The new docs.trellix.com features a modernized UI and AI-powered conversational search. Content is currently available in English, with additional languages expected in November 2026. We hope you enjoy the updated experience.

Monitoring network traffic with the Flink detection engine

Prev Next

The Apache Flink-based engine detects suspicious network and user activity in real time. This engine uses Complex Event Processing (CEP) patterns to analyze security events across multiple traffic types. The engine identifies both single and correlation anomalies.

The system generates an alert when traffic matches specific rule criteria. These alerts appear in the NDR interface. Use these alerts to identify potential malicious attacks against your environment.

The detection process follows a specific pipeline:

  1. Ingestion: The engine collects various network event data from the appliance based on the rule criteria.

  2. Normalization: The engine converts incoming events to a common and standard internal format to reduce processing time.

  3. Baselining: Selected rules learn the standard behavior of an entity (for example, a user or host)

    and flag deviations. Baselining is optional for a rule.

  4. Detection: The system generates alerts when traffic matches a rule pattern or deviates from a learned baseline.

  5. Contextualization: Alerts are enriched with MITRE ATT&CK mappings and risk scores.

  6. Suppression: The system suppresses duplicate alerts for a rule-specific cooldown period to prevent alert fatigue.

The Flink Detection Engine is an internal service and does not require manual configuration from the user interface.

The engine depends on the following requirements:

  • Licensing: The engine runs if the user has an Enterprise or Core license.

    Note

    The engine does not run on an Essential license.

  • Data Transfer Server (DTS) connectivity: After connecting to the DTS, the engine fetches the latest published rules. It periodically fetches new changes every 5 hours. The engine deploys the new or updated rules immediately after a successful fetch.

Note

If the flinkdetection service lacks DTS connectivity, the engine cannot perform detections or run the latest rules.

Use the CLI to confirm the status of detection services and rule versions.

  1. Log in to the Trellix NDR CLI using a terminal window or SSH.

  2. Run this command: show version

  3. Locate the flinkdetection entry in the list.

  4. Review the running version services list.

  5. Verify that the Flink service name and versions appear in the output.

  • Real-time detection: Detection starts when the engine becomes active. The engine does not scan historical traffic or detect activity that occurred before it started.

  • Live data required: Detections depend on NDR sensors and collectors actively forwarding events to the engine. If no events are received, no detections are generated.

  • Alert deduplication: By default, the engine suppresses duplicate alerts for the same rule match during the configured cooldown period. You receive one alert per cooldown period rather than one alert for each matching event.

  • Automatic recovery: If the Normalization job stops unexpectedly, the system automatically restarts it. No manual restart is required.

  • Managed configuration: Rules, thresholds, and suppression settings are managed internally and cannot be configured by users. Per-rule configuration is not available in the user interface.

Best practices

  • Verify that the detection services are running after an appliance upgrade or restart. See Verify Flink engine connectivity and version.

  • Ensure that Trellix NDR sensors and collectors are actively forwarding events before expecting detections.

  • Do not expect the engine to detect historical events. Detection is performed in real time.

  • Do not manually modify rules or internal service configuration. These settings are managed by the system, and manual changes can be overwritten or affect the detection pipeline.

  • Do not manually restart the detection or Flink cluster services to troubleshoot missing detections. The system automatically recovers the services, and unnecessary restarts can interrupt live detection.