A simple Pivotal Container Service (PKS) deployment

This post will walk you through a simplified PKS (Pivotal Container Service) deployment in my lab. The reason why I say this is simplified is because all of the components will be deployed on a single flat network. PKS has a number of network dependencies. These include the bosh agents deployed on the Kubernetes (K8s) VMs being able to reach the BOSH Director, as well as the vCenter server. Let’s not get too deep into the components just yet – these will be explained over the course of the post. So rather than trying to set up routing between multiple different networks, I have deployed everything on a single flat network. Again, like some of my previous posts, this is more of a learning exercise than a best practice. What it will show you are the various moving parts that need to be considered when deploying your own PKS. If you are looking for a more complex network deployment, have a read of the excellent series of posts that my good pal William Lam put together. He highlights all of the steps needed to get PKS deployed with NSX-T. I will be referencing William’s blogs from time to time as they contain some very useful snippets of information. Note that this is going to be quite a long post as there are a lot of pieces to deploy to get to the point where a K8s cluster can be rolled out.

If you want to learn more about what PKS is about, have a read of this blog post where I discuss the PKS announcement from VMworld 2017.

1. Networking overview

Let’s begin by having a quick look at the various components and how they are connected. I put together this diagram to assist with understanding the network dependencies.

To get started, the Pivotal Ops Manager is deployed. This needs to be able to communicate with your vSphere environment. This is responsible for deploying the BOSH Director component and then the Pivotal Container Service (PKS). Once these are deployed, a PKS client VM is setup which contains the UAA (user account and authorization) and PKS command line extensions. These are used to get the correct privileges to roll out the Kubernetes cluster(s). As I mentioned in the opening paragraph, the BOSH agents on the K8s nodes need to reach back to the BOSH director to pull down appropriate packages, as well as reach back to the vSphere environment to request the creation of persistent disks for the master and workers. While this makes the integration with NSX-T really useful, I wanted just look at the steps involved without having NSX-T in the mix. And because I am restricted in my own lab environment, I went with a single flat network as follows:

Now that we know what the networking looks like, lets go ahead and check on what components we need to deploy it.


2. PKS Components

We already mentioned that we will need the Pivotal Ops Manager. This is an OVF which, once deployed, can be used to deploy the BOSH Directory and PKS, the Pivotal Container Service.

We will then need to deploy the necessary components on a VM which we can refer to as the PKS Client. The tools we need to install in this VM are the UAA CLI (for user authentication), the PKS CLI (for creating K8s clusters) and then finally the Kubectl CLI (which provides a CLI interface to manage our K8s cluster).

This is all we need, and with this infrastructure in place, we will be able to deploy K8s clusters via PKS.


3. Deploying Pivotal Ops Manager

This is a straight-forward OVF deploy. You can download the Pivotal Cloud Foundry Ops Manager for vSphere by clicking here. The preference is to use a static IP address and/or FQDN for the Ops Manager. Once deployed, open a browser to the Ops Manager, and you will be presented with an Authentication System to select as follows:

I used “Internal Authentication”. You will then need to populate password fields and agree to the terms and conditions. Once the details are submitted, you will be presented with the login page where you can now login to Ops Manager. The landing page will look something like this, where we see the BOSH Director tile for vSphere. BOSH is basically a deployment mechanism for deploying Pivotal software on various platforms. In this example, we are deploying on VMware vSphere, so now we need to populate a manifest file with all of our vSphere details. To proceed, we simply click on the tile below. Orange means that it is not yet populated; green means that is has been populated.


4. Configuring BOSH Director for vSphere

The first set of details that we need to populate are related to vCenter. As well as credentials, you also need to provide the name of your data center, a datastore and whether you are using standard vSphere network (which is what I am using) or NSX. Here is an example taken from my environment.

