Monitoring
By default, the project will create an Amazon CloudWatch Dashboard. This dashboard is created using the library cdk-monitoring-constructs and it is recommended to update the metrics you track based on your project needs.
The dashboard is created in lib/monitoring/index.ts
During the configuration setup, the advanced monitoring setting will enable the following:
- AWS X-Ray will collect traces that can be viewed by opening the Trace Map from the CloudWatch console.
- Generate a custom metric per LLM model used (Bedrock only) allowing you to track the token usage. These metrics are available in the dashboard and are created using Cloudwatch filters.
- Create sample CloudWatch Alarms.
Cost: Be mindful of the costs associated with AWS resources, as enabling advanced motoring is adding custom metrics, alarms and AWS X-Ray traces.
Recommended changes (Advanced monitoring)
Receive alerts
The default setup is monitoring key resources such as the error rates of the APIs or the dead letter queues (if not empty, the processing of LLM requests failed). All these alarms can be viewed from the Amazon CloudWatch console.
The alarms are part of a composite alarm which will send an event to an SNS Topic if any alarm is active.
To receive notifications, add a subscription (manually or in lib/monitoring/index.ts
) to the topic listed in the CloudFormation output CompositeAlarmTopicOutput
(When deploying).
Update alarms and their thresholds
The alarms listed in lib/monitoring/index.ts
are examples and they should be updated to match your project needs. Please refer to the following project describing how to add/update the alarms.
Review AWS X-Ray sampling
Consider updating the default AWS X-Ray sampling rules to define the amount of data recorded