Merge pull request #498 from linode/fix-flaky-udp-e2etest

[test][fix] reduce flakiness with e2e test

Author: komer3

Makefile

@@ -165,7 +165,7 @@ generate-capl-cluster-manifests:
 create-capl-cluster:
 	# Create a CAPL cluster with updated CCM and wait for it to be ready
 	kubectl apply -f $(MANIFEST_NAME).yaml
-	kubectl wait --for=condition=ControlPlaneReady cluster/$(CLUSTER_NAME) --timeout=600s || (kubectl get cluster -o yaml; kubectl get linodecluster -o yaml; kubectl get linodemachines -o yaml)
+	kubectl wait --for=condition=ControlPlaneReady cluster/$(CLUSTER_NAME) --timeout=600s || (kubectl get cluster -o yaml; kubectl get linodecluster -o yaml; kubectl get linodemachines -o yaml; kubectl logs -n capl-system deployments/capl-controller-manager --tail=50)
 	kubectl wait --for=condition=NodeHealthy=true machines -l cluster.x-k8s.io/cluster-name=$(CLUSTER_NAME) --timeout=900s
 	clusterctl get kubeconfig $(CLUSTER_NAME) > $(KUBECONFIG_PATH)
 	KUBECONFIG=$(KUBECONFIG_PATH) kubectl wait --for=condition=Ready nodes --all --timeout=600s

e2e/test/lb-with-udp-ports-stickiness/chainsaw-test.yaml

@@ -47,12 +47,33 @@ spec:
               echo "Nodebalancer config found, updating config stickiness"
 
               kubectl annotate --overwrite svc svc-test -n $NAMESPACE service.beta.kubernetes.io/linode-loadbalancer-default-stickiness=source_ip
-              sleep 5s
-
-              echo "Verifying that stickiness is set to source_ip"
-              nbconfig=$(LINODE_TOKEN=$LINODE_TOKEN NBID=$nbid ../scripts/get-nb-config.sh)
-              stickiness=$(echo $nbconfig | jq -r '.stickiness')
-              echo "stickiness is $stickiness"
+              
+              echo "Waiting for stickiness to be updated to source_ip..."
+              max_attempts=12
+              for i in $(seq 1 $max_attempts); do
+                echo "Attempt $i/$max_attempts: Checking stickiness..."
+                nbconfig=$(LINODE_TOKEN=$LINODE_TOKEN NBID=$nbid ../scripts/get-nb-config.sh)
+                stickiness=$(echo $nbconfig | jq -r '.stickiness')
+                echo "Current stickiness: $stickiness"
+                
+                if [[ "$stickiness" == "source_ip" ]]; then
+                  echo "Stickiness successfully updated to source_ip"
+                  break
+                fi
+                
+                if [[ $i -eq $max_attempts ]]; then
+                  echo "Timeout: Stickiness was not updated to source_ip after $max_attempts attempts"
+                  exit 1
+                fi
+                
+                echo "Waiting 10 seconds before next check..."
+                sleep 10
+              done
             check:
               ($error == null): true
-              (contains($stdout, 'stickiness is source_ip')): true
+              (contains($stdout, 'Stickiness successfully updated to source_ip')): true
+      catch:
+        - script:
+            content: |
+              echo "Test failed. Fetching CCM logs..."
+              kubectl logs -n kube-system daemonsets/ccm-linode | grep "lb-with-udp-ports-stickiness"

e2e/test/lb-with-udp-ports-stickiness/create-pods-services.yaml

@@ -35,6 +35,8 @@ metadata:
   name: svc-test
   labels:
     app: lb-with-udp-ports-stickiness
+  annotations:
+    service.beta.kubernetes.io/linode-loadbalancer-default-algorithm: ring_hash
 spec:
   type: LoadBalancer
   selector: