Cloud Foundry – SAP Cloud Platform on Azure

Cloud Foundry – SAP Cloud Platform on Azure (Beta)

The main project that kept me busy for the past 7 months was the work with SAP Labs Israel to get a beta version of SAP Cloud Platform running on Azure. During SAP Sapphire, SCP on Azure got now announced together with other public cloud vendors as part of SAP’s multi-cloud strategy (see here and here. It is currently available as public beta for anyone to try on Azure. Here I dig into some first basics…

Note: as outlined in my previous blog post about HANA Express, please note that this is my personal blog which reflects my personal thoughts instead of Microsoft’s official opinion. For offical announcements and statements, please refer to the Microsoft Azure Blog.

Cloud Foundry on Azure and SCP on Azure

Before digging into the details of SAP Cloud Platform (short: SCP), just a quick reminder about how we support Cloud Foundry as a platform on Azure in general.

With SCP, we now support the 2nd commercially available Cloud Foundry flavor on Azure next next to Pivotal Cloud Foundry. But, what makes SCP on Azure really different to the offers we had so far is, that it is a fully managed PaaS… managed by SAP. So using SCP on Azure is exactly the same as on any other public cloud platform which gives you full portability in a multi-cloud world. We do have a few differentiators in my opinion, though:

Finally, I strongly believe that from the work we’ve done with SAP, Cloud Foundry on Azure benefits across the board, in general. We’ve made numerous improvements to the Bosh CPI implementation for Azure as part of the efforts and we fixed several bugs. Examples of that are e.g. the Keep-Alive of unhealthy virtual machines for debugging purposes or support for managed disks (which was on the plans, already, but we needed it earlier:)).

Until General Availabiltiy of SAP Cloud Platform on Azure, you’ll see many more improvements in our CPI which I cannot elaborate in detail at the moment.

SAP Cloud Platform on Azure – Getting Started

To get SCP on Azure, you work through the regular SCP administation cockpit. So your starting point is the same as for any other work you do with SCP. You register or log-in as usual, nothing Azure specific with that. For activating Azure, which is at the time of writing this post available as Beta+Trial, you click on the "Start Cloud Foundry Trial"-button within the cockpit!

https://account.hanatrial.ondemand.com

SCP Cockpit Login and highlight activation button

When clicking the button, the SCP cockpit will ask you for selecting a region in which you want to activate the trail. Now, within SCP, activated Azure regions do just appear as an SCP-region within the cockpit. That means, you would pick an SCP-region for the activation that runs on-top of Azure as shown below:

Pick the SCP region

Until now, I’ve done several trial activations. It usually takes a minute till the tenant is activated and available for you. For Cloud Foundry knowledable folks, what happens is the creation of an organization and a space for you within the SCP Cloud Foundry environment.

Exploring what’s available

After that process is completed, you can navigate to your organization and space within the SCP cockpit. Through the cockpit, you have convenient ways to browse and montior deployed applications or also browse the marketplace to review, which backing services SAP has enabled for the Beta+Trial which we’ve enabled on Azure. Cloud Foundry veterans will find it quite easy to navigate through the portal.

SCP Marketplace

Now, since this is a beta as well as a trial, the backing services are all running as single instances in containers behind the scenes. I think that’s more than fair because you’re not paying for the service as long as it is in beta. And… this is also the general approach followed by SAP for using trial accounts on generally available regions, as well. For GA and production (non-trial) environments, these backing services will run in Virtual Machine clusters, of course.

The exception even for today in that list is… guess what… SAP HANA of course. The HANA backing service offered as part of the Azure SCP beta is a multi-tenant, shared HANA environment that runs in Virtual Machines behind the scenes.

Note: When using any of the backing services from the SCP market place, you indeed use SAP-deployed backing services. That gives you a higher level of portability since SAP and not Azure controls the versions and configurations used for the technologies that are the foundation for those backing services. Now, if you still would prefer to use native Azure services, you can use the Azure Meta Service Broker to hook them up with applications you run in SCP the Cloud Foundry way (you can use them, directly, as well, of course). But in that case, you need to be aware that this has impact on how portable your applications are across multiple cloud vendors if that’s relevant for you.

Using the Cloud Foundry CLI

Now, the cool thing about SCP is, that it is Cloud Foundry. It’s a Cloud Foundry PaaS enriched by many SAP services (ok, it will take a while till we have them all active on Azure as that requires further engineering work). It ultimately means, you manage and work with it as with other Cloud Foundry environments including the CLI and the APIs. For SAP-specific services such as HANA, SAP provides plug-ins for the CLI to make it easier to work with those services such as the HANA MTA plug-in which I think SAP announced during this year’s Sapphire, as well.

For CLI/API interaction purposes, SCP uses different API endpoints for each region. When using the Azure Beta region which we’ve deployed right now, you’d use the following API endpoint:

cf api api.cf.us20.hana.ondemand.com
cf login

From that moment forward, everything looks and feels like Cloud Foundry. You can push apps, you can explore your marketplace etc. The learning curve is quite easy for Cloud Foundry experienced developers, I’d say:

SCP CF CLI in Action

Of course, I was curious what development runtimes SAP has enabled and as a .NET developer I wanted to know about .NET. Looking at the screen shot above, I found it great that .NET Core was enabled. But be aware: SAP typically offers some sort of enterprise support for languages and runtimes in their environment, but from the build packs listed, only a few enjoy that status. .NET Core does not. So, check back with SAP on that… .NET Core only gets community-level support through the Cloud Foundry Foundation.

Using the Azure Meta Service Broker with SAP Cloud Platform

When you’re running on Azure, it makes sense to use Azure-native services if you want to. At the end of the day, that’s when you can unleash the full power of running on a specific cloud platform, right:)? And… you still could implement your "portability" thorugh IoC/DI at the application level to have a more effective/efficient integration into native cloud services but still stay portable to a certain extent.

Fortunately, SAP allows you to enable the Azure Meta Service Broker at the space-level when using SCP. I thought I wanna try that out again now this is in beta (we prototyped that in early phases). Essentially, you really only need to follow the standard-procedure to enable the Azure Meta Service Broker as documented here.

The first step, though, is to create an Azure SQL Database instance and a Service Principal in your Azure Active Directory, as shown below. The Service Principal is needed because the service broker dynamically creates/deletes resources as they are requested through the Cloud Foundry API (or the CLI).

$ az ad app create --display-name=yourazureadappdisplayname --homepage=http://yourhomepage.com --identifier-uris="https://yourazureadserviceprincipalname" --key-type=Password --reply-urls="https://notreallyimportantwhatyouputhere"
AppId                                 DisplayName                  Homepage                 ObjectId                              ObjectType
------------------------------------  ---------------------------  -----------------------  ------------------------------------  ------------
yourazureadserviceprincipalappid      yourazureadappdisplayname    http:/yourhomepage.com   thecreatedappobjectid                 Application
$
$ az ad sp create --id https://yourazureadserviceprincipalname
AppId                                 DisplayName                  ObjectId                              ObjectType
------------------------------------  ---------------------------  ------------------------------------  ----------------
yourazureadappdisplayname             yourazureadappdisplayname    thecreatedappobjectid                 ServicePrincipal
$
$ az ad app update --id=yourazureadserviceprincipalappid --password="yourazureadserviceprincipalpassword"
$

Don’t forget to give the created service principal contributor-rights to the resource group you want to use for the resources of the Meta Service Broker.

After you’ve done the basic creation of the assets mentioned above, the next step is to clone the Azure Meta Service Broker repositiory and adjust the Cloud Foundry Application Manifest for it. The Meta Service Broker essentially is a CF-application that represents the Service Broker for several Azure Services as documented. You can run it wherever you want, but it’s built for being pushed as an application into CF with configuration data about your Azure Subscription to use for the resources created by the Meta Service Broker. Here’s a sample configuration, so that’s how a filled manifest.yml for the Azure Meta Service Broker looks like.

Note that the parameters for SUBSCRIPTION_ID, TENANT_ID, CLIENT_ID and CLIENT_SECRET need to match those you’ve used or retrieved through the Azure CLI commands above as they represent your subscription as well as the servvice principal you’ve created above! Ok, the commands above don’t give you the subscription ID. You can retrieve that by issuing az account list --output=json or by looking it up in the Azure Portal.

---
applications:
- name: meta-azure-service-broker
  buildpack: https://github.com/cloudfoundry/nodejs-buildpack
  instances: 1
  env:
    ENVIRONMENT: AzureCloud
    SUBSCRIPTION_ID: yoursubscriptionid
    TENANT_ID: yourazureadtenantidfromtheserviceprincipal
    CLIENT_ID: yourazureadserviceprincipalappid
    CLIENT_SECRET: yourazureadserviceprincipalpassword
    SECURITY_USER_NAME: usertoauthenticateagainstmetaservicebroker
    SECURITY_USER_PASSWORD: passwordtoauthenticateagainstmetaservicebroker
    AZURE_BROKER_DATABASE_PROVIDER: sqlserver
    AZURE_BROKER_DATABASE_SERVER: yourazuresqldbserver.database.windows.net
    AZURE_BROKER_DATABASE_USER: yourazuresqldbuser@yourazuresqldbserver
    AZURE_BROKER_DATABASE_PASSWORD: yourazuresqldbpassword
    AZURE_BROKER_DATABASE_NAME: yourazuresqldbdatabasename
    AZURE_BROKER_DATABASE_ENCRYPTION_KEY: 32charactersofyourchoicegohereok
    AZURE_SQLDB_ALLOW_TO_CREATE_SQL_SERVER: false
    AZURE_SQLDB_ENABLE_TRANSPARENT_DATA_ENCRYPTION: true

Assuming you have logged-in with cf login as mentioned above, you can just push the meta service broker application into your SCP tenant using cf push. Important tip: since the settings specified above in the manifest are set as environment variables, be careful with special characters used by bash when defining the passwords. Either avoid or escapte them!! A successful deployment of the meta service broker app should end up with the following output on your console:

Service Broker App Pushed

When you’ve pushed the application, you only created the meta service broker but you did not tell Cloud Foundry that this is indeed a Service Broker and not any other type of general application. So, as a next step after the meta service broker has been deployed, successfully, is to register it as a service broker. The steps for that are:

  • Review the deployed application and take note of it’s entry point URL.
  • Create the service broker using that URL. You also need to authenticate against the meta service broker. For that purpose, use the previously specified SECURITY_USER_NAME and SECURITY_USER_PASSWORD parameters from the manifest file above.
$ cf apps
Getting apps in org yourorgname_trial / space dev as your@username.com...
OK

name                        requested state   instances   memory   disk   urls
meta-azure-service-broker   started           1/1         1G       1G     urlfromyourapppush.cfapps.us20.hana.ondemand.com
$
$
$ cf create-service-broker nameofyourbroker usertoauthenticateagainstmetaservicebroker passwordtoauthenticateagainstmetaservicebroker  https://urlfromyourapppush.cfapps.us20.hana.ondemand.com --space-scoped

Important Note: since SCP is a multi-tenant Cloud Foundry environment, you need to know that you are only the administrator of the spaces you’re creating. Permissions to organizations are limited and you don’t have any permissions beyond organizations. That means when activating the service broker, you need to use the --space-scoped switch to activate the broker on your space. When you do that, you are using a so called private broker with private plans which become active, automatically. That means, although the documentation of the Azure Meta Service Broker states it, you do not need to call cf enable-service-access since all services are enabled for your space in that case, by default. A simple cf marketplace confirms that:

cf marketplace results

As a next step, you can start using the services exposed by the Azure Meta Service Broker. At the time of writing this post, those services included: Azure Storage, Azure SQL DB, Azure DocumentDB, Azure Service Bus, Azure Redis Cache and Azure Key Vault services. Means you can start using those services right away. If you have asks/requests for additional services Microsoft should enable, I think you’re best by posting requests/issues into the GitHub Repository for the Azure Meta Service Broker or file a pull request.

The way how you make use of the services follows standard cloud foundry patterns with service brokers. You create a service and then you bind the service to one or many applications using the Cloud Foundry service broker APIs or CLI-commands. Let’s create an Azure Storage account that way so you get a sense of it. The approach is always the same for all services offered through the Azure Meta Service Broker.

First you create a parameters-file specifying some of the provisioning paramters for creating the respective Azure service. An example of such a parameters JSON-file for creating an Azure storage account via the service broker looks as follows:

{
  "resource_group_name": "marioszpScpMetaBroker",
  "storage_account_name": "marioszpscpteststrg",
  "location": "westus",
  "account_type": "Standard_LRS"
}

With this JSON-file in-place, you can leverage the service broker CLI commands or Cloud Foundry APIs to create a new service and make Cloud Foundry aware of it:

$ cf create-service azure-storage standard marioszpscpstoragetest -c .\params.json

$ cf service marioszpscpstoragetest

Service instance: marioszpscpstoragetest
Service: azure-storage
Bound apps:
Tags:
Plan: standard
Description: Azure Storage Service
Documentation url:
Dashboard:

Last Operation
Status: create succeeded
Message: Created the storage account, state: Succeeded
Started: 2017-05-19T11:02:21Z
Updated: 2017-05-19T11:02:40Z
$

Reults of creating a service with the broker

Now, once this is done you can check on the provisioning status and availability of that service using the cf service marioszpscpstoragetest command which gives you all the details about the created service. Finally, by using the cf bind-service command, you can bind the service to any application you’ve deployed. Configuration parameters will then be exposed through environment variables as per the documentation of the respective service in the Azure Meta Service Broker GitHub. For an Azure storage account, an environment variable with a JSON-document will be added to your applications Environment variables that contains e.g. the storage account name and storage account keys you can use to authenticate and execute calls against the created storage account.

What’s really cool now is, that all of what I’ve done above with the Cloud Foundry CLI is visible in the SAP Cloud Platform Cockpit, as well. For example, since I’ve activated the Azure Meta Service Broker, you’ll see all the services in the marketplace in the portal as well. Even the Azure Storage Account which is surfaced as a service as shown below. That means the Azure Meta Service Broker really brings SCP and Azure close together and allows you to look at Azure resources through the SCP as well as the Azure management perspectives.

