|
| 1 | +# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved. |
| 2 | +# SPDX-License-Identifier: Apache-2.0 |
| 3 | + |
| 4 | +""" |
| 5 | +Purpose |
| 6 | +
|
| 7 | +Shows how to use Amazon CloudWatch to monitor infrastructure metrics and send |
| 8 | +alerts via Amazon SNS when thresholds are breached. |
| 9 | +
|
| 10 | +This scenario demonstrates: |
| 11 | +1. Creating an SNS topic and email subscription for alarm notifications |
| 12 | +2. Creating a CloudWatch alarm that monitors a custom metric |
| 13 | +3. Publishing metric data to trigger alarm state changes |
| 14 | +4. Retrieving alarm history to view state transitions |
| 15 | +5. Cleaning up all created resources |
| 16 | +""" |
| 17 | + |
| 18 | +import logging |
| 19 | +import sys |
| 20 | +import time |
| 21 | +from datetime import datetime, timezone |
| 22 | + |
| 23 | +import boto3 |
| 24 | +from botocore.exceptions import ClientError |
| 25 | + |
| 26 | +sys.path.append("../..") |
| 27 | +from demo_tools import question as q |
| 28 | + |
| 29 | +logger = logging.getLogger(__name__) |
| 30 | + |
| 31 | + |
| 32 | +# snippet-start:[python.example_code.sns.Scenario_CloudWatchSnsMonitoring] |
| 33 | +class CloudWatchSnsMonitoringScenario: |
| 34 | + """Runs an interactive scenario demonstrating CloudWatch monitoring with SNS alerts.""" |
| 35 | + |
| 36 | + def __init__(self, sns_resource, cloudwatch_client): |
| 37 | + """ |
| 38 | + :param sns_resource: A Boto3 SNS resource. |
| 39 | + :param cloudwatch_client: A Boto3 CloudWatch client. |
| 40 | + """ |
| 41 | + self.sns_resource = sns_resource |
| 42 | + self.cloudwatch_client = cloudwatch_client |
| 43 | + self.topic = None |
| 44 | + self.subscription = None |
| 45 | + self.alarm_name = None |
| 46 | + self.namespace = None |
| 47 | + self.metric_name = "ErrorCount" |
| 48 | + self.dashboard_name = None |
| 49 | + |
| 50 | + def run_scenario(self): |
| 51 | + """Runs the CloudWatch SNS monitoring scenario.""" |
| 52 | + print("-" * 88) |
| 53 | + print("Welcome to the CloudWatch Monitoring with SNS Alerts Scenario.") |
| 54 | + print("-" * 88) |
| 55 | + print( |
| 56 | + "This scenario demonstrates how to:\n" |
| 57 | + "1. Create an SNS topic for alarm notifications\n" |
| 58 | + "2. Create a CloudWatch alarm that monitors a custom metric\n" |
| 59 | + "3. Publish metric data to trigger the alarm\n" |
| 60 | + "4. Receive email notifications when the alarm state changes\n" |
| 61 | + ) |
| 62 | + |
| 63 | + try: |
| 64 | + self._setup_phase() |
| 65 | + self._publish_metrics_phase() |
| 66 | + self._demonstrate_alarm_phase() |
| 67 | + except Exception as e: |
| 68 | + logger.exception("Scenario failed: %s", e) |
| 69 | + print(f"\nThe scenario encountered an error: {e}") |
| 70 | + finally: |
| 71 | + self._cleanup_phase() |
| 72 | + |
| 73 | + def _setup_phase(self): |
| 74 | + """Setup phase: Create SNS topic, subscription, alarm, and dashboard.""" |
| 75 | + # snippet-start:[python.example_code.sns.CreateTopicForAlarm] |
| 76 | + print("\n" + "-" * 88) |
| 77 | + print("Setup Phase") |
| 78 | + print("-" * 88) |
| 79 | + |
| 80 | + email = q.ask("Enter an email address to receive alarm notifications: ", q.non_empty) |
| 81 | + |
| 82 | + print("\nCreating SNS topic...") |
| 83 | + self.topic = self.sns_resource.create_topic(Name="cloudwatch-alarms-topic") |
| 84 | + print(f"SNS topic created: {self.topic.arn}") |
| 85 | + |
| 86 | + print("Creating email subscription...") |
| 87 | + self.subscription = self.topic.subscribe(Protocol="email", Endpoint=email) |
| 88 | + print("Email subscription created. Please check your email and confirm the subscription.") |
| 89 | + # snippet-end:[python.example_code.sns.CreateTopicForAlarm] |
| 90 | + |
| 91 | + # snippet-start:[python.example_code.cloudwatch.PutMetricAlarmWithSns] |
| 92 | + self.alarm_name = q.ask("\nEnter a name for the CloudWatch alarm: ", q.non_empty) |
| 93 | + |
| 94 | + namespace_input = input("Enter a name for the custom metric namespace (default: CustomApp/Monitoring): ") |
