While serverless computing has a compelling set of benefits, it is no silver bullet. To architect effective solutions, you need to be aware of its strengths and weaknesses, which are shown in the following table:

We will discuss each of its main benefits in detail here.

Firstly, a major strength of serverless computing is its speed of development. The developer doesn't have to concern oneself with, or write any of the code for, the underlying architecture. The process of listening for HTTP messages (or other data inputs/events) and routing is done entirely by the cloud provider, with only some configuration provided by the developer. This allows a developer to simply write code that implements business logic, and deploy it straight to Azure. Each function can then be tested independently.

Secondly, serverless computing has automatic scaling by default. The routing system in a serverless service like Azure Functions will detect how many requests are coming in and deploy the serverless code to more servers when necessary. It will also reduce the number of servers all the way down to zero, if necessary. This is particularly useful if you are running an advertisement or a publicity stunt of some kind—half-time advertisements in the Premier League final, for example. Your backend can instantly scale to handle the massive influx of new users for this brief period of time, before returning to normal very quickly. It's also useful for separating out what would usually be a large, monolithic backend. Some of the parts of the backend will probably be used a lot more than others, but usually, you have to scale the whole thing manually to keep the most-utilized functions responsive. If each function is separated, automatically scaling it allows the developer to identify and optimize the most-used functions.

Continuing on the subject of being able to scale down to zero servers, the cost of serverless code is very flexible. Due to auto-scaling, providers are able to charge according to resource usage. This means that you only pay according to usage, which has benefits for all developers, but for small businesses most of all. If your business has lots of customers one month, you can scale accordingly and pay your higher bill with the higher revenue you have made. If you have a quieter month, your serverless computing bill will be proportionally lower for that lower revenue period.

Finally, serverless code requires very little active management. An application running on a shared server will usually require a significant amount of monitoring and management, for both the server itself and its resource allocation at any given moment. Containerized code, or code running on virtual machines, is better, but still requires either container management software or a person to monitor the usage and then scale down or scale up appropriately, even if the server itself no longer requires active management. Serverless code has no server visible to the developer and will scale according to demand, meaning that it requires monitoring only for exceptions or security issues, with no real active management to keep a normal service running.

Serverless code does have some weaknesses which are described here.

The first weakness is warmup latency. Although the traffic managers of various platforms (AWS, Azure, Google Cloud, and so on) are very good at responding to demand when there are already instances of serverless code running, it's a very different problem when there are no instances running at all. The traffic managers need to detect the message, allocate a server, and deploy the code to it before running it. This is necessarily slower than having a constantly running container or server. One way to combat this is to keep your code small and simple, as the biggest slowdown in this process can be transferring large code files.

Secondly, there is the issue of vendor lock-in. Each cloud provider implements its serverless service differently, so serverless code written for Azure is difficult to port over to AWS. If prices spike heavily in the future, then you will be locked in to that provider for your serverless architecture.

There's also the issue of languages. JavaScript is the only language that is universally available, with patchy service across providers for other languages, like C# and Java. There is a solution to this, however; it is called the serverless framework. This is a framework that you can use to write simple HTTP-triggered functions, which can then be deployed to all of the major cloud providers. Unfortunately, this means that you will miss out on a lot of the best features of Azure Functions, because their real power comes from deep integration with other Azure services.

Finally, there is the issue of a lack of low-level control. If you are writing a low latency trading platform, then you may be used to accessing networking ports directly, manually allocating memory, and executing some commands using processor-specific code. Your own application might require similar low-level access, and this isn't possible in serverless computing. One thing to bear in mind, however, is that it's possible to have part of the application running on a server that you have low-level access to, and background parts of it running in a serverless function.

If an Azure Function isn't executed for a while, the function stops being deployed and the server gets reallocated to other work. When the next request comes in the function needs to deploy the code, warm up the server and execute the code, so it's slower. Inevitably, this leads to latency when the functions are triggered again, making serverless computing somewhat unsuitable for use cases that demand continuous low-latency. Also, by its very nature, serverless computing prevents you from accessing low level commands and the performance benefits that they can give. It's important to emphasize that this doesn't mean serverless computing is unbearably slow; it just means that applications that demand the utmost performance are unlikely to be suitable for serverless computing.

