jpapenfuss/px-deploy

What

This will deploy one or more clusters in the cloud, with optional post-install tasks defined by template.

Supported platforms

Container

• Kubernetes (choose a version < 1.24)
• K3s
• Docker EE
• Openshift 4 (only on AWS at this time)
• EKS (only makes sense on AWS)
• GKE (only makes sense on GCP)
• AKS (only makes sense on Azure)

Cloud

• AWS
• GCP
• Azure
• Vsphere

Getting started

1. Install the CLI for your choice of cloud provider and ensure it is configured:

• AWS: https://docs.aws.amazon.com/cli/latest/userguide/getting-started-install.html
• GCP: https://cloud.google.com/sdk/docs/quickstarts
• Azure: https://docs.microsoft.com/en-us/cli/azure/install-azure-cli?view=azure-cli-latest

2. Install and enable Docker.

3. Install bash-completion (optional), eg:

$ brew install bash-completion

You will need to restart your shell.

4. Run the install script:

curl https://raw.githubusercontent.com/andrewh1978/px-deploy/master/install.sh | bash

This will:

• build the Docker image
• create and populate $HOME/.px-deploy
• provide instructions to configure .bash_profile/.zshrc

It may take 10-20 minutes to complete.

Update your .bash_profile or .zshrc as directed. Source them or login again. Validate it is complete with px-deploy -h.

Review the various cloud settings in ~/.px-deploy/defaults.yml.

5. Deploy something:

If you are using AWS and you have not accepted the CentOS terms, browse to https://aws.amazon.com/marketplace/pp?sku=aw0evgkw8e5c1q413zgy5pjce.

px-deploy create --name=my-deployment --template=clusterpair

This will provision a VPC and some other objects, and deploy into it from the template.

6. Connect via SSH:

px-deploy connect --name my-deployment

7. Execute a command:

px-deploy connect --name my-deployment "storkctl get clusterpair"

8. Tear down the deployment:

px-deploy destroy --name my-deployment

NOTES

The deployments can be listed:

$ px-deploy list
DEPLOYMENT  CLOUD  REGION         PLATFORM  TEMPLATE  CLUSTERS  NODES  CREATED
foo         aws    eu-west-1      k8s       px               1      3  2020-02-11T16:14:06Z
bar         gcp    europe-north1  gcp       <none>           1      3  2020-02-04T09:50:11Z

The templates can be listed:

$ px-deploy templates
NAME                         DESCRIPTION
async-dr                     Deploys 2 clusters with Portworx, sets up and configures a cluster pairing,
                             configures an async DR schedule with a loadbalancer in front of the setup.
backup-restore               Deploys a Kubernetes cluster, Minio S3 storage, Petclinic Application and
                             Backup/Restore config
elk                          Deploys an ELK stack of 3 master nodes, 3 data nodes and 2 coordinator nodes, as per
                             https://docs.portworx.com/portworx-install-with-kubernetes/application-install-with-kubernetes/elastic-search-and-kibana/
harbor                       Deploys a single K8s cluster with Portworx and Harbor (https://goharbor.io/)
metro                        Deploys 2 K8s clusters in AWS with a stretched Portworx cluster. It configures
                             Metro, a GUI and Petclinic, ready for a manual failover demo
migration                    Deploys 2 clusters with Portworx, sets up and configures a cluster pairing, and
                             deploys a set of apps and a migration template.
px-backup                    A single Kubernetes cluster with Portworx and PX-Backup via Helm installed
px-fio-example               An example fio benchmark on a gp2 disk and a Portworx volume on a gp2 disk
px-vs-storageos-fio-example  An example fio benchmark on a gp2 disk, a Portworx volume on a gp2 disk, and a
                             StorageOS volume on a gp2 disk
px                           A single Kubernetes cluster with Portworx installed
storageos                    A single Kubernetes cluster with StorageOS installed
training                     Deploys training clusters

Generate a list of IP address, suitable for training:

