Azure App Services with Private Link

In one of my videos on my YouTube channel, I discuss Azure App Services with Private Link. The video describes how it works and provides an example of deploying the infrastructure with Bicep. The Bicep templates are on GitHub.

If you want to jump straight to the video, here it is:

In the rest of this blog post, I provide some more background information on the different pieces of the solution.

Azure App Service

Azure App Service is a great way to host web application and APIs on Azure. It’s PaaS (platform as a service), so you do not have to deal with the underlying Windows or Linux servers as they are managed by the platform. I often see AKS (Azure Kubernetes Service) implementations to host just a couple of web APIs and web apps. In most cases, that is overkill and you still have to deal with Kubernetes upgrades, node patching or image replacements, draining and rebooting the nodes, etc… And then I did not even discuss controlling ingress and egress traffic. Even if you standardize on packaging your app in a container, Azure App Service will gladly accept the container and serve it for you.

By default, Azure App Service gives you a public IP address and FQDN (Fully Qualified Domain Name) to reach your app securely over the Internet. The default name ends with azurewebsites.net but you can easily add custom domains and certificates.

Things get a bit more complicated when you want a private IP address for your app, reachable from Azure virtual networks and on-premises networks. One solution is to use an App Service Environment. It provides a fully isolated and dedicated environment to run App Service apps such as web apps and APIs, Docker containers and Functions. You can create an internal ASE which results in an Internal Load Balancer in front of your apps that is configured in a subnet of your choice. There is no need to configure Private Endpoints to make use of Private Link. This is often called native virtual network integration.

At the network level, an App Service Environment v2, works as follows:

External ASE
ASE networking (from Microsoft website)

Looking at the above diagram, an ILB ASE (but also an External ASE) also makes it easy to connect to back-end systems such as on-premises databases. The outbound connection to internal resources originates from an IP in the chosen integration subnet.

The downside to ASE is that its isolated instances (I1, I2, I3) are rather expensive. It also takes a long time to provision an ASE but that is less of an issue. In reality though , I would like to see App Service Environments go away and replaced by “regular” App Services with toggles that give you the options you require. You would just deploy App Services and set the options you require. In any case, native virtual network integration should not depend on dedicated or shared compute. One can only dream right? 😉

Note: App Service Environment v3, in preview at the time of this writing, provides a simplified deployment experience and also costs less. See App Service Environment v3 public preview – Azure App Service

As an alternative to an ASE for a private app, consider a non-ASE App Service that, in production, uses Premium V2 or V3 instances. The question then becomes: “How do you get a private IP address?” That’s where Private Link comes in…

Azure Private Link with App Service

Azure Private Link provides connectivity to Azure services (such as App Service) via a Private Endpoint. The Private Endpoint creates a virtual network interface card (NIC) on a subnet of your choice. Connections to the NICs IP address end up at the Private Link service the Private Endpoint is connected to. Below is an example with Azure SQL Database where one Private Endpoint is mapped, via Azure Private Link, to one database. The other databases are not reachable via the endpoint.

Private Endpoint connected to Azure SQL Database (PaaS) via Private Link (source: Microsoft website)

To create a regular App Service that is accessible via a private IP, we can do the same thing:

  • create a private endpoint in the subnet of your choice
  • connect the private endpoint to your App Service using Private Link

Both actions can be performed at the same time from the portal. In the Networking section of your App Service, click Configure your private endpoint connections. You will see the following screen:

Private Endpoint connection of App Service

Now click Add to create the Private Endpoint:

Creating the private endpoint

The above creates the private endpoint in the default subnet of the selected VNET. When the creation is finished, the private endpoint will be connected to App Service and automatically approved. There are scenarios, such as connecting private endpoints from other tenants, that require you to approve the connection first:

Automatically approved connection

When you click on the private endpoint, you will see the subnet and NIC that was created:

Private Endpoint

From the above, you can click the link to the network interface (NIC):

Network interface created by the private endpoint

Note that when your delete the Private Endpoint, the interface gets deleted as well.