Overall, there is a clear benefit to using serverless computing, and particularly, Azure Serverless, especially if you use some of the tips detailed in the weaknesses section. The benefits are strong for both the developer and the business.

The serverless framework (https://serverless.com/) can help with vendor lock-in by making your serverless functions cross-cloud.

In this section, we will create an Azure Function in Visual Studio, debug it locally, and deploy it to an Azure cloud instance. While doing this, we will cover material about the basics of serverless runtime and the high-level benefits and disadvantages of serverless computing.

The core of all serverless products is to get straight into development with minimal setup time, so that's what we'll do in this subtopic. You should have a serverless function in the cloud at the end of this topic, and once you've learned the ropes, you could create and deploy one in a few minutes.

An Azure Function can be written in several languages. At the time of writing, there are three languages with full support: C#, JavaScript, and F#. Generally, the most popular languages are C# and JavaScript. Java is a language in preview, but, being in preview, it is not ready for production yet, so it will not be covered.

Visual Studio 2017 has a comprehensive suite of Azure tools, including Azure Function development. While it's possible to use a combination of any other IDE and the command-line interface to develop Azure Functions, the best tooling will always arrive on Visual Studio first.

Before you begin, confirm that you have Visual Studio 2017 version 15.4 installed; if not, download and install it. To do so, perform the following steps:

1. Open the Visual Studio Installer, which will show you the version of Visual Studio that you have installed and allow you to select the Azure Workflow and install it, if it is missing, then update Visual Studio, if required, to latest version:

2. Click on Modify, select the Azure development workload, and click on Modify again:

Now, we'll create a new Azure Function as a part of our serverless architecture that listens to HTTP requests to a certain address. It will listen to HTTP requests to a certain address as its trigger. Let's begin by implementing the following steps:

3. You now have a solution with two files in it: host.json and local.settings.json. The local.settings.json file is used solely for local development, where it stores all of the details on connections to other Azure services.
   

4. Now, right-click on the project and select Add New Item. Once again, choose the Azure Function template:

5. On the next screen, select Http trigger with parameters and Access rights should be set to Anonymous. Right-click on your solution and select Enable NuGet Package Restore:

6. You will now have a C# file called PostTransactions.cs. It consists of a single method, Run, with an awful lot in the method signature: an attribute and an annotation. Some of this will be familiar to you if you are an experienced C# developer, and it is important to understand this signature.

Configuration as code is an important modern development practice. Rather than having servers reconfigured or configured manually by developers before code is deployed to them, configuration as code dictates that all of the configuration required to deploy an application to production be included in the source code.

This allows for variable replacement by your build/release agent, as you will (understandably) want slightly different settings, depending on your environment.

Azure Functions implement this principle, with a configuration split between the host.json file for app-wide configurations and app settings, and the Run method signature for individual functions. Therefore, you can deploy an Azure Function to production with only the code that you find in the GitHub repository:

Outcome

You created an Azure Function, understood the roles the different files play, and learned about configuration as code.

The FunctionName annotation defines the name of the function within the Function App.

This can be used for triggering your function, or it can be kept separate. The first parameter is an HttpRequestMessage object with an HttpTrigger attribute. This is what varies when you choose different triggers (for example, a timer trigger will have an object with a TimerTrigger attribute).

This attribute has several arguments. The first is the authorization level. Do you remember setting this when you created the function? It was called Access rights in the template. This defines the level of authorization that the function will demand of HTTP requests. The five levels are shown in the following table:

The next parameters are GET and POST, which define the HTTP verbs that will activate the function. Generally, from a microservices architecture point of view, you should only have one verb, to prevent you from having to do bug-prone switching logic inside of the function. You can simply create four separate functions if you want GET, POST, PUT, and DELETE on an artifact.

Finally, there is a string assigned to the property route. This is the only bit of routing logic that the function itself can see, and it simply defies the subpath from the Function App. It accepts WebAPI syntax, which you can see in the curly braces, / {name}. This will assign any text that appears where the curly braces are to a parameter called name.

This completes the HttpTrigger object. Other types of triggers, which we will touch on later, have different objects and different constructors. The three parameters left in the method signature are an HttpRequestMessage object, which allows you to access the HttpRequestMessage that triggered the function; a string parameter called name, which is what the string in the curly braces in the path will get bound to; and a TraceWriter for logging.

