e2e: fix 02 test failure

operator/internal/controller/opsModes.go

@@ -55,6 +55,7 @@ func (r *ElastiServiceReconciler) switchMode(ctx context.Context, req ctrl.Reque
 	return nil
 }
 
+// TODO: Add atomicity to this, all-or-nothing
 func (r *ElastiServiceReconciler) enableProxyMode(ctx context.Context, req ctrl.Request, es *v1alpha1.ElastiService) error {
 	targetNamespacedName := types.NamespacedName{
 		Name:      es.Spec.Service,
@@ -70,16 +71,16 @@ func (r *ElastiServiceReconciler) enableProxyMode(ctx context.Context, req ctrl.
 	}
 	r.Logger.Info("1. Checked and created private service", zap.String("public service", targetSVC.Name), zap.String("private service", PVTName))
 
+	if err = r.createOrUpdateEndpointsliceToResolver(ctx, targetSVC); err != nil {
+		return fmt.Errorf("failed to create or update endpointslice to resolver: %w ", err)
+	}
+	r.Logger.Info("2. Created or updated endpointslice to resolver", zap.String("service", targetSVC.Name))
+
 	// Check if Public Service is present, and has not changed from the values in CRDDirectory
 	if err := r.watchPublicService(ctx, es, req); err != nil {
 		return fmt.Errorf("failed to add watch on public service: %w", err)
 	}
-	r.Logger.Info("2. Added watch on public service", zap.String("service", targetSVC.Name))
-
-	if err = r.createOrUpdateEndpointsliceToResolver(ctx, targetSVC); err != nil {
-		return fmt.Errorf("failed to create or update endpointslice to resolver: %w ", err)
-	}
-	r.Logger.Info("3. Created or updated endpointslice to resolver", zap.String("service", targetSVC.Name))
+	r.Logger.Info("3. Added watch on public service", zap.String("service", targetSVC.Name))
 
 	return nil
 }

tests/e2e/manifest/send_traffic.sh

@@ -61,6 +61,8 @@ NC='\033[0m' # No Color
 
 # log_failure_details: Centralized function to log detailed debugging information on failure
 log_failure_details() {
+    start_time="${1}"
+
     echo "${RED}--- DETAILED FAILURE ANALYSIS ---${NC}"
 
     # Target EndpointSlice Status
@@ -69,24 +71,24 @@ log_failure_details() {
 
     # Ingress Logs
     echo "${CYAN}  Logs from ingress:${NC}"
-    kubectl logs -n istio-system deployments/istiod --tail=15 | sed 's/^/    /' || echo "${YELLOW}    - Could not retrieve resolver logs${NC}"
+    kubectl logs -n istio-system deployments/istiod --since-time="${start_time}" | sed 's/^/    /' || echo "${YELLOW}    - Could not retrieve ingress logs${NC}"
 
     # Resolver Logs
     echo "${CYAN}  Logs from elasti-resolver:${NC}"
-    kubectl logs -n elasti services/elasti-resolver-service --tail=20 | sed 's/^/    /' || echo "${YELLOW}    - Could not retrieve resolver logs${NC}"
+    kubectl logs -n elasti services/elasti-resolver-service --since-time="${start_time}" --all-pods=true | sed 's/^/    /' || echo "${YELLOW}    - Could not retrieve resolver logs${NC}"
 
     # Controller Logs
     echo "${CYAN}  Logs from elasti-controller:${NC}"
-    kubectl logs -n elasti services/elasti-operator-controller-service --tail=20 | sed 's/^/    /' || echo "${YELLOW}    - Could not retrieve controller logs${NC}"
+    kubectl logs -n elasti services/elasti-operator-controller-service --since-time="${start_time}" | sed 's/^/    /' || echo "${YELLOW}    - Could not retrieve controller logs${NC}"
 
     # Target Logs
     echo "${CYAN}  Logs from target (${TARGET_RESOURCE}/${TARGET_NAME}):${NC}"
-    kubectl logs -n "$TARGET_NAMESPACE" "${TARGET_RESOURCE}/${TARGET_NAME}"  --tail=120 | sed 's/^/    /' || echo "${YELLOW}    - Could not retrieve target pod logs${NC}"
+    kubectl logs -n "$TARGET_NAMESPACE" "${TARGET_RESOURCE}/${TARGET_NAME}"  --since-time="${start_time}" | sed 's/^/    /' || echo "${YELLOW}    - Could not retrieve target pod logs${NC}"
 
