as i wrote in last posts regarding snmp-check-timeout, i had install a checkmk-raw standalone without docker to compare check duration between these 2 setup environments.
What did is see? SNMP-Check to a simple printer…
Duration snmp-check standalone: Test 1=16,0sec, Test 2=11,0sec
Duration snmp-check docker: Test 1=29,2sec, Test 2=29,3sec
OK, not scratching the timeoutdeadline… but tooks the double of time, could be a bottleneck
What did i also see? Printers are stale after bulk discovery…
Duration snmp-check standalone: no probelem, services goes ok after few minutes
Duration snmp-check docker: services stale UNTIL I DID A MANUAL RESCHEDULE CHECK
What the hell? I don´t understand this…
Is there anybody out there, who has a same case? Distributed setup, slaves and master are docker-container, version ist raw 1.6.0p9
no other docker users here? here´s a screenshot from a tcpdump, what you can see is, on the right side (cmk as dockerimage) there are delays between every set of get response an get bulkrequest.
On the left side (cmk as standalone vm), no delays.
In my view, the amount of the delaytime could be the point for timeouts and stale services.
This behavior depends only on your docker setup. If i run a CMK container on a host with limited resources then it looks like your output. If the machine is powerful and you assign enough CPU resources it should be quicker.
thanks for your input. For the first time we used checkmk as dockerimage, we did the ressource setup equal to the normal vm which had no performace issues till now.
In front of your comment i gave the server more cores and memory. From 4 to 8 cores and 6 GB Ram to 12 GB Ram. From Friday till now, there are no stale services on this slave.
So far so good but one of my colluege is not really happy with this, he say this setup is too heavy for simple monitoring use case. How did you see that? Are there any recommendations for performance setups in docker environments?
I would use the docker only for testing things like build own mkp’s where i need a clean setup.
Docker containers have a very big problem to react quick if you need more resources from one second to the next. And this is the typical things happen inside a monitoring instance. At the moment it does nothing and at the next moment you want to check 10 hosts parallel. This will not work in a container setup.
Productive systems are VMs for smaller installations, VMs with dedicated host for bigger things or real hardware for really big installations
