When an ECS instance experiences performance degradation or abnormal behavior, you can use the Status and Alarms page to assess its operational health. This page aggregates information about instance health status, system health status, and monitoring alarms. You can review triggered alarms and use the ECS AI Assistant for root cause analysis and remediation suggestions.
Procedure
This feature is currently in invite-only preview.
-
Log on to the ECS console - Instances page. At the top of the page, select a resource group and region.
-
Find the target instance and click its ID to go to the instance details page.
-
Click the Monitoring tab. The Status and Alert subtab is displayed by default.
This tab contains the Status Check and Monitoring Alarms sections. The Status Check section shows the reachability results for system and instance status checks. The Monitoring Alarms section displays information such as the alarm name, alert rule, alarm type, alarm period, and number of alarms in a table, along with action links for View Monitoring Data and Troubleshooting.
How it works
Status check
Underlying probes perform real-time checks on the status of the physical host and the network stack reachability of the operating system.
System status check
-
Normal: The host and physical network are running correctly.
-
Impaired: The host has a hardware failure or is undergoing maintenance. We recommend that you restart or migrate the instance.
-
Not Detected: The instance is in the Stopped or Starting state, so the system cannot perform a check.
Instance status check
-
Normal: The operating system's network stack is responding correctly.
-
Impaired: There is an internal error in the operating system, such as a system crash or incorrect network adapter configuration.
-
Not Detected: The instance is in the Stopped or Starting state, so the system cannot perform a check.
For more information, see Instance health status.
Monitoring alarms
The Monitoring Alarms list aggregates two types of information: performance risk events and CloudMonitor alarms.
-
Performance risk events: These are automatically detected by the underlying Alibaba Cloud system. This feature is enabled by default. An event is triggered when a running instance metric reaches its instance specification limit. For example, the instance's CPU or storage performance reaches its specification limit.
Trigger mechanism: An event is typically triggered if the specification limit is reached twice within the last three minutes.
-
CloudMonitor alarms: These are alarms created based on metric data from CloudMonitor. After you create an alert rule, CloudMonitor displays alarms in the Monitoring Alarms list when a metric exceeds its configured threshold.
Root cause analysis with the ECS AI Assistant
When a performance risk event or an alarm appears in the Monitoring Alarms list, you can use the ECS AI Assistant to perform an automated, in-depth diagnosis.
-
Start the AI diagnosis.
Click Suggestions in the Actions column. The ECS AI Assistant starts an interactive diagnostic process.
After the ECS AI Assistant starts the interactive diagnostic process, a blue notification bar at the top displays an alarm based on the cpuUsage metric, indicating that the instance has reached its CPU specification limit twice in the last three minutes. The Monitoring and O&M card shows that the instance monitoring data was successfully retrieved. Below, the CPU Utilization (Base) line chart, with a one-minute data granularity, shows the average value. The chart indicates that CPU utilization was consistently near 100% before dropping to about 75%.
-
Authorize the AI Assistant to perform a deep inspection.
In the AI Assistant's dialog box, when a request to Send Remote Commands appears, click Immediate execution.
The AI Assistant analyzes existing CloudMonitor metrics. If the available data is insufficient to identify the problem, it requests to run a remote command to gather real-time information from within the operating system.
-
Get the root cause analysis and solution.
The AI Assistant automatically parses the command's output, presents the root cause in natural language and provides actionable solutions.
-
Solution 1: Run a remediation command. For example, to address a process with an abnormally high load, the AI Assistant might suggest a
killcommand.ImportantBefore you run any command to terminate a process, always confirm that the process is not critical to your business to prevent service interruptions or data loss. We recommend first trying to gracefully terminate the process with the
kill -15 <PID>command. -
Solution 2: Adjust resource configuration. If the issue is caused by insufficient resources, the AI Assistant recommends actions such as upgrading the instance specifications or expanding a cloud disk.
-
FAQ
AI Assistant stuck on "Thinking..."
Analyzing complex issues can take tens of seconds. If there is no response for an extended period, especially when it involves running a remote command, follow these steps to troubleshoot:
-
Check the status of the Cloud Assistant Agent: Make sure the Cloud Assistant Agent on the instance is online. The AI Assistant relies on Alibaba Cloud Cloud Assistant to run remote commands. If the Cloud Assistant Agent is not working correctly, the AI Assistant cannot perform a deep inspection. In this case, first check the status of the Cloud Assistant Agent.
-
Check security group rules: Confirm that the instance's security group allows outbound access to Alibaba Cloud API endpoints. The AI Assistant needs this network path to communicate with the instance.