Attach an alert to a resource

Attach an alert to a running resource so it starts watching, and detach it when you want it to stop.

Before you begin

You need an alert for the same kind of resource, in the same project and region. Attach is available only while the resource is running: an Instance, VM, or Inference service in the running state, or a Training Cluster or Load Balancer in the Running status. On a Training Cluster in any other status, the Alerts tab says alerts are not available.

An alert that uses a GPU metric attaches only to a resource on a GPU plan. The GPU metrics are GPU Temperature, GPU Power, GPU Memory Used, and XID Error. The network and storage metrics of a VM attach only to a VM.

Open the alerts panel

Resource	Where
Instance	Compute > Containers (Instances), select the instance, then its Alert tab.
VM	Compute > Virtual Machines, select the VM, then its Alerts tab.
Inference service	Open the service, select the Settings tab, and find Alerts.
Training Cluster	Under Training Cluster, open the cluster, then its Alerts tab.
Load Balancer	Compute > Load Balancer, select the load balancer, then its Alerts tab.

Attach the alert

1. Open Select Alerts. It lists the project’s alerts for this kind of resource, each with its name, service type, and severity. An alert already attached to this resource reads “Attached to:” followed by the resource name, since an alert attaches to a resource only once. If the project has no alert for this kind of resource, the list says that no alerts are available. Create one with the link in the panel.
2. Choose the alert.
3. Select Attach. The alert starts watching the resource.

On a Load Balancer, you can attach only unattached alerts whose metric suits this load balancer’s type. Each alert serves one load balancer, so create a separate alert for each.

To create an alert from here, select the link in the panel. The new alert has its service already set to this resource, ready to attach.

Read an alert’s status

Each attached alert is listed with its severity, threshold, and status. Firing means the condition is met now; OK means it is not.

Detach an alert

1. In the row, open Action and select Detach alert. You need update permission on the resource.
2. Tick I understand and want to detach this alert.
3. Select Detach.

The alert stops watching this resource and keeps watching the others. It stays in Alert Management. To remove the alert everywhere, delete it.

Confirm the alert is attached

The alert is in the table with the status OK, and its Status in Alert Management reads Attached. If the alert is OK while you expect a notification, see Troubleshooting Alert Management.

Default alerts on a Load Balancer

When a load balancer becomes Running, TIR creates and attaches default alerts to it, using every monitoring integration the project has at that moment. Each alert name ends with the load balancer’s name in brackets.

Alert	Metric	Condition	Severity
Healthy backends below 50%	Healthy Backends %	Less Than 50	Critical
5xx error rate above 5%	Error Rate (5xx)	Greater Than 5	Warning
TLS certificate expires within 14 days	Certificate Expiry (days)	Less Than 14	Warning

The certificate alert is created only for a load balancer that serves HTTPS. TIR creates the default alerts only when the project has a monitoring integration at the moment the load balancer becomes Running. When you delete a load balancer, its alerts are removed with it.