SCP Cockpit with Azure Services

Final Words

This is just the beginning of SAP Cloud Platform on Azure. Note that we’ve announced Beta. During our path forward for becoming generally available, we’ll get to see several improvements in Azure’s Bosh CPI and optimized backing services (currently all except HANA run in single instance containers in the Beta/Trial).

There’s one question a lot of people ask when looking at something like this, though: how does that relate to native Azure PaaS services such as Azure App Services or Service Fabric? The answer is simple: Azure is an open Cloud Platform. Microsoft’s strategy is clearly to enable as many relevant platforms as possible in addition to Microsoft’s own technology. This gives customers and partners choice and allows them to leverage the skills and knowledge they have available in their portfolio, already. Cloud Foundry and it’s flaviours do fit very well into this strategy. If you prefer Cloud Foundry based PaaS platforms, then Azure is a place where you can go… and if you want you can even integrate with all the native goodness of Azure including especially Azure Active Directory for Single Sign-On or all of the other services in addition to those offered with the Azure Meta Service Broker.

Developed an SAP HANA Express Azure Quick Start Template

Context

This is an exciting week for me… although I am usually not that much into attending Business-oriented conferences, over the past two years I did so by attending SAP Sapphire.

Up until know, that mainly is caused by the fact that I’ve contributed some aspects to what was announced with regards to the partnership between Microsoft and SAP at the conference. Last year, my part was mainly about Office 365 with the work I’ve supported for Concur, Ariba, Fieldglass and SuccessFactors as well as the Sports Basement Demo which was shown in the Key Note to highlight the HANA One Certification for Azure in our DS14 VM-Series at that time.

While I cannot write about the major project I supported for this year, yet, here’s a little nugget to get started with – a Quick Start Template for SAP HANA Express!

Important Note: While I am working for Microsoft, this blog summarizes my personal opinions and my personal understanding of topics. This means that all I am writing here is not related to Microsoft’s official opinion, at all! If you want to get that view, look at the Azure Blog or Jason Zander’s Blog for official announcements!

Important Note: The Pull-Request into the official Microsoft Azure Quick Start Templates GitHub repository is not completed, yet. Therefore, the link still redirects to the working branch in my GitHub repository. I’ll update the blog-post as soon as the pull request is completed (there’s currently a general issue with the Travis CI pipeline used for validation on the Azure Quick Start Templates GitHub repository that caused an unexpected delay for the pull request to go through in time for Sapphire).

SAP HANA Express

As you might know, already, SAP HANA is SAP’s in-memory Database Engine which is supposed to back all of the major future releases of SAP’s core business suites centered around S/4HANA. But, HANA can be used as a stand-alone database system for developing custom solutions, as well. The Sports Basement Demo on Azure DS14 Instances from last year’s Sapphire Key Note is an example of that. It was a plain HANA Database fronted by a Java Web Application.

Now, within last year’s Sapphire and now, SAP released a version of SAP HANA that is free for development and testing purposes up to 32GBs of RAM called HANA Express.

For more details, you should navigate to the SAP HANA Express Homepage to get the full picture and the official view:

https://www.sap.com/developer/topics/sap-hana-express.html

Azure Quick-Start-Templates

Now, shortly before Sapphire some folks from Microsoft and SAP approached me for creating an Azure Marketplace Image for HANA Express. That’s something we’re working on, but it is not something that can be done in just a few days. Too short… but since HANA Express clearly addresses developers, I thought a good solution that can be implemented within a few days is an Azure Quick-Start-Template.

For those of you who are new to Azure: Azure Quick-Start-Templates are open source based Azure Resource Manager Templates and Deployment Scripts which can be used to quickly spin-up Solutions on Azure. Many of those are used as a learning resource, but some of them can definitely be used for dev/test scenarios or even as a starting point for production scenarios.

All of these are available under the following two links:

Now, the main point with these quick start templates is, that they’re automating most of the setup/provisioning procedure by using Scripts and Templates so you can get started, quickly.

SAP HANA Express Quick Start Template

I decided to build such a template for HANA Express and make it available as part of the Quick-Start-Templates. This Quick-Start works with SAP HANA Express 2.0 SPS1 and it also should work with other version, but I’ve only tested it with this one.

http://aka.ms/sap-hana-express-quickstart

There’s just one caveat: since SAP requires you to accept the EULA for SAP HANA Express, you first need to go to SAP’s HANA Express Home Page, register and download the SAP HANA Express Setup Images manually before being able to start using the Quick Start Template I’ve created. But from there on you can start using the template. The basic workings of the template are:

  1. First you register with SAP and download the HANA Express Setup Packge.
  2. Then you use the quick start template to upload the setup package into your private Azure Storage Account.
  3. From that moment forward you can use the Azure Resource Manager Template included in the template to deploy as many HANA Express Instances into your own subscription as you want.

Starting with Step 2, everything is automated with Scripts and Templates. That means, only the first step – downloading the setup packges and accepting the EULA at the SAP HANA Express Setup Homepage is something you need to do manually. Sure, a Marketplace Image would be more convenient, but we’ll work with SAP on that…

All the details are explained in the sections below in this blog post including how you can validate, that the installation really went well at the end of the entire process.

Important Note: Please don’t forget, that quick start templates are not something that is backed by Microsoft Support of any means. They are here to help you getting started on your own, they are full open source and maintained at a best effort basis!

Requirements

Before moving on, all you need on your local machine are the following assets/tools:

Register Download HANA Express Setup from SAP

Yes, that’s the first step. It needs to be done for two reasons: first, you need to accept SAP’s EULA and second SAP will inform you about important updates and service releases for HANA when registering at the registration page.

That said, the first thing you do is navigating to the SAP HANA Express Home Page to register with SAP and Accept the EULA:

HANA Express

HANA Express Download Manager Option

Next, you’ll need to use the SAP Download Manager to download the SAP HANA Express Setup packages. The Download Manager exists as native version for Linux or Windows or as a Cross-Platform Version for Java (distributed as JAR-package). I’ve used the JAR-package version, but it should not matter at all.

What matters is selecting the right type of setup packages. The quick-start-template I’ve built is tested for the Server-Version, only, without XA Advanced services. So you should select the following for download when using the SAP HANA Express Download Manager:

HANA Express Server Only Download Option

The downloaded setup package will appear as TAR-Archive in your local Downloads-folder (or wherever you downloaded it to). It should be called something along the lines of hxe.tgz.

Upload the HANA Express to your Azure Storage Account

For automating the setup procedure of SAP HANA Express inside of an Azure Virtual Machine, the HANA Express Setup packages need to be available for download through an automation script! To avoid uploading the setup package (~ 1.6GBs of data) with each new VM, the best approach is uploading that into an Azure storage account and using a shared access signature to feed the setup files into the provisioning script for Virtual Machines.

Now, performing those steps can be done manually by using a tool such as the Azure Storage Explorer. But I decided to automate the procedure using the Azure CLI 2.0.

Assuming that you have your SAP HANA Express Setup packages downloaded to the folder /mnt/c/temp/hxe.tgz, you can execute the following command:

# Login into your Azure Subscription using the Azure CLI 2.0
az login

# Create a resource group for the storage account
az group create --name "sampleresourcegroupname" --location "westeurope"

# Upload the HANA Express Setup Files to your Azure Storage Account
./prepare-hxe-setup-files.sh sampleresourcegroupname samplestorageaccountname samplecontainer westeurope /home/mydirectory/hxe.tgz

Now, as I mentioned before, the automated setup of HANA Express happens in a script that runs in the post-provisioning phase of the Azure VM. That means, this script needs to have access to those setup files for an automated download without user interaction. To enable that scenario, the prepare-hxe-setup-files.sh-Script of my Quick Start uploads the setup packages for HANA Express to an Azure Storage Account and generates a Shared Access Signature URL which allows to simply download the packages using that Signature as a means of authentication by using wget or any similar shell-tool.

The following Screen-Shot shows the output of the prepare-hxe-setup-files-sh-script. You should especially take note of the Shared Access Signature URL the script outputs at the end!

Output of prepare-hxe-setup-files.sh

Deploy an SAP HANA Express using the templates

With the Setup packages for SAP HANA Express uploaded to an Azure Storage Account and the Storage Shared Access Signature generated as mentioned above, you can deploy as many SAP HANA Express Virtual Machines as you need to.

Important Note: the script prepare-hxe-setup-files.sh above generates Shared Access Signatures that are valid for a year. That means, after a year, you need to run the script, again, to generate a new Shared Access Signature. Note that the script is smart in detecting, if the files have been uploaded, already, and if so it generates the signature for the existing blob instead of uploading, again!

When using the quick-start-template, you can either use the “Deploy-To-Azure”-Button presented on the landing page of the Quick Start or you fill out the parameters in the azuredeploy.parameters.json-file as shown below and deploy the template via PowerShell or the Azure CLI:

Parameters filled in the azuredeploy.parameters.json

After you’ve filled out the parameters – the screen shot above shows the minimum ones you need to fill out – you can move ahead and deploy the template using code similar to the following:

# Create a resource group for your HANA Express VM Resources
az group create --name "samplehanaexpressgroup" --location "westeurope"

# Deploy the template with the filled parameters file
az group deployment create --resource-group="samplehanaexpressgroup" \
                           --template-file="azuredeploy.json" \
                           --parameters="@azuredeploy.sample.parameters.json" \
                           --name="samplehanaexpress"

The output of that script should look similar to the following screen shot:
Parameters filled in the azuredeploy.parameters.json

Validating the Installation

So far so good, if the output looks similar to the screen shot above, then you should be all set! But you can of course validate your installation – in two ways: using regular HANA tools to see if your instance is responsive or look at the installation logs from the provisioning process.

For that, you need to understand some background. I am using Azure Custom Script Extensions for Linux to automatically execute the HANA Installation with all required pre-requisites during the post-VM provisioning phase of the Azure Virtual Machine. That is expressed in the Azure Resource Manager Template with the following code:

... REST OF THE TEMPLATE ...

    {
        "type": "extensions",
        "name": "hxeinstallextension",
        "apiVersion": "2016-04-30-preview",
        "location": "[resourceGroup().location]",
        "dependsOn": [
            "[concat('Microsoft.Compute/virtualMachines/', parameters('vmNamePrefix'))]"
        ],
        "properties": {
            "publisher": "Microsoft.Azure.Extensions",
            "type": "CustomScript",
            "typeHandlerVersion": "2.0",
            "autoUpgradeMinorVersion": true,
            "settings": {
                "fileUris": [
                    "[parameters('hxeInstallScriptUrl')]"
                ]
            },
            "protectedSettings": {
                "commandToExecute": "[concat('sudo ./', parameters('hxeInstallScriptName'), ' \"', parameters('hxeSetupFileUrl'), '\" \"', parameters('hxeMasterPwd'), '\" && exit 0')]"
            }
        }
    }

... REST OF THE TEMPLATE ...

This part of the template shows, that after the Virtual Machine Resource has been provisionied, it uses the Azure Virtual Machine Agent to run the script specified in the template. This script is downloaded from the Quick-Start-Templates GitHub repository, directly. So no further steps needed to enable this.

If you now want to validate, whether the installation script from SAP HANA Express ran, successfully, you first should review the deployment logs within the Azure Portal similar to what’s shown in the following screen shot:

Azure Portal Deployment Log

If you still need to see more, then you just need to SSH into the virtual machine created (refer to the DNS name specified in the azuredeploy.parameters.json as per the screen shots above) and output the content of the stdout and stderr files within the /var/lib/waagent/custom-script/download/0 directory similar to what’s shown in the following screen shot:

Virtual Machine Deployment Log

When you look at the output, you’ll quickly realize, how much this really accelerates you. The setup script automatically performs the following steps for you:

  • Install the needed Oracle JDK on your machine
  • Install required library packages using zypper
  • Download and Extract the HANA Express Setup Packages to the VM
  • Install HANA Express on the VM

Final Words

The post above went quite into the details of every step of how the quick-start works. But the essence can literally be done within about 30min. depending on how fast your Internet connection is:

  • Register for an Azure Subscription if you don’t have one, yet
  • Setup the Azure CLI 2.0 on your machine if not done so, yet
  • Register with SAP
  • Download the SAP HANA Express Setup Packages
  • Execute the script prepare-hxe-setup-files.sh and note the generated Shared Access Signature
  • Click the “Deploy-to-Azure Button” or update the parameters file and execute az group deployment create

All of the needed pre-requisites and of course SAP HANA Express itself gets set up for you and within about 30min. you have an instance of it running in Microsoft Azure. I hope you find this more valuable and helps you to accelerate the setup of dev/test environments with SAP HANA Express since you don’t need to walk through all of the needed setup steps, manually.

Azure & Cloud Foundry – Setting up a Multi-Cloud Environment

This week I was presenting at the CloudFoundry Summit 2016 Europe in Frankfurt, of course about running CloudFoundry on Azure and Azure Stack. It was greate being here, especially because one of my two main Global ISV partners I am working with on the engineering side, have been here as well and are even a Gold-sponser of the event. It was indeed an honor and great pleasure for me to be part of this summit here … and great to finally have a technical session at a non-Microsoft conference, again:)

Indeed, one reason for that blog-post is because I ran out of time during my session and was able to show only small parts of the last demo.

Anyways, let’s get to the more techncial part of this blog-post. My session was all about running CF in Public, Private as well as Hybrid Clouds with Azure being involved in some way. This is highly relevant since most enterprises are driving a multi-cloud strategy of some way:

  • Either they are embracing Hybrid cloud and run deployments in the public cloud as well as in their own data centers for various reasons or
  • they want to distribute and minimize risk by running their solutions across two (or more) public cloud providers.

Despite the fact my session was focused on running Cloud Foundy on Azure, a lot of the concepts and architectural insights presented, can be re-used for other kinds of deployments with other cloud vendors or private clouds, as well.

The basics – Running Cloud Foundry on Azure and Pivotal

Microsoft has developed a Bosh CPI that enables bosh-based deployments of Cloud Foundry on Azure. The CPI is entirely developed as an Open Source Project and contributed to the Cloud Foundry Incubator project on GitHub.