Great! Now we have an IP address that we can use to reach the App Service. If you use the default name of the web app, in my case https://web-geba.azurewebsites.net, you will get:

Oops, no access on the public name (resolves to public IP)

Indeed, when you enable Private Link on App Service, you cannot access the website using its public IP. To solve this, you will need to do something at the DNS level. For the default domain, azurewebsites.net, it is recommended to use Azure Private DNS. During the creation of my Private Endpoint, I turned on that feature which resulted in:

Private DNS Zone for privatelink.azurewebsites.net

You might wonder why this is a private DNS zone for privatelink.azurewebsites.net? From the moment you enable private link on your web app, Microsoft modifies the response to the DNS query for the public name of your app. For example, if the app is web-geba.azurewebsites.net and you query DNS for that name, it will respond with a CNAME of web-geba.privatelink.azurewebsites.net. If that cannot be resolved, you will still get the public IP but that will result in a 403.

In my case, as long as the DNS servers I use can resolve web-geba.privatelink.azurewebsites.net and I can connect to 10.240.0.4, I am good to go. Note however that the DNS story, including Private DNS and your own DNS servers, is a bit more complex that just checking a box! However, that is not the focus of this blogpost so moving on… 😉

Note: you still need to connect to the website using https://web-geba.azurewebsites.net in your browser

Outbound connections to internal resources

One of the features of App Service Environments, is the ability to connect to back-end systems in Azure VNETs or on-premises. That is the result of native VNET integration.

When you enable Private Link on a regular App Service, you do not get that. Private Link only enables private inbound connectivity but does nothing for outbound. You will need to configure something else to make outbound connections from the Web App to resources such as internal SQL Servers work.

In the network configuration of you App Service, there is another option for outbound connectivity to internal resources – VNet integration.

VNET Integration

In the Networking section of App Service, find the VNet integration section and click Click here to configure. From there, you can add a VNet to integrate with. You will need to select a subnet in that VNet for this integration to work:

Outbound connectivity for App Service to Azure VNets

There are quite some things to know when it comes to VNet integration for App Service so be sure to check the docs.

Private Link with Azure Front Door

Often, a web app is made private because you want to put a Web Application Firewall (WAF) in front of the app. Typically, that goal is achieved by putting Azure Application Gateway (AG) with WAF in front of an internal App Services Environment. As as alternative to AG, you can also use virtual appliances such as Barracuda WAF for Azure. This works because the App Services Environment is a first-class citizen of your Azure virtual network.

There are multiple ways to put a WAF in front of a (non-ASE) App Service. You can use Front Door with the App Service as the origin, as long as you restrict direct access to the origin. To that end, App Services support access restrictions.

With Azure Front Door Premium, in preview at the time of this writing (June 2021), you can use Private Link as well. In that case, Azure Front Door creates a private endpoint. You cannot control or see that private endpoint because it is managed by Front Door. Because the private endpoint is not in your tenant, you will need to approve the connection from the private endpoint to your App Service. You can do that in multiple ways. One way is Private Link Center Pending Connections:

Pending Connections

If you check the video at the top of this page, this is shown here.

Conclusion

The combination of Azure networking with App Services Environments (ASE) and “regular” App Services (non-ASE) can be pretty confusing. You have native network integration for ASE, private access with private link and private endpoints for non-ASE, private DNS for private link domains, virtual network service endpoints, VNet outbound configuration for non-ASE etc… Most of the time, when I am asked for the easiest and most cost-effective option for a private web app in PaaS, I go for a regular non-ASE App Service and use Private Link to make the app accessible from the internal network.

Adding Authentication and Authorization to an Azure Static Web App

In a previous post, we created a static web app that retrieves documents from Cosmos DB via an Azure Function. The Azure Function got deployed automatically and runs off the same domain as your app. In essence, that frees you from having to setup Azure Functions separately and configuring CORS in the process.