The next screen is the Director configuration. There are only a handful of items to add here. The first is an NTP server. The others are check-boxes, shown below. The interesting one is the “Enable Post Deploy Scripts”. If this is enabled, BOSH will automatically deploy K8s applications on your Kubernetes cluster immediately after the cluster is deployed. This includes the K8s dashboard. If this checkbox is not checked, then you will not have any applications deployed and the cluster will be idle after it is deployed.

This brings us to Availability Zones (AZ). This is where you can define multiple pools of vSphere resources that are associated with a network. When a K8s cluster is deployed, one can select a particular network for the K8s cluster, and in turn the resources associated with the AZ. Since I am going with one large flat network, I will also create a single AZ which is the whole of my cluster.

Now we come to the creation of a network and the assigning of my AZ to that network. I am just going to create two networks, one for my BOSH Director and PKS and another for K8s. But these will be on the same segment since everything is going to sit on one flat network. And this is the same network that my vCenter server resides on. As William also points out in his blog, the Reserved IP Ranges need some explaining. This entry defines which IP addresses are already in use in the network. So basically, you are blocking out IP ranges from BOSH and K8s, and anything that is not defined can be consumed by BOSH and K8s. In effect, for BOSH, we will required 2 IP address to be free – the first for BOSH Director and the second for PKS which will deploy shortly. So we will need to block all IP addresses except for the two we want to use. For K8s, we will require 4 IP addresses for each cluster we deploy, one for the master and 3 workers. So block all IP addresses except for the 4 you want to use. This is what my setup looks like, with my 2 networks:

Finally, we associate an Availability Zone and a Network with the BOSH Director. I chose the network with 2 free IP addresses created previously, and my AZ is basically the whole of my vSphere cluster, also created previously.

I can now return to my installation dashboard (top left hand corner in the above screenshot), and see that my BOSH Director tile has now turned green, meaning it has been configured. I can see on the right hand side that there are “Pending Changes” which is to install the BOSH Director. If I click on Apply changes, this will start the roll out of my BOSH Director VM in vSphere.

You can track the progress of the deployment by clicking on the Verbose Output link in the top right hand corner:

And if everything is successful, you will hopefully see a deployment message to highlight that the changes were successfully deployed.

You should now be able to see your BOSH Director VM in vSphere. The custom attributes of the VM should reveal that it is a BOSH director. That now completes the deployment of the BOSH director. We can now turn our attention to PKS.


5. Adding the PKS tile and necessary stem cell

Now we install the Pivotal Container Service. This will create a new tile once imported. We then add a “stemcell” to this tile. A “stemcell” is a customized operating system image containing the filesystem for BOSH-managed virtual machines. For all intents and purposes, vSphere admins can think of this as a template.  PKS can be downloaded from this location. On the same download page, you will notice that there are a list of available stemcells. The stemcell version needs to match the PKS version. The required version will be obvious when PKS is deployed.

To begin the deploy, click on “Import a Product” on the left hand side of the Ops Manager UI. Browse to the PKS download, and if the import is successful, you will see the PKS product available on the left hand side as well, along with the version.

Click on the + sign to add the PKS tile, and it will appear in an orange un-configured state, as shown below.

You will also notice that this tile requires a stemcell, as highlighted in red in the PKS tile. Click on this to import the tile.

Here you can also see the required version of stemcell (3468.21) so this is the one that you need to download from Pivotal. This will be different depending on the version of PKS that you choose. From the PKS download page, if we click on the 3468.21 release, we get taken to this download page where we can pick up the Ubuntu Trusty Stemcell for vSphere 3468.21. Once the stemcell is imported and applied, we can return to the Installation Dashboard and start to populate the configuration information needed for PKS by clicking on the tile.


6. Configure PKS

The first configuration item is Availability Zones and Networks. As I have only a single AZ and a single flat network, that is easy. For the Network selection, the first network placement is for the PKS VM and the second is for the K8s VMs. I will place PKS on the same network as my BOSH Director, and the Service Network will be assigned the network with the 4 free IP address (for master and 3 workers). Now in my flat network setup, these are all on the same segment and VLAN, but in production environments, these will most likely be on separate segments/VLANs, as shown in my first network diagram. If that is the case, then you will have to make sure that the Service Network has a route back to the BOSH Director and PKS VMs, as well as your vCenter server and ESXi hosts. In a future blog, I’ll talk about network dependencies and the sort of deployment issues you will see if these are not correct.

The next step is to configure access to the PKS API. This will be used later when we setup a PKS client VM with various CLI components. This certificate will be generated based on a domain, which in my case is rainpole.com. Populate the domain, then click generate:

And once generated, click on Save.

Next step is to populate the Plans. These decide the resources assigned to the VMs which are deployed when your Kubernetes clusters are created. You will see later how to select a particular plan when we create a K8s cluster. These can be left at the default settings; the only step in each of the plans is to select the Availability Zone. Once that is done, save the plans. Plan 1 is small, Plan 2 is medium. Plan 3 (large) can be left inactive.

This brings us to the K8s Cloud Provider. Regular readers of this blog might remember posts regarding Project Hatchway, which is a VMware initiative to provider persistent storage for containers. PKS is leveraging this technology to provide “volumes” to cloud native applications running in K8s. This has to be able to communicate with vSphere, so this is where these details are added. You will also need to provide a datastore name as a VM folder. I matched these exactly with the settings in the vCenter configuration for the BOSH Director. I’m not sure if this is necessary (probably not) but I didn’t experience any issues reusing them here for PKS.

Networking can be left at the default of Flannel rather than selecting NSX.

The final configuration step in the PKS tile is the UAA setting. This is the user account and authorization part, and this is how we will manage the PKS environment, and basically will define who can manage and deploy K8s clusters. This takes a DNS entry and once PKS is deployed, that DNS entry will need to point to the PKS VM, once it is deployed. I used uaa.rainpole.com.

If we now return to the installation dashboard, we should once again see a set of pending changes, including changes to BOSH Director and the installation of Pivotal Container Service (PKS). Click on Apply changes as before. The changes can be tracked via the verbose output link as highlighted previously.

If the changes are successful, the deployment dashboard should now look something like this:

So far so good. The next step is to set up a PKS client with the appropriate CLI tools so that we can now go ahead and roll out K8s clusters.


7. Configuration a PKS Client

I’m not going to spend too much time on the details here. William already does a great job on how to deploy the various components (uaac, pks and kubectl) on an Ubuntu VM in his blog post here. If you’d rather not use Ubuntu, we already saw where the CLI components can be downloaded previously in this post. The CLI components are in the same download location as PKS. When the components are installed, it would now be a good time to do your first DNS update. You will need to add uaa.rainpole.com to your DNS to match the same IP as the PKS VM (or add it to the PKS client /etc/hosts file).


8. Deploy your first K8s cluster

First step is to retrieve the secret token for your UAA Admin. Select the Pivotal Container Service tile in Pivotal Ops Manager, then select the credentials tab and then click on the Link to Credential. Here you will find the secret needed to  allow us to create an admin user that can then be used to create K8s clusters via PKS. In the final command, we include the role “pks.clusters.admin” which will give us full admin rights to all PKS clusters.

  • uaac target https://uaa.rainpole.com:8443 –skip-ssl-validation
  • uaac token client get admin -s
  • uaac user add admin –emails admin@rainpole.com -p
  • uaac member add pks.clusters.admin admin


Before we create our K8s cluster, there is another very useful set of bosh CLI commands. In order to run these commands however, we need to authenticate against our Pivotal Operations Manager. Here are the two “om” (short for ops manager) commands to do that (you will replace the pivotal-ops-mgr.rainpole.com with your own ops manager):

