Monday, March 16, 2015

Application Modeling with VM Roles, DSC and SMA

Earlier this year, I started to go deep into DSC to learn more about the concept, possibilities and most important, how we can improve what we already have and know, using this new approach of modeling.

For more information and as an introduction to this blog post, you can read my former blog post on the subject:

Desired State Configuration is very interesting indeed – and to fully embrace it you need to be comfortable with Powershell. Having that said, Desired State Configuration can give you some of what you are requiring today, but not everything.

Let me spend some minutes trying to explain what I am actually saying here.

If you want to use DSC as your primary engine, the standard solution to configure and deploy applications and services across clouds throughout the life cycle, there is nothing there to stop you from doing so.
However, given the fact that in many situations, you won’t be the individual who’s ordering the application, server and dependencies, it is important that we can make this available in a world full of tenants with a demand for self-servicing.

Looking back at how we used to do things before to handle the life-cycle management of applications and infra, I think it is fair to say it was something like this (in context of System Center):

1)      We deployed a Virtual Machine based on a VM Template using SCVMM
We either
a)      Manually installed and configured applications and services within the guest post VM deployment
b)      Used SCCM to install agents, letting the admin interact with the OS to install and configure applications using a central management solution
2)      If we wanted to provide monitoring, we then used SCOM to roll out the agents to our servers and configured them to report to their management group
3)      Finally yet importantly, we also wanted to be secure and have a reliable set of data. That’s why we also added backup agents to our servers using SCDPM

In total, we are talking about 4 agents here (SCVMM, SCCM, SCOM and SCDPM).
That is a lot.

Also note that I didn’t specify any version of System Center, so this was probably even before we started to talk about Private Clouds (introduced with System Center 2012).

And that’s the next topic, all of this in the context of cloud computing.

If we take a walk down the memorial lane, we can see some of Microsoft’s least proud moments, all the attempts in order to bring the private cloud a fully functional self-service portal.

-        We’ve had several self-service portals for VMM that later was replaced by different solutions, such as Cloud Service Process Pack and App Controller
-        Cloud Service Process Pack – which was introduced with SC 2012 – where all the components were merged into a single license, giving you out-of-the-box functionality related to IaaS.
The solution was one of the worst we have seen, and the complexity to implement it was beyond what we have seen ever since.
-        AppController was based on Silverlight and gave us the “single-pane of glass” vision for cloud management. With a connector to Azure subscriptions (IaaS) and to private and service provider clouds (using SPF), you could deploy and control your services and virtual machines using this console

Although it is common knowledge that AppController will be removed in vNext of System Center ( ), AppController introduced us to a very interesting thing: self-service of service templates.

The concept of service templates was introduced in System Center 2012 – Virtual Machine Manager, and if we go back to my list of actions we needed to perform, we could say that service templates at some point would replace the need of SCCM.
Service Templates was an extension to the VM template. It gave us the possibility to design, configure and deploy multi-tier applications – and deploy it to our private clouds.
However, I have to admit that back then; we did not see much adoption of service templates. Actually, we did not see some serious adoption before Microsoft started to push some pre-configured service templates on their own, and that happened last year – at the same time as their Gallery Items for Azure Pack was released.

To summarize, the service template concept (which was based on XML) gave the application owners and the fabric administrators a chance to interact to standardize and deploy complex applications into the private clouds, using AppController. In the same sentence there we found AppController (Silverlight) and XML.

If we quickly turn to our “final destination”, Microsoft Azure, we can see that those technologies aren’t the big bet in any circumstances.

VM Roles are replacing service templates in the private cloud through Windows Azure Pack.

A VM Role is based on JSON – and define a virtual machine resource that tenants can instantiate and scale according to their requirements.

We have in essence two JSON files. One for the resource definition (RESDEF) and one for the resource extension (RESEXT).
The resource definition describes the virtual machine hardware and instantiation restrictions, while the resource extension definition describes how a resource should be provisioned.

In order to support user input in a user friendly way, we also have a third JSON file – the view definition (VIEWDEF), which provides the Azure Pack details about how to let the user customize the creation of a VM Role.

These files are contained in a package, along with other files (custom resources, logo’s etc) that describe the entire VM Role.

You might ask yourself why I am introducing you to something you already know very well, or why I am starting to endorse JSON. The answer lays in the clouds.

If you have every played around with the Azure preview portal, you have access to the Azure Resource Manager.
ARM introduced an entirely new way of thinking about you resources. Instead of creating and managing individual resources, you are defining a resource model of your service – to create a resource group with different resources that are logically managed throughout the entire life cycle.

-        And guess what?

The Azure Resource Manager Templates is based on JSON, which describes the resources and associated deployment parameters.

So to give you a short summary so far:

Service Templates was great when it came with SCVMM 2012. However, based on XML and AppController for self-service, it wasn’t flexible enough, nor designed for the cloud.

Because of a huge focus on consistency as part of the Cloud OS vision by Microsoft, Windows Azure Pack was brought on-premises and should help organizations to adopt the cloud at a faster cadence. We then got VM Roles that should be more aligned with the public cloud (Microsoft Azure), compared to service templates.

So we might (so far) end up with a conclusion that VM Roles is here to stay, and if you are focusing too much on service templates today, you need to reconsider that investment.

The good, the bad and the ugly

So far, the blog post has been describing something similar to a journey. Nevertheless, we have not reached the final destination yet.

I promised you a blog post about DSC, SMA and VM Roles, but so far, you have only heard about the VM Roles.
Before we proceed, we need to be completely honest about the VM Roles to understand the requirement of engineering here. To better understand what I am talking about, I am comparing a VM Role with a stand-alone VM based on a VM Template:

As you can see, the VM Role gives us very much more compared to a stand-alone VM from a VM template. A VM Role is our preferred choice when we want to deploy applications in a similar way as a service template, but only as single tiers. We can also service the VM Role and scale it on demand.

A VM on the other hand, lacks all these fancy features. We can purely base a stand-alone VM on a VM Template, giving us a pre-defined HW template in VMM with some limited settings at the OS level.
However, please note that the VM supports probably the most important things for any production scenarios: backup and DR.
That is correct. If you use backup and DR together with a VM Role, you will end up in a scenario where you have orphaned objects in Azure Pack. This will effectively break the relationship between the VM Role (CloudService in VMM) and its members. There is currently no way to recover from that scenario.

This got me thinking.

How can we leverage the best from both worlds? Using VM Role as the engine that drives and creates the complexity here, supplemented by SMA and Desired State Configuration to perform the in-guest operations into normal VM templates?

I ran through the scenario with a fellow MVP, Stanislav Zhelyazkov and he nodded and agreed. “-This seems to be the right thing to do moving forward, you have my blessing” he said.

The workflow

This is where it all makes sense. To combine the beauty of VM Roles, DSC and SMA to achieve the following scenario:

1)      A tenant logs on to the tenant portal. The subscription includes the VM Cloud resource provider where the cloud administrator has added one or more VM Roles.
2)      The VM Role Gallery shows these VM Roles and provides the tenant with instructions on how to model and deploy the application.
3)      The tenant provides some input during the VM Role wizard and the VM Role deployment starts
4)      In the background, a parent runbook (SMA) that is linked to the event in the portal kicks in, and based on the VM Role the tenant chose, it will invoke the correct child runbook.
5)      The child runbook will deploy the (stand-alone) VMs necessary for the application specified in the VM Role, join them to the proper domain (if specified) and automatically add them to the tenant subscription.
6)      Once the stand-alone VMs are started, the VM Role resource extension kicks in (which is the DSC configuration, using push) that based on the parameters and inputs from the tenant is able to deploy and model the application entirely.
7)      Once the entire operation has completed, the child runbook will clean-up the VM Role and remove it from the subscription

In a nutshell, we have achieved the following with this example:

1)      We have successfully been able to deploy and model our applications using the extension available in VM Roles, where we are using Desired State Configuration to handle everything within the guests (instead of normal powershell scripts).
2)      We are combining the process in WAP with SMA Runbooks to handle everything outside of the VM Role and the VMs.
3)      We are guaranteed a supported life-cycle management of our tenant workloads

Here you can see some screenshots from a VM Role that will deploy Windows Azure Pack on 6 stand-alone VMs, combining DSC and SMA.

In an upcoming blog post, we will start to have a look at the actual code being used, the challenges and workarounds.

I hope that this blog post showed you some interesting things about application modeling with VM Roles, SMA and DSC, and that the times are a-changing compared to what we used to do in this space.

Monday, March 2, 2015

DSC with Azure and Azure Pack

Every now and then, it comes a time when I really need to ramp up on certain things.
It can be a new technology, a new product, or a new way of doing things.

This kind of journey is never easy, and I am that kind of person who doesn’t stop before I have a certain level of satisfaction. I expect a lot from myself and have a crazy self-discipline.

Starting early this year, I went deep into DSC to learn more about something that will be impossible to avoid in the next couple of months.

Before continuing, I just want you to know that this will not be yet another blog post that explains the importance of Powershell, which you need to learn ASAP or else you will "flip burgers in the future".

A result of have working with Azure Pack and Azure for the last years has made me much more creative.
Instead of having our out-of-the-box products where we were limited by the actions provided by the GUI, we can now easily create our own custom solutions where integrating several APIs, modules and so on to create new opportunities for our business.

Let us stop for a second on Azure. Microsoft Azure.
We have been talking about the Cloud OS and cloud consistency for over a year now and we should all be very familiar with MS vision and strategy around this topic.
Especially “Mobile first, Cloud first” will give us a hint that whatever comes will appear in Microsoft Azure first.

In the context of DSC, we can see that we can leverage some Azure VM Extensions and Features in our IaaS VMs today.
And that is really the background of this blog post.

Microsoft Azure provides us with several VM Extensions, either directly by Microsoft or some third-parties to enable security, runtime, debugging, management and other features that will boost your productivity working with IaaS VMs in Azure.

When you deploy a virtual machine in the Azure portal, you can decide whether or not the VM Extension should be enabled.

