Realtime Compute for Apache Flink:Configure monitoring and alerts

Job event alerts

You can subscribe to system event alerts for jobs by setting conditions and configure batch alerts for events.

1. Log on to the Cloud Monitor console.

2. In the navigation pane on the left, choose Event Center > Event Subscription.

3. On the Subscription Policy tab, you can click Create Subscription Policy.

4. On the Create Subscription Policy page, you can configure the parameters. For details about other parameters, see Manage event subscriptions.

   

   • For Subscription Type, select System Events.

   • For Product, select Realtime Compute for Apache Flink.

   • Supported events include JOB_FAILED (not supported in ARMS), ECS.SystemFailure, and ECS.SystemMaintenance.

   • Event Content: Enter the following Flink information to configure alerts for specific jobs or in batches.

     • Workspace ID: This applies the alert rule to all jobs across all namespaces in the target workspace. For more information, see How do I view information such as the workspace ID?.

     • Namespace name: This applies the alert rule to all jobs in the target namespace.

     • Deployment name: This applies the alert rule to specific job deployments. You can separate multiple names with commas (,). If duplicate deployment names exist, use the deployment ID.

     • Deployment ID: Apply event alerting to specific jobs. Separate multiple IDs with commas (,). You can find the Deployment ID on the Deployment Details tab of your Flink job, in the Deployment Job ID field.

   Note

   If you leave Application group, Event Content, or Event Resources blank, the subscription applies to all workspaces in your account.

Workflow event alerts

You can configure conditions to subscribe to system event alerts for Flink workflows. Batch configuration is supported for event alerting. For more information about workflows, see Manage workflows.

1. Obtain the resource ID of the workflow task.

   1. Log on to the Cloud Monitor console.

   2. In the navigation pane on the left, choose Event Center > System Events.

   3. On the Event Monitoring tab, set Product to Realtime Compute for Apache Flink, Event Name to Workflow Task State Change, and click Search.

   4. Identify the resource ID of workflow nodes below.

      

      The resource ID is in the acs:flink:cn-hangzhou:<AlibabaCloudAccountID>:resourceId/workspaceId/<workspaceId-namespaceId>#workflowDefinitionName/<workflowDefinitionName>#taskDefinitionName/<taskDefinitionName> format. You can also construct the resource ID directly using this format.

      Parameter	Description
      <AlibabaCloudAccountID>	The Alibaba Cloud account ID that owns the Flink workspace.
      <workspaceId-namespaceId>	Consists of workspaceId and namespaceId, joined by a hyphen (-). workspaceId: Your workspace ID. For more information, see Workspace and namespace FAQ. namespaceId: Your namespace name.
      <workflowDefinitionName>	The workflow name.
      <taskDefinitionName>	The name of the workflow node.

      Note

      Workflow state change events in Cloud Monitor may be delayed by a few minutes.

2. Subscribe to event notifications.

   1. In the left navigation pane, choose Event Center > Event Subscription.

   2. On the Subscription Policy tab, click Create Subscription Policy.

   3. On the Create Subscription Policy page, configure the policy. For more information, see Manage event subscriptions (Recommended).

      • Subscription type: System Event.

      • Product: Select Realtime Compute for Apache Flink.

      • Event Name: Select Pipeline Task State Change.

      • Event Content: Filter by transition state, such as toState: FAILED (workflow failed), toState: SUCCESS (workflow succeeded), or fromState: SCHEDULED, toState: RUNNING (workflow transitioned from scheduled to running).

      • Event Resources: Enter the resource ID that you obtained in Step 1. You can separate multiple IDs with commas (,).

      • Event Type, Event Level, and Application group: Leave these parameters empty.

Single job configuration (Realtime Compute console)

You can create alert rules for target jobs or use existing alert rule templates to accelerate the configuration.

1. Go to the alert configuration page.

   1. Log on to the Realtime Compute console. Then, in the Actions column for your workspace, click Console.

   2. On the Operation Center > Job O&M page, click the name of the job that you want to manage.

   3. Click the Alert Configuration tab.

2. On the Alert Rules tab, choose Add Rule > Custom Rule.

   You can also choose Add Rule > Rule Template to create alert rules directly or to modify them slightly after applying a template.

