Types of anomalous behavior

  • Release version: Xanadu
  • Updated August 1, 2024
  • 2 minutes to read
  • Summarize
    Summarized using AI
    This content was generated using new OpenAI-powered functionality. Results are provided on an as is basis and are not guaranteed to be accurate or complete.

    Summary of Types of Anomalous Behavior

    Anomalous behavior in a CI or service indicates potential issues, such as abnormal spikes in message frequency. The system monitors log streams to establish baselines for expected behavior, which can vary over different time periods (hourly, daily, weekly, or unlimited). Any behavior that deviates from these baselines is classified as anomalous.

    Show full answer Show less

    Key Features

    • Log Property Types:
      • Pattern: Repeating values or rates in text, time, or relationships.
      • Meter: Numeric or text values (e.g., status or response codes).
      • Gauge: Continuously reported numerical values indicating resource consumption (e.g., CPU or memory usage).
    • Anomaly Card: Displays anomalous activity in the Service Operations Workspace, highlighting recent anomalies against learned baselines.

    Key Outcomes

    Identifying various types of anomalies allows users to quickly recognize significant changes in behavior, which can be critical for maintaining service integrity. Key types of anomalies include:

    • New Behavior: A previously unseen pattern.
    • Signal Dead: No data from a source for at least five minutes.
    • Signal Alive: Previously inactive data is now appearing again.
    • Above/Below Average: Activity that diverges from expected baselines.
    • Baseline Changes: Significant increases or decreases in log property values compared to established baselines.
    • Severity and Keyword Correlation: Increases in volume related to specific severity levels or keywords.

    These insights enable proactive monitoring and management of services, helping customers address potential issues before they escalate.

    Anomalous behavior in a CI or a service can indicate an important issue. For example, a spike in the frequency or number of messages of a particular type can indicate a problem.

    Understanding anomalies

    To build models of expected behavior, the system monitors the log stream to learn baselines for patterns, metrics, and gauges over various time periods. Time periods can be hourly, daily, weekly, or unlimited. Behavior that departs from the learned models is considered anomalous behavior.

    Types of log property

    Pattern
    A pattern is a value or rate that repeats, whether in text, time, or relationships.
    Meter
    A meter property is a numeric or text value. For example, a status code, a response code, an action, or a pattern.
    Gauge
    A gauge property has a numerical value that is reported continuously. Gauge properties represent operations that consume resources. For example, CPU usage, memory usage, or response time.

    How anomalies appear in the Service Operations Workspace

    The Anomaly card illustrates the anomalous activity that led to the alert.
    • The blue line shows the recent anomalous activity.
    • On some charts, the lightly shaded area indicates the expected (learned baseline) behavior.

      A peach-shaded area represents the baseline values for the same hour one day earlier. A pink-shaded area shows the values for the same period in the previous week.

    • Click the information icon to see how the anomaly was identified: Information icon.
    In this example, the peach-shaded area shows the same data for the same hour one day earlier. The spike in the metric value (events per minute) is clearly visible.
    Figure 1. Anomaly card
    Anomaly card identifies and illustrates anomalous behavior.

    Kinds of anomalies

    Table 1. Some of the kinds of anomalies
    Behavior Description
    New behavior A pattern that has not ever been seen. The New Behavior alert type does not display a chart.
    Signal dead/Stopped appearing All pattern or log data from a source has stopped. There has been no signal for at least five minutes.
    Signal alive/Appearing again A pattern or log data from a "dead" source is appearing again​. For a baseline of one hour, a pattern is "dead" if it appears less than once per minute.
    Anomaly above average or below average Activity that deviates from expected baseline behavior for pattern or meter or gauge metrics, such as keywords metrics or severity metrics.
    Baseline reference​ increase or decrease An increase or decrease in the value or volume of a log property as compared to the one-hour or one-week baseline.
    Correlation of severity and keyword alerts An increase in the volume of a severity level or keyword.