We have several extensions available, all depending on what we are trying to achieve.
The extensions I find most interesting belongs to the category of “Deployment and Configuration Management”.

First, let us talk about a VM extension for “MSEnterpriseApplication”.
Using this Extension, we will effectively implements features that supports VM Roles resource extensions, the same we can leverage on-premises with Azure Pack and Service Provider Foundation.
To add this extension, the VM must already exist in Azure and have the Azure Guest Agent pre-installed.

Running the following cmdlet using the Azure module gives us more details about the extension

With this extension enabled in the VM, we can use the VM Role Authoring tool to author our resource extension (that is the package that we normally import to VMM which contains the application payload). The latest version let us deploy directly to Azure.
If you rather want to use Powershell, you should view the Powershell functionality of the tool and save only the portion of the script that assigns a value to $plainSettings in a text file.

From here, you can store the text file in a variable ($plainSettings) and update your VM with the following cmdlet:

$VM = Set-AzureVMExtension –ExtensionName “MSEnterpriseApplication” –Publisher “Microsoft.SystemCenter” –Version “1.0” –PrivateConfiguration $plainSettings –VM $vmcontext.VM

Next, update your VM directly using the following cmdlet:

Update-AzureVM –ServiceName “ServiceName” –VM $VM –Name “VMName”

So, given the fact that we now have a single tool where we can author and deploy our resource extensions (application payload) to IaaS VMs in both WAP and Azure is good news, however, it is not idempotent.

This is where Desired State Configuration comes into the picture.
Been built on the Common Information Model (CIM) and uses Windows Remote Management (WinRM) as the communication mechanism, DSC is like putting steroids into your Powershell scripts.

I know I will get a lot of Powershell experts on my neck here, but that is at least one way to visualize what DSC is.
Let us say you create a script, deploy it to a node and then you are done.
If someone makes any changes to that configuration afterwards, the Powershell script would not care nor notice.
A Desired State Configuration can ensure that there won’t be any configuration drift and apply and monitor (for example) the configuration.
This is handled by the Local Configuration Manager (LCM) which you can consider as an “agent”, although it is not an agent per definition.

So, looking at the capabilities of DSC, we can quickly understand how important this will be for any in-guest management solution moving forward.

The requirement of using Azure Powershell DSC VM extension is that you must have Azure Powershell module installed. The DSC extension handler has a dependency on Windows Management Framework (WMF) version 5 – which is currently in preview and only supported by 2012 R2. WMF 5.0 will automatically be installed in your IaaS VM as a Windows Update once enabled, and require a reboot.

The following cmdlets are specific to DSC:

Publish-AzureVMDscConfiguration – will upload a DSC script to Azure blob storage, that later will be applied to your IaaS VMs using the Set-AzureVMDscExtension cmdlet

Get-AzureVMDscExtension – Gets the settings of the DSC extension on a particular VM
Remove-AzureVMDscExtension – Will remove the DSC extension from a VM

Set-AzureVMDscExtension – Configures the DSC extension on a VM

Here’s a very easy example on how to apply a DSC script to your VM in Azure, assuming you have the script already created.

Publish-AzureVMDscConfiguration –ConfigurationPath “c:\folder\DSCscript.ps1”

That will create a ZIP package which will be uploaded to a blob storage in Azure.

Next, we will add the config to the VM (which we assume is already stored in the variable named $VM ) )

$VM = Set-AzureVMDscExtension –VM $VM –ConfigurationArchive “” –ConfigurationName “DSCscript”

Once this cmdlet is executed, the following will happen within the VM:

1)      WMF 5.0 is downloaded and installed (the latest version) on the server
2)      The extension handler looks in the specified Azure container (which is defined when you connect with your subscription) for the .zip file
3)      Then the archive is unpacked and any dependent modules are moved into the PS Module path and runs the specified configuration function

Adding that this will also accept parameters gives you an understanding of how flexible, dynamic and powerful the DSC VM Extension will be.

Now, this was all about Microsoft Azure.
What about the things that are taking place in Azure Pack?

I briefly mentioned the VM Role Authoring Tool in this blog post which will be playing an important role in this setting.
The research I have been doing this year isn’t easy to put within a single blog post, especially not if I should describe all the errors and mistakes I have done as part of this journey J

I have been trying to simulate the Azure experience in Windows Azure Pack, but unfortunately, that is an impossible challenge as we don’t have the same possibilities when it comes to the interaction through the API. I am only able to achieve some of the good parts, but that again will qualify for some blog posts in the near future.

Before you start thinking “no, it is not that hard to simulate the exact experience”, I would like to remind you about that everything I do in this context, will always be using Network Virtualization with NVGRE, so there is no data-channel from the datacenter into the tenant environment what so ever.

If you think this is interesting, to learn more about DSC with Azure and Azure Pack, I have to point out the spectacular blog post series by Ben Gelens, where he has done a very good job explaining the complete setup of an entire DSC environment (using Pull) including the authoring of the required VM Role.

I will focus on the Push method in my examples, given the fact that the tenants are isolated and should be able to perform certain actions through self-service.

See you soon!