Instance performance monitoring and troubleshooting-Elastic Compute Service(ECS)-阿里云帮助中心

Performance monitoring and alerting

Troubleshoot and resolve issues

CPU

Memory

Troubleshoot and resolve high memory usage on Linux instances
Troubleshoot and resolve high memory usage on Windows instances
- How to resolve insufficient virtual memory on ECS instances that run Windows Server 2003 or 2008?
- How to configure virtual memory for a Windows instance?

Disk

Network

Are ECS monitoring and GPU monitoring installed by default?

The CloudMonitor agent is installed on servers by default. In the CloudMonitor console, on the Cloud Resource Monitoring > Host Monitoring page, the Auto Install Agent switch is enabled by default, so servers automatically display monitoring data. To manually install or uninstall the agent, see Install and uninstall the Cloud Monitor agent.

How do I uninstall the CloudMonitor agent from an ECS instance?

The CloudMonitor agent (process name: argusagent) collects OS-level metrics such as memory, network, and disk data from ECS instances.

After you uninstall the CloudMonitor agent, the CloudMonitor console stops collecting OS-level metrics for the instance (historical data is not affected). Basic monitoring in the ECS console (such as CPU utilization and network traffic) does not rely on this agent and remains available after uninstallation.

Linux

Log on to the instance as the root user and run the following commands in sequence.

# Stop the CloudMonitor agent
bash /usr/local/cloudmonitor/cloudmonitorCtl.sh stop
# Uninstall the CloudMonitor agent
bash /usr/local/cloudmonitor/cloudmonitorCtl.sh uninstall
# Remove the directory
rm -rf /usr/local/cloudmonitor

Windows

Log on to the instance as the Administrator, open Command Prompt (Win+R, then type cmd), and run the following commands in sequence.

# Stop the CloudMonitor agent
net stop argusagent
# Uninstall the CloudMonitor agent
"C:\Program Files\Alibaba\cloudmonitor\bin\argusagent_service.exe" uninstall
# Remove the directory
cd C:\Program Files\Alibaba
rd /s /q cloudmonitor

For more information, see Install and uninstall the Cloud Monitor agent.

How do I troubleshoot access latency on an ECS instance?

Access latency on an ECS instance can be caused by multiple factors. We recommend that you troubleshoot the issue systematically by performing the following steps:

Use the one-click diagnostics tool (recommended)

Alibaba Cloud provides a self-service diagnostics tool that can quickly detect common issues related to the network, security groups, and system configurations.

Log on to the diagnostics page in the ECS console.
Select the region where the instance resides and select the target ECS instance.
Select Network Connectivity Diagnostics and view the diagnostic report and optimization suggestions.

Manual troubleshooting steps

If the self-service tool cannot resolve the issue, troubleshoot in the following order:

Network: Test the access speed from different regions.
Security group configuration: Check whether the security group restricts specific IP addresses or ports. You can temporarily open all ports for testing (be sure to restore the original configuration after the test).
System resources: Log on to the instance and check the CPU, memory, and disk I/O usage. For more information, see View instance monitoring data.
Application: Check the web server configuration (Nginx/Apache), database connection pool settings, and errors or warnings in application logs.

Performance optimization suggestions

Insufficient bandwidth: Upgrade the public bandwidth or use CDN acceleration.
High-load scenarios: Configure Server Load Balancer (SLB) with multiple ECS instances for horizontal scaling.
Disk performance: Use ESSDs for the system disk and select ESSDs at an appropriate performance level (PL) for data disks.
Network optimization: Deploy related services in the same region to reduce cross-region access latency.

How do the use cases of CloudMonitor and atop differ?

The CloudMonitor agent is required to collect OS-level metrics, such as memory usage, disk I/O details, and process-level metrics. If the agent is not installed, CloudMonitor cannot collect memory or process-level data. This limits the coverage of alert rules and reduces troubleshooting efficiency. For information about the metrics that can be collected without the agent, see View instance monitoring information.

atop complements CloudMonitor and is suitable for analyzing historical loads when an instance is inaccessible. CloudMonitor provides real-time online monitoring, whereas atop provides offline historical data analysis. We recommend that you deploy atop and configure it to start at boot while the instance is running normally.