Instead of allowing anonymous users to call the api at https://yourwebapp/api/device, I only want to allow specific users to do so. In this post, we will explore how that works.

You can find the source code of the static web app and the API on GitHub: https://github.com/gbaeke/az-static-web-app.

More into video tutorials? Then check out the video below. I recommend 1.2x speed! 😉

Full version about creating the app and protecting the API

Create a routes.json

To define the protected routes, you need routes.json in the root of your project:

routes.json to protect /api/*

The routes.json file serves multiple purposes. Check out how it works here. In my case, I just want to protect the /api/* routes and allow the Authenticated users role. The Authenticated role is a built-in role but you should create custom roles to protect sensitive data (more info near the end of this post). For our purposes, the platform error override is not needed and be removed. These overrides are useful though as they allow you to catch errors and act accordingly.

Push the above change to your repository for routes.json to go in effect. Once you do, access to /api/* requires authentication. Without it, you will get a 401 Unauthorized error. To fix that, invite your users and define roles.

Inviting Users

In Role Management, you can invite individual users to your app:

User gbaeke (via GitHub) user identity added

Just click Invite and fill in the blanks. Inviting a user results in an invitation link you should send the user. Below is an example for my Twitter account:

Let’s invite myself via my Twitter account

When I go to the invite link, I can authorize the app:

Authorizing Static Web Apps to access my account

After this, you will also get a Consent screen:

Granting Consent (users can always remove their data later; yeah right 😉)

When consent is given, the application will open with authentication. I added some code to the HTML page to display when the user is authenticated. The user name can be retrieved with a call to .auth/me (see later).

App with Twitter handle shown

In the Azure Portal, the Twitter account is now shown as well.

User added to roles of the web app

Note: anyone can actually authenticate to your app; you do not have to invite them; you invite users only when you want to assign them custom roles

Simple authentication code

The HTML code in index.html contains some links to login and logout:

  • To login: a link to /.auth/login/github
  • To logout: a link to /.auth/logout

Microsoft provides these paths under /.auth automatically to support the different authentication scenarios. In my case, I only have a GitHub login. To support Twitter or Facebook logins, I would need to provide some extra logic for the user to choose the provider.

In the HTML, the buttons are shown/hidden depending on the existence of user.UserDetails. The user information is retrieved via a call to the system-provided /.auth/me with the code below that uses fetch:

async  getUser() {
     const response = await fetch("/.auth/me");
     const payload = await response.json();
     const { clientPrincipal } = payload;
     this.user = clientPrincipal;

user.UserDetails is just the username on the platform: gbaeke on GitHub, geertbaeke on Twitter, etc…

The combination of the routes.json file that protects /api/* and the authentication logic above results in the correct retrieval of the Cosmos DB documents. Note that when you are not authorized, the list is just empty with a 401 error in the console. In reality, you should catch the error and ask the user to authenticate.

One way of doing so is redirecting to a login page. Just add logic to routes.json that serves the path you want to use when the errorType is Unauthenticated as shown below:

"platformErrorOverrides": [
    {
      "errorType": "NotFound",
      "serve": "/custom-404.html"
    },
    {
      "errorType": "Unauthenticated",
      "serve": "/login"
    }
  ]

The danger of the Authenticated role

Above, we used the Authenticated role to provide access to the /api/* routes. That is actually not a good idea once you realize that non-invited users can authenticate to your app as well. As a general rule: always use a custom role to allow access to sensitive resources. Below, I changed the role in routes.json to reader. Now you can invite users and set their role to reader to make sure that only invited users can access the API!

"routes": [
      {
        "route": "/api/*",
        "allowedRoles": ["reader"]
      }

      
    ]

Below you can clearly see the effect of this. I removed GitHub user gbaeke from the list of users but I can still authenticate with the account. Because I am missing the reader role, the drop down list is not populated and a 401 error is shown:

Authenticated but not in the reader role

Conclusion

In this post, we looked at adding authentication and authorization to protect calls to our Azure Functions API. Azure Static Web Apps tries to make that process as easy as possible and we all now how difficult authentication and authorization can be in reality! And remember: protect sensitive API calls with custom roles instead of the built-in Authenticated role.

