Merge pull request #1010 from chandrams/missing_metrics

Included tests for validating the behavior when metrics section is missing

tests/scripts/common/common_functions.sh

@@ -232,16 +232,16 @@ function deploy_autotune() {
 			namespace="monitoring"
 		fi
 		echo "Namespace = $namespace"
-		service="autotune"
 		if [ ${target} == "crc" ]; then
 			service="kruize"
-		fi
-		autotune_pod=$(kubectl get pod -n ${namespace} | grep ${service} | cut -d " " -f1)
-		echo "autotune_pod = $autotune_pod"
-		if [ ${target} == "crc" ]; then
+			autotune_pod=$(kubectl get pod -n ${namespace} | grep ${service} | grep -v kruize-ui | cut -d " " -f1)
+			echo "autotune_pod = $autotune_pod"
 			echo "kubectl -n ${namespace} logs -f ${autotune_pod} > "${AUTOTUNE_POD_LOG}" 2>&1 &"
 			kubectl -n ${namespace} logs -f ${autotune_pod} > "${AUTOTUNE_POD_LOG}" 2>&1 &
 		else
+			service="autotune"
+			autotune_pod=$(kubectl get pod -n ${namespace} | grep ${service} | cut -d " " -f1)
+			echo "autotune_pod = $autotune_pod"
 			echo "kubectl -n ${namespace} logs -f ${autotune_pod} -c autotune > "${AUTOTUNE_POD_LOG}" 2>&1 &"
 			kubectl -n ${namespace} logs -f ${autotune_pod} -c autotune > "${AUTOTUNE_POD_LOG}" 2>&1 &
 		fi

tests/scripts/remote_monitoring_tests/json_files/missing_metrics_jsons/update_results_missing_metrics_all_containers.json

@@ -0,0 +1,25 @@
+[
+  {
+    "version": "1.0",
+    "experiment_name": "quarkus-resteasy-kruize-min-http-response-time-db",
+    "interval_start_time": "2022-01-23T18:25:43.511Z",
+    "interval_end_time": "2022-01-23T18:40:43.570Z",
+    "kubernetes_objects": [
+      {
+        "type": "deployment",
+        "name": "tfb-qrh-deployment",
+        "namespace": "default",
+        "containers": [
+	  {
+            "container_image_name": "kruize/tfb-db:1.15",
+            "container_name": "tfb-server-0"
+          },
+          {
+            "container_image_name": "kruize/tfb-qrh:1.13.2.F_et17",
+            "container_name": "tfb-server-1"
+          }
+        ]
+      }
+    ]
+  }
+]

tests/scripts/remote_monitoring_tests/json_files/missing_metrics_jsons/update_results_missing_metrics_single_container.json

@@ -0,0 +1,114 @@
+[
+  {
+    "version": "1.0",
+    "experiment_name": "quarkus-resteasy-kruize-min-http-response-time-db",
+    "interval_start_time": "2022-01-23T18:25:43.511Z",
+    "interval_end_time": "2022-01-23T18:40:43.570Z",
+    "kubernetes_objects": [
+      {
+        "type": "deployment",
+        "name": "tfb-qrh-deployment",
+        "namespace": "default",
+        "containers": [
+	  {
+            "container_image_name": "kruize/tfb-db:1.15",
+            "container_name": "tfb-server-0"
+          },
+          {
+            "container_image_name": "kruize/tfb-qrh:1.13.2.F_et17",
+            "container_name": "tfb-server-1",
+            "metrics": [
+              {
+                "name": "cpuRequest",
+                "results": {
+                  "aggregation_info": {
+                    "sum": 4.4,
+                    "avg": 1.1,
+                    "format": "cores"
+                  }
+                }
+              },
+              {
+                "name": "cpuLimit",
+                "results": {
+                  "aggregation_info": {
+                    "sum": 2.0,
+                    "avg": 0.5,
+                    "format": "cores"
+                  }
+                }
+              },
+              {
+                "name": "cpuUsage",
+                "results": {
+                  "aggregation_info": {
+                    "min": 0.14,
+                    "max": 0.84,
+                    "sum": 0.84,
+                    "avg": 0.12,
+                    "format": "cores"
+                  }
+                }
+              },
+              {
+                "name": "cpuThrottle",
+                "results": {
+                  "aggregation_info": {
+                    "sum": 0.19,
+                    "max": 0.09,
+                    "avg": 0.045,
+                    "format": "cores"
+                  }
+                }
+              },
+              {
+                "name": "memoryRequest",
+                "results": {
+                  "aggregation_info": {
+                    "sum": 250.85,
+                    "avg": 50.21,
+                    "format": "MiB"
+                  }
+                }
+              },
+              {
+                "name": "memoryLimit",
+                "results": {
+                  "aggregation_info": {
+                    "sum": 500,
+                    "avg": 100,
+                    "format": "MiB"
+                  }
+                }
+              },
+              {
+                "name": "memoryUsage",
+                "results": {
+                  "aggregation_info": {
+                    "min": 50.6,
+                    "max": 198.50,
+                    "sum": 198.50,
+                    "avg": 40.1,
+                    "format": "MiB"
+                  }
+                }
+              },
+              {
+                "name": "memoryRSS",
+                "results": {
+                  "aggregation_info": {
+                    "min": 50.6,
+                    "max": 123.6,
+                    "sum": 123.6,
+                    "avg": 31.91,
+                    "format": "MiB"
+                  }
+                }
+              }
+            ]
+          }
+        ]
+      }
+    ]
+  }
+]

tests/scripts/remote_monitoring_tests/rest_apis/test_update_results.py

@@ -16,6 +16,14 @@
     ("valid_plus_30s", "2022-01-23T18:41:13.511Z")
 ]
 
