Centralising monitoring

The monitoring server and its probes: collection by polling or by pushing, SNMP, the four states of a service and confirmation, dependencies between hosts, escalation, Prometheus and its exporters, and who watches the monitor.

  • 34 cards
  • 5 topics
  • Free

The outline, topic by topic

Each topic is weighted the way the exam weighs it.

  1. Concepts
  2. Questions
  3. On the diagram
  4. Scenarios
  5. Understand

A few cards from the deck

  • Probe: what is it?

    Show the answer

    An automatic test that queries a host or a service and returns a state: OK, warning, critical or unknown.

    A probe is only as good as what it checks: a test that the process exists says nothing about the page the user sees.

  • Nagios allows 3 attempts. A probe returns OK, then critical, then critical. What is the service's state?

    • Soft critical (2/3), no notification
    • Hard critical, the notification goes out
    • Warning, until the third attempt has happened
    • OK, since there are not yet three failures
    Show the answer

    Soft critical (2/3), no notification

    The service is already critical, but the state is only confirmed at the third attempt in a row. Until then it is soft: it shows, but nobody is notified.

  • To avoid false alarms, a critical result first stays … until the maximum number of attempts; it then becomes … and the notification goes out.

    Show the answer

    To avoid false alarms, a critical result first stays soft until the maximum number of attempts; it then becomes hard and the notification goes out.

    A lost packet or a one-second load spike then wakes nobody up. The price is a delay: after the first failure, it takes another (number of attempts − 1) × retry interval before you are told.

  • Monitoring agent: what is it?

    Show the answer

    A small program installed on the monitored machine, which measures locally (processor, disk, processes) and sends the results to the monitoring server.

    It sees what an outside probe cannot, such as a nearly full disk, but it is one more piece of software to install and update on every machine.

  • All the monitoring lights are green while the site no longer responds. What is the most likely cause?

    • The probe checks the process exists, not that a page shows
    • The alert threshold is set too low and drowns real alerts
    • The maximum number of attempts is too low to confirm the state
    • The SNMP community changed and the probes return "unknown"
    Show the answer

    The probe checks the process exists, not that a page shows

    A process can be running, stuck, serving nobody. A useful probe imitates the user: it loads the page and checks its content.