root@pks-cli:~# om –target https://pivotal-ops-mgr.rainpole.com-u admin -p -k curl -p /api/v0/certificate_authorities | jq -r ‘.certificate_authorities | select(map(.active == true))[0] | .cert_pem’ > /root/opsmanager.pem
Status: 200 OK
Cache-Control: no-cache, no-store
Connection: keep-alive
Content-Type: application/json; charset=utf-8
Date: Mon, 16 Apr 2018 10:17:31 GMT
Expires: Fri, 01 Jan 1990 00:00:00 GMT
Pragma: no-cache
Server: nginx/1.4.6 (Ubuntu)
Strict-Transport-Security: max-age=15552000
X-Content-Type-Options: nosniff
X-Frame-Options: SAMEORIGIN
X-Request-Id: 8dea4ce9-a1d5-4b73-b673-fadd439d4689
X-Runtime: 0.034777
X-Xss-Protection: 1; mode=block
root@pks-cli:~# om –target https://pivotal-ops-mgr.rainpole.com-u admin -p -k curl -p /api/v0/deployed/director/credentials/bosh2_commandline_credentials -s | jq -r ‘.credential’


Ignore the BOSH_CA_CERT output, but take the rest of the output from the command and update your ~/.bash_profile. Add the following entries:

  • export BOSH_CLIENT=ops_manager
  • export BOSH_CLIENT_SECRET=<whatever-the command-returned-to-you>
  • export BOSH_CA_CERT=/root/opsmanager.pem
  • export BOSH_ENVIRONMENT=<whatever-the-BOSH-ip-returned-to-you>


Source your ~/.bash_profile so that the entries take effect. Now we can run some PKS CLI commands to login to the UAA endpoint configured during the PKS configuration phase.

  • pks login -a uaa.rainpole.com -u admin -p -k
  • pks create-cluster k8s-cluster-01  –external-hostname pks-cluster-01  –plan small –num-nodes 3

Note that there are two cluster references in the final command. The first, k8s-cluster-01, is how PKS identifies the cluster. The second, pks-cluster-01, is simply a K8s thing – essentially the expectation that there is some external load-balancer front-end sitting in front of the K8s cluster. So once again, we will need to edit our DNS and add this entry to coincide with the IP address of the K8s master node, once it is deployed.The plan entry relates to the plans that we set up in PKS earlier. One of the plas was labeled “small” which is what we have chosen here. Lastly, the number of nodes refers to the number of K8s worker nodes. In this example, there will be 3 workers, alongside the master.

Here is the output of the create command:

root@pks-cli:~# pks create-cluster k8s-cluster-01 –external-hostname pks-cluster-01 –plan small –num-nodes 3
Name: k8s-cluster-01
Plan Name: small
UUID: 72540a79-82b5-4aad-8e7a-0de6f6b058c0
Last Action: CREATE
Last Action State: in progress
Last Action Description: Creating cluster Kubernetes Master
Host: pks-cluster-01 Kubernetes Master
Port: 8443
Worker Instances: 3
Kubernetes Master IP(s): In Progress


And while that is running, we can run the following BOSH commands (assuming you have successfully run the om commands above):

root@pks-cli:~# bosh task
Using environment ‘’ as client ‘ops_manager’

Task 404

Task 404 | 10:53:36 | Preparing deployment: Preparing deployment (00:00:06)
Task 404 | 10:53:54 | Preparing package compilation: Finding packages to compile (00:00:00)
Task 404 | 10:53:54 | Creating missing vms: master/a975a9f9-3f74-4f63-ae82-61daddbc78df (0)
Task 404 | 10:53:54 | Creating missing vms: worker/64a55231-cdb6-4ce5-b62c-83cc3b4b233d (0)
Task 404 | 10:53:54 | Creating missing vms: worker/c3c9b49a-f89a-41e6-a1af-36ea0416f3c3 (1)
Task 404 | 10:53:54 | Creating missing vms: worker/e6fd330e-4e95-4d46-a937-676a05a32e5e (2) (00:00:59)
Task 404 | 10:54:54 | Creating missing vms: worker/c3c9b49a-f89a-41e6-a1af-36ea0416f3c3 (1) (00:01:00)
Task 404 | 10:54:58 | Creating missing vms: worker/64a55231-cdb6-4ce5-b62c-83cc3b4b233d (0) (00:01:04)
Task 404 | 10:55:00 | Creating missing vms: master/a975a9f9-3f74-4f63-ae82-61daddbc78df (0) (00:01:06)
Task 404 | 10:55:00 | Updating instance master: master/a975a9f9-3f74-4f63-ae82-61daddbc78df (0) (canary) (00:02:03)
Task 404 | 10:57:03 | Updating instance worker: worker/64a55231-cdb6-4ce5-b62c-83cc3b4b233d (0) (canary) (00:01:25)
Task 404 | 10:58:28 | Updating instance worker: worker/c3c9b49a-f89a-41e6-a1af-36ea0416f3c3 (1) (00:01:28)
Task 404 | 10:59:56 | Updating instance worker: worker/e6fd330e-4e95-4d46-a937-676a05a32e5e (2) (00:01:28)

