Werk #1563

KomponenteChecks & Agents
TitelReworked configuration of process monitoring
Datum2014-12-03 12:41:26
Check_MK EditionCheck_MK Raw Edition (CRE)
Check_MK Version1.2.6b1
Level2 - Wichtige Änderung
KlasseBug Fix
KompatibilitätIncompatible - Manual interaction might be required

The configuration of the monitoring of processes has been reworked. This was mainly to overcome a bug: when the process check was used on a cluster then the check parameters would be taken randomly from one of the nodes.

The new process monitoring works like this:

Static checks

Statically configured checks are like before. They are configured via the ruleset State and count of processes.

Checks created by service discovery

The ruleset Process Inventory now does not set parameters (levels) for the created services anymore. Only the settings for matching process names and user remain - and of course the service description.

The new ruleset State and count of processes in the section Parameters for inventorized checks (please do not mix up with static checks here) is used for defining levels for the count, CPU usage, averaging and all of the other parameters. The default is to make the check OK if at least one process is being found. So from now on if you are using process inventory with custom levels, you need to make configurations in both rulesets.

On the other hand that can make things easier for you. You might have a lot of rulesets for detecting various processes but then only need a few rules for setting levels for these services.

In order to make the migration easier services that have already been created with previous versions of Check_MK will remain to work as they did. But after a new service discovery the levels will vanish and then you have to configure them in the new extra ruleset.

Please note: the ruleset State and count of processes does contain the settings for matching the process and the user even when used for discovered checks. The reason is that it is the same ruleset as for manual checks. Setting these parameters do not make sense here in most cases. Simply leave them unchecked.

Performance data - ps versus ps.perf

The check ps.perf is now deprecated. The normal ps check now does always create performance data. For compatibility reasons the old ps.perf is still defined but does exactly the same as ps