Monitoring Rekognition with Amazon CloudWatch

With CloudWatch, you can get metrics for individual Rekognition operations or global Rekognition metrics for your account, You can use metrics to track the health of your Rekognition-based solution and set up alarms to notify you when one or more metrics fall outside a defined threshold. For example, you can see metrics for the number of server errors that have occurred, or metrics for the number of faces that have been detected. You can also see metrics for the number of times a specific Rekognition operation has succeeded. To see metrics, you can use Amazon CloudWatch, Amazon AWS Command Line Interface, or the CloudWatch API.

You can also see aggregated metrics, for a chosen period of time, by using the Rekognition console. For more information, see Exercise 4: See aggregated metrics (console).

Using CloudWatch metrics for Rekognition

To use metrics, you must specify the following information:

• The metric dimension, or no dimension. A dimension is a name-value pair that helps you to uniquely identify a metric. Rekognition has one dimension, named Operation. It provides metrics for a specific operation. If you do not specify a dimension, the metric is scoped to all Rekognition operations within your account.

• The metric name, such as UserErrorCount.

You can get monitoring data for Rekognition using the AWS Management Console, the AWS CLI, or the CloudWatch API. You can also use the CloudWatch API through one of the Amazon AWS Software Development Kits (SDKs) or the CloudWatch API tools. The console displays a series of graphs based on the raw data from the CloudWatch API. Depending on your needs, you might prefer to use either the graphs displayed in the console or retrieved from the API.

The following list shows some common uses for the metrics. These are suggestions to get you started, not a comprehensive list.

How Do I?	Relevant Metrics
How do I track the numbers of faces recognized?	Monitor the Sum statistic of the DetectedFaceCount metric.
How do I know if my application has reached the maximum number of requests per second?	Monitor the Sum statistic of the ThrottledCount metric.
How can I monitor the request errors?	Use the Sum statistic of the UserErrorCount metric.
How can I find the total number of requests?	Use the ResponseTime and Data Samples statistic of the ResponseTime metric. This includes any request that results in an error. If you want to see only successful operation calls, use the SuccessfulRequestCount metric.
How can I monitor the latency of Rekognition operation calls?	Use the ResponseTime metric.
How can I monitor how many times IndexFaces successfully added faces to Rekognition collections?	Monitor the Sum statistic with the SuccessfulRequestCount metric and IndexFaces operation. Use the Operation dimension to select the operation and metric.

You must have the appropriate CloudWatch permissions to monitor Rekognition with CloudWatch. For more information, see Authentication and Access Control for Amazon CloudWatch.

Access Rekognition metrics

The following examples show how to access Rekognition metrics using the CloudWatch console, the AWS CLI, and the CloudWatch API.

To view metrics (console)

1. Open the CloudWatch console at https://console.aws.amazon.com/cloudwatch/.

2. Choose Metrics, choose the All Metrics tab, and then choose Rekognition.

3. Choose Metrics with no dimensions, and then choose a metric.

   For example, choose the DetectedFace metric to measure how many faces have been detected.

4. Choose a value for the date range. The metric count displayed in the graph.

To view metrics successful DetectFaces operation calls have been made over a period of time (CLI).

• Open the AWS CLI and enter the following command:

  aws cloudwatch get-metric-statistics --metric-name SuccessfulRequestCount --start-time 2017-1-1T19:46:20 --end-time 2017-1-6T19:46:57 --period 3600 --namespace AWS/Rekognition --statistics Sum --dimensions Name=Operation,Value=DetectFaces --region us-west-2

  This example shows the successful DetectFaces operation calls made over a period of time. For more information, see get-metric-statistics.

To access metrics (CloudWatch API)

• Call GetMetricStatistics. For more information, see the Amazon CloudWatch API Reference.

Create an alarm

You can create a CloudWatch alarm that sends an Amazon Simple Notification Service (Amazon SNS) message when the alarm changes state. An alarm watches a single metric over a time period you specify, and performs one or more actions based on the value of the metric relative to a given threshold over a number of time periods. The action is a notification sent to an Amazon SNS topic or an Auto Scaling policy.

Alarms invoke actions for sustained state changes only. CloudWatch alarms do not invoke actions simply because they are in a particular state. The state must have changed and been maintained for a specified number of time periods.

To set an alarm (console)

1. Sign in to the AWS Management Console and open the CloudWatch console at https://console.aws.amazon.com/cloudwatch/.

2. Choose Create Alarm. This launches the Create Alarm Wizard.

3. In the Metrics with no dimensions metric list, choose Rekognition Metrics, and then choose a metric.

   For example, choose DetectedFaceCount to set an alarm for a maximum number of detected faces.

4. In the Time Range area, select a date range value that includes face detection operations that you have called. Choose Next

5. Fill in the Name and Description. For Whenever, choose >=, and enter a maximum value of your choice.

6. If you want CloudWatch to send you email when the alarm state is reached, for Whenever this alarm:, choose State is ALARM. To send alarms to an existing Amazon SNS topic, for Send notification to:, choose an existing SNS topic. To set the name and email addresses for a new email subscription list, choose Create topic CloudWatch saves the list and displays it in the field so you can use it to set future alarms.

