All Products
Search
Document Center

Auto Scaling:Event-triggered tasks of the system monitoring type

Last Updated:Apr 01, 2026

Event-triggered tasks of the system monitoring type use scaling-group-level metrics to determine when to scale. The metric value is the average across all instances in the scaling group and updates automatically as instances are added or removed.

This topic lists the metrics supported for each resource type (ECS instances, Elastic Container Instances, and ALB backend servers), along with each metric's unit, statistical period, and network requirements.

ECS metrics

The following tables list the metrics available for Elastic Compute Service (ECS) instances.

Metrics collected from ECS instances

Metric (Auto Scaling console)Metric (Auto Scaling API)UnitStatistical periodSupported network
(ECS) CPU UtilizationCpuUtilization%15 seconds, 1 minute, 2 minutes, 5 minutes, 15 minutesVPC and classic network
(ECS) ECS Concurrent ConnectionsConcurrentConnectionsVPC and classic network
(ECS) Internal Network Outbound TrafficIntranetTxKb/minVPC and classic network
(ECS) Internal Network Inbound TrafficIntranetRxKb/minVPC and classic network
(ECS) Public Network Outbound TrafficVpcInternetTxKb/minVPC and classic network
(ECS) Public Network Inbound TrafficVpcInternetRxKb/minVPC and classic network
(ECS) System Disk Read (BPS)SystemDiskReadBpsByte/sVPC and classic network
(ECS) System Disk Write (BPS)SystemDiskWriteBpsByte/sVPC and classic network
(ECS) System Disk Read (IOPS)SystemDiskReadOpsVPC and classic network
(ECS) System Disk Write (IOPS)SystemDiskWriteOpsVPC and classic network

Metrics collected from CloudMonitor (agent-based)

These metrics require the CloudMonitor agent to be installed on each instance. When you select CloudMonitor as the collection source, Auto Scaling installs the agent on all instances in the scaling group. To make sure new instances are also covered, enable Automatically Install CloudMonitor Agent on Newly Purchased ECS Instances in the CloudMonitor console.

Metric (Auto Scaling console)Metric (Auto Scaling API)UnitStatistical periodSupported network
(Agent) CPU UtilizationCpuUtilizationAgent%15 seconds, 1 minute, 2 minutes, 5 minutes, 15 minutesVPC and classic network
(Agent) GPU UtilizationGpuUtilizationAgent%VPC and classic network
(Agent) Idle GPU Memory PercentageGpuMemoryFreeUtilizationAgent%VPC and classic network
(Agent) GPU Memory UsageGpuMemoryUtilizationAgent%VPC and classic network
(Agent) MemoryMemoryUtilization%VPC and classic network
(Agent) Average System LoadLoadAverageVPC and classic network
(Agent) Total TCP ConnectionsTcpConnectionVPC and classic network
(Agent) Established TCP ConnectionsTcpConnectionVPC and classic network
(Agent) Internal Network Packets SentPackagesNetOutCount/sVPC and classic network
(Agent) Internal Network Packets ReceivedPackagesNetInCount/sVPC and classic network
(Agent) Public Network Packets SentPackagesNetOutCount/sClassic network
(Agent) Public Network Packets ReceivedPackagesNetInCount/sClassic network
To use GPU metrics, install GPU drivers on your GPU-accelerated instances first. For instructions, see Installation guideline for NVIDIA drivers.

Elastic Container Instance metrics

The following table lists the metrics available for Elastic Container Instances. All ECI metrics are collected via CloudMonitor and support only Virtual Private Cloud (VPC) networks. The 15-second statistical period is not available for ECI metrics.

Metric (Auto Scaling console)Metric (Auto Scaling API)UnitStatistical periodSupported network
CPU UtilizationEciPodCpuUtilization%1 minute, 2 minutes, 5 minutes, 15 minutesVPC
MemoryEciPodMemoryUtilization%VPC

ALB metrics

If your scaling group (ECS type or Elastic Container Instance type) is associated with an Application Load Balancer (ALB) instance, use the (ALB) QPS per Backend Server metric to trigger scaling based on request load.

Metric (Auto Scaling console)Metric (Auto Scaling API)UnitStatistical periodSupported network
(ALB) QPS per Backend ServerLoadBalancerRealServerAverageQpsCount/s15 seconds, 1 minute, 2 minutes, 5 minutes, 15 minutesVPC
The 15-second statistical period is available only for scaling groups of the ECS type.

Usage notes

CloudMonitor agent requirement

Agent-based ECS metrics (prefixed with "Agent") require the CloudMonitor agent on every instance.

GPU metrics requirement

GPU metrics (GpuUtilizationAgent, GpuMemoryFreeUtilizationAgent, GpuMemoryUtilizationAgent) are only available on GPU-accelerated instances with GPU drivers installed.

Classic network support

All ECS and agent-based metrics support both VPC and classic network; ECI and ALB metrics support VPC only.