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: http://kristiannese.blogspot.no/2015/03/dsc-with-azure-and-azure-pack.html

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 (https://technet.microsoft.com/en-us/library/dn806370.aspx?f=255&MSPPError=-2147217396 ), 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.

No comments: