GPU-accelerated instances are managed as Elastic Compute Service (ECS) instances. Use the ECS console or API operations to stop them.
Stopping an instance interrupts all services running on it. Proceed with caution.
Before you stop
Billing and resource behavior differ by instance type. Review the relevant row before stopping.
| Instance type | Billing after stop | Other effects |
|---|---|---|
| Spot instance | N/A — may be reclaimed | Cannot restart if resources are insufficient or your bid is below the market price. For details, see Spot instance overview. |
| Subscription instance | Continues | Expiration time is not affected. |
| Pay-as-you-go instance | Continues until you release the instance | To stop billing, release the instance instead. |
Stop an instance in the ECS console
Go to the ECS console — Instances.
In the top navigation bar, select the region and resource group of the instance.
Click the instance ID to open the instance details page. In the upper-right corner, click All Actions, then search for and click Stop.
Set the Stopped By or Stop Mode parameter, then click Confirm.
Stop method (Stopped By)
Choose how the instance is stopped based on your situation:
| Option | Description | Data impact |
|---|---|---|
| Stop | Normal shutdown — the operating system shuts down gracefully. Use this in most cases. | None under normal conditions. |
| Force Stop | Forcefully stops the instance. Equivalent to a physical shutdown. | May cause data loss if instance data has not been written to disks. |
If your instance remains in the Stopping state, see Instance stuck in the Stopping state.
Stop mode (pay-as-you-go instances in VPCs only)
| Option | Billing after stop | Resource behavior |
|---|---|---|
| Standard Mode | All resources continue to be billed. | Resources are retained. |
| Economical Mode (Formerly Known as No Fees for Stopped Instances) | Billing stops for: vCPUs, memory, images, and public bandwidth of the static public IP address (pay-by-bandwidth). Billing continues for: system disk, data disks, and public bandwidth of the elastic IP address (EIP) (pay-by-bandwidth). | vCPUs, memory, and the static public IP address are reclaimed. The private IP address is retained. |
Economical Mode introduces two restart risks:
Restart may fail. vCPUs and memory are reclaimed when the instance stops. The restart can fail due to insufficient resources or overdue payments. If resources are insufficient, try again later or change the instance type. If payments are overdue, add funds to your account. For details, see Economical mode risks and Overdue payments.
The static public IP address may change. The auto-assigned public IP address is reclaimed when the instance stops. When the instance restarts, a new public IP address is assigned, which may differ from the original.
When the instance stops, its status changes to Stopped.
Stop instances using API operations
Call one of the following API operations to stop instances programmatically. For authentication and request setup, see Integration overview.
| API operation | Description |
|---|---|
| StopInstance | Stops a single instance in the Running state. |
| StopInstances | Stops one or more instances in the Running state. |
Troubleshooting
Instance stuck in the Stopping state
If the instance remains in the Stopping state for an extended period, the operating system may be mid-process (normal) or blocked by an error (abnormal).
Diagnose the issue:
View the instance screenshot to check the OS shutdown progress. For details, see View system logs and screenshots.
If the screenshot is inconclusive, connect via Virtual Network Computing (VNC) to observe the shutdown progress directly. For details, see Connect to an instance by using VNC.
Act based on what you see:
| OS stop status | Action |
|---|---|
| Normal stop (installing patches, slow response due to low specs) | Wait for the shutdown to complete. |
| Abnormal stop (OS error, shutdown blocked) | Connect via VNC and resolve the issue manually. |
If waiting is not an option, use Force Stop to immediately stop the instance.
Force Stop is equivalent to a physical shutdown and may cause data loss if instance data is not written to disks.