     # Verbose Curl Request
     echo "${CYAN}  Attempting verbose request for more details...${NC}"
     kubectl exec -n "$CURL_NAMESPACE" "$CURL_POD_NAME" -- curl --max-time 10 -v -s "$URL" 2>&1 | head -20 | sed 's/^/    /' || echo "${YELLOW}  - Verbose request failed${NC}"
-    
+
     echo "${RED}-----------------------------------${NC}"
 }
 
@@ -121,25 +123,32 @@ echo "${CYAN}--- Starting Traffic Test ---${NC}"
 failure_count=0
 
 for i in $(seq 1 $MAX_RETRIES); do
+    if [ $i -lt 2 ]; then
+        sleep 5
+    fi
+
     echo "\n${CYAN}--- Request $i/$MAX_RETRIES ---${NC}"
     echo "  ${CYAN}Time:${NC} $(date)"
 
-    echo "  ${CYAN}Executing: kubectl exec -n $CURL_NAMESPACE $CURL_POD_NAME -- curl --max-time $TIMEOUT -s -o /dev/null -w \"%{http_code}\" \"$URL\""
+    echo "  ${CYAN}Executing: kubectl exec -n $CURL_NAMESPACE $CURL_POD_NAME -- curl --max-time $TIMEOUT --retry 2 --retry-delay 1 -s -o /dev/null -w \"%{http_code}\" \"$URL\""
     start_time=$(date +%s)
+    start_time_rfc=$(date -u +"%Y-%m-%dT%H:%M:%SZ")
     code=$(kubectl exec -n "$CURL_NAMESPACE" "$CURL_POD_NAME" -- curl \
         --max-time "$TIMEOUT" \
+        --retry 2 \
+        --retry-delay 1 \
         -s \
         -o /dev/null \
         -w "%{http_code}" \
         "$URL" 2>&1)
     result=$?
     end_time=$(date +%s)
     duration=$((end_time - start_time))
-    
+
     echo "  ${CYAN}Curl exit code:${NC} $result"
     echo "  ${CYAN}HTTP status code:${NC} $code"
     echo "  ${CYAN}Request duration:${NC} ${duration}s"
-    
+
     # Detailed error analysis
     if [ "$result" != "0" ]; then
         echo "ERROR: Curl command failed with exit code $result"
@@ -153,11 +162,11 @@ for i in $(seq 1 $MAX_RETRIES); do
             *) echo "  - Unknown curl error" ;;
         esac
 
-        log_failure_details
+        log_failure_details "${start_time_rfc}"
         failure_count=$((failure_count + 1))
         continue
     fi
-    
+
     if [ "$code" != "200" ]; then
         echo "ERROR: Expected HTTP 200, got $code"
         case $code in
@@ -170,13 +179,13 @@ for i in $(seq 1 $MAX_RETRIES); do
             *) echo "  - HTTP error response" ;;
         esac
 
-        log_failure_details
+        log_failure_details "${start_time_rfc}"
         failure_count=$((failure_count + 1))
         continue
     fi
 
     echo "${GREEN}SUCCESS: Request $i completed successfully (HTTP $code in ${duration}s)${NC}"
-    
+
     if [ $i -lt $MAX_RETRIES ]; then
         sleep 1
     fi

tests/e2e/tests/02-enable-serve-via-traffic/02-assert.yaml

@@ -0,0 +1,6 @@
+apiVersion: elasti.truefoundry.com/v1alpha1
+kind: ElastiService
+metadata:
+  name: target-elastiservice
+status:
+  mode: proxy

tests/e2e/tests/02-enable-serve-via-traffic/04-assert.yaml → tests/e2e/tests/02-enable-serve-via-traffic/05-assert.yaml

@@ -4,8 +4,3 @@ metadata:
   name: target-elastiservice
 status:
   mode: serve
----
-apiVersion: kuttl.dev/v1beta1
-kind: TestAssert
-# give operator time — 60s is reasonable here
-timeout: 60

tests/e2e/tests/02-enable-serve-via-traffic/07-assert.yaml

@@ -0,0 +1,6 @@
+apiVersion: elasti.truefoundry.com/v1alpha1
+kind: ElastiService
+metadata:
+  name: target-elastiservice
+status:
+  mode: proxy

tests/e2e/tests/02-enable-serve-via-traffic/08-assert.yaml → tests/e2e/tests/02-enable-serve-via-traffic/10-assert.yaml