$ px-deploy status --name trainingDeployment
master-1 34.247.219.101 ec2-34-247-219-101.eu-west-1.compute.amazonaws.com
master-2 34.254.155.6 ec2-34-254-155-6.eu-west-1.compute.amazonaws.com

Generate kubeconfig files so you can run kubectl from your laptop:

$ px-deploy kubeconfig --name exampleDeployment
$ kubectl get nodes --kubeconfig $HOME/.px-deploy/kubeconfig/exampleDeployment.1
NAME                                           STATUS   ROLES    AGE   VERSION
ip-192-168-5-19.eu-west-1.compute.internal     Ready    <none>   11m   v1.21.5-eks-9017834
ip-192-168-56-14.eu-west-1.compute.internal    Ready    <none>   11m   v1.21.5-eks-9017834
ip-192-168-74-141.eu-west-1.compute.internal   Ready    <none>   11m   v1.21.5-eks-9017834

Note this is currently only tested with EKS.

The defaults.yml file sets a number of deployment variables:

• aws_ebs - a list of EBS volumes to be attached to each worker node. This is a space-separated list of type:size pairs, for example: "gp2:30 standard:20" will provision a gp2 volume of 30 GB and a standard volume of 20GB
• aws_region - AWS region
• aws_tags - a list of tags to be applied to each node. This is a comma-separate list of name=value pairs, for example: "Owner=Bob,Purpose=Demo"
• aws_type - the AWS machine type for each node
• cloud - the cloud on which to deploy (aws, gcp, azure or vsphere)
• clusters - the number of clusters to deploy
• k8s_version - the version of Kubernetes to deploy
• stop_after - stop the intances after this many hours
• post_script - script to run on each master after deployment, output will go to stdout
• quiet - if "true", hide provisioning output
• auto_destroy - if set to true, destroy deployment immediately after deploying (usually used with a post_script to output the results of a test or benchmark)
• nodes - the number of worker nodes on each cluster
• platform - can be set to either k8s, k3s, none, dockeree, ocp4, eks, gke, aks or nomad
• px_version - the version of Portworx to install
• gcp_disks - similar to aws_ebs, for example: "pd-standard:20 pd-ssd:30"
• gcp_region - GCP region
• gcp_type - the GCP machine type for each node
• gcp_zone - GCP zone
• gke_version - GKE k8s Version
• azure_disks - similar to aws_ebs, for example: "20 30"
• azure_type - the Azure machine type for each node
• vsphere_host - endpoint
• vsphere_compute_resource - compute resource
• vsphere_user - user with which to provision VMs
• vsphere_password - password
• vsphere_template - full path to CentOS 7 template
• vsphere_datastore - datastore prefix
• vsphere_folder - folder for vSphere VMs
• vsphere_disks - similar to aws_ebs, for example: "20 30" (NOTE: these are not provisioned as static block devices, but they are used as clouddrives)
• vsphere_network - vSwitch or dvPortGroup for cluster ex: Team-SE-120
• vsphere_memory - RAM in GB
• vsphere_cpu - number of vCPUs
• ocp4_domain - domain that has been delegated to route53
• ocp4_version - eg 4.3.0
• ocp4_pull_secret - the pull secret '{"auths" ... }'

There are two ways to override these variables. The first is to specify a template with the --template=... parameter. For example:

$ cat templates/px-fio-example.yml
description: An example fio benchmark on a gp2 disk and a Portworx volume on a gp2 disk
scripts: ["install-px", "px-wait", "px-fio-example"]
clusters: 1
nodes: 3
cloud: aws
aws_ebs: "gp2:150 gp2:150"
post_script: cat
auto_destroy: true
env:
  px_suffix: "s=/dev/nvme1n1"
  cat: "/tmp/output"

More on scripts below.

