Azure Support

[cforce]

Please provide an story how to ruin kubo together with pcf on azure (also if azure container services might not be available)

[cf-gitbot]

We have created an issue in Pivotal Tracker to manage this. Unfortunately, the Pivotal Tracker project is private so you may be unable to view the contents of the story.

The labels on this github issue will be updated when the story is started.

[glestaris]

@cforce Azure support has been on our roadmap but unfortunately fairly low on priority ATM.

However, I'd like to understand more about your use case and expectation for Kubo on Azure. E.g. what workloads would you want to run? Would you need multiple AZs? Which load balancer? Would you need persistent volumes? etc.

[cforce]

Mostly the ide is to run statefull services so actually it is needed to support persistent volumes.
I thought that there is no concept of AZ on Azure but availability sets (AS) , which should be supported by any load on Azure, doesn't it? Anyway to fulfill HA (we definitely need) we require that AS support.
I would prefer to use the same LB as i use for PCF on Azure, isn't that the reference for integrated setup of PCF and Kubo/CFCS?

[glestaris]

Mostly the ide is to run statefull services so actually it is needed to support persistent volumes.
I thought that there is no concept of AZ on Azure but availability sets (AS) , which should be supported by any load on Azure, doesn't it? Anyway to fulfill HA (we definitely need) we require that AS support.

Thanks @cforce, that's useful context.

I would prefer to use the same LB as i use for PCF on Azure, isn't that the reference for integrated setup of PCF and Kubo/CFCS?

Yes, we are supporting using the same LB.

[glestaris]

@cforce we started looking into Azure support

[cforce]

1.) In the meantime we are using a lot k8 loads and i think the only future promising approach in the cloud (also in azure) is to make use the managed kubernetes service of the cloud provider isntead managing your onw stack on TOP of cloud IAS. Not only from moving the responsibiblty of non/functional requirments and efforts (like scalability, cloud IAS integration tests of supported base containers, zero downtime maitenance/updates, integration of cloud services into kuberntes service catalogue/service brokering) but also from risk/financial perspective ( financial backed availability, only pay for workload not for management nodes.
2.) I would really hope to see PCF itsetlf would introduce a new CPI layer, that uses Kuberntes (and therfore also the cloud provider offering) as foundation. So instead of managing vm's you manage pod and container/workers to let PCF fill the gap between CAAS and application platform for devops.

[svrc]

I've made some progress here, #285

[grenzr]

Definitely keeping an eye out for this. Currently we're using acs-engine to deploy k8s in Azure, but I find it quite fiddly especially in the area of deployment lifecycle changes. I also love bosh from past experience deploying cf etc and would like nothing more than to switch to it asap! I'm willing to assist with the progress of this ticket, either with development or testing in the field.

We currently deploy k8s with flannel support, using vm scale sets (support for this added fairly recently: kubernetes/kubernetes#43287) although I assume bosh would want to control the vm deployments itself, but Azure AZ's do support vms or vm scale sets (see: https://docs.microsoft.com/en-us/azure/availability-zones/az-overview) which would be useful for cross-az deployments.

We also have a vpn gateway to on-prem, so we need to configure k8s services to be on a specific azure subnet which has permission to interop with on-prem machines. Terraform sets all of that side of things up though.

Nothing too crazy so far, but plan to make it multi-az and AD based RBAC soon.

[alex-slynko]

It has been releases as an alpha in CFCR 0.23!
We will keep improving support for this.