Based on this CPI, there are two main ways for deploying deploying Cloud Foundry clusters on Microsoft Azure:

There’s a very detailed guidance on all of those GitHub repositories available that do explain all the details, I would suggest to follow this one since it is by far the easiest one: Deploy Cloud Foundry on Azure and always follow the via ARM templates suggestions of the docs.

Finally, in addition to Azure, to completly follow this post you need a 2nd CF cluster running in another cloud. The by far easiest way is to setup a trial account on Pivotal Cloud, which provides you with some sort of "CloudFoundry-as-a-Service". Follow these steps here for doing so…

A Multi-Cloud CF Architecture with Azure on one side

There are many reasons for multi-cloud environments. Some might include running parts in private clouds because of legal and compliance reasons while others including spreading risk across multiple cloud providers for disaster recovery reasons. The example in this post is focused exactly around the multi-cloud DR case since it covers two public cloud providers:

architecture

  • Azure Traffic Manager acts as a DNS-based load balancer. We will configure traffic manager with a Priority-Policy, which essentially leads traffic based on priority and if one cloud has a failure, Traffic Manager will route traffic to the other cloud.
  • The Azure Load Balancer is a component you get "for free" in Azure and don’t really need to take care off. It balances traffic across the front-nodes of your CF cluster and is automatically configured for you if you follow the guidance above for deploying CF on Azure.
  • Inside of each CF cluster, we need to make sure to register the DNS names used by Traffic Manager and configure the CF routers to route to our apps in the CF cluster, apropriately.

Setting up traffic manager

Let’s start with setting up the Azure Traffic Manager since we’ll need it’s domain name for the configuration of the apps in both Cloud Foundry targets. You can just add Azure Traffic Manager as a Resource to the Resource Group of your Cloud Foundry deployment or any other resource group. In my case, I deployed the Traffic Manager in another resource group as shown in the following screen shot:

Traffic Manager Setup

The important piece to take for now is the Domain Name of your traffic manager end-points. The actual end-points for traffic manager do not need to be configured at this point in time – we will look at it later.

Deploying the sample app to Pivotal Web Services

As a next step, we need to deploy the sample application to Pivotal web services and need to take note of the (probably random) domain name it has associated ot the application.

$pivotalApiEndpoint="api.run.pivotal.io"
cf login -a $pivotalApiEndpoint
cf target -o $pivotalOrg -s $pivotalSpace
cf push -f ./sampleapp/manifest.yml -p ./sampleapp
cf set-env multicloudapp REGION "Pivotal Cloud"
cf restage multicloudapp

To get the domain name and IP, just execute a cf app multicloudapp and take note of the domain name as shown in the following figure:

Pivotal App Domain Name

Deploying the App into Cloud Foundry on Azure

The deployment of the sample app into Azure goes exactly the same way, except that we’ll need to use different API end-points, organization names and spaces inside of Cloud Foundry:

$azureCfApiEndpoint="api.$azureCfPublicIp.xip.io"
cf login -a $azureCfApiEndpoint
cf target -o $azureOrg -s $azureSpace
cf push -f ./sampleapp/manifest.yml -p ./sampleapp
cf set-env multicloudapp REGION "Microsoft Azure"
cf restage multicloudapp

The Cloud Foundry API end-point I used above is the one that is registered by default when using the ARM-based deployment of open source Cloud Foundry with the Azure Quickstart Templates. The DNS-registration mechanism used there is documented here.

Also note the environment variables I am setting in the scripts above using cf set-env multicloudapp REGION "xyz". Indeed, that is used by our sample application (which is written with Ruby in this case) to output, in which region we are running the app. That way, we can see, if we are directed to the app deployed in Azure or in Pivotal Web Services.

Finally, if you’re new to Azure, the best way to find out the public IP which has been created for your CF cluster, is looking up a public IP address in the Azure Portal which has been created inside of the Resource Group for your Cloud Foundry cluster. Another way – if you are a Shell Scripter – would be to use the following command with the Azure Cross Platform CLI:

azure network public-ip show --resource-group YOUR-RESOURCE-GROUP YOUR-IP-NAME
info:    Executing command network public-ip show
+ Looking up the public ip "YOUR-IP-NAME"
data:    Id                              : /subscriptions/YOUR-SUBSCRIPTION-ID/resourceGroups/YOUR-RESOURCE-GROUP/providers/Microsoft.Network/publicIPAddresses/mszcfbasics-cf
data:    Name                            : YOUR-IP-NAME
data:    Type                            : Microsoft.Network/publicIPAddresses
data:    Location                        : northeurope
data:    Provisioning state              : Succeeded
data:    Allocation method               : Static
data:    IP version                      : IPv4
data:    Idle timeout in minutes         : 4
data:    IP Address                      : 52.169.87.212
data:    IP configuration id             : /subscriptions/YOUR-SUBSCRIPTION-ID/resourceGroups/marioszpCfSimple/providers/Microsoft.Network/networkInterfaces/SOME-ID/ipConfigurations/ipconfig1
data:    Domain name label               : marioszpcfsimple
data:    FQDN                            : marioszpcfsimple.northeurope.cloudapp.azure.com
info:    network public-ip show command OK

Configuring Traffic Manager Endpoints

Next, we need to tell Azure Traffic Manager the endpoints it should direct request which do approach on the DNS record registered with Traffic Manager to.

In our case, we use a simple Priority-based policy which means, Traffic Manager tries to always direct requests to an endpoint with the more important priority except that endpoint is not responsive. For a full documentation about policy routes, please refer to the Azure Traffic Manager docs.

Traffic Manager Endpoints

As you can see from the above, we have two endpoints:

  • Azure Endpoint which goes against the Public IP that the scripts and Bosh deployed for us when we deployed Cloud Foundry on Azure at the beginning.
  • External Endpoint which goes against the domain name for the app that Pivotal Web Services has registered for us (something like multicloudapp-xyz-abc.cfapps.io).

Let’s give it a try…

Now, in the previous configuration for Traffic Manager, we defined that the Pivotal Deployment has priority #1 and therefore will be preferred by Traffic Manager for Traffic routing. So, let’s open up a browser and navigate to the Traffic Manager DNS name for your deployment (in my screen shots and at my CF session that is marioszpcfsummithybrid.trafficmanager.net):

not working

Of course, a Cloud Foundry veteran spots immediately, what that means. I am not a veteran in that area, so I was falling into the trap…

Configuring Routes in Cloud Foundry

What I forgot when setting this up, originally, was configuring routes for the Traffic Manager Domain in my Cloud Foundry clusters. Otherwise, Cloud Foundry will reject requests coming in through that domain as it does not know about it.

We need to configure the routes on both ends to make it working, as shown below, we’re adding the traffic manager domain to the routes and ensure, CF routes traffic from those domains to our multi-cloud sample app:

$trafficMgrDomain=marioszpcfsummithybrid.trafficmanager.net

#
# First do this for Pivotal
#
cf login -a $pivotalApiEndpoint
cf target -o $pivotalOrg -s $pivotalSpace

cf create-domain $pivotalOrg $trafficMgrDomain
cf create-route $pivotalSpace $trafficMgrDomain
cf map-route multicloudapp $trafficMgrDomain

#
# Then do this for the CF Cluster on Azure
#
$azureCfApiEndpoint="api.$azureCfPublicIp.xip.io"
cf login -a $azureCfApiEndpoint
cf target -o $azureOrg -s $azureSpace

cf create-domain $azureOrg $trafficMgrDomain
cf create-route $azureSpace $trafficMgrDomain
cf map-route multicloudapp $trafficMgrDomain

Now let’s give it a try, again, and see what happens. This time we should see our Ruby sample app running and showing that it runs in Pivotal since we defined the priority for the Pivotal-based deployment within Azure Traffic Manager.
it works

Fixing Routes on Azure with Traffic Manager

After I indeed did the route mapping on Azure, Traffic Manager still claimed that the Azure-side of the house is Degraded, despite having the route configured. Initially, I didn’t understand why.

I didn’t have this problem when I initially tried this setup before. But when I initially tried this, I did not have assigned a DNS name to the Cloud Foundry Public IP in Azure. I’ve changed that because I tried something else in between and assigned a DNS name to the Azure Public IP for the CF Cluster. This lead traffic manager to route against that DNS name instead of the IP.

For troubleshooting that, I initated a fail-over and stopped the app on the Pivotal side (see next section) to make sure, Traffic Manager would try to route to Azure. A tracert finally told me, what was going on:

C:\code\github\mszcool\cfMultiCloudSample [master ≡]> tracert marioszpcfsummithybrid.trafficmanager.net

Tracing route to marioszpcfsimple.northeurope.cloudapp.azure.com [52.169.87.212]
over a maximum of 30 hops:

  1     5 ms     5 ms     4 ms  10.10.16.4
  2     2 ms     1 ms     1 ms  80.146.218.2
  3     2 ms     1 ms     2 ms  62.156.233.185
  4     5 ms     5 ms     5 ms  87.190.232.17
  5     8 ms     7 ms     7 ms  f-ed1-i.F.DE.NET.DTAG.DE [62.154.14.118]

When looking at the selected routes, we immediately spot, that the traffic manager domain gets resolved to the .cloudapp.net domain of the Azure Public IP. So my route on the CF-side of the house was just wrong. The route for Azure should not go against the traffic manager, but rather on the custom domain assigned to the cloud foundry cluster’s public IP in Azure:

cf map-route multicloudapp marioszpcfsimple.northeurope.cloudapp.azure.com

C:\code\github\mszcool\cfMultiCloudSample [master ≡]> cf routes
Getting routes for org default_organization / space dev as admin ...

space   host   domain                                            port   path   type   apps            service
dev            52.169.87.212
dev            marioszpcfsimple.northeurope.cloudapp.azure.com                        multicloudapp
dev            marioszpcfsummithybrid.trafficmanager.net                              multicloudapp

Testing a failover

Of course, we want to test if our failover strategy really works. For this purpose, we kill the App on the Pivotal-environment by executing the following commands:

cf login -a $pivotalApiEndpoint
cf target -o $pivotalOrg -s $pivotalSpace
cf stop multicloudapp

After that, we need to wait a while until traffic manager detects, that the application is not healthy. It then also might take a few seconds or minutes until the DNS record updates are propagated until we see the failover working (the smallest DNS TTL you can set, is 300s as of today).

So watch, what goes on, the simplest way is looking at the Azure Portal and opening up the Azure Traffic Manager configuration. At some point in time we should see, that one of the endpoints changes its status from Online to Degraded. When opening up a browser and trying to navigate to the traffic manager URL, we should no get redirected to the Azure-based deployment (which we see given our App is outputing the content of the environment variable we did set different for each of the deployments, before):

failover test

Final Words

I hope this gives you a nice start in setting up a Multi-Cloud Cloud Foundry environment across Azure and a 3rd-party cloud or your own data center. I will try to continue this conversation on my blog, for sure. There are tons of other cool things to explore with Cloud Foundry in relationship to Azure, and I’ll at least try to cover some of those. Let me know what you think by contacting me through twitter.com/mszcool!

As usual – all the code is available on my GitHub in the following repository:

https://github.com/mszcool/cfMultiCloudSample

Azure Virtual Machines – A Solution for Instance Metadata in Linux (and Windows) VMs

At SAP Sapphire we announced the availabiltiy of SAP HANA on Azure. My little contribution to this was working on case that was shown as a demo in the key note at SAP Sapphire 2016: Sports Basement with HANA on Azure. It was meant as a show-case and proof for running HANA One workloads in Azure DS14 VMs and it was the first case of HANA on Azure productive outside of the SAP HANA on Azure Large Instances.

While we proved we can run HANA One in DS14, what’s still missing is the official Marketplace image. We are working on that on-boarding of HANA One into the Azure Marketplace at the time I am writing this post here. This post is about a very specific challenge which I know is needed by many others, as well. While Azure will have a built-in solution, it is not available, today (August 2016), so this might be of help for you!

Scenario: A VM reading and modifying data about itself

This is a very common scenario. HANA One needs it as well. On other cloud platforms, especially AWS, a Virtual Machine can query information about itself without any hurdles through an instance metadata service. On Azure, as powerful as it is, we don’t have such a service available, yet (per August 2016). To be precise, we do, but it currently delivers information about regular maintenance, only. See here for further details. While such a service is in the works, it is not available, yet.

Instance metadata is especially interesting for software providers which want to offer their solutions through the marketplace. The metadata can be used for various aspects including association and validation of licenses or protection of software assets inside of the VM.

But what if a VM needs to modify settings through Cloud Provdier Management APIs, automatically? Even with an instance metadata service available, such requirements need a more advanced approach.

Solution: A possible approach outlined (and code on my GitHub Repo)

Based on that I started thinking about this challenge, prototyping it and sharing it with the broader technical community. With Azure having the concept of Service Principals available, I tried the following path:

  1. If we could pass in a Service Principal at the creation of the VM, we’d have all we need to call into Azure Resource Manager APIs.
  2. The VM can identify itself through it’s “Unique VM ID”. So we could query into Azure Resource Manager APIs and find the VM based on this ID.
  3. For Marketplace use cases it is necessary, that the user is FORCED to enter the credentials. So an ARM template with mandatory parameters for passing in the details for the Service Credential is needed.

With this in place we can solve both problems with a single solution: with the right permissions equipped, a Service Principal can query instance metadata through Azure Resource Manager APIs and modify virtual machine settings at the same time. Indeed, the Azure Cloud Foundry Bosh solution uses that approach as well, although it does not need to “identify” virtual machines. It just creates and deletes them…

For most Marketplace Vendors incl. the case above, the VM needs to change details about itself. So their would need to be a way for the VM to find itself through the VM Unique ID. Since nobody was able to answer the quesiton if that’s possible, I prototyped it with the Azure CLI.

Important Note: This is considered to be a prototype to proof if what is outlined above generally works. For production scenarios you’d need to code this in professional frameworks, better protect secrets by using those and build this into your product.

GitHub Repository: I’ve prototyped the entire solution and published it on my GitHub Repository here:

–>> https://github.com/mszcool/azureSpBasedInstanceMetadata

Step #1: Create a Service Principal

The first step is creating a Service Principal. That is not an easy task, especially when you think about offerings in a Marketplace where business people want to have fast and simple on-boarding.

