalk8s-monitoring",
-        kwargs.pop("kubeconfig", None),
+        **kwargs,
     )
 
-    url = "http://{}:{}/{}".format(
-        endpoint["ip"],
-        endpoint["ports"]["web"],
-        route,
-    )
+    try:
+        ip = endpoints["ips"][0]
+        port = endpoints["ports"]["http-web"]
+        url = f"http://{ip}:{port}/{route}"
+    except (IndexError, KeyError) as exc:
+        raise CommandExecutionError(
+            "Unable to get proper Alertmanager API endpoint: "
+            f"Available endpoints: {endpoints}"
+        ) from exc
 
     try:
         session = __utils__["metalk8s.requests_retry_session"]()
         response = session.request(method, url, **kwargs)
     except Exception as exc:
         raise CommandExecutionError(
-            "Unable to query Alertmanager API on {}".format(url)
+            f"Unable to query Alertmanager API on {url}"
         ) from exc
 
     try:
         json = response.json()
     except ValueError as exc:
         if response.status_code != requests.codes.ok:
-            error = "Received HTTP code {} when querying Alertmanager API on {}".format(
-                response.status_code, url
+            error = (
+                f"Received HTTP code {response.status_code} when "
+                f"querying Alertmanager API on {url}"
             )
         else:
             error = (
-                "Malformed response returned from Alertmanager API: {!s}: {}".format(
-                    exc, response.text
-                )
+                "Malformed response returned from Alertmanager API: "
+                f"{exc}: {response.text}"
             )
         raise CommandExecutionError(error) from exc
 
     if json["status"] == "error":
-        raise CommandExecutionError("{}: {}".format(json["errorType"], json["error"]))
+        raise CommandExecutionError(f"{json['errorType']}: {json['error']}")
 
     return json.get("data")

salt/_pillar/metalk8s_endpoints.py

@@ -38,7 +38,7 @@ def ext_pillar(minion_id, pillar, kubeconfig):  # pylint: disable=unused-argumen
                 try:
                     service_endpoints = __salt__[
                         "metalk8s_kubernetes.get_service_endpoints"
-                    ](service, namespace, kubeconfig)
+                    ](service, namespace, kubeconfig=kubeconfig)
                 except CommandExecutionError as exc:
                     errors.append(str(exc))
 

salt/tests/unit/modules/files/test_metalk8s_kubernetes_utils.yaml

@@ -278,3 +278,132 @@ get_service_endpoints:
       subsets: null
     raises: True
     result: "Unable to get kubernetes endpoints for my_service in namespace my_namespace"
+
+get_service_ips_and_ports:
+  # Multiples IPs and multiple ports
+  - obj:
+      spec:
+        clusterIPs:
+          - 10.0.0.1
+          - 10.0.0.2
+        clusterIP: 10.0.0.1
+        ports:
+          - name: http-web
+            port: 1234
+          - name: reloader-web
+            port: 1235
+    result:
+      ips:
+        - 10.0.0.1
+        - 10.0.0.2
+      ports:
+        http-web: 1234
+        reloader-web: 1235
+  # ClusterIP not in clusterIPs and multiple ports
+  - obj:
+      spec:
+        clusterIPs:
+          - 10.0.0.2
+          - 10.0.0.3
+        clusterIP: 10.0.0.1
+        ports:
+          - name: http-web
+            port: 1234
+          - name: reloader-web
+            port: 1235
+    result:
+      ips:
+        - 10.0.0.1
+        - 10.0.0.2
+        - 10.0.0.3
+      ports:
+        http-web: 1234
+        reloader-web: 1235
+  # Single IP and multiple ports
+  - obj:
+      spec:
+        clusterIPs:
+          - 10.0.0.1
+        clusterIP: 10.0.0.1
+        ports:
+          - name: http-web
+            port: 2345
+          - name: reloader-web
+            port: 2346
+    result:
+      ips:
+        - 10.0.0.1
+      ports:
+        http-web: 2345
+        reloader-web: 2346
+  # Only clusterIP and multiple ports
+  - obj:
+      spec:
+        clusterIP: 10.0.0.1
+        ports:
+          - name: http-web
+            port: 2345
+          - name: reloader-web
+            port: 2346
+    result:
+      ips:
+        - 10.0.0.1
+      ports:
+        http-web: 2345
+        reloader-web: 2346
+  # No IPs and multiple ports
+  - obj:
+      spec:
+        ports:
+          - name: http-web
+            port: 1234
+          - name: reloader-web
+            port: 1235
+    result:
+      ports:
+        http-web: 1234
+        reloader-web: 1235
+  # Named and unnamed ports
+  - obj:
+      spec:
+        ports:
+          - name: http-web
+            port: 1234
+          - port: 1235
+    result:
+      ports:
+        http-web: 1234
+        unnamed-0: 1235
+  # Unnamed ports
+  - obj:
+      spec:
+        ports:
+          - port: 1234
+          - port: 1235
+    result:
+      ports:
+        unnamed-0: 1234
+        unnamed-1: 1235
+  # No ports
+  - obj:
+      spec:
+        ports: []
+    result: "No ports for service my_service in namespace my_namespace"
+    raises: True
+  # Empty spec
+  - obj:
+      spec: {}
+    result: "No ports for service my_service in namespace my_namespace"
+    raises: True
+  # Empty object returned
+  - obj: {}
+    result: "Unable to get kubernetes service my_service in namespace my_namespace"
+    raises: True
+  # Object is False
+  - obj: False
+    result: "Unable to get kubernetes service my_service in namespace my_namespace"
+    raises: True
+  # No object returned
+  - obj: null
+    result: "Unable to get kubernetes service my_service in namespace my_namespace"
+    raises: True