We're occasionally getting close to maxing out our NFS server's CPU. I'd rather find a way to scale out storage instead of scale up. I have been looking at Portworx this last week as it would appear to meet our needs however, I've run into a couple of issues.
Guide 1 - https://docs.portworx.com/scheduler/kubernetes/gke.html
Seems to fail when trying to install into the cluster as GKE doesn't open it's etcd implementation up to outside access. I suppose I could run an etcd deployment, but haven't looked into this.
Guide 2 - https://github.com/portworx/terraporx/tree/master/gcp/kubernetes_ubuntu16
Does work - but creates a four machine cluster all in instance space with no autoscaling of nodes. It's also separate from GCP, in that GCP doesn't even know it's a K8S cluster. GCP takes care of things like LoadBalancers - so am not sure, if I stuck with Guide 2, how to get my traffic into my services.
So.. Am after ways to make either Guide 1 actually work - or a way to autoscale out the cluster in Guide 2?
I managed to get this working - am now running through hoops to get NFS mounts going.
Essentially, Guide 1 is missing information before "Go to the kubernetes install page to continue"
GKE doesn't expose access to etcd, so you have to build etcd as a deployment on your cluster, or a second cluster before portworx will install.
I followed the GKE etcd guide here:
https://github.com/heneise/k8s-etcd-cluster
Once the service was brought up, I pointed the install at the service public ip address to get going.
this is a bit late but better late than never. Yes, I have run Portworx on GKE (disclaimer i work for portworx). The updated documentation can be found here:
Yes, its recommended to install your own ETCd. However, you can run the etcd-operator (etcd on Kubernetes) and point portworx at that so it is self healing and gains benefits of Kubernetes.
Or you can use our installer and click "built in (Beta)" to have the key value storage be a part of the portworx cluster.
Also feel free to stop by portworx slack for more help.