Guess for what I’ve created this solution-prototype on my GitHub repository (with a blog-post followed). The idea of this prototype is to provide a ready-to-use service that creates Service Principals in your own subscription.

I still run this on my Azure Subscription, so if you need a Service Principal and you don’t like scripting, just use my tool for creating it. Note: please use in-private browsing and sign-in with a Global Admin (or get a Global Admin who does an Admin-Consent for my tool in your tenant).

If you love scripting, then you can use tools such as the Azure PowerShell or the Azure Cross Platform CLI. In my prototype, I built the entire set of scripts with the Azure CLI and tested it on Ubuntu Linux (14.04 LTS). Even cooler, I indeed developed and debugged all the Scripts on the new Bash on Ubuntu on Windows:
Bash on Windows

The script createsp.sh shows a sample-script which creates a Service Principal and assigns the needed roles to the Service Principal to read VM metadata in the subscription (it would be better to just target the resource group in which you want to create the VM… I just kept it like that for convenience).

# Each Service Principal in Azure AD is backed by an 'Application-registration'
azure ad app create --name "$servicePrincipalName" \
                    --home-page "$servicePrincipalIdUri" \
                    --identifier-uris "$servicePrincipalIdUri" \
                    --reply-urls "$servicePrincipalIdUri" \
                    --password $servicePrincipalPwd

# I use JQ to extract data out of JSON results such as the AppId
createdAppJson=$(azure ad app show --identifierUri "$servicePrincipalIdUri" --json)
createdAppId=$(echo $createdAppJson | jq --raw-output '.[0].appId')

azure ad sp create --applicationId "$createdAppId"

Note: I created the App and the Service Principal separately since the AppID is needed to login using Azure CLI with the Service Principal, anyways. Therefore I separated those steps since I needed to read the App and the Service Principal Object IDs, anyways.

Note: JQ is really a handy command line tool to extract data from the neat JSON-responses of the Azure CLI. Take a look at further details here.

After the Service Principal and the App are both created, I can assign the roles to the Service Principal so that he can query the VM Metadata in my subscription:

# If I would create the resource group earlier, I could use the
# --resource-group switch instead of the --subscription switch here to scope
# permissions to the resource group of the VM to-be-created, only.
azure role assignment create --objectId "$createSpObjectId" \
                             --roleName Reader \
                             --subscription "$subId" 

Finally, to complete the work, I needed the Tenant ID of the Azure AD Tenant for the target subscription which is also needed for the Login with a Service Principal with the Azure CLI. Indeed the following code-snippet is at the very beginning of the createsp.sh-Script:

# Get the entry for the target subscription
accountsJson=$(azure account list --json)

# The Subscription ID is needed throughout the script
subId=$(echo $accountsJson | jq --raw-output --arg pSubName $subscriptionName '.[] | select(.name == $pSubName) | .id')

# Finally get the TenantID of the Azure AD tenant which is associated to the Azure Subscription:
tenantId=$(echo $accountsJson | jq --raw-output --arg pSubName $subscriptionName '.[] | select(.name == $pSubName) | .tenantId')

With those data-assets above in place, the tenantId, the appId and the password selected for the app-creation, we can log-in with the service principal using the Azure CLI as follows:

azure telemetry --disable
azure config mode arm
azure login --username "$appId" --service-principal --tenant "$tenantId" --password "$pwd"

Note: Since we want to login in a script that runs automated in the VM to extract the metadata for an application at provisioning-time (in my sample – in the real world this could happen on a regular basis with a cron-job or something similar), we need to make sure to avoid any user prompts. The latest versions of Azure CLI prompt for telemetry data collection on the first call after installation. In an automation script you should always turn this off with the first command (azure telemetry --disable) in your script.

Step #2: A Metadata Extraction Script

Okay, now we have a Service Principal that could be used from backend jobs to extract metadata for the VM in an automated way, e.g. with the Azure CLI. Next we need a script to do exactly that. For my prototpye, I’ve created a shell script (readmeta.sh) that does exactly that. For this prototype I injected this script through the Custom Script Extension for Linux.

Note: Since the SAP HANA One team uses Linux as their primary OS, I just developed the entire prototype with Shell-Scripts for Linux. But fortunately, due to the Bash on Ubuntu on Windows 10, you can also run those from your Windows 10 machine right away (if you have the 2016 Anniversary Update installed).

You can dig into the depths of the entire readmeta.sh-Script if you’re interested. I just extract VM and Networking details in their to show, how to crack the VM UUID and to show, how-to extract related items which are exposed as separate resources in ARM attached to the VM.

Let’s start with first things first: the script requires the Azure Cross Platform CLI installed. On a newly provisioned Azure VM, that’s not there. So the script starts with installing stuff:

sudo mkdir /home/metadata
export HOME=/home/metadata

#
# Install the pre-requisites using apt-get
#

sudo apt-get -y update
sudo apt-get -y install build-essential
sudo apt-get -y install jq

curl -sL https://deb.nodesource.com/setup_4.x | sudo -E bash -
sudo apt-get -y install nodejs

sudo npm install -g azure-cli

Important Note: Since the script will run as a Custom Script extension, it does not have things like a user HOME directory set. To make NodeJS and NPM work, we need a Home-Directory. Therefore I set the HOME to /home/metadata to which I also save all the metadata JSON responses during the script.

The next hard thing was cracking the VM Unqiue ID. This Unique ID is available for some time in Azure and it identifiers a Virtual Machine for its entire lifetime in Azure. That ID changes when you take the VM off from Azure or delete it and re-create it. But as long as you just provision/de-provision or start/shutdown/start the VM, this ID remains the same.

But, the key question is whether you can use that ID to find a VM using ARM REST APIs to read metadata about itself, or even change its settings through Azure Resource Manager REST APIs. Obviously, the answer is yes, otherwise I would not write this post:). But the VM ID presented in responses from Azure Resource Manager REST APIs is different from what you get when reading it inside of the VM out of its asset tags – due to Big Endian bit ordering differences, also documented here.

So in my Bash-script for reading the metadata, I had to convert the VM ID before trying to use it to find my VM through the ARM REST APIs as follows:

#
# Read the VMID from the BIOS asset tag (skip the prefix, i.e. the first 6 characters)
#
vmIdLine=$(sudo dmidecode | grep UUID)
echo "---- VMID ----"
echo $vmIdLine
vmId=${vmIdLine:6:37}
echo "---- VMID ----"
echo $vmId

#
# Now switch the order due to encoding differences between the Windows and Linux World
#
vmIdCorrectParts=${vmId:20}
vmIdPart1=${vmId:0:9}
vmIdPart2=${vmId:10:4}
vmIdPart3=${vmId:15:4}
vmId=${vmIdPart1:7:2}${vmIdPart1:5:2}${vmIdPart1:3:2}${vmIdPart1:1:2}-${vmIdPart2:2:2}${vmIdPart2:0:2}-${vmIdPart3:2:2}${vmIdPart3:0:2}-$vmIdCorrectParts
vmId=${vmId,,}
echo "---- VMID fixed ----"
echo $vmId

That did the trick to get a VM ID which I can use to find my VM through ARM REST APIs, or through the Azure CLI since I am using bash-scripts here:

#
# Login, and don't forget to turn off telemetry to avoid user prompts in an automation script.
#
azure telemetry --disable
azure config mode arm
azure login --username "$appId" --service-principal --tenant "$tenantId" --password "$pwd"

#
# Get the details for the VM and save it
#
vmJson=$(azure vm list --json | jq --arg pVmId "$vmId" 'map(select(.vmId == $pVmId))')
echo $vmJson > /home/metadata/vmmetadatalist.json
echo "---- VM JSON ----"
echo $vmJson

What you see above is, that there’s today (as of August 2016) no way to query Azure Resource Manager REST APIs by using the VM Unique ID. Only attributes such as resource group and VM name can be used. Of course that applies to the Azure CLI, as well. Therefore I retrieve a list of VMs and filter it down using JQ by the VM ID… which fortunately is delivered as an attribute in the JSON response from the ARM REST APIs.

Now we have our first metadata asset, a simple list entry for the VM in which we are runnign with basic attributes. But what if you need more details. The obvious way is to execute an azure vm show --json command to get the full VM-JSON. But even that will not include all details. E.g. lets say you need the public or the private IP address assigned to the VM. What you need to do then is, navigating through relationships between those Azure Resource Manager Assets (the VM and the Network Interface Card resource, in specific). That is where it gets a bit tricky:

#
# Get the detailed VM JSON with relationship attributes (e.g. the NIC identified through its unique Resource ID)
#
vmResGroup=$(echo $vmJson | jq -r '.[0].resourceGroupName')
vmName=$(echo $vmJson | jq -r '.[0].name')
vmDetailedJson=$(azure vm show --json -n "$vmName" -g "$vmResGroup")
echo $vmDetailedJson > /home/metadata/vmmetadatadetails.json

#
# Then get the NIC for the VM through ARM / Azure CLI
#
vmNetworkResourceName=$(echo $vmJson | jq -r '.[0].networkProfile.networkInterfaces[0].id')
netJson=$(azure network nic list -g $vmResGroup --json | jq --arg pVmNetResName "$vmNetworkResourceName" '.[] | select(.id == $pVmNetResName)')
echo $netJson > /home/metadata/vmnetworkdetails.json

#
# The private IP is contained in the previously received NIC config (netJson)
#
netIpConfigsForVm=$(echo $netJson | jq '{ "ipCfgs": .ipConfigurations }')
echo $netIpConfigsForVm > /home/metadata/vmipconfigs.json

#
# But the public IP is a separate resource in ARM, so you need to navigate and execute a further call
#
netIpPublicResourceName=$(echo $netJson | jq -r '.ipConfigurations[0].publicIPAddress.id')
netIpPublicJson=$(azure network public-ip list -g $vmResGroup  --json | jq --arg ipid $netIpPublicResourceName '.[] | select(.id == $ipid)')
echo $netIpPublicJson > /home/metadata/vmipconfigspublicip.json

This should give you enough of the needed concepts to get all sorts of VM Metadata for your own VM using Bash-scripting. If you want to translate this to your Java, .NET, NodeJS or whatsoever code, then you need to look at the management libraries for the respective runtimes/languages.

Step #3: Putting it all together – the ARM template

Finally we need to put this all together! That happens in an ARM template and the parameters this ARM template requests from the user to be entered on provisioning. An ARM-template similar to this could be built for a solution template based Marketplace Offer.

On my GitHub repository for this prototype, the ARM template and its parameters are baked into the files azuredeploy.json and azuredeploy.parameters.json. I won’t go through all details of these templates. The most important aspects are in the parameters-section and in the VM creation section where I hook up the Service Principal with the Script and attach it as a Custom Script Extension. Start with an excerpt of the “parameters”-section of the template:

"parameters": {
    "storageAccountName": {
      "type": "string"
    },
    "dnsNameForPublicIP": {
      "type": "string"
    },
    "adminUserName": {
      "type": "string"
    },
    "adminPassword": {
      "type": "securestring"
    },
    "azureAdTenantId": {
      "type": "string"
    },
    "azureAdAppId": {
      "type": "string"
    },
    "azureAdAppSecret": {
      "type": "securestring"
    },
    ...
  },
...

The important parameters are the azureAdTenantId, azureAdAppId and azureAdAppSecret parameters. Those together form the sign-in details for the Service Principal as it is used in the script described in the previous section to read out the metadata for the VM on provisioning, automatically.

Reading the metadata is initiated through specifying my readmeta.sh-script as a custom script extension for the VM in the ARM template as below:

...
    {
      "type": "Microsoft.Compute/virtualMachines/extensions",
      "name": "[concat(parameters('vmName'),'/writemetadatajson')]",
      "apiVersion": "2015-06-15",
      "location": "[parameters('location')]",
      "dependsOn": [
        "[concat('Microsoft.Compute/virtualMachines/', parameters('vmName'))]"
      ],
      "properties": {
        "publisher": "Microsoft.OSTCExtensions",
        "type": "CustomScriptForLinux",
        "typeHandlerVersion": "1.5",
        "settings": {
          "fileUris": [
            "[concat('https://', parameters('storageAccountName'), '.blob.core.windows.net/customscript/readmeta.sh')]"
          ]
        },
        "protectedSettings": {
          "commandToExecute": "[concat('bash readmeta.sh ', parameters('azureAdTenantId'), ' ', parameters('azureAdAppId'), ' ', parameters('azureAdAppSecret'))]"
        }
      }
    }
...

Since the Azure Linux Custom Script extension prints a lot of diagnostics details about what it is doing, we need to at least make sure that our sensitive data, especially the Service Principal’s password is NOT included in that diagnostics logs to keep it protected (well… as good as possible:)). Therefore the commandToExecute-setting is put into the protectedSettings-section which is NOT disclosed in any diagnostics-logs from the Custom Script Extension.

Important Note: On the Azure Quickstarts Template Gallery are many templates that are using the custom script extension version 1.2. For having the commandToExecute-setting in the protectedSettings-section, you have to use a newer version. For me, the latest version 1.5 at the time of writing the post worked. With the previous versions it just didn’t call the script.

Step #4: Trying it out…

Before you can try things out, there’s one thing you need to prepare: create the storage account and upload the readmeta.sh-script into that account (argh, next time I just write the scripts to clone my GitHub-repository:)). To make it easy, I created a script called deploy.sh with 10 parameters that does everything:

  1. Create the Resource group
  2. Create the storage account
  3. Upload the script to the storage account
  4. Update the parameters in azuredeploy.parameters.json to reflect your service principal attributes
  5. Start the deployment with the template and the updated template parameters.

And while trying I thought the 10 parameters make it flexible, but it’s still a hard start if you’d love to just quickly try this. So I created another bash-script called getstarted.sh. That asks you for all the data interactively and calls the createsp.sh and deploy.sh scripts based on the input you interactively entered. Just like below:

Getting Started

Final Words

With this in place, you have a solution that allows you to do both, reading instance metadata of the VM in which your software runs and also (with the right permissions set on the Service Principal) modify aspects of the VM through Azure Resource Manager APIs or Command Line Interfaces.