First Look at Azure Static Web Apps

Note: part 2 looks at the authentication and authorization part.

At Build 2020, Microsoft announced Azure Static Web Apps, a new way to host static web apps on Azure. In the past, static web apps, which are just a combination of HTML, JavaScript and CSS, could be hosted in a Storage Account or a regular Azure Web App.

When you compare Azure Static Web Apps with the Storage Account approach, you will notice there are many more features. Some of those features are listed below (also check the docs):

  • GitHub integration: GitHub actions are configured for you to easily deploy your app from your GitHub repository to Azure Static Web Apps
  • Integrated API support: APIs are provided by Azure Functions with an HTTP Trigger
  • Authentication support for Azure Active Directory, GitHub and other providers
  • Authorization role definitions via the portal and a roles.json file in your repository
  • Staging versions based on a pull request

It all works together as shown below:

SWAdiagram.png
Azure Static Web Apps (from https://techcommunity.microsoft.com/t5/azure-app-service/introducing-app-service-static-web-apps/ba-p/1394451)

As a Netlify user, this type of functionality is not new to me. Next to static site hosting, they also provide serverless functions, identity etc…

If you are more into video tutorials…

Creating the app and protecting calls to the API

Let’s check out an example to see how it works on Azure…

GitHub repository

The GitHub repo I used is over at https://github.com/gbaeke/az-static-web-app. You will already see the .github/workflows folder that contains the .yml file that defines the GitHub Actions. That folder will be created for you when you create the Azure Static Web App.

The static web app in this case is a simple index.html that contains HTML, JavaScript and some styling. Vue.js is used as well. When you are authenticated, the application reads a list of devices from Cosmos DB. When you select a device, the application connects to a socket.io server, waiting for messages from the chosen device. The backend for the messages come from Redis. Note that the socket.io server and Redis configuration are not described in this post. Here’s a screenshot from the app with a message from device01. User gbaeke is authenticated via GitHub. When authenticated, the device list is populated. When you log out, the device list is empty. There’s no error checking here so when the device list cannot be populated, you will see a 404 error in the console. 😉

Azure Static Web App in action

Note: Azure Static Web Apps provides a valid certificate for your app, whether it uses a custom domain or not; in the above screenshot, Not secure is shown because the application connects to the socket.io server over HTTP and Mixed Content is allowed; that is easy to fix with SSL for the socket.io server but I chose to not configure that

The API

Although API is probably too big a word for it, the devices drop down list obtains its data from Cosmos DB, via an Azure Function. It was added from Visual Studio Code as follows:

  • add the api folder to your project
  • add a new Function Project and choose the api folder: simply use F1 in Visual Studio Code and choose Azure Functions: Create New Project… You will be asked for the folder. Choose api.
  • modify the code of the Function App to request data from Cosmos DB

To add an Azure Function in Visual Studio Code, make sure you install the Azure Functions extension and the Azure Function Core Tools. I installed the Linux version of Core Tools in WSL 2.

Adding the function (JavaScript; HTTP Trigger, anonymous, name of GetDevice) should result in the following structure:

Function app as part of the static web app (api folder)

Next, I modified function.json to include a Cosmos DB input next to the existing HTTP input and output:

{
  "bindings": [
    {
      "authLevel": "anonymous",
      "type": "httpTrigger",
      "direction": "in",
      "name": "req",
      "methods": [
        "get",
        "post"
      ],
      "route": "device"
    },
    {
      "type": "http",
      "direction": "out",
      "name": "res"
    },
    {
      "name": "devices",
      "type": "cosmosDB",
      "direction": "in",
      "databaseName": "geba",
      "collectionName": "devices",
      "sqlQuery": "SELECT c.id, c.room FROM c",
      "connectionStringSetting": "CosmosDBConnection"    
    }
  ]
}

In my case, I have a Cosmos DB database geba with a devices collection. Device documents contain an id and room field which simply get selected with the query: SELECT c.id, c.room FROM c.

