Manage detailed monitoring for your EC2 instances
Amazon CloudWatch provides two categories of monitoring: basic monitoring and detailed monitoring. By default, your instance is configured for basic monitoring. You can enable detailed monitoring to receive the default instance metrics in 1-minute periods. You can also associate metric categories with an instance to publish additional metrics. You can manage detailed monitoring and metric categories at launch or when the instance is running or stopped.
Enabling detailed monitoring on an instance does not affect the monitoring of its attached EBS volumes. For more information, see Amazon CloudWatch metrics for Amazon EBS.
The following table shows the differences between basic monitoring and detailed monitoring for your instances.
| Monitoring type | Description | Charges |
|---|---|---|
| Basic monitoring |
Status check metrics are available in 1-minute periods. All other default metrics are available in 5-minute periods. |
No charge. |
| Detailed monitoring |
Default instance metrics are available in 1-minute periods. Metric categories can add selected graphics processing unit (GPU) metrics in 1-minute periods and CPU utilization in 10-second periods. |
For CloudWatch Metrics, CloudWatch charges for each metric. For OpenTelemetry metrics,
CloudWatch charges by the volume of data ingested. Standard CloudWatch pricing applies. For more
information about pricing, see Paid tier in
Amazon CloudWatch
pricing |
Contents
Required permissions
To enable detailed monitoring or associate metric categories with an existing instance, you must have permission to use the MonitorInstances API action. To turn off detailed monitoring or remove metric category associations, you must have permission to use the UnmonitorInstances API action.
To associate metric categories when you launch an instance, your IAM policy must allow
ec2:RunInstances for the metric category resources.
To manage metric category resources, you need permission to use the
CreateMetricCategory, DescribeMetricCategories,
ModifyMetricCategory, and DeleteMetricCategory API actions. If
you add tags when you create a category, you also need permission to use the
CreateTags API action.
Use metric categories for additional metrics
A metric category is a Regional Amazon EC2 resource that contains a reusable set of
additional metric names. Each category has an ID that starts with mc- and can
contain up to 20 metrics. You can have up to 100 active metric categories in each Region,
and deleted categories do not count toward this quota. Associate up to 10 category IDs with
an instance that has detailed monitoring enabled.
Each metric name selects a publication format and period. The CloudWatch Metrics and OpenTelemetry names for the same measurement are separate selections. If you include both names, Amazon EC2 publishes both formats and charges apply to both. You can query the OpenTelemetry metrics using Prometheus Query Language (PromQL).
The following metrics are available for metric categories. For each metric's description, unit, and supported instance types, see Instance metrics and Accelerator metrics.
| Metric | Metric type | Period |
|---|---|---|
instance.cpu.utilization |
OpenTelemetry | 10 seconds |
instance.gpu.utilization |
OpenTelemetry | 1 minute |
instance.gpu.sm_utilization |
OpenTelemetry | 1 minute |
instance.gpu.memory_utilization |
OpenTelemetry | 1 minute |
instance.gpu.context_accumulated_time |
OpenTelemetry | 1 minute |
instance.gpu.temperature_percent_of_limit |
OpenTelemetry | 1 minute |
instance.gpu.memory_temperature_percent_of_limit |
OpenTelemetry | 1 minute |
instance.gpu.pcie_utilization |
OpenTelemetry | 1 minute |
instance.gpu.nvlink_utilization |
OpenTelemetry | 1 minute |
CPUUtilization10Sec |
CloudWatch | 10 seconds |
GPUUtilization |
CloudWatch | 1 minute |
GPUStreamingMultiprocessorUtilization |
CloudWatch | 1 minute |
GPUMemoryUtilization |
CloudWatch | 1 minute |
GPUContextAccumulatedTime |
CloudWatch | 1 minute |
GPUTemperaturePercentOfLimit |
CloudWatch | 1 minute |
GPUMemoryTemperaturePercentOfLimit |
CloudWatch | 1 minute |
GPUPCIeUtilization |
CloudWatch | 1 minute |
GPUNVLinkUtilization |
CloudWatch | 1 minute |
The GPU metrics are available only for supported accelerated computing instance types. If an instance does not support a selected metric, Amazon EC2 does not publish that metric for the instance.
Create a metric category
Metric category management
Use the following commands to manage your metric categories.
-
To list all metric categories, use
describe-metric-categorieswith no parameters:aws ec2 describe-metric-categories -
To replace the complete metric set in a category, use
modify-metric-category:aws ec2 modify-metric-category \ --metric-category-idmc-0123456789abcdef0\ --metrics GPUUtilization GPUMemoryUtilization GPUNVLinkUtilization -
To delete a category, use
delete-metric-category:aws ec2 delete-metric-category \ --metric-category-idmc-0123456789abcdef0
To update your running or stopped instances after
modify-metric-category, call monitor-instances with the
category ID.
Amazon EC2 also updates every associated instance asynchronously after you modify or delete a metric category. For more information, see Eventual consistency.
Enable detailed monitoring at launch
Use the following procedures to enable detailed monitoring at launch. By default, your instance uses basic monitoring.
Manage detailed monitoring
Use the following procedures to manage detailed monitoring for a running or stopped instance.