Error getting cmk-config-<node-name> Configmap

[hanyin-intel]

Hi guys,

I may find several bugs towards the management of Configmap in the latest 1.5.2 version and also try to find a workaround to fix them. My patch file is attached here.
hanyin-intel@e8d5903

Here are the details:

1. The node-report feature can't get the cmk-config- Configmap since error calling "get_config" function in class "Config". Since this function is not defined in the "Config" class, I use the constructor function and "lock/unlock" function as a workaround. But it seems that the "node-report" container continuously restarts since race on such "lock".

2. The describe feature can't get the cmk-config- Configmap since the "namespace" attribute is not passed when construct the "Config" class. As a result, I use the "/var/run/secrets/kubernetes.io/serviceaccount/namespace" file in container to get the namespace and pass it to the construction function of "Config" class.

3. The isolate feature also misses the "namespace" attribute when getting the cmk-config- Configmap.

4. When enabling the "hostNetwork" in the container, the environment variable "HOSTNAME" in container has been changed to the node name rather than the pod name. Thus, the isolate feature which always be used as the wrapper of the real workload can not get the correct pod name. Then, in the workaround, I use the environment variable "PODNAME" instead of the "HOSTNAME" in isolate.py. And also, the user needs to add the "PODNAME" into their containers.

Thanks

[pbrownlow7]

Hi @hanyin-intel,

Thanks for the several workarounds. I'll have a look at your patch and if everything seems okay you can make a merge request if that sounds good?

Thanks,
Philip

[hanyin-intel]

Hi Philip, Sorry for the late reply, I can raise a pull request. But the problem is that the workaround of node-report makes the node-report container continuously restarts. When failing acquiring the lock of the core-pool “configmap”, the “except” branch will do “sys.exit(1)”. And in such case, the workaround in node-report.py will crash since we do “Config.unlock” in the finally branch. I’m not sure how to solve this problem and whether this problem is allowed. ***@***.***D736D0.C4761750] Thanks, Yin Han From: pbrownlow7 ***@***.***> Sent: Tuesday, April 13, 2021 10:11 PM To: intel/CPU-Manager-for-Kubernetes ***@***.***> Cc: Yin, Han ***@***.***>; Mention ***@***.***> Subject: Re: [intel/CPU-Manager-for-Kubernetes] Error getting cmk-config-<node-name> Configmap (#297) Hi @hanyin-intel<https://github.com/hanyin-intel>, Thanks for the several workarounds. I'll have a look at your patch and if everything seems okay you can make a merge request if that sounds good? Thanks, Philip — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#297 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AOXIIOVZXEF7XZYZYI4L7R3TIRGG7ANCNFSM42INDR7A>.