Server monitoring is monitoring the system and application status with measurements and reporting situations that require intervention to the responsible person. A good alert layout makes the problem affecting the user and the work to be done visible, rather than just producing a lot of graphs.
Which metrics are evaluated together?
CPU, memory and disk values are helpful in investigating the cause; But it is also necessary to check that the application can actually be used. Indicators such as web request, login, database connection, or job queue are selected based on the application.
Example scenario: The server receives a ping response, but the offer form does not create a registration. Even if network access is successful, the user's job is unsuccessful. The control must represent the actual function.
Should an alarm be turned on for every high value?
Persistent deterioration must be distinguished from transient elevations. Immediately reporting every measurement change can make it difficult to notice important alerts. Prometheus recommends setting fewer and more meaningful alerts based on user impact and symptoms that require action.
Technical reference: Prometheus — Alarm design.
Threshold, waiting time and priority are determined by the normal behavior of the relevant application. It is not a sufficient design to consider a fixed CPU percentage taken from the Internet as critical in all systems.
What information should be included in an alarm?
- Affected service and observed symptom.
- Time of onset and whether it continues.
- Link to the relevant chart, record or control result.
- First review step and responsible team.
- Takeover person and notification channel if necessary.
How to control the monitoring system itself?
Stopping the measurement collecting system or disrupting the notification path can create an invisible gap. Test with a controlled experiment that the alarm occurs, reaches the right person, and enters the request process.
In the monthly evaluation, recurring problems, false alerts and actions that do not close should be addressed rather than how many alarms come. With your app list and current notification layout server management and monitoring You can clarify its scope.
Related resources
Official sources
Technical documentation and legislation may change; The relevant version and current text should be checked before application.
Frequently Asked Questions
Does monitoring automatically correct the fault?
Some processes can be automated; monitoring essentially detects the situation. Automatic intervention should also be designed and tested.
Can monitoring be done only during working hours?
Measurement collection and human intervention are different. Notification and intervention hours should be clearly stated within the scope of service.