Sure, this reads like a complex, long thing. It would be much easer for Instance Metadata if you could do it without authentication and Service Principals. All I can say is that this will change and will become easier. But for now, that’s a solution and I hope I provide you with valuable assets that make the story less complex for you to achieve this goal!

And even when we have a simpler solution for instance metadata available in Azure, the content above shows you some advanced scripting concepts of which I hope you can learn from. The coolest thing of it: since Windows 10 Anniversary Update you can run all of the above on both, Windows and Ubuntu Linux, BECAUSE all is written as Bash scripts.

For me the nice side-effect of this was experiencing, how mature the Linux Subsystem for Windows seems to be. What really surprised me is, that I even can run Node Version Manager and build-essential on it (I even tried compiling v5 of my Node.JS version using it and it ran through and works).

Anyways – if you have any questions, reach out to me on Twitter.

A Deep Dive into Azure AD Multi-Tenant Apps, OAuth/OpenIdConnect Flows, Admin Consent and Azure AD Graph API

I am currently working with one if my main Global Independent Software Vendor (ISV) partners for on-boarding their solution into the Azure Marketplace. The main challenge that we face there is, that the solution needs to do some post-provisioning steps in the end-customer’s target subscription as well as Azure Active Directory tenant:

  • Creating a Service Principal that can be used by the Software inside of the provisioned VM in the end-customer’s target directory.
  • Using that service principal to read data from the end-customer’s Azure Subscription.

Note: the end customer in this case is the customer, who purchases the product published by the ISV in the store!

Such cases typically require the creation of "multi-tenant" Azure Active Directory applications. And this application then needs to access the end-customer’s target directory using the Azure AD Graph API. At the same time, creating service principals is not an easy task.

A Multi-Tenant Web App to create Service Principals as Sample

To make this as practical as possible, I decided to create a web app that creates service principals in the target Azure Active Directory of an end-customer that’s using the web app.

This shows, how the general multi-tenancy challenge can be solved and at the same time provides a handy tool for creating Service Principals, which is a harder task on its own.

All the details for using the app and for cloning the source code are available on my GitHub-repository under the link below. In addition, I also run the app on my Azure Subscription as a free-tier Azure Web App.

The documentation shows, how-to register a multi-tenant application in your Azure AD tenant to make such an application available as a multi-tenant application. It shows, how such an application is reflected in a customer’s target Azure AD tenant and how-to manage access to it.

The sample also demonstrates the various OAuth- and OpenIdConnect-flows which are needed in a simple yet practical and useful scenario. All of this should be easy to reflect to your own scenarios and I found that, despite Microsoft has decent docs for Azure Active Directory out there, such a sample is not easy to find in an end-2-end and focused way. That’s what I tried to create.

The basic/initial OpenIdConnect-Flow for Signing-In

So, let’s start with digging into the OAuth details. First of all, all the theory is well-explained on the official Microsoft Azure and MSDN documentation pages (see last section of the article).

I just get down at the protocol-trace level so that it’s easy for developers to understand what’s going on and how simple those protocols are, indeed. Also it should help configuring/using other frameworks on all sorts of platforms appropriately to fit into this model.

For all of the below I am using the real deployment of my Service Principal Web App Demo mentioned above (note: I might remove that deployment at any point in time since I’ve guidance on my GitHub-repo for how-to deploy it in your own Azure AD tenant, as well).

  1. First the user browses to the target application which is secured by Azure AD.

  2. That typically ends up in a redirect to Azure AD as an IDP to get an initial token. A typical Redirect Request for an OAuth Sign-In flow looks as follows (using line-breaks to make it easier to read):

    GET https://login.microsoftonline.com/common/oauth2/authorize?
        client_id=---your client id from azure ad app registration---
        &response_mode=form_post
        &response_type=code+id_token
        &scope=openid+profile
        &state=OpenIdConnect.AuthenticationProperties%3dW1HmJdRTdYw...
        &redirect_uri=https%3a%2f%2flocalhost%3a44330%2f HTTP/1.1
    Host: login.microsoftonline.com
    Connection: keep-alive
    Upgrade-Insecure-Requests: 1
    User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/51.0.2704.103 Safari/537.36
    Accept: text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,*/*;q=0.8
    Accept-Encoding: gzip, deflate, sdch, br
    Accept-Language: en-US,en;q=0.8
    
    
    • The client_id-parameter reflects the Client ID that is configured in Azure Active Directory for that application.
    • The scope-parameter contains various additional items used for token validation.
    • The nonce is used to protect against token replay attacks (typically). It’s value provided in the request must match the response and is unique per user session, typically.
  3. When the user (assume Admin) signs in for the first time, a consent dialog is displayed. This is part of the OAuth Authorization flow and gives the user a chance to "Accept" or decline the permissions the app needs. Since that is handled by Azure AD as an IdP, we don’t look into the details of the requests issued there.

    Consent

  4. Once the user accepted this consent, Azure AD posts a token to a target URL which was specified in the earlier request with the redirect_uri parameter. Let’s look at the details (again with newlines for readability):

    POST https://localhost:44330/ HTTP/1.1
    Host: localhost:44330
    Connection: keep-alive
    Content-Length: 2428
    Cache-Control: max-age=0
    Origin: https://login.microsoftonline.com
    Upgrade-Insecure-Requests: 1
    User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/51.0.2704.103 Safari/537.36
    Content-Type: application/x-www-form-urlencoded
    Accept: text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,*/*;q=0.8
    Referer: https://login.microsoftonline.com/common/Consent/Grant
    Accept-Encoding: gzip, deflate, br
    Accept-Language: en-US,en;q=0.8
    Cookie: OpenIdConnect.nonce.Z9f6E8u...
    
    code=AAABA...
    

    That post contains an OAuth-Authorization code in the body. This Code can be used to request tokens from Azure AD for downstream API-calls of APIs which are also secured by Azure AD. Of course, the code will only work for APIs to which the app has been given permissions in the Azure AD portal.

    Permissions of the App

    For the "Service Principal Demo App" those permissions are highlighted in the screen shot above. The Code therefore would work for requests of tokens for the Azure Active Directory Graph API (identified as https://graph.windows.net) and the Azure Service Management and Resource Manager APIs (identified as https://management.core.windows.net).

  5. When the Service Principal Web App receives the request, it actually uses it to request an additional token that permits the app to call into Azure Active Directory Graph APIs. This is another token-request which the app tries to execute when it received the post above.

    POST https://login.microsoftonline.com/common/oauth2/token HTTP/1.1
    Accept: application/json
    x-client-last-request: a5db36d8-ab46-4dfc-b96e-9dc31cf06a5c
    x-client-last-response-time: 1284
    x-client-last-endpoint: token
    x-client-SKU: PCL.Desktop
    x-client-Ver: 3.10.0.0
    x-client-CPU: x64
    x-client-OS: Microsoft Windows NT 10.0.10586.0
    x-ms-PKeyAuth: 1.0
    client-request-id: 1eb9034c-e02c-4e7b-8c4f-0fe5e2faabfe
    return-client-request-id: true
    Content-Type: application/x-www-form-urlencoded
    Host: login.microsoftonline.com
    Content-Length: 1079
    Expect: 100-continue
    
    resource=https%3A%2F%2Fgraph.windows.net&client_id=---your client id from azure ad app registration---&client_secret=---your client secret configured in the azure ad portal&grant_type=authorization_code&code=---previously received authorization code---&redirect_uri=https%3A%2F%2Flocalhost%3A44330%2F
    

    Such a request would the respond with a new OAuth Bearer Token that would permit us to call into the Azure AD Graph APIs. This token needs to be added to the HTTP Authorize header on each request, then. Here’s an example response for the request above:

    HTTP/1.1 200 OK
    Cache-Control: no-cache, no-store
    Pragma: no-cache
    Content-Type: application/json; charset=utf-8
    Expires: -1
    Server: Microsoft-IIS/8.5
    Strict-Transport-Security: max-age=31536000; includeSubDomains
    X-Content-Type-Options: nosniff
    x-ms-request-id: fb2db119-ca9e-421d-8007-6ae7e97d163e
    client-request-id: 1eb9034c-e02c-4e7b-8c4f-0fe5e2faabfe
    x-ms-responsehealth: TargetId=ESTSFE_IN_329;Action=None;Category=None;Health=0;Load=9;
    P3P: CP="DSP CUR OTPi IND OTRi ONL FIN"
    Set-Cookie: esctx=AAABAA ...; domain=.login.microsoftonline.com; path=/; secure; HttpOnly
    Set-Cookie: x-ms-gateway-slice=productionb; path=/; secure; HttpOnly
    Set-Cookie: stsservicecookie=ests; path=/; secure; HttpOnly
    X-Powered-By: ASP.NET
    Date: Wed, 29 Jun 2016 21:31:37 GMT
    Content-Length: 3826
    
    {
      "token_type": "Bearer",
      "scope": "Directory.AccessAsUser.All Directory.ReadWrite.All Group.ReadWrite.All User.Read",
      "expires_in": "3599",
      "ext_expires_in": "3600",
      "expires_on": "1467239498",
      "not_before": "1467235598",
      "resource": "https://graph.windows.net",
      "access_token": "eyJ0eXAiOiJK...",
      "refresh_token": "AAABAAAAiL9Kn2..."
    }
    

    The response is a JSON-response containing some helpful details about the issued token as well as a refresh-token to renew the actual access token. Note: if you need to get a new access token with the refresh token, you still need to have the Client ID and the App Secret available in that refresh-request.

OAuth Admin Consent for Multi-Tenant Azure AD Apps

Yikes, the biggest challenge I faced with the tool when building it was, that ordinary Azure AD Users (role = ‘User’) where not able to use it. You had to be a ‘Global Admin’ to execute it.

The main reason for that was, that my app requires "acting as the Signed-in User" against Azure AD Graph API. And for that, the Azure AD team changed the default behavior for a good reason a while ago (well, in March 2015): https://blogs.msdn.microsoft.com/aadgraphteam/2015/03/18/update-to-graph-api-consent-permissions/.

So, to enable ordinary users to make use of such applications, a Global Admin first needs to "approve" the application for the target directory by running through an OAuth Admin Consent. This is a special type of consent that asks the Global Admin if he wants to make the permissions the App requires available to ordinary users inside of the Organization (technically: in the target directory against the multi-tenant app tries to work depending on the signed-in user).

The steps are:

  1. The Global Admin needs to Sign-in into the application.

  2. The application needs to provide the appropriate "on-boarding"-function, which essentially initiates the Admin-Consent against the target directory of the signed-in user. I did this by just adding a button to my app that starts the Admin Consent.

    Admin Consent Function

  3. All that button does is composing a URL that goes against the Azure Active Directory OAuth endpoints to walk through the Admin Consent. This leads to the following request that initiates the Admin Consent:

    GET https://login.windows.net/yourazureadtenantid/oauth2/authorize?
        api-version=1.0
        &response_type=code
        &client_id=yourazureadappid
        &resource=https://management.core.windows.net/
        &redirect_uri%20=https://mszcoolserviceprincipal.azurewebsites.net/Home/CatchConsentResult
        &prompt=admin_consent 
        HTTP/1.1
    Host: login.windows.net
    Connection: keep-alive
    Cache-Control: max-age=0
    Upgrade-Insecure-Requests: 1
    User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/51.0.2704.103 Safari/537.36
    Accept: text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,*/*;q=0.8
    Referer: https://mszcoolserviceprincipal.azurewebsites.net/
    Accept-Encoding: gzip, deflate, sdch, br
    Accept-Language: en-US,en;q=0.8    
    

    The really important aspect of that request is the query-string parameter prompt=admin_consent which does the work. I combine the request with issuing an authorization code right away, but I do think that’s optional (would need to read back in the specs:)).

  4. After that initial admin-consent is completed, every other ordinary user (role = ‘user’) can sign-in to the application and make use of it. The admin-consent literally approved the application through an administrator for the organization for security reasons.

The Graph API calls with the issued tokens

Finally, with that Access Token we can make calls into the Azure AD Graph API. The sample-calls for creating a Service Principal are similar to the following types of requests.

  1. First, the app tries to find if the needed "Application" for the Service Principal has been created in Azure AD, already:

    GET https://graph.windows.net/yourazureadtenantid/applications()?$filter=identifierUris/any(iduri:iduri%20eq%20'http%3A%2F%2Fyourappidurienteredinthescreen')&api-version=1.6 HTTP/1.1
    DataServiceVersion: 3.0;NetFx
    MaxDataServiceVersion: 3.0;NetFx
    Accept: application/json;odata=minimalmetadata
    Accept-Charset: UTF-8
    DataServiceUrlConventions: KeyAsSegment
    User-Agent: Microsoft Azure Graph Client Library 2.1.1
    Authorization: Bearer eyJ0eXAiOiJK...
    X-ClientService-ClientTag: Office 365 API Tools 1.1.0612
    Host: graph.windows.net
    Connection: Keep-Alive
    
    

    The request above looks, if an Application is registered in the target tenant yourazureadtenantid with the App ID URI http://yourappidurienteredinthescreen. The HTTP-response will have an OData-based JSON with the resulting elements in it if an App exists, already.

    {
      "odata.metadata": "https://graph.windows.net/yourazureadtenantid/$metadata#directoryObjects/Microsoft.DirectoryServices.Application",
      "value":[
        ...
      ]
    }
    
  2. If no application exists, it actually creates the application by posting an ApplicationEntity into the Graph API:

    POST https://graph.windows.net/yourazureadtenantid/applications?api-version=1.6 HTTP/1.1
    DataServiceVersion: 3.0;NetFx
    MaxDataServiceVersion: 3.0;NetFx
    Content-Type: application/json;odata=minimalmetadata
    Accept: application/json;odata=minimalmetadata
    Accept-Charset: UTF-8
    DataServiceUrlConventions: KeyAsSegment
    User-Agent: Microsoft Azure Graph Client Library 2.1.1
    Authorization: Bearer eyJ0eXAiOiJKV1...
    X-ClientService-ClientTag: Office 365 API Tools 1.1.0612
    Host: graph.windows.net
    Content-Length: 201
    Expect: 100-continue
     
    {
     "odata.type": "Microsoft.DirectoryServices.Application",
     "displayName": "YourAppDisplayName",
     "identifierUris@odata.type": "Collection(Edm.String)",
     "identifierUris": [
         "http://YourAppIdUri"
     ]
    }
    

    This post will return with a detailed JSON object which contains all the details about the created App including it’s AppId.

