CloudWatch Concepts

CloudWatch Concepts

Namespaces

• A namespace is a container for metrics, isolating metrics from different applications.

• AWS services follow a convention: AWS/service (e.g., AWS/EC2).

Metrics

• Metrics represent a time-ordered set of data points (e.g., EC2 instance CPU usage).

• Metrics are unique per Region and expire after 15 months without new data.

• Custom metrics can have high-resolution (1 second) or standard-resolution (1 minute).

• Metrics have dimensions (name/value pairs) that identify specific metrics (e.g., InstanceId for EC2).

Time Stamps

• Each metric data point must include a time stamp in UTC.

• CloudWatch can accept time stamps up to 2 weeks in the past or 2 hours into the future.

Retention

• Retention depends on data resolution:

  • < 1 minute: Available for 3 hours

  • 1 minute: Available for 15 days

  • 5 minutes: Available for 63 days

  • 1 hour: Available for 15 months

Dimensions

• Dimensions help categorize metrics (e.g., Server=Prod, Domain=Frankfurt).

• CloudWatch aggregates metrics only for matching dimension sets.

Resolution

• Metrics can be:

  • Standard resolution: 1-minute granularity.

  • High resolution: 1-second granularity.

Statistics and Aggregation

• Statistics summarize metrics over specific time periods (e.g., Min, Max, Average).

• Percentiles (e.g., p95) help detect anomalies in data distribution.

Alarms

• Alarms monitor metrics and perform actions (e.g., Auto Scaling or SNS notifications).

• Alarms act only on sustained state changes.

• High-resolution alarms have periods of 10 or 30 seconds but come with additional costs.