Managing Instances

Instance States and Controls

AtlasCloud Serverless instances can be managed to optimize costs and resource usage. You have full control to start and stop your instances as needed.

Instance States

Your serverless instance can be in one of the following states:

Starting a Stopped Instance

When your serverless instance is stopped, you can quickly resume it:

When to Start

• Your instance shows 0/1 Running status
• You need to make the service available for requests
• You want to resume a previously paused service

How to Start

1. Locate your stopped instance in the dashboard - it will show 0/1 Running with a black Start button
2. Click the Start button (▶️)
3. Confirm the action when prompted with "Are you sure to start this endpoint?"
4. Wait for activation - the instance will transition to running state

What Happens

• The instance state changes from Stopped to Active
• Worker pods are created (takes 30-60 seconds)
• The status indicator changes to ✅ Ready
• Billing resumes for the running instance
• Your API endpoint becomes available for requests

Stopping a Running Instance

You can stop running instances to save costs when they're not needed:

When to Stop

• Your instance shows 1/1 Running with ✅ Ready status
• You want to pause the service to save costs
• The service is not currently needed but you plan to use it later

How to Stop

1. Locate your running instance - it will show 1/1 Running with ✅ Ready status
2. Click the Stop button (⏹️)
3. Wait for the instance to stop - the transition happens quickly

What Happens

• The instance state changes from Active to Stopped
• Worker pods are terminated (takes 5-15 seconds)
• The display updates to show 0/1 Running
• Billing stops immediately - no charges while stopped
• Your API endpoint URL remains valid but won't respond to requests

Cost Management

Active Instance

• Billing: Charged per second based on GPU type and worker count
• Cost Formula: Duration × Unit Price × GPUs per Worker
• Auto-scaling: Can scale between Active Workers and Max Workers

Stopped Instance

• Billing: No charges - 100% cost savings
• Persistent: Your configuration and endpoint URL are preserved
• Quick Resume: Can be restarted anytime with the Start button

Cost Optimization Tips

1. Stop during idle periods: Stop instances when not actively in use
2. Monitor usage patterns: Identify optimal times to stop/start
3. Overnight savings: Consider stopping development instances outside business hours
4. Weekend optimization: Stop non-production instances during weekends
5. Quick resume: Instances restart in under a minute when needed

Important Considerations

Service Availability

• Stopped instances: The URL remains valid but returns no response
• Starting latency: Expect 30-60 seconds for the instance to become fully available
• No data loss: Stopping an instance doesn't affect your configuration or mounted storage

Auto-scaling Behavior

• Stopped instances: No auto-scaling (effectively scaled to zero)
• Active instances: Full auto-scaling between Active Workers and Max Workers
• Scale-to-zero: Active instances with Active Workers set to 0 can auto-scale to zero during idle periods

Logs and Monitoring

• Metrics: Available for both stopped and active instances
• Historical logs: Preserved across start/stop cycles
• Events: Start/stop operations are logged in your instance's event history

Troubleshooting

Instance Won't Start

• Ensure you have sufficient account balance
• Check that the instance is in Stopped state
• Verify you have permission to manage the instance
• Wait a moment and refresh if the UI hasn't updated

Instance Won't Stop

• Ensure the instance is in Active/Running state
• Allow current requests to complete (graceful shutdown)
• Check that you have permission to manage the instance

Status Inconsistency

• Refresh the page to see the latest status
• Allow 30-60 seconds for state transitions to complete
• Check the instance logs for any error messages

Best Practices

1. Plan your usage: Stop instances during known idle periods
2. Monitor before stopping: Check if there are active requests
3. Graceful shutdown: Allow ongoing requests to complete
4. Development workflow: Stop development instances when not coding
5. Testing optimization: Start for testing, stop immediately after
6. Production planning: Use monitoring data to optimize start/stop schedules