Task 404 Started Tue Apr 24 10:53:36 UTC 2018
Task 404 Finished Tue Apr 24 11:01:24 UTC 2018
Task 404 Duration 00:07:48
Task 404 done



root@pks-cli:~# bosh vms
Using environment ‘’ as client ‘ops_manager’

Task 409
Task 410
Task 409 done

Task 410 done

Deployment ‘pivotal-container-service-e7febad16f1bf59db116’

Instance Process State AZ IPs VM CID VM Type Active
pivotal-container-service/d4a0fd19-e9ce-47a8-a7df-afa100a612fa running CH-AZ vm-54d92a19-8f98-48a8-bd2e-c0ac53f6ad70 micro false

1 vms

Instance Process State AZ IPs VM CID VM Type Active
master/a975a9f9-3f74-4f63-ae82-61daddbc78df running CH-AZ vm-1e239504-c1d5-46c0-85fc-f5c02bbfddb1 medium false
worker/64a55231-cdb6-4ce5-b62c-83cc3b4b233d running CH-AZ vm-a0452089-b2cc-426e-983c-08a442d15f46 medium false
worker/c3c9b49a-f89a-41e6-a1af-36ea0416f3c3 running CH-AZ vm-54e3ff52-a9a0-450a-86f4-e176afdb47ff medium false
worker/e6fd330e-4e95-4d46-a937-676a05a32e5e running CH-AZ vm-fadafe3f-331a-4844-a947-2390c71a6296 medium false

4 vms



root@pks-cli:~# pks clusters

Name Plan Name UUID Status Action
k8s-cluster-01 small 72540a79-82b5-4aad-8e7a-0de6f6b058c0 succeeded CREATE


root@pks-cli:~# pks cluster k8s-cluster-01

Name: k8s-cluster-01
Plan Name: small
UUID: 72540a79-82b5-4aad-8e7a-0de6f6b058c0
Last Action: CREATE
Last Action State: succeeded
Last Action Description: Instance provisioning completed
Kubernetes Master Host: pks-cluster-01
Kubernetes Master Port: 8443
Worker Instances: 3
Kubernetes Master IP(s):



The canary steps are interesting. This is where it creates a test node with the new components/software and if that all works, we can use the new node in place of the old node rather than impact running environment. You will see one for the master, and one for the workers. If the worker one is successful, then we know it will work for all workers so no need to repeat it for all workers.

The very last command has returned the IP address of the master. We can now add the DNS entry for the pks-cluster-01 with this IP address.


9. Using kubectl

Excellent – now you have your Kubernetes cluster deployed. We also have a K8s CLI utility called kubectl, so let’s run a few commands and examine our cluster. First, we will need to authenticate. We can do that with the following command:

root@pks-cli:~# pks get-credentials k8s-cluster-01

Fetching credentials for cluster k8s-cluster-01.
Context set for cluster k8s-cluster-01.

You can now switch between clusters by using:
$kubectl config use-context <cluster-name>


root@pks-cli:~# kubectl config use-context k8s-cluster-01
Switched to context “k8s-cluster-01”.