The second way to override the defaults is to specify on the command line. See px-deploy create -h for a full list. For example, to deploy petclinic into the foo deployment:

px-deploy create --name=foo --clusters=5 --template=petclinic --nodes=6

This example is a mixture of both methods. The template is applied, then the command line parameters are applied, so not only is the template overriding the defaults, but also the parameters are overriding the template.

scripts is a list of scripts to be executed on each master node. For example:

$ cat ~/.px-deploy/scripts/petclinic
# Install petclinic on each cluster
kubectl apply -f /assets/petclinic.yml

These variables are passed to the script:

• $nodes
• $clusters
• $px_version
• $k8s_version

You can also select a different defaults file with:

$ DEFAULTS=/path/to/other/defaults.yml px-deploy create ...

All files in ~/.px-deploy/assets will be copied to /assets on the master nodes. They are then available to be used by the script, as above.

In addition to these, there are some more variables available:

• $cluster - cluster number

There is also an optional cluster object. At the moment, all it can be used for is defining cluster-specific scripts. These will be executed after the scripts above. For example:

cluster:
- id: 1
  scripts: ["script-1", "script-2"]
- id: 2
  scripts: ["script-3", "script-4"]

post_script is a script that will be run on each master node after all of the scripts have completed, and the output will go to stdout. The default is to display the external IP address of master-1, but it could be used to show benchmark outputs, for example.

Last, environment variables can be define in templates or defaults.yml, and these are also available to scripts:

$ cat templates/metro.yml
...
env:
  licenses: "XXXX-XXXX-XXXX-XXXX-XXXX-XXXX-XXXX-XXXX XXXX-XXXX-XXXX-XXXX-XXXX-XXXX-XXXX-XXXX"

Enviroment variables can also be defined on the command line:

px-deploy create -n foo -t migration -e install_apps=true,foo=bar

The install-px script looks for an environment variable called cloud_drive. If it exists, it will deploy Portworx using a clouddrive rather than looking for all attached devices. Note that this is a requirement for Openshift 4. For example:

px-deploy create -n foo -t px -e cloud_drive=type%3Dgp2%2Csize%3D150
px-deploy create -n bar -t px --platform ocp4 -e cloud_drive=type%3Dgp2%2Csize%3D150
px-deploy create -n baz -t px --platform gke --cloud gcp -e cloud_drive="type%3Dpd-standard%2Csize%3D150"
px-deploy create -n qux -t px --platform aks --cloud azure -e cloud_drive="type%3DPremium_LRS%2Csize%3D150"

Notes for vSphere

Before you can start deploying in vSphere, a template must be built. The command

$ px-deploy vsphere-init

will read the vsphere variables from defaults.yml and provision a template at the path defined in vsphere_template.

Notes for OCP4 + AWS

A "master" node will be provisioned for each cluster. This is not really a master node - it is just where openshift-install is run. The root user will have a kubeconfig, so it can be treated as a master node for the purposes of the scripts used in the templates.

A subdomain must be delegated to Route53 on the same AWS account, so you will need to be able to create records for your own domain:

1. Login to the AWS console and go to Route53.

2. Click on "Hosted Zones". "Click on Created hosted zone".

3. Enter the subdomain, eg openshift.example.com and click "Created hosted zone". It will give you 4 authoritative nameservers for the subdomain.

4. Login to your DNS provider.

5. Create an NS record for each of the nameservers for the subdomain, eg:

$ host -t ns openshift.example.com
openshift.example.com name server ns-1386.awsdns-45.org.
openshift.example.com name server ns-1845.awsdns-38.co.uk.
openshift.example.com name server ns-282.awsdns-35.com.
openshift.example.com name server ns-730.awsdns-27.net.

6. Wait a few minutes for the changes to be reflected. Then validate all is well in Route53:

$ host -t soa openshift.example.com
openshift.example.com has SOA record ns-730.awsdns-227.net. awsdns-hostmaster.amazon.com. 1 7200 900 1209600 86400

Bugs

• The Azure Vagrant plugin will fail when provisioning VMs in parallel, so px-deploy disables parallel provisioning. This is really slow, and if a template uses a script that will not terminate until another VM is up, then it will never finish provisioning.
• Provisioning multiple deployments in an Azure region at the same time gives DNS errors and fails.