GPU-accelerated instances are a type of Elastic Compute Service (ECS) instances, managed the same way as standard ECS instances. You can restart a GPU-accelerated instance from the ECS console or through API calls. This topic covers user-initiated restarts only.
Impacts
Service interruption: Restarting an instance interrupts all running services on that instance.
In-memory data loss (force restart only): A force restart is equivalent to a power-off. Any cached data in memory that has not been flushed to the storage device is lost, and file system corruption may occur.
Before you begin
Before you restart an instance, make sure that you have:
Scheduled the restart during a low-traffic window to minimize service disruption
Stopped application services within the operating system (OS) to allow all in-flight requests and data writes to complete
Verified that critical business applications are configured to start automatically on boot
Created a snapshot to back up your data
Snapshots are a paid feature. For pricing details, see Snapshots.
Restart an instance from the console
Go to the ECS console - Instances page. In the top-left corner, select the target region and resource group.
Click the instance ID to open its details page. In the upper-right corner, click Restart.
In the dialog box, select a restart mode:
Normal restart (default): Leave the Force Restart checkbox unselected. The OS attempts to shut down all processes gracefully before restarting.
Force Restart: Select the Force Restart checkbox. This is equivalent to a power-off and risks in-memory data loss and file system corruption. Use this option only when the instance is unresponsive.
Choose when to restart:
Restart immediately: Click Confirm.
Schedule the restart: Select Set Timed Execution, configure the time and the required role, then click Confirm. After the task is created, go to the CloudOps Orchestration Service (OOS) console - Timed Task page to modify the task configuration if needed.
During a restart, the OS releases processes, CPU, and memory resources, and the virtualization layer releases its associated resources. This takes 3 to 5 minutes on average but can take up to 20 minutes.
Restart an instance through API
The following API operations restart one or more ECS instances:
| API operation | Description |
|---|---|
| RebootInstance | Restart a single instance in the Running state |
| RebootInstances | Restart one or more instances in the Running state |
| ReActivateInstances | Reactivate and restart an expired or overdue instance |
To schedule a restart through API, see the ACS-ECS-ScheduleToRebootInstances public template in OOS.
FAQ
Why do my changes not take effect after I run the reboot command?
Certain changes -- such as modifying instance resources, updating the hostname, or resizing a cloud disk -- are applied through the console or API and stored on the underlying virtualization platform. A console or API restart reloads the full instance configuration from this platform, so these changes take effect.
The reboot command runs inside the OS and is unaware of configuration changes made outside the OS. To apply such changes, restart the instance from the console or through API instead.
What permissions does a RAM user need to restart an instance?
A Resource Access Management (RAM) user needs the ecs:DescribeInstances and ecs:RebootInstance permissions.