Note: with route set to device, the API will need to be called with /api/device instead of /api/GetDevice.

The actual function in index.js is kept as simple as possible:

module.exports = async function (context, req) {
    context.log('Send devices from Cosmos');
  
    context.res = {
        // status: 200, /* Defaults to 200 */
        body: context.bindings.devices
    };
    
};

Yes, the above code is all that is required to retrieve the JSON output of the Cosmos DB query and set is as the HTTP response.

Note that local.settings.json contains the Cosmos DB connection string in CosmosDBConnection:

{
  "IsEncrypted": false,
  "Values": {
    "AzureWebJobsStorage": "",
    "FUNCTIONS_WORKER_RUNTIME": "node",
    "CosmosDBConnection": "AccountEndpoint=https://geba-cosmos.documents.a...;"
  }
}

You will have to make sure the Cosmos DB connection string is made known to Azure Static Web App later. During local testing, local.settings.json is used to retrieve it. local.settings.json is automatically added to .gitignore to not push it to the remote repository.

Local Testing

We can test the app locally with the Live Server extension. But first, modify .vscode/settings.json and add a proxy for your api:

"liveServer.settings.proxy": {
        "enable": true,
        "baseUri": "/api",
        "proxyUri": "http://172.28.242.32:7071/api"
    }

With the above setting, a call to /api via Live Server will be proxied to Azure Functions on your local machine. Note that the IP address refers to the IP address of WSL 2 on my Windows 10 machine. Find it by running ifconfig in WSL 2.

Before we can test the application locally, start your function app by pressing F5. You should see:

Function App started locally

Now go to index.html, right click and select Open with Live Server. The populated list of devices shows that the query to Cosmos DB works and that the API is working locally:

Test the static web app and API locally

Notes on using WSL 2:

  • for some reason, http://localhost:5500/index.html (Live Server running in WSL 2) did not work from the Windows session although it should; in the screenshot above, you see I replaced localhost with the IP address of WSL 2
  • time skew can be an issue with WSL 2; if you get an error during the Cosmos DB query of authorization token is not valid at the current time, perform a time sync with ntpdate time.windows.com from your WSL 2 session

Deploy the Static Web App

Create a new Static Web App in the portal. The first screen will be similar to the one below:

Static Web App wizard first screen

You will need to authenticate to GitHub and choose your repository and branch as shown above. Click Next. Fill in the Build step as follows:

Static Web App wizard second screen

Our app will indeed run off the root. We are not using a framework that outputs a build to a folder like dist so you can leave the artifact location blank. We are just serving index.html off the root.

Complete the steps for the website to be created. You GitHub Action will be created and run for the first time. You can easily check the GitHub Action runs from the Overview screen:

Checking the GitHub Action runs

Here’s an example of a GitHub action run:

A GitHub Action run

When the GitHub Action is finished, your website will be available on a URL provided by Azure Static Web Apps. In my case: https://polite-cliff-01b6ab303.azurestaticapps.net.

To make sure the connection to Cosmos DB works, add an Application Setting via Configuration:

Adding the Cosmos DB connection string

The Function App that previously obtained the Cosmos DB connection string from local.settings.json can now retrieve the value from Application Settings. Note that you can also change these settings via Azure CLI.

Conclusion

In this post, we created a simple web app in combination with an function app that serves as the API. You can easily create and test the web app and function app locally with the help of Live Server and a Live Server proxy. Setting up the web app is easy via the Azure Portal, which also creates a GitHub Action that takes care of deployment for you. In a next post, we will take a look at enabling authentication via the GitHub identity provider and only allowing authorized users to retrieve the list of devices.