The current logic of the Function App can be seen in the following example, and you should see that it will take whatever name is put into it and send back an HTTP response saying Hello to that name. We will test this out in the next subtopic.

You now have a working Azure Function that can be deployed to Azure or run locally. We will first host and debug the function locally, to show the development cycle in action. At the moment, its logic is very basic and not particularly useful, so we will be developing this to store some data in a cloud-hosted database over the rest of the chapter.

In this section, we'll run an Azure Function locally and debug it. We'll be developing new functions and test the functionality before deploying to public cloud. And to ensure that it happens correctly, we'll require the single function created directly from the HTTP trigger with parameters template.

Currently, your machine does not have the correct runtime to run an Azure Function, so we need to download it:

1. Click the Play button in Visual Studio, and a dialog box should ask you if you want to download Azure Functions Core Tools—click on Yes. A Windows CMD window will open, with the lightning bolt logo of Azure Functions:

2. Open up Postman app and copy and paste the endpoint into it, selecting either a POST or GET verb.

3. Create a debug point in the Run method by clicking in the margin to the left of the code:

4. Use Postman to send the request:

5. You are now able to use standard Visual Studio debugging features and inspect the different objects as shown in the following screenshot:

6. Set your verb to POST , and add a message in the payload. See if you can find the verb in the HttpRequestMessage object in debug mode. It should be in the method property.
   

In this activity, we will add a JSON payload to the request and write code to parse that message into a C# object.

Prerequisites

You will require a function created from the HTTP trigger with the parameters template.

Scenario

You are creating the start of a personal finance application that allows users to add their own transactions, integrate with other applications and perhaps allow their credit card to directly log transactions. It will be able to scale elastically to any number of users, saving us money when we don't have any users.

Aim

Parse a JSON payload into a C# object, starting your RESTful API.

Steps for Completion

1. Change the Route to transactions.
2. Remove the get verb. Remove the String parameter called name:
   

3. Add the Newtonsoft.json package, if it isn't already present. You can do this by right-clicking the Solution | Manage NuGet packages | Browse | Newtonsoft.Json.

4. Right-click on the project and add a folder called Models, then add a C# class called Transaction. Add two properties to this class: a DateTime property called ExecutionTime, and a Decimal property called Amount:
   

5. Use JsonConvert.DeserializeObject<Transaction>(message).Result() to deserialize the HttpRequestMessage into an instantiation of this class. To do that, you need to import the Models namespace and Newtonsoft.Json. This will parse the JSON payload and use the Amount property to file the corresponding property on the Transaction object:
   

6. Change the return message to use a property of the new Transaction object, for example, You entered a transaction of £47.32! Go to Postman and open the Body tab and select raw.

7. Enter in the following JSON object:
   

{
  Amount: 47.32,
  ExecutionTime: "2018-01-01T09:00:00Z"
} 

8. Run locally to test. Make sure you change the endpoint to /transactions in Postman.
   

Outcome

You have learned how to access the HttpRequestMessage, and you will have a function that can read a JSON message and turn it into a C# object.

During this subtopic, you debugged an Azure Function. Visual Studio only allows this through downloading azure-functions-core-tools. Unfortunately, it doesn't make it available on the general command line—only through command windows started in Visual Studio. If you want to use it independently, then you have to download it using npm. If you need to download azure-functions-core-tools separately, you can use npm to get it—npm install -g azure-functions-core-tools for version 1 (fully supported) and npm install -g azure-functions-core-tools@core for version 2 (beta). We will go into the differences in versions later in this chapter. You can then use the debug setup to set Visual Studio to call an external program with the command func host start when you click on the Debug button.

This package is a lot more than just a debug environment, however; it actually has a CLI for everything you could possibly need in Azure Function development. Open up a command window (in Visual Studio, if you haven't downloaded it independently) and type func help; you should see a full list of everything the CLI can do. Notable commands are func host start, which starts the local debug environment, and func azure {functionappname} fetch-app-settings, which lets you download the app settings of a function deployed to Azure so that you can test integration locally, as well. These need to be run in the same folder as the host.json file.