  3. Then the application does the same for checking if a Service Principal exists for the Application previously created, already.

    GET https://graph.windows.net/yourazureadtenantid/servicePrincipals()?$filter=appId%20eq%20'b3ccae52-19bc-45a1-a4e4-f572f6963213'&api-version=1.6 HTTP/1.1
    DataServiceVersion: 1.0;NetFx
    MaxDataServiceVersion: 3.0;NetFx
    Accept: application/json;odata=minimalmetadata
    Accept-Charset: UTF-8
    DataServiceUrlConventions: KeyAsSegment
    User-Agent: Microsoft Azure Graph Client Library 2.1.1
    Authorization: Bearer eyJ0eXAiOiJKV1...
    X-ClientService-ClientTag: Office 365 API Tools 1.1.0612
    Host: graph.windows.net
    

    The response will again contain an OData JSON document with the service principal if it exists, already. I am skipping the details for now…

  4. Finally, if the Service Principal does not exist, the app creates one with a password credential attached to it. That means this principal can be used by service- and backend-applications.

    POST https://graph.windows.net/yourazureadtenantid/servicePrincipals?api-version=1.6 HTTP/1.1
    DataServiceVersion: 3.0;NetFx
    MaxDataServiceVersion: 3.0;NetFx
    Content-Type: application/json;odata=minimalmetadata
    Accept: application/json;odata=minimalmetadata
    Accept-Charset: UTF-8
    DataServiceUrlConventions: KeyAsSegment
    User-Agent: Microsoft Azure Graph Client Library 2.1.1
    Authorization: Bearer eyJ0eXAiOiJKV1...
    X-ClientService-ClientTag: Office 365 API Tools 1.1.0612
    Host: graph.windows.net
    Content-Length: 627
    Expect: 100-continue
     
    {
     "odata.type": "Microsoft.DirectoryServices.ServicePrincipal",
     "accountEnabled": true,
     "appId": "b3ccae52-19bc-45a1-a4e4-f572f6963213",
     "displayName": "tttttteeeeeeeessssstttt",
     "passwordCredentials@odata.type": "Collection(Microsoft.DirectoryServices.PasswordCredential)",
     "passwordCredentials": [
         {
             "customKeyIdentifier": null,
             "endDate": "2017-06-29T21:43:15.6654372Z",
             "keyId": "0259571d-a663-4507-94e9-9381629e2116",
             "startDate": "2016-06-29T21:43:15.6639533Z",
             "value": "pass@word1"
         }
     ],
     "servicePrincipalNames@odata.type": "Collection(Edm.String)",
     "servicePrincipalNames": [
         "b3ccae52-19bc-45a1-a4e4-f572f6963213",
         "http://tttttteeeeeeeessssstttt"
     ]
    }
    

Note: One piece missing is to assign appropriate roles for executing on Service Management Operations for Azure Resource Manager Rolebased Access Control so that the Service Principal can execute the needed operations against the management APIs.

Do you really need to know all of these details?

With that we went through all the protocol details for the OAuth, OpenIdConnect and Graph API calls that are needed to accomplish an end-2-end task. It’s actually a very practical look at how all these "sequence diagrams" that are talking about OAuth are looking in the real world.

My intent to show these details was, to help people which are working with programming languages and runtimes that do not have nice SDKs available for encapsulating those protocol details to at least have a high-level overview and starting-point without reading the OAuth and OpenIdConnect specs. I know it’s high-level, but it’s practical.

OAuth and OpenId Connect Azure AD Resources

The following links do explain all the different query string parameters of the OAuth/OpenIdConnect flows with Azure AD. They are a great resource to better understand the http-requests I’ve outlined above.

SDKs for languages and Runtimes

Fortunately, if you are a .NET, Java, Node.js, PHP or Python developer, there are numerous examples and resources available. Also for Azure AD’s Graph API there’s a nice tool available to dig into all the JSON and protocol details.

Here are the most important links:

For Graph API there are also good samples and SDKs out there:

I hope that was helpful and gives you a great background or even a handy tool to create Service Principals. My partner needed the understanding of how-to build such multi-tenant Azure AD applications that do access the Azure AD Graph API and they needed to create Service Principals out of such a multi-tenant web application. So I thought it’s worth spending the additional time and getting it documented!

NServiceBus, Azure Service Bus and Service Bus for Windows Server – A PoC for a Hybrid Cloud / Portable Solution

NServiceBus is a very popular messaging and workflow framework for .NET developers across the globe. This week a few peers of mine and I were working for one of our global ISV partners to evaluate, if NServiceBus can be used for Hybrid Cloud and portable solutions, that can be moved seamless from On-Premises to the Public Cloud and vice-versa.

My task was to evaluate, if NServiceBus can be used with both, Microsoft Azure Service Bus in the public cloud as well as Service Bus 1.1 for Windows Server in the private cloud. It was a very interesting collaboration and finally I got to write some prototype code for one of our partners, again. How cool is that – doing interesting stuff and at the same time it helps a partner. That’s how it should be!

Part #1: On-Premises Service Bus 1.1 Environment

The journey and prototyping began with setting up an On-Premises Service Bus 1.1 environment in my home lab. Fortunately there are some good instructions out there, but of course nothing goes without any pitfalls. Here’s a good set of instructions to start with – note that I did setup an entire Azure Pack express setup, which is clearly optional. But it makes things more convenient, especially for presentations, since it provides the nice, good old Azure Management Portal experience for your Service Bus on-premises. Here’s where you should look at how-to setup things:

  • Install the Azure Pack Express Setup
    • This shows, how-to setup a basic Azure Pack environment using Web Platform Installer.
    • I did not run into any problems installing it on a Hyper-V Box on my Home Lab. So it should be fairly straight forward.
    • You can install all on a single machine. What you need is SQL Server (Express is sufficient) pre-installed.
  • Install Service Bus for Windows Server
    • Again this happens via Web Platform Installer. With this I had a little challenge. Unfortunately as of writing this article, the link to the required version of the Windows Fabric in the Web Platform Installer was broken. I’ve uploaded it
      on my public OneDrive for convenience. You find it here. But I’ve had conversations with the product team, they will fix the broken link. So when you try it, it might work, already.
  • Configure Service Bus using the Wizard.
    • After installing you need to configure Service Bus for Windows Server. That happens through a Wizard. It essentially allows you to configure endpoints, ports and certificates used for security purposes for Service Bus 1.1 for Windows Server.
    • The configuration failed on the first attempt because something went wrong with installing the Service Bus patch. Uninstall and re-install did solve the problem.
  • Configure Service Bus for Windows Server for the Azure Pack Portal
    • That’s the final step to get the Azure Pack management portal experience for Service Bus 1.1 for Windows Server.
    • If you are fine with managing Service Bus through PowerShell, you can skip the entire Azure Pack Express stuff, start with Service Bus 1.1 right away and manage it through PowerShell.

At the end of this journey, which took me about half a day overall starting from scratch and figuring out the little gotchas mentioned above, I had a lab environment to test against. I am a fan of using Royal TS, hence the screen shot with my on-premises Service Bus as web pages embedded in Royal TS:

Part #2: NServiceBus and Azure Service Bus

The second part of the challenge was easy – figure out if NServiceBus supports Azure Service Bus, already. Because that would give us a good starting point, wouldn’t it!? Here are the docs for the NServiceBus transport extension for the Azure Serivce Bus: source code.

But the point is, the earliest version that supports Azure seems to be NServiceBus v5.0.0 and they started with a Microsoft.ServiceBus.dll above 3.x in the code-base. This version of the library is not compatible with Service Bus 1.1 for Windows Server. So I had to dig into the source code and back-port the library. Fortunately, Particular is open sourcing most of the framework’s bits and pieces on GitHub – so it does for the NServiceBus.AzureServiceBus connector here.

Note that I am referring to the version 6.2 of the implementation, directly, since that works with NServiceBus 5.0.0 which our global ISV partner is using at this point in time. I also tried back-porting the current development branch, but that turned
out to be way more complex and risky. And it was not needed for the partner, either:)

Part #3: Back-Porting to Microsoft.ServiceBus.dll v2.1

So the needed step is to back-port to a Service Bus SDK library that also works with Service Bus 1.1. Service Bus for Windows Server recently received a patch to work with .NET 4.6.1, but it has not received any major updates since its original release.
So it is behind with regards to its APIs compared to Service Bus in Azure.

I’ve done all of the steps below on my GitHub repository in a fork of the original implementation. Note that you should only look at my work in the branch ‘support-6.2’ which is the one that works with NServiceBus 5.0.0. The rest is considered to be experiments as we speak right now:)

Here is the link to my GitHub repo and the fork!!

The first step for doing so was to remove the NuGet package and replace it with one that works with Service Bus for Windows Server. Microsoft fortunately released a separate NuGet package that contains the version that is compatible with Service Bus for Windows Server:

The rest was all about looking where the NServiceBus-implementation is using features that are not available in version 2.1 of the Service Bus SDK and testing it against my Service Bus 1.1 for Windows Server lab setup. I think the best place to look at
what actually changed is by looking at the change-logs on my GitHub-Repository:

  • Initial back-port with most code changes (click here to open details on GitHub)
    • Update the NuGet Package to “ServiceBus.v1_1” instead of “WindowsAzure.ServiceBus”.
    • Remove EnablePartitioning because that’s not supported on SB 1.1.
    • Use MessagingFactory.CreateFromConnectionString() instead of MessagingFactory.Create() because the latter one does not assume different ports on different EndPoints for different APIs Service Bus is exposing. But that’s typically the case on default-setups of Service Bus for Windows Server (see my first ScreenShot).
    • I also added some regular expressions to detect, if a Service Bus Connection String is one for on-premises or the public cloud to keep most of the default behaviors when connecting against the public cloud true. See the code-snippet below. It might not be complete or perfect, but it fulfills the basic needs.
  • Added some samples (click here to open details on GitHub)
    • This contains a basic Sender and Receiver implementation that uses the transport.
    • You need to set the Environment Variable AzureServiceBus.ConnectionString in a command prompt and start Visual Studio from that one to successfully execute. Btw. that’s also needed if you need to run the tests. In that case you
      also need to set AzureServiceBus.ConnectionString.Fallback with an alternate Service Bus connection string.

Here is the little code-snippet that checks if the code is used for on-premises Service Bus services or for Azure Service Bus instances:

class CreatesMessagingFactories : ICreateMessagingFactories
{
    #region mszcool 2016-04-01

    // mszcool - Added Connection String parsing to detect whether a public or private cloud Service Bus is addressed!
    public static readonly string Sample = "Endpoint=sb://[namespace name].servicebus.windows.net;SharedAccessKeyName=[shared access key name];SharedAccessKey=[shared access key]";
    private static readonly string Pattern =
        "^Endpoint=sb://(?<namespaceName>[A-Za-z][A-Za-z0-9-]{4,48}[A-Za-z0-9]).servicebus.windows.net/?;SharedAccessKeyName=(?<sharedAccessPolicyName>[\\w\\W]+);SharedAccessKey=(?<sharedAccessPolicyValue>[\\w\\W]+)$";

    public static readonly string OnPremSample = "Endpoint=[namespace name];StsEndpoint=[sts endpoint address];RuntimePort=[port];ManagementPort=[port];SharedAccessKeyName=RootManageSharedAccessKey;SharedAccessKey=[shared access key]";
    private static readonly string OnPremPattern =
        "^Endpoint=sb\\://(?<serverName>[A-Za-z][A-Za-z0-9\\-\\.]+)/(?<namespaceName>[A-Za-z][A-Za-z0-9]{4,48}[A-Za-z0-9])/?;" +
        "StsEndPoint=(?<stsEndpoint>https\\://[A-Za-z][A-Za-z0-9\\-\\.]+\\:[0-9]{2,5}/[A-Za-z][A-Za-z0-9]+)/?;" +
        "RuntimePort=[0-9]{2,5};ManagementPort=[0-9]{2,5};" +
        "SharedAccessKeyName=(?<sharedAccessPolicyName>[\\w\\W]+);" +
        "SharedAccessKey=(?<sharedAccessPolicyValue>[\\w\\W]+)$";

    private bool DetectPrivateCloudConnectionString(string connectionString)
    {
        if (Regex.IsMatch(OnPremPattern, connectionString, RegexOptions.IgnoreCase))
            return true;
        else if (Regex.IsMatch(Pattern, connectionString, RegexOptions.IgnoreCase))
            return false;
        else {
            throw new ArgumentException($"Invalid Azure Service Bus connection string configured. " +
                                        $"Valid examples: {Environment.NewLine}" +
                                        $"public cloud: {Pattern} {Environment.NewLine}", 
                                        $"private cloud (SB 1.1): {OnPremPattern}");
        }
    }

    #endregion

    ICreateNamespaceManagers createNamespaceManagers;
    // ... rest of the implementation ...
}

The part where I needed this detection most was to decide, how-to instantiate the MessagingFactory. This is the relevant piece of code – note that MessagingFactory.Create() with the NamespaceManager-Address passed in does only work in the public cloud, not with Service Bus 1.1 on Windows Server:

class CreatesMessagingFactories : ICreateMessagingFactories
{
    // ... earlier stuff in that class including 'DetectPrivateCloudConnectionString' ...

