Best Practices for Live Process Monitoring

Cette page n'est pas encore disponible en français, sa traduction est en cours.
Si vous avez des questions ou des retours sur notre projet de traduction actuel, n'hésitez pas à nous contacter.

Overview

With the Live Processes product, you can monitor the number of running processes across your entire infrastructure. Live process monitors are most useful for adding observability to non-containerized processes.

Use live process monitors to:

  • Ensure that you have enough replicas of a process to serve customers.
  • Alert when a specific process is running.

An improperly configured monitor is prone to false positives. This guide covers the recommended best practices for creating a reliable live process monitor. For a detailed overview of the monitor creation process, see Live process monitor creation.

Best practices

Scoping the monitor

Datadog recommends scoping your monitor to no more than a few thousand processes. Because text search is fuzzy, tags are the most accurate way to adjust the scope of your monitor.

Example workflow:

  1. Navigate to the Monitors > New Monitor > Live Process page.

  2. Add tags to the monitor in the by tags field.

    • For example, use command:puma to monitor processes associated with the puma command.
A live process monitor that has been scoped using a tag
  1. Optionally, refine the monitor’s scope by adding search text to the by text field. In the example below, only processes whose command line matches cluster worker are included.
A live process monitor that has been scoped using text search
  1. If your monitor’s scope still exceeds a few thousand processes total across all monitor groups, use additional tags to break it into multiple monitors.
    • For example, you can use the env tag to create separate monitors for prod and staging.

Choosing a time frame

A common misconception is that increasing the evaluation window leads to slower responses or missed alerts, but a monitor continuously assesses data no matter what query evaluation interval you choose. The evaluation interval only determines how many datapoints are used to decide whether an anomaly exists.

By increasing the evaluation window, you can ensure that you are only alerted if a behavior is happening consistently, not temporarily.

  • To avoid false positives, use a minimum interval of 5 minutes.
  • If your monitor uses tags that come from a cloud provider integration, use a minimum interval of 15 minutes.
  • To avoid delayed alerts, use a maximum interval of 1 hour.

For additional guidelines, see Best Practices to Prevent Alert Fatigue.

PREVIEWING: rtrieu/product-analytics-ui-changes