MLS-C01 Question 259
Select 3A machine learning team deployed a model on Amazon SageMaker and noticed a sudden spike in prediction response times. They suspect unauthorized API calls might be affecting the endpoint's performance. How can the team use AWS CloudTrail and Amazon CloudWatch together to investigate this issue?
- A
Use AWS CloudTrail to track API calls made to the SageMaker endpoint and identify any unauthorized usage.
- B
Enable detailed monitoring in Amazon CloudWatch to visualize SageMaker endpoint latency metrics.
- C
Set up a CloudWatch alarm to monitor the SageMaker endpoint's invocation count and trigger an alert when it exceeds a threshold.
- D
Use AWS CloudTrail logs to identify the IAM role or user responsible for any anomalous API activity.
- E
Configure a CloudWatch dashboard to display real-time CPU and memory usage of the SageMaker endpoint.
Show answer and explanation
Correct answers: A, B, D
Explanation
To investigate the sudden spike in latency, the team can use AWS CloudTrail to track API calls and identify any unauthorized access. They can also enable detailed monitoring in Amazon CloudWatch to analyze latency metrics. Additionally, analyzing CloudTrail logs can help identify the IAM role or user responsible for any anomalous behavior. Combining these tools provides a comprehensive approach to diagnosing the issue.
- A. Correct.
Correct. AWS CloudTrail provides a history of API calls, including details such as the IAM user or role and the source IP address, which can help identify unauthorized access.
- B. Correct.
Correct. Detailed monitoring in Amazon CloudWatch provides latency metrics for SageMaker endpoints, which can help the team analyze the performance spike.
- C. Incorrect.
Incorrect. While CloudWatch alarms can monitor invocation counts, this does not directly help in identifying unauthorized API calls or latency spikes.
- D. Correct.
Correct. CloudTrail logs can be analyzed to pinpoint the specific IAM user or entity responsible for anomalous API behavior.
- E. Incorrect.
Incorrect. CloudWatch does not provide real-time CPU and memory metrics for SageMaker endpoints, as SageMaker abstracts such details.