Merge pull request #234 from jsitu777/release-v1.2.0

[Cherry-pick] Enable Cloudwatch metrics for Canary Runs

Author: jsitu777

test/canary/canary.buildspec.yaml

@@ -19,8 +19,15 @@ phases:
     commands:
       # Run tests
       - docker run --name ack-canary $(env | cut -f1 -d= | sed 's/^/-e /') --mount type=bind,source="$(pwd)/",target="/${SERVICE}-controller/" ${ECR_CACHE_URI}:latest
-      
+  
+  post_build:
+    commands:
+      - docker cp ack-canary:/sagemaker-controller/test/canary/integration_tests.xml /tmp/results.xml || true
       # Push test image to cache ECR repo
       - docker push ${ECR_CACHE_URI}:latest || true
 
-      
+reports:
+  IntegrationTestReport:
+    files:
+      - "results.xml"
+    base-directory: "/tmp"   

test/canary/scripts/push_stats_to_cloudwatch.py

@@ -0,0 +1,102 @@
+import boto3
+from datetime import datetime
+import xml.etree.ElementTree as ET
+import os
+
+
+xml_path = "../integration_tests.xml"
+
+def readXML_and_publish_metrics_to_cw():
+    if os.path.isfile(xml_path):
+        tree = ET.parse(xml_path)
+        testsuite = tree.find("testsuite")
+        failures = testsuite.attrib["failures"]
+        tests = testsuite.attrib["tests"]
+        successes = int(tests) - int(failures)
+        success_rate = (successes/int(tests))*100
+    else:
+        print("f{xml_path} does not exists.")
+        print(os.getcwd())
+        failures = 0
+        successes = 0
+        tests = 0
+        success_rate = 0
+
+    timestamp = datetime.now().strftime("%Y-%m-%dT%H:%M:%S")
+
+    print(f"Failures: {failures}")
+    print(f"Total tests: {tests}")
+    print(f"Success: {successes}")
+
+    # push to cloudwatch
+    cw_client = boto3.client("cloudwatch")
+    project_name = os.getenv("PROJECT_NAME")
+
+    # Define the metric data
+    metric_data = [
+        {
+            "MetricName": "failures",
+            "Timestamp": timestamp,
+            "Dimensions": [
+                {"Name": "CodeBuild Project Name", "Value": project_name},
+            ],
+            "Value": int(failures),
+            "Unit": "Count",
+        },
+        {
+            "MetricName": "total_tests",
+            "Timestamp": timestamp,
+            "Dimensions": [
+                {"Name": "CodeBuild Project Name", "Value": project_name},
+            ],
+            "Value": int(tests),
+            "Unit": "Count",
+        },
+        {
+            "MetricName": "successes",
+            "Timestamp": timestamp,
+            "Dimensions": [
+                {"Name": "CodeBuild Project Name", "Value": project_name},
+            ],
+            "Value": int(successes),
+            "Unit": "Count",
+        },
+        {
+            "MetricName": "success_rate",
+            "Timestamp": timestamp,
+            "Dimensions": [
+                {"Name": "CodeBuild Project Name", "Value": project_name},
+            ],
+            "Value": int(success_rate),
+            "Unit": "Percent",
+        },
+    ]
+
+    # Use the put_metric_data method to push the metric data to CloudWatch
+    try:
+        response = cw_client.put_metric_data(
+            Namespace="Canary_Metrics", MetricData=metric_data
+        )
+        if response["ResponseMetadata"]["HTTPStatusCode"] == 200:
+            print("Successfully pushed data to CloudWatch")
+            # return 200 status code if successful
+            return 200
+        else:
+            # raise exception if the status code is not 200
+            raise Exception(
+                "Unexpected response status code: {}".format(
+                    response["ResponseMetadata"]["HTTPStatusCode"]
+                )
+            )
+    except Exception as e:
+        print("Error pushing data to CloudWatch: {}".format(e))
+        # raise exception if there was an error pushing data to CloudWatch
+        raise
+
+
+def main():
+    readXML_and_publish_metrics_to_cw()
+
+
+if __name__ == "__main__":
+    main()

test/canary/scripts/run_test.sh

@@ -28,6 +28,11 @@ function print_controller_logs() {
 }
 
 function cleanup {
+  #push to metrics to cloudwatch
+  echo "Pushing Codebuild stats to Cloudwatch."
+  cd $SCRIPTS_DIR
+  python push_stats_to_cloudwatch.py
+
   echo "Cleaning up resources"
   set +e
   kubectl delete monitoringschedules --all
@@ -66,6 +71,7 @@ function cleanup {
 }
 trap cleanup EXIT
 
+
 # Update kubeconfig
 aws --region $CLUSTER_REGION eks update-kubeconfig --name $CLUSTER_NAME
 
@@ -87,7 +93,7 @@ pushd $E2E_DIR
 
   # run tests
   echo "Run Tests"
-  pytest_args=( -n 15 --dist loadfile --log-cli-level INFO )
+  pytest_args=( -n 15 --dist loadfile --log-cli-level INFO --junitxml ../canary/integration_tests.xml)
   if [[ $SERVICE_REGION =~ ^(eu-north-1|eu-west-3)$  ]]; then
     # If select_regions_1 true we run the notebook_instance test
     pytest_args+=(-m "canary or select_regions_1")