    ICreateNamespaceManagers createNamespaceManagers;

    public CreatesMessagingFactories(ICreateNamespaceManagers createNamespaceManagers)
    {
        this.createNamespaceManagers = createNamespaceManagers;
    }

    public MessagingFactory Create(Address address)
    {
        var potentialConnectionString = address.Machine;
        var namespaceManager = createNamespaceManagers.Create(potentialConnectionString);

        // mszcool - Updated to detect if Service Bus 1.1 for Windows Server is used
        if (DetectPrivateCloudConnectionString(potentialConnectionString))
        {
            // mszcool - Need to use this approach because different ports for control and transport endpoints are used
            return MessagingFactory.CreateFromConnectionString(potentialConnectionString);
        }
        else {
            var settings = new MessagingFactorySettings
            {
                TokenProvider = namespaceManager.Settings.TokenProvider,
                NetMessagingTransportSettings =
            {
                BatchFlushInterval = TimeSpan.FromSeconds(0.1)
            }
            };
            return MessagingFactory.Create(namespaceManager.Address, settings);
        }
    }
}

Finally with those fixes incorporated, I was able to get almost all things working and all except two tests passing for now. For the proof-of-concept that is successful for now, since it proofs that the partner could achieve what they need to achieve.

Part #4: See it in Action

Now comes the cool part – testing it out and seeing it in Action. The Samples I’ve added to the git-repository are simple messaging examples which I’ve modified from the NServiceBus Samples repository as well. Note that I’ve taken the Non-Durable MSMQ sample to proof since starting-point for the partner was MSMQ and I wanted something super-simple to start with. That is just how far I got, eventually I try other samples (but no promise at this time:)). Below the code-snippet of the sender – the receiver looks nearly identical and you can look it up on my repository on GitHub:

static void Main()
{
    string connStr = System.Environment.GetEnvironmentVariable("AzureServiceBus.ConnectionString");

    Console.Title = "Samples.MessageDurability.Sender";
    #region non-transactional
    BusConfiguration busConfiguration = new BusConfiguration();
    busConfiguration.Transactions()
        .Disable();
    #endregion
    busConfiguration.EndpointName("Samples.MessageDurability.Sender");
    busConfiguration.ScaleOut().UseSingleBrokerQueue();
    busConfiguration.UseTransport<AzureServiceBusTransport>()
        .ConnectionString(connStr);
    busConfiguration.UseSerialization<JsonSerializer>();
    busConfiguration.EnableInstallers();
    busConfiguration.UsePersistence<InMemoryPersistence>();

    using (IBus bus = Bus.Create(busConfiguration).Start())
    {
        bus.Send("Samples.MessageDurability.Receiver", new MyMessage());
        Console.WriteLine("Press any key to exit");
        Console.ReadKey();
    }
}

Here’s the code actually in action and working. You see what it produced on my on-premises Service Bus as well as the log output from the console windows. Note that the message handler part of the receiver outputs that it received a message.

One thing I played around with was having two receivers, that’s why you see two output-lines for one single message in my receiver. To clarify, here’s the code of MyHandler.cs from the Receiver project which outputs those lines:

public class MyHandler : IHandleMessages<MyMessage>
{
    static ILog logger = LogManager.GetLogger<MyHandler>();

    public void Handle(MyMessage message)
    {
        logger.Info("Hello from MyHandler");
    }
}

public class MyHandler2 : IHandleMessages<MyMessage>
{
    static ILog logger = LogManager.GetLogger<MyHandler2>();

    public void Handle(MyMessage message)
    {
        logger.Info("Hello from MyHandler2!");
    }
}

Final Words

I think this Proof-Of-Concept we’ve built alongside with other aspects we covered for that Global Software Vendor partner in UK demonstrates several aspects:

  • That it is possible to have a solution that works (nearly) seamless on-premises and in the public cloud with largely the same code base.
    • The situation should DRAMATICALLY improve once Microsoft has released the Azure Stack, which is the successor of what I’ve used here (which was the Azure Pack).
    • We can expect that the Azure Stack will deliver a much more up2date and consistent experience with Azure in the public cloud once it is fully available incl. Service Bus.
  • That with NServiceBus one of the most important 3rd-party middle-ware frameworks plays very well together with Azure and that it also can be used for Service Bus On-Premises with some caveats (like back-porting the transport-library).
    • An alternative, which I also tried to demonstrate with the simple sample, would be to use the MSMQ NServiceBus transport on-premises and use the AzureServiceBus Transport from NServiceBus for public cloud deployments. As long as only features
      supported on both sides are used, that might be the preferred way since with that you can fully rely in code delivered for you by NServiceBus without any changes.

Note that my attempts are meant to be a Proof-of-Concept, only. You can look at them, try them and even apply them for your solutions fully at your own risk:)

I think it was a great experience working with the team in UK on this part of a larger Proof-of-Concept (which also included e.g. Azure Service Fabric for software that
needs to be portable between on-premises and the public cloud but wants to make use of a true Platform-as-a-Service foundation).

I hope you enjoyed reading this and found it interesting and useful.

Detecting if a Virtual Machine Runs in Azure – Part 2 – Updates for Linux VMs

A few months ago I did blog-post about how-to detect whether a virtual machine runs in Azure or not. This is vital for many independent software vendors who are planning to offer their own software through the Azure Marketplace for Virtual Machines.

The main detection strategy (Windows, Ubuntu)

In the post I did explain a few tricks on detecting whether the VM runs in Azure or not for both, Windows and Linux. Still the most reliable check known as of today is to check if the DHCP option “unknown-245” is set for in the DHCP-lease options for
a virtual machine.

  • Ubuntu Linux: I’ve posted a bash script in my previous blog. I generally stated that this works for Linux all up without considering that other Linux distributions might have different configuration files for storing DHCP lease details. Hence the following script works on Ubuntu-Linux based flavors, only:
      if `grep -q unknown-245 /var/lib/dhcp/dhclient.eth0.leases`; then
          echo “Running in an Azure VM”
      fi
    

Detecting if a CentOS VM runs on Azure

My peer and colleague Arsen Vladimirskiy pointed out that on CentOS the file for DHCP leases is stored on a different location. Hence the detection strategy for the DHCP-lease option I’ve explained in my original post does not work in CentOS-based virtual machines.

For CentOS based virtual machines the DHCP lease options are indeed stored in the path /var/lib/dhclient/dhclient.leases (or in case of multiple network interfaces dhclient-eth0.leases whereas the part eth0 needs to be replaced with the networking interface device you’re going to check against).

Therefore in a default configuration with just one ethernet adapter the script needs to be updated as follows to work inside of a CentOS virtual machine:

# manually start dhclient (seems to be a workaround)
dhclient

# then check against the lease files
if `grep -q unknown-245 /var/lib/dhclient/dhclient.leases`; then
   echo "Running in Azure VM"
fi

Note: There was one weird issue I ran into when trying the approach above, hence the script starts with launching dhclient. On a fresh deployed CentOS 7 VM in Azure from the marketplace stock image dhclient is not started by default. Therefore files such as dhclient.leases or dhclient-*.leases do not exist by default under /var/lib/dhclient/.

Only after manually executing the command sudo dhclient for starting the DHCP-client the files where created successfully and the check works. Well, now someone could think that this might be related to static IP addresses – but in Azure that’s not correct since IP addresses are always assigned by the Azure DHCP server. In case you want to have static IPs you configure those through the Azure Portal or Management APIs so that the Azure DHCP server always assigns the same, static IP address to the VM in the private, virtual network. So that cannot be the reason.

A more Complete Story for detecting DHCP unknown-245 in Linux

Well, now the distributions above are very common ones but are by var not all of the supported ones on Azure. The source code for the Azure Linux Agent contains all the secrets currently valid. If you really want to be on the save side across multiple Linux distributions. A few hints in the Python-based source code are:

  • Line 99-100 do show the directories you should consider for your detection strategy
      VarLibDhcpDirectories = 
         ["/var/lib/dhclient", "/var/lib/dhcpcd", "/var/lib/dhcp"]
      EtcDhcpClientConfFiles = 
         ["/etc/dhcp/dhclient.conf", "/etc/dhcp3/dhclient.conf"]
    
  • Further down in the code starting at line 5107 there is a section that makes use of option 245 as well:
      # ... other code before
      elif option == 3 or option == 245:
          # ...
      else:
          # ...
      # ... more code goes here
    

This code has been updated to version 2.0.15 24 days before writing/publishing this post. So it should still be safe to leverage option 245 for your detection strategy. As soon as there’s something better available, I’ll definitely post another update for this blog-post!

Final Disclaimer

The approaches outlined above did work on both, Ubuntu and CentOS 7 based VMs in Resource Manager based deployments (using the new ARM-template approach introduced by the Azure teams earlier this year) at the time of publishing this post (2015-09) during my tests. When I published the original post I did test them with classic service management based VMs, of course.

Therefore and as there is still no better way introduced at the time of publishing this post, yet, the options outlined in this and my original post are still valid and eventually the best you can get so far for detecting if your VM runs inside of Microsoft Azure or not…

If you found better options don’t hesitate to contact me via my twitter feed

Azure VMs – SQL Server AlwaysOn Setup across multiple Data Centers fully automated (Classic Service Management)

Last December I started working with two of my peers, Max Knor and Igor Pagliai, with a partner in Madrid on implementing a Cross-Data Center SQL Server AlwaysOn availability group setup for a financial services solution which is supposed to be provided to 1000s of banks across the world running in Azure. Igor posted about our setup experience which we partially automated with Azure PowerShell and Windows PowerShell – see here.

At the moment the partner’s software still requires SQL Server in VMs as opposed to Azure SQL Databases because of some legacy functions they use from full SQL Server – therefore this decision.

One of the bold goals was to fully enable the partner and their customers to embrace DevOps and continuous delivery across multiple environments. For this purpose we wanted to FULLY AUTOMATE the setup of their application together with an entire cross-data-center SQL Server AlwaysOn environment as outlined in the following picture:

In December we did a one-week hackfest to start these efforts. We successfully did setup the environment, but partially automated, only. Over the past weeks we went through the final effort to fully automate the process. I’ve published the result on my github repository here:

Deployment Scripts Sample Published on my GitHub Repository

Note: Not Azure Resource Groups, yet

Since Azure Resource Manager v2 which would allow us to dramatically improve the performance and reduce the complexity of the basic Azure VM environment setup is still in Preview, we were forced to use traditional Azure Service Management.

But about 50%-60% of the efforts we have done are re-usable given the way we built up the scripts. E.g. all the database setup and custom service account setup which is primarily built on-top of Azure Custom Script VM Extensions can be re-used after the basic VM setup is completed. We are planning to create a next version of the scripts that does the fundamental setup using Azure Resource Groups because we clearly see the advantages.

Basic Architecture of the Scripts

Essentially the scripts are structured into the following main parts which you would need to touch if you want to leverage them or understand them for learning purposes as shown below:

  • Prep-ProvisionMachine.ps1 (prepare deployment machine)
    A basic script you should execute on a machine before starting first automated deployments. It installs certificates for encrypting passwords used as parameters to Custom Script VM Extensions as well as copying the basic PowerShell modules into the local PowerShell module directories so they can be found.
  • Main-ProvisionConfig.psd1 (primary configuration)
    A nice little trick by Max which is nice to provide at least some sort of declarative configuration was to build a separate script file that creates an object-tree with all the configuration data typically used for building up the cluster. It contains cluster configuration settings, node configuration settings and default subscription selection data.
  • Main-ProvisionCrossRegionAlwaysOn.ps1 (main script for automation)
    This is the main deployment script. It performs all the actions to setup the entire cross-region cluster including the following setups:
    • Setup your subscription if requested
    • Setup storage accounts if they do not exist, yet
    • Upload scripts required for setup inside of the VMs to storage
    • Setup cloud services if requested
    • Create Virtual Networks in both regions (Primary/Secondary)
    • Connect the Virtual Networks by creating VPN Gateways
    • Set the primary AD Forest VM and the Forest inside of the VM
    • Setup secondary AD DC VMs including installing AD
    • Provision SQL Server VMs
    • Setup the Internal Load Balancer for the AlwaysOn Listener
    • Configure all SQL VMs to have AlwaysOn enabled
    • Configure the Primary AlwaysOn node with the initial database setup
    • Join secondary AlwaysOn nodes and restore databases for sync
    • Configure a file-share based witness in the cluster
  • VmSetupScripts Folder
    This is essentially a folder with a series of PowerShell scripts that do perform single installation/configuration steps inside of the Virtual Machines. They are downloaded with a Custom Script VM Extension into the Virtual Machines and executed through VM Extensions, as well.

Executing the Script and Looking at the Results

Before executing the main command make sure to execute .\Prep-ProvisionMachine.ps1 to setup certificates or import the default certificate which I provide as part of the sample. If you plan to seriously use those scripts, please create your own certificate. Prep-ProvisionMachine.ps1 provides you with that capability assuming you have makecert.exe somewhere on your machines installed (please check Util-CertsPasswords for the paths in which I look for makecert.exe).

# To install a new certificate
.\Prep-ProvisionMachine.ps1

# To install a new certificate (overwriting existing ones with same Subject Names)
.\Prep-ProvisionMachine.ps1 -overwriteExistingCerts

# Or to install the sample certificate I deliver as part of the sample:
.\Prep-ProvisionMachine.ps1 -importDefaultCertificate

Then everything should be fine to execute the main script. If you don’t specify the certificate-related parameters as shown below I assume you use my sample default certificate I include in the repository to encrypt secrets pushed into VM Custom Script Extensions.

# Enter the Domain Admin Credentials
$domainCreds = Get-Credential

# Perform the main provisioning

.\Main-ProvisionCrossRegionAlwaysOn.ps1 -SetupNetwork -SetupADDCForest -SetupSecondaryADDCs -SetupSQLVMs -SetupSQLAG -UploadSetupScripts -ServiceName "mszsqlagustest" -StorageAccountNamePrimaryRegion "mszsqlagusprim" -StorageAccountNameSecondaryRegion "mszsqlagussec" -RegionPrimary "East US" -RegionSecondary "East US 2" -DomainAdminCreds $domainCreds -DomainName "msztest.local" -DomainNameShort "msztest" -Verbose

After executing a main script command such as the following, you will get 5 VMs in the primary region and 2 VMs in the secondary region acting as a manual failover. 

The following image shows several aspects in action such as the failover cluster resources which are part of the AlwaysOn availability group as well as SQL Server Management Studio accessing the AlwaysOn Availability Group Listener as well as SQL Nodes, directly. Click on the image to enlarge it and see all details.

Please note that the failover in the secondary region needs to happen MANUALLY by executing either a planned manual failover or a forced manual failover as documented on MSDN. Failover in the primary region (from the first to the second SQL Server) is configured to happen automatically.

