Open ylhsiehitri opened 5 years ago
It seems I misunderstood... According to the section 2 in installation guide [1], it means to add those things in 2.1 and 2.2 to the default kube-scheduler and replace the default kube-scheduler, right?
If yes, I tried to modify /etc/kubernetes/manifests/kube-scheduler.yaml (then "$kubectl get po -n kube-system" will see the pod "kube-scheduler-leadtek-gs4820" restarts. Here "leadtek-gs4820" is the node name), but then I examine the running config (by "kubectl edit po -n kube-system kube-scheduler-leadtek-gs4820"), there's no update.
[1] https://github.com/AliyunContainerService/gpushare-scheduler-extender/blob/master/docs/install.md
Hello, it is indeed to modify the default kube-scheduler file, but I deployed the GPU shared pod after the configuration is completed, found no effect, then I tried to restart the machine, to achieve the desired effect
@Vae1997 , why can't we create one more scheduler instead of modifying the default one?
@ide8 Hello, I am just a beginner, the specific principle is not very clear, so I can only say my thoughts:
If you modify the default scheduler, k8s should be re-introduced by internal related mechanisms without affecting the overall environment of the cluster. Deploy the modified scheduler so that your changes take effect.
Conversely, if you deploy a scheduler yourself, you may conflict with the default scheduling mechanism. On the other hand, your changes will not take effect. The worst case is that the original scheduling mechanism of the cluster will not work.
Of course, there should be a way to delete the default scheduler and redeploy as needed. It should be the same as when building a cluster with binary files. (But I haven't tried to build a k8s cluster from a binary file. It's not clear if this operation will affect the existing cluster.)
Hello, it is indeed to modify the default kube-scheduler file, but I deployed the GPU shared pod after the configuration is completed, found no effect, then I tried to restart the machine, to achieve the desired effect
Similarly to @Vae1997 that restarting the machine, for some reason I happened to re-install OS, then the problem just disappeared...
Hi,
I tried to create the kube-scheduler with
kubectl create -f https://github.com/AliyunContainerService/gpushare-scheduler-extender/blob/master/config/kube-scheduler.yaml
, but failed.Even if I remove the
livenessProbe
section in the aforementioned kube-scheduler.yaml, the kubectl logs still shows the same error.What's going wrong...?
Thanks!