| 95 | + self.namespace = namespace_input.strip() if namespace_input.strip() else "CustomApp/Monitoring" |
| 96 | + |
| 97 | + print(f"\nCreating CloudWatch alarm '{self.alarm_name}'...") |
| 98 | + print(f"Alarm will trigger when {self.metric_name} >= 10 for 1 evaluation period (1 minute).") |
| 99 | + |
| 100 | + self.cloudwatch_client.put_metric_alarm( |
| 101 | + AlarmName=self.alarm_name, |
| 102 | + ComparisonOperator="GreaterThanOrEqualToThreshold", |
| 103 | + EvaluationPeriods=1, |
| 104 | + MetricName=self.metric_name, |
| 105 | + Namespace=self.namespace, |
| 106 | + Period=60, |
| 107 | + Statistic="Average", |
| 108 | + Threshold=10.0, |
| 109 | + ActionsEnabled=True, |
| 110 | + AlarmActions=[self.topic.arn], |
| 111 | + AlarmDescription="Alarm when error count exceeds threshold", |
| 112 | + ) |
| 113 | + print("CloudWatch alarm created successfully.") |
| 114 | + # snippet-end:[python.example_code.cloudwatch.PutMetricAlarmWithSns] |
| 115 | + |
| 116 | + # snippet-start:[python.example_code.cloudwatch.PutDashboard] |
| 117 | + self.dashboard_name = "monitoring-dashboard" |
| 118 | + print(f"\nCreating CloudWatch dashboard '{self.dashboard_name}'...") |
| 119 | + |
| 120 | + dashboard_body = { |
| 121 | + "widgets": [ |
| 122 | + { |
| 123 | + "type": "metric", |
| 124 | + "properties": { |
| 125 | + "metrics": [[self.namespace, self.metric_name]], |
| 126 | + "period": 60, |
| 127 | + "stat": "Average", |
| 128 | + "region": self.cloudwatch_client.meta.region_name, |
| 129 | + "title": "Error Count Monitoring" |
| 130 | + } |
| 131 | + } |
| 132 | + ] |
| 133 | + } |
| 134 | + |
| 135 | + import json |
| 136 | + self.cloudwatch_client.put_dashboard( |
| 137 | + DashboardName=self.dashboard_name, |
| 138 | + DashboardBody=json.dumps(dashboard_body) |
| 139 | + ) |
| 140 | + |
| 141 | + region = self.cloudwatch_client.meta.region_name |
| 142 | + print(f"Dashboard '{self.dashboard_name}' created successfully.") |
| 143 | + print(f"View at: https://console.aws.amazon.com/cloudwatch/home?region={region}#dashboards:name={self.dashboard_name}") |
| 144 | + print("-" * 88) |
| 145 | + # snippet-end:[python.example_code.cloudwatch.PutDashboard] |
| 146 | + |
| 147 | + def _publish_metrics_phase(self): |
| 148 | + """Publish metric data to demonstrate alarm triggering.""" |
| 149 | + # snippet-start:[python.example_code.cloudwatch.PutMetricDataForAlarm] |
| 150 | + print("\n" + "-" * 88) |
| 151 | + print("Publishing Metric Data") |
| 152 | + print("-" * 88) |
| 153 | + |
| 154 | + print("Publishing normal metric data (ErrorCount = 5)...") |
| 155 | + self.cloudwatch_client.put_metric_data( |
| 156 | + Namespace=self.namespace, |
| 157 | + MetricData=[ |
| 158 | + { |
| 159 | + "MetricName": self.metric_name, |
| 160 | + "Value": 5.0, |
| 161 | + "Unit": "Count", |
| 162 | + "Timestamp": datetime.now(timezone.utc) |
| 163 | + } |
| 164 | + ] |
| 165 | + ) |
| 166 | + print("Metric data published successfully.") |
| 167 | + # snippet-end:[python.example_code.cloudwatch.PutMetricDataForAlarm] |
| 168 | + |
| 169 | + # snippet-start:[python.example_code.cloudwatch.DescribeAlarmsState] |
| 170 | + time.sleep(5) |
| 171 | + print("\nChecking alarm state...") |
| 172 | + alarms = self.cloudwatch_client.describe_alarms(AlarmNames=[self.alarm_name]) |
| 173 | + if alarms["MetricAlarms"]: |
| 174 | + alarm = alarms["MetricAlarms"][0] |
| 175 | + print(f"Current alarm state: {alarm['StateValue']}") |
| 176 | + print(f"Alarm reason: {alarm['StateReason']}") |
| 177 | + # snippet-end:[python.example_code.cloudwatch.DescribeAlarmsState] |
| 178 | + |
| 179 | + print("\nPublishing high metric data to trigger alarm (ErrorCount = 15)...") |
| 180 | + self.cloudwatch_client.put_metric_data( |
| 181 | + Namespace=self.namespace, |
| 182 | + MetricData=[ |
| 183 | + { |
| 184 | + "MetricName": self.metric_name, |
| 185 | + "Value": 15.0, |
| 186 | + "Unit": "Count", |
| 187 | + "Timestamp": datetime.now(timezone.utc) |
| 188 | + } |
| 189 | + ] |
| 190 | + ) |
| 191 | + print("Metric data published successfully.") |
| 192 | + |
