Baseline creation

Baseline creation involves collecting sufficient performance metrics using the mmhealth anomaly collect command, which enables the system to train the anomaly detection algorithm and establish a baseline for monitoring and identifying abnormal behavior.

About this task

The mmhealth anomaly collect command collects the performance metrics defined in the all_pdisks_spc_rule configuration for the previous 12 hours. If the collected data is sufficient, Anomaly Detection automatically creates a worker process to train the detection algorithm using these metrics and the configuration settings. The baseline, derived from this historical data, is a prerequisite for anomaly detection monitoring. Once training is complete, the worker uses the established baseline to continuously monitor metrics and identify deviations that indicate anomalous behavior.

Important: Ensure that performance monitoring is correctly configured and running.

Procedure

  1. Run the following command to collect baseline data for the required configuration:
    mmhealth anomaly collect <configuration>
    Example:
    # mmhealth anomaly collect all_pdisks_spc_rule
    This command collects performance metrics defined in the specified configuration for the previous 12 hours.
  2. Review the command output to verify baseline creation.
    • If all elements have adequate data as shown in the following example:
      Collect baseline data:
      
      config                  result                                status
      good_pdisks_spc_rule    5/5 element(s) have adequate data     started and baseline is OK
      
      Use subcommand 'show config(s)' for details and 'start config(s)' if enough elements have a good baseline.
      Baseline files are in /var/mmfs/anomaly/baselines'
      The output indicates that a total of five elements were detected. All five elements provide data of sufficient quality for training the Anomaly Detection algorithm. As a result, anomaly detection monitoring starts automatically. Because all required elements have a valid baseline, you can ignore the message shown at the end of the output.
    • If the baseline is empty as shown in the following example:
      Collect baseline data:
      
      config                result              status
      all_pdisks_spc_rule   baseline is empty   no baseline
      
      Use subcommand 'show config(s)' for details and 'start config(s)' if enough elements have a good baseline.
      Baseline files are in /var/mmfs/anomaly/baselines'
      
      The mmhealth anomaly collect command reports an empty baseline. You do not have any workload on any pdisk, no data is collected for any pdisk or no data can be queried from the ZIMON database. Except the situation that no IO exists, the root cause of the problem is related to performance monitoring.
  3. Resolve issues and repeat baseline collection if required.