3. Enter the alert rule information.

   Category	Parameter	Description
   Specific Rules	Name	Start with a letter. Use lowercase letters, digits, and underscores (_). Length: 3 to 64 characters.
   	Description	Notes about the rule.
   	Content	Configure conditions that trigger the alert. After configuration, the system compares metric values against the threshold at regular intervals and triggers an alert if conditions are met. Metric: Restart Count in 1 Minute: Number of JobManager restarts per minute. Checkpoint Count in 5 Minutes: Number of successful checkpoints per 5 minutes. Emit Delay: Business latency, measured in seconds as the time difference between data generation and egress from the source operator. Important Emit Delay accuracy depends on accurate timestamps from external systems. If external systems lack timestamps or record incorrect ones, Emit Delay will be inaccurate and fail to reflect true latency. To better reflect actual latency, configure composite metric alerts. For details, see Recommended monitoring configurations. IN RPS: Input records per second. OUT RPS: Output records per second. Source Idle Time: Duration the source operator was idle, in milliseconds. Job Failed: Job failure. Time Interval: The length of the historical data window the system queries backward during each check, in minutes. Comparator: Supports >= and <=. Threshold: Value compared against the metric. If you select >=, the system compares the maximum metric value against the threshold. An alert triggers if max_value >= threshold. If you select <=, the system compares the minimum metric value against the threshold. An alert triggers if min_value <= threshold. Example: Monitor Checkpoint Count in 5 Minutes with a 10-minute time interval, threshold of 2, and comparator <=. The system checks every minute over the last 10 minutes. An alert triggers if any 5-minute period has a checkpoint count ≤ 2.
   	Effective Time	When the alert rule is active. Defaults to 24/7. You can restrict it to business hours (9 AM to 6 PM).
   	Alert Rate	Minimum interval between alerts. Supports 1 minute to 1440 minutes (24 hours).
   Notification Method	Notification Method	Select one or more notification methods: DingTalk. Email: SMS: text message. Webhook: Specifies the service endpoint. Phone: Phone number. Ensure the recipient's phone number is verified, or the notification will not be effective. You can click Notification Recipient Management below, and if the Unverified tag appears in the Phone column for the target contact on the Contacts tab, click the tag to complete the verification. Important Ensure your chosen notification method has a corresponding notification object configured. For DingTalk, this means adding a DingTalk robot notification object after selecting DingTalk as the method.
   	Notification Object	You can notify multiple Notification Recipients simultaneously. You can either select or search for them directly. Before you select Notification Recipients, you need to click Notification Recipient Management on the right to create them. For more information, see Storage Management and Operations.
   	Alert Noise Reduction	Click Advanced Settings and turn on Alert Noise Reduction. When enabled, alerts are sent only when thresholds are met continuously. This prevents alerts for temporary issues, such as brief failovers caused by cluster scheduling or auto-tuning.
   	No Data Alerts	Click Advanced Settings, turn on No Data Alerts, and specify the duration. Triggers an alert if no monitoring data is reported within the selected time period. Common causes: JobManager exceptions, job stops, or reporting link failures.

4. You can click OK.

   After you save the alert rule, it is enabled by default and appears in the alert rule list. You can then stop, edit, or delete the rule.

Single job/multi-job configuration (ARMS console)

1. Log on to the Realtime Compute console.

2. In the Actions column for your workspace, click More > Monitoring Indicator Configuration to access the ARMS console.

   The workspace name, workspace ID, and the corresponding Prometheus instance name appear at the top of the page.

   

3. In the navigation pane on the left, click Alert Rules to create an alert rule.

   • Detection type: supports metric-based alerting through static thresholds and custom PromQL, except for metrics that Flink already supports for alerting.

   • Filter Conditions enable batch alert configuration. For Namespace, enter the namespace name. Select All to apply to all namespaces in the workspace. For Deployment, enter the Deployment Job ID for the target job within the namespace (you can find this ID on the Deployment Details tab of the Flink job). Select All to apply to all jobs in the namespace.

   For more information about the other configuration parameters, see Create a Prometheus alert rule. You can also create a Prometheus alert rule template. For more information, see Create a Prometheus alert rule template.