| 193 | + print("\nWaiting for alarm state to change (this may take up to 1 minute)...") |
| 194 | + for _ in range(12): |
| 195 | + time.sleep(5) |
| 196 | + alarms = self.cloudwatch_client.describe_alarms(AlarmNames=[self.alarm_name]) |
| 197 | + if alarms["MetricAlarms"]: |
| 198 | + alarm = alarms["MetricAlarms"][0] |
| 199 | + if alarm["StateValue"] == "ALARM": |
| 200 | + print(f"Alarm state changed to: {alarm['StateValue']}") |
| 201 | + print(f"Alarm reason: {alarm['StateReason']}") |
| 202 | + print(f"\nAn email notification has been sent to your email address.") |
| 203 | + print("Check your email for the alarm notification.") |
| 204 | + break |
| 205 | + print("-" * 88) |
| 206 | + |
| 207 | + def _demonstrate_alarm_phase(self): |
| 208 | + """Retrieve and display alarm history.""" |
| 209 | + # snippet-start:[python.example_code.cloudwatch.DescribeAlarmHistory] |
| 210 | + print("\n" + "-" * 88) |
| 211 | + print("Alarm History") |
| 212 | + print("-" * 88) |
| 213 | + |
| 214 | + print("Retrieving alarm history...") |
| 215 | + history = self.cloudwatch_client.describe_alarm_history( |
| 216 | + AlarmName=self.alarm_name, |
| 217 | + HistoryItemType="StateUpdate", |
| 218 | + MaxRecords=5 |
| 219 | + ) |
| 220 | + |
| 221 | + if history["AlarmHistoryItems"]: |
| 222 | + print("\nRecent alarm state changes:") |
| 223 | + for i, item in enumerate(history["AlarmHistoryItems"], 1): |
| 224 | + timestamp = item["Timestamp"].strftime("%Y-%m-%dT%H:%M:%SZ") |
| 225 | + print(f"{i}. {timestamp}: {item['HistorySummary']}") |
| 226 | + else: |
| 227 | + print("No alarm history available yet.") |
| 228 | + print("-" * 88) |
| 229 | + # snippet-end:[python.example_code.cloudwatch.DescribeAlarmHistory] |
| 230 | + |
| 231 | + def _cleanup_phase(self): |
| 232 | + """Cleanup phase: Delete all created resources.""" |
| 233 | + # snippet-start:[python.example_code.cloudwatch.DeleteMonitoringResources] |
| 234 | + if not self.alarm_name: |
| 235 | + return |
| 236 | + |
| 237 | + print("\n" + "-" * 88) |
| 238 | + print("Cleanup") |
| 239 | + print("-" * 88) |
| 240 | + |
| 241 | + delete_resources = q.ask( |
| 242 | + "Delete all resources created by this scenario? (y/n) ", q.is_yesno |
| 243 | + ) |
| 244 | + |
| 245 | + if delete_resources: |
| 246 | + try: |
| 247 | + if self.dashboard_name: |
| 248 | + print("\nDeleting CloudWatch dashboard...") |
| 249 | + self.cloudwatch_client.delete_dashboards(DashboardNames=[self.dashboard_name]) |
| 250 | + print("Dashboard deleted successfully.") |
| 251 | + |
| 252 | + print("\nDeleting CloudWatch alarm...") |
| 253 | + self.cloudwatch_client.delete_alarms(AlarmNames=[self.alarm_name]) |
| 254 | + print("Alarm deleted successfully.") |
| 255 | + |
| 256 | + if self.subscription: |
| 257 | + print("\nUnsubscribing from SNS topic...") |
| 258 | + self.subscription.delete() |
| 259 | + print("Subscription removed.") |
| 260 | + |
| 261 | + if self.topic: |
| 262 | + print("\nDeleting SNS topic...") |
| 263 | + self.topic.delete() |
| 264 | + print("SNS topic deleted successfully.") |
| 265 | + |
| 266 | + print("\nAll resources cleaned up successfully.") |
| 267 | + except ClientError as e: |
| 268 | + print(f"Error during cleanup: {e}") |
| 269 | + else: |
| 270 | + print("\nResources will remain active.") |
| 271 | + print(f"SNS Topic ARN: {self.topic.arn if self.topic else 'N/A'}") |
| 272 | + print(f"Alarm Name: {self.alarm_name}") |
| 273 | + |
| 274 | + print("-" * 88) |
| 275 | + print("CloudWatch Monitoring with SNS Alerts scenario completed.") |
| 276 | + # snippet-end:[python.example_code.cloudwatch.DeleteMonitoringResources] |
| 277 | + |
| 278 | + |
| 279 | +# snippet-end:[python.example_code.sns.Scenario_CloudWatchSnsMonitoring] |
| 280 | + |
| 281 | +if __name__ == "__main__": |
| 282 | + logging.basicConfig(level=logging.INFO, format="%(levelname)s: %(message)s") |
| 283 | + |
| 284 | + scenario = CloudWatchSnsMonitoringScenario( |
| 285 | + boto3.resource("sns"), |
| 286 | + boto3.client("cloudwatch") |
| 287 | + ) |
| 288 | + scenario.run_scenario() |
0 commit comments