In addition on Azure it means to take the IP cluster resource for the secondary region online which by default is offline in the cluster setup as you can see on the previous image.

Customizing the Parts you Should Customize

As you can see in the image above, the script creates sample databases which it sets up for the AlwaysOn Availability Group to be synchronized across two nodes in the main. This happens based on *.sql scripts you can add to your configuration. To customize the SQL Scripts and Databases affected, you need to perform the following steps:

  • Create *.sql scripts with T-SQL code that creates the databases you want to create as part of your AlwaysOn Availability Group.
  • Copy the *.sql Files into the VmSetupScripts directory BEFORE starting the execution of the main script. That leads to have them included into the package that gets pushed to the SQL Server VMs
  • Open up the main configuration file and customize the database list based on the databases created with your SQL scripts as well as the list of SQL Scripts that should be pushed into osql.exe/sqlcmd.exe as part of the setup process for creating the databases.
  • Also don’t forget to customize the subscription name if you plan to not override it through the script-parameters (as it happens with the example above).

The following image shows those configuration settings highlighted (in our newly released Visual Studio Code editor which also has basic support for PowerShell):


Fundamental Challenges

The main script can primarily be seen as a PowerShell workflow (we didn’t have the time to really implement it as a Workflow, but that would be a logical next step after applying Azure Resource Groups).

It creates one set of Azure VMs after another and joins them to the virtual networks it has created before. It then executes scripts on the Virtual Machines locally which are doing the setup by using Azure VM Custom Script Extensions. Although custom script extensions are cool, you have two main challenges with them for which the overall package I published provides re-usable solutions:

  • Passing “Secrets” as Parameters to VM Custom Script Extensions such as passwords or storage account keys in a more secure way as opposed to clear-text.
  • Running Scripts under a Domain User Account as part of Custom Script Extensions that require full process level access to the target VMs and Domains (which means PowerShell Remoting does not work in most cases even with CredSSP enabled … such as for Cluster setups).

For these two purposes the overall script package ships with some additional PowerShell Modules I have written, e.g. based on a blog-post from my colleague Haishi Bai here.

Running Azure VM Custom Script Extensions under a different User

Util-PowerShellRunAs.psm1 includes a function called Invoke-PoSHRunAs which allows you to run a target script with its parameters under a different user account as part of a custom script VM Extension. A basic invocation of that script looks as follows:

$scriptName = [System.IO.Path]::Combine($scriptsBaseDirectory, "Sql-Basic01-SqlBasic.ps1") 
Write-Host "Calling into $scriptName"
try {
    $arguments = "-domainNameShort $domainNameShort " + `
                 "-domainNameLong $domainNameLong " +  `
                 "-domainAdminUser $usrDom " +  `
                 "-dataDriveLetter $dataDriveLetter " +  `
                 "-dataDirectoryName $dataDirectoryName " +  `
                 "-logDirectoryName $logDirectoryName " +  `
                 "-backupDirectoryName $backupDirectoryName " 
    Invoke-PoSHRunAs -FileName $scriptName -Arguments $arguments -Credential $credsLocal -Verbose:($IsVerbosePresent) -LogPath ".\LogFiles" -NeedsToRunAsProcess
} catch {
    Write-Error $_.Exception.Message
    Write-Error $_.Exception.ItemName
    Write-Error ("Failed executing script " + $scriptName + "! Stopping Execution!")
    Exit
}

This function allows you to either run through PowerShell remoting or in a separate process. Many setup steps of the environment we setup do actually not work through PowerShell remoting because they rely on impersonation/delegation or do PowerShell Remoting on their own which imposes several limitations.

Therefore the second option this script provides is executing as a full-blown process. Since Custom Script Extensions to run as local system, it is nevertheless not as simple as just doing a Start-Process with credentials being passed in (or a System.Diagnostics.Process.Start() with different credentials). Local System does not have those permissions, unfortunately. So the work-around is to use the Windows Task Scheduler. For such cases the function performs the following actions:

  • Schedule a task in the Windows Task Scheduler with the credentials needed to run the process as.
  • Manually start the task using PowerShell cmdLets
    • (Start-ScheduledTask -TaskName $taskName)
  • Wait for the task to be finished from running
  • Look at the exit code
  • Throw an Exception if the exit code is non-zero, otherwise assume success
  • Delete the task again from the task scheduler

This “work-around” helped us to completely execute the entire setup steps successfully. We were also discussing with the engineers building the SQL AlwaysOn single-data-center Azure Resource Group template that is available for single-data-center deployments in the new Azure Portal, today. They are indeed doing the same thing, details are just a bit different.

Encrypting Secrets Passed to Custom Script VM Extensions

Sometimes we were just required to pass secret information to custom script extensions such as storage account keys. Since Azure VM Custom Script Extensions are logged very verbose, it would be a piece of cake to get to that secret information by doing a Get-AzureVM and looking at the ResourceExtensionStatusList member which contains the status and detailed call information for all VM Extensions.

Therefore we wanted to encrypt secrets as they are passed to Azure VM Extensions. The basic (yet not perfect) approach works based on some guidance from a blog post from Haishi Bai as mentioned earlier.

I’ve essentially written another PowerShell module (Util-CertsPasswords) which can perform the following actions:

  • Create a self-signed certificate as per guidance on MSDN for Azure.
  • Encrypt Passwords using such a certificate and return a base64-encoded, encrypted version.
  • Decrypt Passwords using such a certificate and return the clear-text password.

In our overall workflow all secrets including passwords and storage account keys which are passed to VM Custom Script Extensions as parameters are passed as encrypted values using this module.

Using Azure CmdLets we make sure that the certificates are published with the VM as part of our main provisioning script as per Michael Washams guidance from the Azure Product group.

Every script that gets executed as part of a custom VM Script Extension receives an encrypted password and uses the module I’ve written to decrypt it and use it for the remaining script such as follows:

#
# Import the module that allows running PowerShell scripts easily as different user
#
Import-Module .\Util-PowerShellRunAs.psm1 -Force
Import-Module .\Util-CertsPasswords.psm1 -Force

#
# Decrypt encrypted passwords using the passed certificate
#
Write-Verbose "Decrypting Password with Password Utility Module..."
$localAdminPwd = Get-DecryptedPassword -certName $certNamePwdEnc -encryptedBase64Password $localAdminPwdEnc 
$domainAdminPwd = Get-DecryptedPassword -certName $certNamePwdEnc -encryptedBase64Password $domainAdminPwdEnc 
Write-Verbose "Successfully decrypted VM Extension passed password"

The main provisioning script encrypts the passwords and secrets using that very same module before being passed into VM Custom Script Extensions as follows:

$vmExtParamStorageAccountKeyEnc = `
Get-EncryptedPassword -certName $certNameForPwdEncryption `             -passwordToEncrypt ($StorageAccountPrimaryRegionKey.Primary)

That way we at least make sure that no un-encrypted secret is visible in the Azure VM Custom Script Extension logs that can easily be retrieved with the Azure Service Management API PowerShell CmdLets.

Final Words and More…

As I said, there are lots of other re-usable parts in the package I’ve just published on my Github Repository which even can be used to apply further setup and configuration steps on VM environments which have entirely been provisioned with Azure Resource Groups and Azure Resource Manager. A few examples:

  • Execute additional Custom Script VM Extensions on running VMs.
  • Wait for Custom Script VM Extensions to complete on running VMs.
  • A ready-to-use PowerShell function that makes it easier to Remote PowerShell into provisioned VMs.

We also make use of an AzureNetworking PowerShell module published on the Technet Gallery. But note that we also made some bug-fixes in that module (such as dealing with “totally empty VNET configuration XML files”).

Generally the experience of building these ~2500 lines of PowerShell code was super-hard but a great learning experience. I am really keen to publish the follow-up post on this that demonstrates how much easier Azure Resource Group templates to make such a complex setup.

Also I do hope that we will have such a multi-data-center template in the default gallery soon since it is highly valuable for all partners and customers that do need to provide high-availability across multiple data centers using SQL Server Virtual Machines. In the meantime we will try to provide a sample based on this work above as soon as we can have time/resources for implementation.

Finally – thanks to Max Knor and Igor Pagliai – without their help we would not have achieved these goals at this level of completeness!

Detecting if a Virtual Machine Runs in Microsoft Azure (Linux & Windows) to Protect your Software when distributed via the Azure Marketplace

Our team started working more and more with software vendors we categorize as “Enablers” at a global scale. Such companies are providing building block services which can be used to build finished software services that do run in the cloud (or on-premises).

For such “Enablers” the Azure Marketplace is a key-instrument to gain visibility and traction as well as for instantiating their services in their customer’s Microsoft Azure Subscriptions.

At the moment most of the partners are working with us to deploy offerings based on templates with single or multiple Virtual Machines that do run their software. Later down the path we will also enable on-boarding of “Application Services” where customers do not have to instantiate and manage Virtual Machines, anymore.

One of the main challenges our partners do face when putting their software into Virtual Machine templates which can be instantiated and/or purchased through the Azure Marketplace is protecting their software from being operated outside of Azure since this would enable malicious people to operate the software without charging for it.

Customers have full control to VMs provisioned via the Marketplace

Since when end-customers create Virtual Machines via the Azure Marketplace they have full control of the resulting, instantiated VMs after they provisioned them, many of our partners start asking the following obvious question: How can I detect if a Virtual Machine runs in Azure so that my software can block itself from being started when not running in Azure?

Unfortunately as of today there’s no good and simple answer to that. There are various approaches out there which I would like to summarize below. I think the best possible way as of today (April 2015) is a combination of all of these approaches to make it as hard as possible running your software outside of an Azure VM.

Query for DHCP Option 245

The first option is one that originally came up by a fellow peer from our Azure support engineering team. It has been provided for Windows Virtual Machines as a PowerShell script and essentially performs the following two actions:

  1. Check if the VMBus driver from Hyper-V is active.
  2. If so, check the DHCP lease attributes for option “unknown-245”

The option “unknown-245” is an Azure-proprietary option which only gets issued by an Azure DHCP server. Since in Azure you always will get an address via DHCP (static IPs are also managed by the DHCP and with the REST management API) you will always (and in theory only) get this option as part of the DHCP lease attributes when your machine runs in Azure.

For Windows there is a ready-made PowerShell CmdLet that allows you to detect if a VM runs in Azure: https://gallery.technet.microsoft.com/scriptcenter/Detect-Windows-Azure-aed06d51

For Linux you can create a bash-script such as the following one to detect if the option unknown-245 is available to have a first indicator of whether you run in Azure or not:

if `grep -q unknown-245 /var/lib/dhcp/dhclient.eth0.leases`; then
    echo “Running in an Azure VM”
fi

This is currently considered to be the most used and simplest approach to detect if you’re running on Azure that is “good enough”. But for some partners it is understandably not enough, yet…

Use the Azure Agent as Detection-Strategy

On Linux VMs in specific, another approach is reading the configuration from the Microsoft Azure Agent which is always installed on a Linux VM and try to reach it’s ping-counterpart on the host-agent side. If a VM does not run in Azure, trying to reach the host-agent end-point would always result in a timeout. Here’s a sample script for doing so:

curl –connect-timeout 1 `grep FullConfig /var/lib/waagent/GoalState.1.xml | perl -pe ‘s/<.?FullConfig>//g; s/\s//g’` && echo azure || echo no-azure

On Windows VMs there’s only an agent available when you explicitly select the VM Agent for VM Extensions to be installed. Some partners are checking if that agent is available and explicitly document for their customers that they MUST install the VM Agent when provisioning a Marketplace Image from the Azure Marketplace for their software to work correctly.

Checking your external IP Address

If you Virtual Machine has a public endpoint attached, you can also verify the public IP address your VM is using when trying to access other services and compare it against the IP address ranges that are reserved for Azure data centers.

The Azure data center IP address ranges can be downloaded from the Microsoft Download Center here: http://www.microsoft.com/en-us/download/details.aspx?id=41653

Valuable services to get your publicly visible IP address are services such as the http://ifconfig.me/ip which can also be used in a PowerShell script or bash-script easily:

function Get-ExternalIP {
    (Invoke-WebRequest ifconfig.me/ip).Content
}
Get-ExternalIP

A more complex script can then even automatically download the Azure IP ranges from the Microsoft Download center via the direct URL (which are stored as XML) and try to check if the ranges match.

Leveraging the Azure REST Management API or CLI-interfaces

Finally you could also ship your VM with either Azure PowerShell CmdLets or the Azure Cross-Platform CLI shipped and query details about your VM through the rest management API.

But – that has one big step you need to take upfront: this requires you to force the user somehow to provide credentials or a management certificate that gives the VM access to the customer’s subscription in which the VM is deployed so that you can query the details about the VM (which belongs to the customer’s subscription and is owned by the customer… and not you as the provider/creator of the marketplace VM template and offering).

To get this done you need to do e.g. one of the following things:

  • Write a very explicit documentation for your customers that explains what they need to do after they provisioned the VM from the Azure Marketplace into their subscription before they can use your software in that VM or VMs.
  • Or e.g. write a little provisioning web application which is shipped as part of the VM image that the user needs to browse to immediately after provisioning the VM from the marketplace to enter the remaining details that enable your software or “bootstrapping-scripts” to use the Azure Service Management API or CLI to query additional information about your VM and use that for detecting if you run in Azure or not (e.g. query your public and internal IP and compare with what your VM reports etc.).Of course you need to make sure that this “provisioning-app” you need to build is only active in the provisioned instance after the initial creation from the marketplace to avoid any kind of security issues.

At some point in time in the near future, the Azure Marketplace service will enable publishers of images to require the user providing additional details through the Azure portal as part of the provisioning/creation-process, already. But as long as that’s not possible you need to look at approaches such as the ones I’ve outlined above.

Final Words

The approaches outlined above are all used by publishers of VM templates in the marketplace today and they work. I know they are not optimal, but we also know that the product group is aware of the challenges and will work on better solutions in the future. For now, the approaches I outlined above are easy and pragmatic ways that at least give you some level of guarantee for detecting of whether a VM and your software runs in Microsoft Azure (public cloud) or not…