@@ -4,8 +4,3 @@ metadata:
   name: target-elastiservice
 status:
   mode: serve
----
-apiVersion: kuttl.dev/v1beta1
-kind: TestAssert
-# give operator time — 60s is reasonable here
-timeout: 60

tests/e2e/tests/02-enable-serve-via-traffic/12-assert.yaml

@@ -3,9 +3,4 @@ kind: ElastiService
 metadata:
   name: target-elastiservice
 status:
-  mode: serve
----
-apiVersion: kuttl.dev/v1beta1
-kind: TestAssert
-# give operator time — 60s is reasonable here
-timeout: 60
+  mode: proxy

tests/e2e/tests/02-enable-serve-via-traffic/16-assert.yaml → tests/e2e/tests/02-enable-serve-via-traffic/15-assert.yaml

@@ -4,8 +4,3 @@ metadata:
   name: target-elastiservice
 status:
   mode: serve
----
-apiVersion: kuttl.dev/v1beta1
-kind: TestAssert
-# give operator time — 60s is reasonable here
-timeout: 60

tests/e2e/tests/02-enable-serve-via-traffic/17-assert.yaml

@@ -0,0 +1,6 @@
+apiVersion: elasti.truefoundry.com/v1alpha1
+kind: ElastiService
+metadata:
+  name: target-elastiservice
+status:
+  mode: proxy

tests/e2e/tests/02-enable-serve-via-traffic/20-assert.yaml

@@ -0,0 +1,6 @@
+apiVersion: elasti.truefoundry.com/v1alpha1
+kind: ElastiService
+metadata:
+  name: target-elastiservice
+status:
+  mode: serve

tests/e2e/tests/03-enable-serve-via-manual/03-assert.yaml

@@ -4,8 +4,3 @@ metadata:
   name: target-elastiservice
 status:
   mode: serve
----
-apiVersion: kuttl.dev/v1beta1
-kind: TestAssert
-# give operator time — 60s is reasonable here
-timeout: 60

tests/e2e/tests/06-resolver-endpoint-sync/06-scale-to-1-resolver.yaml

@@ -0,0 +1,5 @@
+apiVersion: kuttl.dev/v1beta1
+kind: TestStep
+commands:
+  - command: kubectl scale deployment/elasti-resolver -n elasti --replicas=1
+  - command: kubectl rollout status deployment/elasti-resolver -n elasti --timeout=400s

tests/e2e/tests/07-scale-down-with-filter/05-assert.yaml

@@ -4,8 +4,3 @@ metadata:
   name: target-elastiservice
 status:
   mode: proxy
----
-apiVersion: kuttl.dev/v1beta1
-kind: TestAssert
-# give operator time — 60s is reasonable here
-timeout: 60

tests/e2e/tests/08-resolver-multiple-endpoints/06-assert-mode-proxy.yaml

@@ -4,8 +4,3 @@ metadata:
   name: target-elastiservice
 status:
   mode: proxy
----
-apiVersion: kuttl.dev/v1beta1
-kind: TestAssert
-# give operator time — 60s is reasonable here
-timeout: 60

tests/e2e/tests/08-resolver-multiple-endpoints/08-assert-mode-serve.yaml

@@ -4,8 +4,3 @@ metadata:
   name: target-elastiservice
 status:
   mode: serve
----
-apiVersion: kuttl.dev/v1beta1
-kind: TestAssert
-# give operator time — 60s is reasonable here
-timeout: 60

tests/e2e/tests/09-statefulset-serve-via-traffic/02-assert.yaml

@@ -4,8 +4,3 @@ metadata:
   name: target-elastiservice
 status:
   mode: proxy
----
-apiVersion: kuttl.dev/v1beta1
-kind: TestAssert
-# give operator time — 60s is reasonable here
-timeout: 60

tests/e2e/tests/09-statefulset-serve-via-traffic/05-assert.yaml

@@ -4,8 +4,3 @@ metadata:
   name: target-elastiservice
 status:
   mode: serve
----
-apiVersion: kuttl.dev/v1beta1
-kind: TestAssert
-# give operator time — 60s is reasonable here
-timeout: 60