You can now start using kubectl commands to examine the state of your cluster.

root@pks-cli:~# kubectl get nodes
2ca99275-244c-4e21-952a-a2fb3586963e Ready <none> 15m v1.9.2
7bdaac2e-272e-45ae-808c-ada0eafbb967 Ready <none> 18m v1.9.2
a3d76bbe-31da-4531-9e5d-72bdbebb9b96 Ready <none> 16m v1.9.2

root@pks-cli:~# kubectl get nodes -o wide
2ca99275-244c-4e21-952a-a2fb3586963e Ready <none> 15m v1.9.2 Ubuntu 14.04.5 LTS 4.4.0-116-generic docker://1.13.1
7bdaac2e-272e-45ae-808c-ada0eafbb967 Ready <none> 18m v1.9.2 Ubuntu 14.04.5 LTS 4.4.0-116-generic docker://1.13.1
a3d76bbe-31da-4531-9e5d-72bdbebb9b96 Ready <none> 17m v1.9.2 Ubuntu 14.04.5 LTS 4.4.0-116-generic docker://1.13.1

root@pks-cli:~# kubectl get pods
No resources found.


The reason I have no pods is that in this deployment, I omitted the “Enable Post Deploy Scripts” option when setting up the Director initially. If I had checked this, I would have the K8s dashboard running automatically. No dig deal – I can deploy it manually.

root@pks-cli:~# kubectl create -f https://raw.githubusercontent.com/kubernetes/dashboard/master/src/deploy/recommended/kubernetes-dashboard.yaml
secret “kubernetes-dashboard-certs” created
serviceaccount “kubernetes-dashboard” created
role “kubernetes-dashboard-minimal” created
rolebinding “kubernetes-dashboard-minimal” created
deployment “kubernetes-dashboard” created
service “kubernetes-dashboard” created

root@pks-cli:~# kubectl get pods –all-namespaces
kube-system kubernetes-dashboard-5bd6f767c7-z2pql 1/1 Running 0 2m

root@pks-cli:~# kubectl get pods –all-namespaces -o wide
kube-system kubernetes-dashboard-5bd6f767c7-z2pql 1/1 Running 0 3m a3d76bbe-31da-4531-9e5d-72bdbebb9b96

Let’s run another application as well – a simple hello-world app.

root@pks-cli:~# kubectl run hello-node –image gcr.io/google-samples/node-hello:1.0
deployment “hello-node” created

root@pks-cli:~# kubectl get pods –all-namespaces
default hello-node-6c59d566d6-85m5s 0/1 ContainerCreating 0 2s
kube-system kubernetes-dashboard-5bd6f767c7-z2pql 1/1 Running 0 30m

root@pks-cli:~# kubectl get pods –all-namespaces
default hello-node-6c59d566d6-85m5s 1/1 Running 0 3m
kube-system kubernetes-dashboard-5bd6f767c7-z2pql 1/1 Running 0 30m


And now we have the K8s dashboard running. Now, we are not able to point directly at our master node to access this dashboard due to authorization restrictions. However William once again saves the day with this steps on how to access the K8s dashboard via a tunnel and the kubectl proxy. Once you have connected to the dashboard and uploaded your K8s config file for the PKS client, you should now be able to access the K8s dashboard and see any apps that you have deployed (in my case, the simple hello app).

There you  have it. Now you have infrastructure in place to allow you to very simply and very simply deploy K8s clusters for your developers. I’ll follow up with a post on some of the challenges I met, especially with the networking, as this should help anyone looking to roll this out in production. But for now, I think this post is already long enough. Thanks for reading to the end. And kudos once more to William Lam and his great blog which provided a lot of guidance on how to successfully deploy PKS and K8s.

  1. Very good blog, thanks.

    FYI: The first 2 diagrams call out “PKS Operations Manager PKS UI” however I think they should say “Pivotal Operations Manager” as Ops Man runs independently of PKS.

Leave a Reply