   Note

   If you use Create topic to create a new Amazon SNS topic, the email addresses must be verified before the intended recipients receive notifications. Amazon SNS sends email only when the alarm enters an alarm state. If this alarm state change happens before the email addresses are verified, intended recipients do not receive a notification.

7. Preview the alarm in the Alarm Preview section. Choose Create Alarm.

To set an alarm (AWS CLI)

• Open the AWS CLI and enter the following command. Change value of the alarm-actions parameter to reference an Amazon SNS topic that you previously created.

  aws cloudwatch put-metric-alarm --alarm-name UserErrors --alarm-description "Alarm when more than 10 user errors occur" --metric-name UserErrorCount --namespace AWS/Rekognition --statistic Average --period 300 --threshold 10 --comparison-operator GreaterThanThreshold --evaluation-periods 2 --alarm-actions arn:aws:sns:us-west-2:111111111111:UserError --unit Count

  This example shows how to create an alarm for when more than 10 user errors occur within 5 minutes. For more information, see put-metric-alarm.

To set an alarm (CloudWatch API)

• Call PutMetricAlarm. For more information, see Amazon CloudWatch API Reference.

CloudWatch metrics for Rekognition

This section contains information about the Amazon CloudWatch metrics and the Operation dimension available for Amazon Rekognition.

You can also see an aggregate view of Rekognition metrics from the Rekognition console. For more information, see Exercise 4: See aggregated metrics (console).

CloudWatch metrics for Rekognition

The following table summarizes the Rekognition metrics.

Metric	Description
SuccessfulRequestCount	The number of successful requests. The response code range for a successful request is 200 to 299. Unit: Count Valid statistics: Sum,Average
ThrottledCount	The number of throttled requests. Rekognition throttles a request when it receives more requests than the limit of transactions per second set for your account. If the limit set for your account is frequently exceeded, you can request a limit increase. To request an increase, see AWS Service Limits. Unit: Count Valid statistics: Sum,Average
ResponseTime	The time in milliseconds for Rekognition to compute the response. Units: Count for Data Samples statistics Milliseconds for Average statistics Valid statistics: Data Samples,Average Note The ResponseTime metric is not included in the Rekognition metric pane.
DetectedFaceCount	The number of faces detected with the IndexFaces or DetectFaces operation. Unit: Count Valid statistics: Sum,Average
DetectedLabelCount	The number of labels detected with the DetectLabels operation. Unit: Count Valid statistics: Sum,Average
ServerErrorCount	The number of server errors. The response code range for a server error is 500 to 599. Unit: Count Valid statistics: Sum, Average
UserErrorCount	The number of user errors (invalid parameters, invalid image, no permission, etc). The response code range for a user error is 400 to 499. Unit: Count Valid statistics: Sum,Average
MinInferenceUnits	The minimum number of inference units specified during the StartProjectVersion request. Unit: Count Valid statistics: Average
MaxInferenceUnits	The maximum number of inference units specified during the StartProjectVersion request. Unit: Count Valid statistics: Average
DesiredInferenceUnits	The number of inference units to which Rekognition is scaling up or down. Unit: Count Valid statistics: Average
InServiceInferenceUnits	The number of inference units that the model is using. Unit: Count Valid statistics: Average It is recommended that you use the Average statistic to obtain the 1 minute average of how many instances are used.

CloudWatch metrics for Rekognition Streaming

Rekognition also has a second namespace used for streaming operations, "Rekognition Streaming". The following table summarizes the Rekognition Streaming metrics.

Metric	Description
SuccessfulRequestCount	The number of successful requests. The response code range for a successful request is 200 to 299. Unit: Count Valid statistics: Sum,Average
CallCount	The number of specified operations performed in your account. Valid statistics: Sum,Average
ThrottledCount	The number of throttled requests. Rekognition throttles a request when it receives more requests than the limit of transactions per second set for your account. If the limit set for your account is frequently exceeded, you can request a limit increase. To request an increase, see AWS Service Limits. Unit: Count Valid statistics: Sum,Average
ServerErrorCount	The number of server errors. The response code range for a server error is 500 to 599. Unit: Count Valid statistics: Sum, Average
UserErrorCount	The number of user errors (invalid parameters, invalid image, no permission, etc). The response code range for a user error is 400 to 499. Unit: Count Valid statistics: Sum,Average

CloudWatch dimension for Rekognition

To retrieve operation-specific metrics, use the Rekognition namespace and provide an operation dimension.

For more information about dimensions, see Dimensions in the Amazon CloudWatch User Guide.

CloudWatch dimension for Rekognition Custom Labels

The following table displays the CloudWatch dimensions available for use with Rekognition Custom Labels:

Dimension	Description
ProjectName	The name of the Rekognition Custom Labels project you created with CreateProject.
VersionName	The name of the Rekognition Custom Labels project version you created with CreateProjectVersion.

For more information about dimensions, see Dimensions in the Amazon CloudWatch User Guide.