+missing_metrics = [
+    ("Missing_metrics_single_res_single_container", "../json_files/missing_metrics_jsons/update_results_missing_metrics_single_container.json", "Out of a total of 1 records, 1 failed to save", "kubernetes_objects : Metric data is not present for container : tfb-server-0 for experiment: quarkus-resteasy-kruize-min-http-response-time-db"),
+    ("Missing_metrics_single_res_all_containers", "../json_files/missing_metrics_jsons/update_results_missing_metrics_all_containers.json", "Out of a total of 1 records, 1 failed to save", "kubernetes_objects : Metric data is not present for container : tfb-server-0 for experiment: quarkus-resteasy-kruize-min-http-response-time-db. Metric data is not present for container : tfb-server-1 for experiment: quarkus-resteasy-kruize-min-http-response-time-db"),
+    ("Missing_metrics_bulk_res_single_container", "../json_files/missing_metrics_jsons/bulk_update_results_missing_metrics_single_container.json", "Out of a total of 100 records, 1 failed to save", "kubernetes_objects : Metric data is not present for container : tfb-server-1 for experiment: quarkus-resteasy-kruize-min-http-response-time-db"),
+    ("Missing_metrics_bulk_res_few_containers", "../json_files/missing_metrics_jsons/bulk_update_results_missing_metrics_few_containers.json", "Out of a total of 100 records, 2 failed to save", "kubernetes_objects : Metric data is not present for container : tfb-server-1 for experiment: quarkus-resteasy-kruize-min-http-response-time-db"),
+    ("Missing_metrics_bulk_res_few_containers_few_individual_metrics_missing", "../json_files/missing_metrics_jsons/bulk_update_results_missing_metrics_few_containers_few_individual_metrics_missing.json", "Out of a total of 100 records, 4 failed to save", "Metric data is not present for container")
+]
+
 
 @pytest.mark.negative
 @pytest.mark.parametrize(
@@ -131,6 +139,54 @@ def test_update_results_invalid_tests(test_name, expected_status_code, version,
     response = delete_experiment(input_json_file)
     print("delete exp = ", response.status_code)
 
+@pytest.mark.negative
+@pytest.mark.parametrize("test_name, result_json_file, expected_message, error_message", missing_metrics)
+def test_update_results_with_missing_metrics_section(test_name, result_json_file, expected_message, error_message, cluster_type):
+    """
+    Test Description: This test validates update results for a valid experiment
+                      by updating results with entire metrics section missing
+    """
+    input_json_file = "../json_files/create_exp.json"
+
+    form_kruize_url(cluster_type)
+    response = delete_experiment(input_json_file)
+    print("delete exp = ", response.status_code)
+
+    # Create experiment using the specified json
+    response = create_experiment(input_json_file)
+
+    data = response.json()
+    assert response.status_code == SUCCESS_STATUS_CODE
+    assert data['status'] == SUCCESS_STATUS
+    assert data['message'] == CREATE_EXP_SUCCESS_MSG
+
+    # Update results for the experiment
+    response = update_results(result_json_file)
+
+    data = response.json()
+    assert response.status_code == ERROR_STATUS_CODE
+    assert data['status'] == ERROR_STATUS
+    print("**************************")
+    print(data['message'])
+    print("**************************")
+
+    # add assertion of expected message
+    assert data['message'] == expected_message 
+
+    # add assertion of expected error message
+    msg_data=data['data']
+    for d in msg_data:
+        error_data=d["errors"]
+        for err in error_data:
+            actual_error_message = err["message"]
+            if test_name == "Missing_metrics_bulk_res_few_containers" and d["interval_end_time"] == "2023-04-13T23:29:20.982Z":
+                error_message = "kubernetes_objects : Metric data is not present for container : tfb-server-1 for experiment: quarkus-resteasy-kruize-min-http-response-time-db. Metric data is not present for container : tfb-server-0 for experiment: quarkus-resteasy-kruize-min-http-response-time-db"
+            if test_name == "Missing_metrics_bulk_res_few_containers_few_individual_metrics_missing" and d["interval_end_time"] == "2023-04-13T23:44:20.982Z" or d["interval_end_time"] == "2023-04-13T23:59:20.982Z":
+                error_message = "Performance profile: Missing one of the following mandatory parameters for experiment - quarkus-resteasy-kruize-min-http-response-time-db : [cpuUsage, memoryUsage, memoryRSS]"
+            assert error_message in actual_error_message
+
+    response = delete_experiment(input_json_file)
+    print("delete exp = ", response.status_code)
 
 @pytest.mark.sanity
 def test_update_valid_results_after_create_exp(cluster_type):