An Azure Function is obviously geared towards being hosted on the Azure cloud, rather than locally or on your own computer. Visual Studio comes with a complete suite of tools to deploy and manage Azure services, and this includes full support for Azure Functions. The azure-functions-core-tools CLI that you downloaded to provide a local debug environment also has a set of tools for interacting with Azure Functions in the cloud, if you prefer CLIs. We will focus on using Visual Studio, but we will mention some of the equivalent methods in the CLI.

In this section, we'll deploy our first function to the public cloud, and learn how to call it. We'll be going live with our Azure Function, starting to create our serverless architecture. And to ensure that it happens correctly, we'll need a function project and a valid Azure subscription. Let's begin by implementing the following steps:

1. Right-click on your project and select Publish.... Now select Azure Function App | Create New as shown in the following screenshot:
   

2. Enter a memorable name and create a resource group and a consumption app service plan to match the following:
   

3. Click on the Publish button to publish your function.
4. Open a browser, navigate to http://portal.azure.com, and find your function. You can use the search bar and search the name of your function. Click on your Function App, then click on the function name. Click the Get function URL in the upper-right corner paste the address of your function into Postman, and test it. Please bear in mind that if you have a paid subscription, these executions will cost a small amount of money—you are only charged for the compute resources that you actually use. On a free account, you get a million executions for free.

Outcome

You now have a fully deployed and working Azure Function in the cloud.

So, we successfully created an Azure Function, ran it on a local environment, and deployed it to the cloud.

We'll cover the two runtimes in more depth now, but the key takeaway is that there is a fully-supported, Windows-based version, and also a beta cross-platform version. Microsoft are pushing towards the cross-platform version.

Version 1 is based on Azure WebJobs, which is written in C# and .NET. Azure WebJobs is essentially a service with a set of adapters for Azure products. These adapters usually go a lot deeper than a standard API. When code is deployed into it, Azure WebJobs reads host.json to create an environment with the same context every time. When you downloaded azure-functions-core-tools, it contained a full production Azure Functions environment, exactly the same as the ones used in Azure. That is why we can only develop on Windows machines for version 1.

Version 2 is based on the new Azure WebJobs, written in C# and .NET Core. It can run on any server that supports .NET Core, which includes all major distributions of Linux, as well as Windows.

The containers that Azure Functions run in are inherently short-lived, with a maximum execution time of 10 minutes allowed (this was set in host.json, as seen earlier), but it is generally advisable to execute HTTP requests in two minutes. This short timeframe is part of what allows for the flexibility of Azure Functions, so if you need longer execution times but still want most of the convenience of serverless computing, then look at using Azure WebJobs SDK, running on a VM or physical machine that you control, or Azure Data Lake Analytics, if you are processing large datasets.

The resources that Azure allocates to functions is set at the Function App level, split equally among functions inside that app. Most of the time, this will have absolutely no practical impact on your work, and generally, you should keep logically connected functions in the same app. However, in cases where you suddenly demand massive scaling for one of your functions but not for the others, this can cause issues. If your suddenly popular function is one of twenty in an app, it's going to take longer to allocate enough resources to supply it. On the other hand, if you have a seldom-used function that you need better latency from, consider adding it to the Function App of a commonly-used one, so that the function doesn't have to warm up from the cold.

Prerequisites

You will need Visual Studio installed, with Azure development workload.

Scenario

Creating a personalized application for users.

Aim

Create a function that stores user details.

Steps for Completion

1. Create a new function, called PostUser, using the HttpTrigger template. If you are competent in another language, like JavaScript or F#, you can try creating it in that language.
2. Create a new model called User. Think about what properties you'd like to include, but as a minimum, you will need an email address and a unique identifier (these can be the same property).
3. Modify the function to decode a JSON-formatted request into this model.
   

Outcome

You have a function ready to store user details.

In this chapter, you covered the benefits and disadvantages of serverless computing. Next, you learned how to create, debug, and deploy an Azure Function, Azure's primary serverless product. This is the basic building block of serverless applications, and the first step to creating fully serverless architectures. Finally, you learned about the technical basis of Azure Functions, and how you can utilize your knowledge of it to improve the performance of your functions.

In the next chapter, we will be covering integrations with other Azure services and Azure-based logging and security solutions.