Governance
Amazon CloudWatch
Monitoring service for metrics, logs, alarms, and dashboards across AWS resources and applications, including AI workloads.
Key points
- Collects metrics and logs and raises alarms when thresholds are crossed.
- Anomaly detection learns a metric's normal pattern and can alarm when the metric moves outside the expected band.
- Amazon Bedrock publishes runtime metrics such as invocations, latency, and token counts to CloudWatch.
- Bedrock model invocation logging can deliver logs to CloudWatch Logs.
- Dashboards give operational visibility across services.
When to use it
- Choose CloudWatch to monitor latency, errors, and usage of AI applications.
- Use anomaly detection when fixed thresholds are hard to set by hand.
- Use alarms to notify teams or trigger automated responses.
Exam tips
- CloudWatch monitors metrics, logs, and alarms; CloudTrail records who made which AWS API calls.
- AWS lists CloudWatch anomaly detection as an alternative to the retired Amazon Lookout for Metrics.
- Operational monitoring alone does not evaluate model fairness or answer quality.