feat(#739): unsupported probes

  - refactor exceptions
  - refactor unit tests
  - add post event for on impair

Author: liquidiert

operator/gefyra/bridge/carrier/__init__.py

@@ -1,6 +1,7 @@
 import json
 from typing import Any, Dict, List, Optional
-from gefyra.utils import BridgeException, exec_command_pod
+from gefyra.bridge.exceptions import BridgeInstallException
+from gefyra.utils import exec_command_pod
 import kubernetes as k8s
 
 from gefyra.bridge.abstract import AbstractGefyraBridgeProvider
@@ -33,12 +34,7 @@ def __init__(
 
     def install(self, parameters: Optional[Dict[Any, Any]] = None):
         parameters = parameters or {}
-        try:
-            self._patch_pod_with_carrier(
-                handle_probes=parameters.get("handleProbes", True)
-            )
-        except BridgeException as be:
-            raise BridgeException from be
+        self._patch_pod_with_carrier(handle_probes=parameters.get("handleProbes", True))
 
     def _ensure_probes(self, container: k8s.client.V1Container) -> bool:
         probes = self._get_all_probes(container)
@@ -144,11 +140,9 @@ def _patch_pod_with_carrier(
                             self._get_all_probes(container),
                         )
                     ):
-                        self.logger.error(
-                            "Not all of the probes to be handled are currently"
-                            " supported by Gefyra"
+                        raise BridgeInstallException(
+                            message="Not all of the probes to be handled are currently supported by Gefyra"
                         )
-                        raise BridgeException()
                 if (
                     container.image
                     == f"{self.configuration.CARRIER_IMAGE}:{self.configuration.CARRIER_IMAGE_TAG}"
@@ -162,8 +156,8 @@ def _patch_pod_with_carrier(
                 container.image = f"{self.configuration.CARRIER_IMAGE}:{self.configuration.CARRIER_IMAGE_TAG}"
                 break
         else:
-            raise RuntimeError(
-                f"Could not found container {self.container} in Pod {self.pod}"
+            raise BridgeInstallException(
+                message=f"Could not found container {self.container} in Pod {self.pod}"
             )
         self.logger.info(
             f"Now patching Pod {self.pod}; container {self.container} with Carrier"

operator/gefyra/bridge/exceptions.py

@@ -0,0 +1,5 @@
+from gefyra.exceptions import BridgeException
+
+
+class BridgeInstallException(BridgeException):
+    pass

operator/gefyra/bridgestate.py

@@ -10,7 +10,8 @@
 from gefyra.base import GefyraStateObject, StateControllerMixin
 from gefyra.configuration import OperatorConfiguration
 
-from gefyra.utils import BridgeException
+from gefyra.bridge.exceptions import BridgeInstallException
+from gefyra.exceptions import BridgeException
 
 
 class GefyraBridgeObject(GefyraStateObject):
@@ -38,7 +39,7 @@ class GefyraBridge(StateMachine, StateControllerMixin):
 
     install = (
         requested.to(installing, on="_install_provider")
-        | installing.to(error)
+        | error.to(installing)
         | installing.to.itself(on="_wait_for_provider")
     )
     set_installed = (
@@ -124,8 +125,9 @@ def _install_provider(self):
         """
         try:
             self.bridge_provider.install()
-        except BridgeException:
-            self.send("impair")
+        except BridgeInstallException as be:
+            self.logger.debug(f"Encountered: {be}")
+            self.send("impair", exception=be)
 
     def _wait_for_provider(self):
         if not self.bridge_provider.ready():
@@ -193,3 +195,11 @@ def on_remove(self):
     def on_restore(self):
         self.bridge_provider.uninstall()
         self.send("terminate")
+
+    def on_impair(self, exception: BridgeException | None):
+        self.logger.error(f"Failed from {self.current_state}")
+        self.post_event(
+            reason=f"Failed from {self.current_state}",
+            message=exception.message,
+            _type="Warning",
+        )

operator/gefyra/exceptions.py

@@ -0,0 +1,5 @@
+class BridgeException(Exception):
+    message: str
+
+    def __init__(self, message: str):
+        self.message = message

operator/gefyra/utils.py

@@ -13,10 +13,6 @@
 logger = logging.getLogger("gefyra.utils")
 
 
-class BridgeException(Exception):
-    pass
-
-
 def get_label_selector(labels: dict[str, str]) -> str:
     return ",".join(["{0}={1}".format(*label) for label in list(labels.items())])
 

operator/tests/e2e/test_create_bridge.py

@@ -1,7 +1,5 @@
 import json
 import logging
-import subprocess
-import pytest
 from pytest_kubernetes.providers import AClusterManager
 from .utils import GefyraDockerClient
 
@@ -148,7 +146,7 @@ def test_c_fail_create_not_supported_bridges(
     k3d.wait("ns/demo-failing", "jsonpath='{.status.phase}'=Active")
     k3d.apply("tests/fixtures/demo_pods_not_supported.yaml")
     k3d.wait(
-        "pod/backend",
+        "pod/frontend",
         "condition=ready",
         namespace="demo-failing",
         timeout=60,
@@ -164,11 +162,9 @@ def test_c_fail_create_not_supported_bridges(
     )
 
     # applying the bridge shouldn't have worked
-    with pytest.raises(subprocess.TimeoutExpired):
-        k3d.wait(
-            "pod/frontend",
-            "jsonpath=.status.containerStatuses[0].image=docker.io/library/"
-            + carrier_image,
-            namespace="demo-failing",
-            timeout=60,
-        )
+    k3d.wait(
+        "pod/frontend",
+        "condition=ready",
+        namespace="demo-failing",
+        timeout=60,
+    )

operator/tests/fixtures/demo_pods_not_supported.yaml

@@ -1,23 +1,7 @@
 apiVersion: v1
-kind: Pod
+kind: Namespace
 metadata:
-  name: backend
-  namespace: demo-failing
-  labels:
-    app: backend
-spec:
-  securityContext:
-    runAsUser: 1000
-    runAsGroup: 1000
-    fsGroup: 1000
-  containers:
-    - name: backend
-      image: quay.io/gefyra/gefyra-demo-backend
-      imagePullPolicy: IfNotPresent
-      ports:
-        - name: web
-          containerPort: 5002
-          protocol: TCP
+  name: demo-failing
 ---
 apiVersion: v1
 kind: Pod
@@ -41,41 +25,14 @@ spec:
       livenessProbe:
         exec:
           command:
-            - cat
-            - /tmp/healthy
+            - ls
+            - /tmp
         initialDelaySeconds: 30
         periodSeconds: 10
       readinessProbe:
         exec:
           command:
-            - cat
-            - /tmp/ready
+            - ls
+            - /tmp
         initialDelaySeconds: 5
-        periodSeconds: 5
----
-apiVersion: v1
-kind: Service
-metadata:
-  name: backend
-  namespace: demo-failing
-spec:
-  selector:
-    app: backend
-  ports:
-  - protocol: TCP
-    port: 5002
-    targetPort: 5002
----
-apiVersion: v1
-kind: Service
-metadata:
-  name: frontend
-  namespace: demo-failing
-spec:
-  selector:
-    app: frontend
-  ports:
-  - protocol: "TCP"
-    port: 80
-    targetPort: 5003
-  type: LoadBalancer
+        periodSeconds: 5