Create an Alarm Subscription

Alarm subscriptions define which alerts are sent, who receives them, and how to reduce duplicate notifications. When an enabled subscription matches an alert, the platform sends a message through the associated notification targets, contact groups, or third-party applications.

​

Subscription and Activation Logic

An alarm subscription typically includes the alert type, subscription scope, notification targets, and noise control rules. Test sending only verifies connectivity to the notification target; it does not guarantee that actual alerts will be sent successfully. Actual delivery also depends on the subscription scope, whether an alert is triggered, and noise control.

​

Create an Event Alarm Subscription

Event alerts are triggered by business events and are suitable for monitoring status changes such as training job failures, image build results, service exceptions, orders, quota, and storage.

​

1. Open the Event Subscription Form

1. Go to Alarm > Alarm Subscription.
2. On the My Subscriptions tab, click Create.

​

2. Configure the Event Subscription

​

Select Alert Type

1. Enter the Subscription Name.
2. Set Alarm Source Type to Event.
3. In the alert type list, expand the business category and select the alert types to subscribe to.
4. Confirm your selections in Selected Alarm Types.

Common business categories include LLM inference, VSCS, JupyterLab, general service, orders, quota management, storage, and CPU. Available categories and alert types depend on the current console. To subscribe to multiple alerts, select multiple alert items at once to reduce repetitive configuration.

​

Configure Subscription Scope

Click Next. In the Narrow Scope step, specify the objects matched by the subscription as needed. This step is optional. If left blank, all events under the selected alert type will be subscribed to.

​

Configure Notify Targets and Noise Control

1. Click Next.
2. In the Notify Targets step, select at least one notification target or contact group.
3. If no suitable target is available, click Create to create a notification target or contact group.
4. Click Next.
5. In the Noise Control step, configure group aggregation and mute periods as needed.

   Configuration	Description
   Group Aggregation	When enabled, the platform merges alerts by aggregation field within the aggregation interval. Suitable for similar alerts triggered repeatedly within a short period.
   Aggregation Interval (Seconds)	The time window for merging alerts, for example, 300 seconds. It can be extended appropriately in scenarios with frequent fluctuations, but it is not recommended to set it too long initially, as this may delay awareness of important changes.
   Aggregation Fields	Used to determine which alerts belong to the same group, such as Owner, Cluster, Aggregate ID, Event Type, and SubStatus.
   Silence Period	When enabled, matching alerts are not sent during the specified period. Suitable for maintenance windows, release windows, or nighttime do-not-disturb periods.

   Event-type Trigger alerts ignore threshold and interval settings.

​

3. Create the Event Subscription

Click Create. Once created, the subscription takes effect immediately when enabled. To avoid exact duplicates, the platform consolidates subscriptions with the same user, alert item, and filter criteria. Creating another subscription with these criteria updates the existing one instead of adding a new one. To create a separate subscription, adjust the filter criteria or subscription target.

​

Create a Metric Alarm Subscription

Metric alerts are triggered based on Prometheus metric thresholds and are suitable for monitoring resource utilization, service metrics, or custom business metrics.

​

1. Open the Metric Subscription Form

1. Go to Alarm > Alarm Subscription.
2. On the My Subscriptions tab, click Create.

​

2. Configure the Metric Subscription

​

Configure Metric Conditions

1. Enter the Subscription Name.
2. Set Alarm Source Type to Metric.
3. Select the Metric Source.

   Metric Source	Description
   Basic Metrics	Common monitoring metrics provided by the platform, suitable for quickly configuring alerts for CPU, GPU, memory, network, and other resources.
   Custom Metrics	Metrics exposed by businesses or services, suitable for monitoring business metrics not covered by the platform’s built-in list. Metric names may contain only letters, numbers, underscores, and colons.

   Continue configuring the metric conditions:

   Configuration Item	Description
   Metric Category	Used to filter metrics.
   Metric Name	The specific metric to monitor. Before configuration, it is recommended to use the metric preview to confirm that the target metric has data.
   Metric Preview	View recent metric trends to help determine whether data is available and whether the threshold is reasonable.
   Aggregation Method	Aggregates or calculates metrics, such as time-window averages, maximum values, and rates.
   Operator	The threshold comparison method, such as >, >=, <, or <=.
   Threshold	The value used to determine whether to trigger an alert. It is recommended to configure it based on the metric preview and business impact.
   Duration	The length of time that the metric condition must remain satisfied, such as 5m, to reduce false alerts caused by transient fluctuations.
   Severity	The severity of the alert, such as Critical, Warning, or Info.

​

Using Advanced Mode

If basic mode cannot express the conditions, click Switch to Advanced Mode and use a PromQL expression to configure an alert. Advanced mode is suitable for combining multiple conditions, filtering custom labels, or calculating complex metrics. Before using it, ensure that the PromQL expression includes the required namespace or orgname isolation label so the query meets the platform’s isolation requirements.

​

Configure Subscription Scope, Notify Targets, and Noise Control

Click Next to continue configuring Narrow Scope, Notify Targets, and Noise Control. Metric alerts can be further limited to applicable objects based on the selected workload scope, such as all workloads, specified workload types, or specified resource pools. Notify targets and noise control are the same as those for event alerts.

​

3. Create the Metric Subscription

After confirming the configuration, click Create. Once created, the subscription takes effect immediately when enabled.

​

Manage Subscriptions and Noise Control Rules

After locating the target subscription in Alarm > Alarm Subscription, you can adjust its configuration or enabled status as needed. To manage noise control across multiple subscriptions, use the Global Noise Control Rules tab. Global rules apply only to the current user and do not affect other users. Subscription-level rules take precedence over global rules. Muting and group aggregation are independent rule types; the system checks each type separately to decide whether to use a subscription-level rule or fall back to a global rule. If a subscription has a rule of a given type, the global rule of that type generally will not take over, even when the subscription-level rule is disabled. To continue inheriting the global rule, do not create a disabled rule of the same type in the subscription.

​

View Results

After the subscription is created, you can verify it in the following ways:

• In the My Subscriptions list, confirm that the subscription is enabled.
• Perform a test send to the bound notification target to confirm that the receiving endpoint is reachable.
• After an actual alert occurs, view the alert record, delivery method, and delivery result in Alarm Overview.

If no actual alert is received, first confirm whether the alert occurred, then check the subscription scope, notification target enabled status, and noise control.

​

Related Documentation

• View Alert History
• Manage Alert Notification Targets