Let UNIX agents execute sections/plugins/local checks in parallel

130 votes

As systems are getting more complex and larger, it becomes increasingly difficult to keep the agent's runtime below 60 seconds, as each check is unnecessarily processed sequentially.

In bigger env. it's very likely that you will reach your limits just by having 2-3 plugins running on the same host.

Here are a few runtimes from real customer environments:
35s cmk_agent
25s mk_docker
20s mk_mysql
20s mk_postgres
25s mk_logwatch
90s mk_oracle (even in parallel mode which is supported inside the plugin itself)
60s mk_inventory (big Hardware Unix Systems)
40s mk_filestats

Simply running everything asynchronously is only a makeshift solution. On larger systems, starting multiple checks in parallel is not a problem, as they are designed to handle large loads.

Give the customer the option to choose running the Checks in parallel (wait for Result), async (don't wait) or sequentially (default) and with different Intervals and Timeouts.
Make the agent run only as long as the slowest plugin

Under consideration Checks&Agents Suggested by: Daniel Roettgermann (07 Jun, '22) • Upvoted: 12 Apr • Comments: 9

Comments: 9

13 Jun, '22
Andrea Kaufmann
Important for a professional environment
23 Jun, '22
Lars Sörensen
- add the same basic functions to all unix agents (linux, aix, solaris, hp...), It's really exhausting to ask this for every OS you have again and again.
- add waitmax function to all checks (local, mrpe, plugins, ...) like in the windows agent
24 Jun, '22
Mike1098
We fully support this but as mentioned by Lars this should be available for all Agents on all supported OS. Nevertheless the agent must remain smart and efficient. If it consumes to much resources it will not be accepted.

We have the experience that some runtime issues could be fixed on the monitored system.
1
05 Aug, '22
JPH
Oh yes please make all, including Windows, agents consistent in all behaviours an general settings and loggings!
Agents has to be handled all together not per OS.
1
15 Oct, '22
Lars Sörensen
In addition to this feature request:
https://features.checkmk.com/suggestions/336976/add-waitmax-in-all-unix-agents-to-the-local-mrpe-plugin-and-run_cached-checks
14 Aug, '23
Florian
A clear upvote! We have some checks running (especially mk_sap) which we have divided into several plugins, just here we urgently need the parallel execution!
1
12 Oct, '23
Christian
Same at our site! mk_sap, mk_sap_hana, mk_oracle are running very long. Parallel execution would be very, very appreciated.
2
19 Dec, '23
Lars Sörensen
Parallel execution would also have a positive effect on the core in large environments, as the core would have to wait much less time for the agent to finish.

I think this Feature could be accomplished relatively easily by copying the run_cached functionality and adapting it accordingly so that the checks are simply executed in parallel. The new function could be extended with a configurable delay between the start of individual checks so that the server is not stressed too much and a configurable timeout per server so that long runners do not block the agent as is the case today.
20 Dec, '23
Niklas Pulina Admin
Hi Lars, thanks a lot for your suggestion. We will take it into consideration when we start working on this functionality.