Webhook requests

A webhook monitoring integration sends an HTTP POST request to your endpoint each time an alert fires or is resolved. Use it to start automated workflows in other tools as alerts happen. You set the URL and headers in Create a monitoring integration. The body has the same structure for every event. The values in the example are sample data.

Request

Part	Value
Method	POST.
Content-Type	application/json, sent with every request.
Authorization	Optional. Add it as a custom header, for example Bearer <token>, if your endpoint needs it.
Custom headers	Optional. You can add up to 10 key and value pairs to an integration.

Body

{
  "version": "v1",
  "monitoring_group_id": 0,
  "status": "firing",
  "alert": {
    "id": 0,
    "name": "Sample Alert",
    "resource_type": "notebook",
    "resource_type_display": "Instance",
    "metric_type": "cpu",
    "metric_type_display": "CPU Utilization",
    "operator": ">",
    "operator_display": "Greater Than",
    "threshold_value": 80,
    "severity": "Warning"
  },
  "project": {
    "id": 0,
    "name": "Sample Project",
    "location": "sample-location"
  },
  "affected_resources": [
    {
      "id": 0,
      "name": "sample-resource",
      "replica": "sample-replica",
      "gpu_uuid": "GPU-SAMPLE"
    }
  ]
}

The fields resource_type_display, metric_type_display, and operator_display hold the names the portal shows, such as CPU Utilization for the metric. The sample that Test URL sends has the same structure but leaves out these three fields.

Fields

Top level

Field	Type	Meaning
version	string	Version of the body format. The current value is v1.
monitoring_group_id	integer	The ID of the monitoring integration that sends this request.
status	string	firing or resolved.

The alert object

Field	Type	Meaning
id	integer	The alert’s ID.
name	string	The alert’s name.
resource_type	string	The kind of resource, such as notebook for an Instance or gateway_service for a Load Balancer.
resource_type_display	string	The same, as the portal names it.
metric_type	string	The metric, such as cpu.
metric_type_display	string	The metric as the portal names it.
operator	string	One of >, <, >=, <=, ==, !=.
operator_display	string	The operator as the portal names it.
threshold_value	number	The threshold.
severity	string	Critical, Warning, or Info.

The project object

Field	Type	Meaning
id	integer	The project’s ID.
name	string	The project’s name.
location	string	The region of the alert.

The affected_resources list

Each entry is a resource the alert is about.

Field	Type	Meaning
id	integer	The resource’s ID.
name	string	The resource’s name.
replica	string	The replica the alert is about, for Inference and training services. Empty otherwise, including for a Load Balancer.
gpu_uuid	string	The GPU’s ID when the alert is about a GPU. Empty otherwise.

When one resource has several affected replicas or GPUs, the field lists them separated by a comma and a space.

What your endpoint must do

• Accept POST requests with a JSON body.
• Answer with a 2xx status, and answer quickly. TIR waits up to 10 seconds, ignores the answer, and sends each request once. For slow work, answer at once and process afterward.
• If you added an Authorization header, check it before you read the body.

A failed request is not sent again, and the portal does not list it. To check your endpoint, use Test URL in the integration form, which sends a sample alert. See Troubleshooting webhook alerts.