Azure Functions anatomy – Part 2 – What is inside a Function App?

In the previous post (Azure Functions 2.0 Anatomy- Part 1), we saw the structure of a function app. In this part, I will explain the different components of an Azure Function.

Components

Every Azure Function 2.0 consists of the following components:

  • Trigger
  • Bindings (one of which is the trigger). Binding can be input or output
  • Value Converter/Binder
  • Listener

A binding can be a Blob file, a queue message, an event from event hub or event grid or any custom binding developed by anyone. One of these bindings can initiate the function call and in this case it is called a Trigger.
A binding can also be Input or Output. Input binding delivers data to the function and output binding is used by the function to write data.

Now, assume we have the following function and its configuration

        [FunctionName("SampleFunction")]
        public static void Run([QueueTrigger("myqueue-items", Connection = "AzureWebJobsStorage")]string myQueueItem, [Blob("sample-output")]Stream output, TraceWriter log)
        {
            log.Info($"C# Queue trigger function processed: {myQueueItem}");
        }

This function has 2 binding parameters. myQueueItem which is both an input binding and a trigger, and output which is an output binding.

How Azure Functions call this method and convert the queue message to a string and the stream to a Blob file?

Binding, Listener, Value Provider and Converter

When the azure function app starts, it scans all functions that exist in the app directory like we saw in the previous post, and for each function it reads the input, output and trigger from the function config JSON file.

The scanning process uses a descriptor provider which creates the input binding, output binding, trigger and an invoker

Once all functions are loaded, it tried to load the types for each function that was developed outside the portal. It uses the scriptFile and EntryPoint as you can see below. The entry point in the previous code snippet is FunctionApp1.Function1.Run method that exist in the assembly FunctionApp1.dll

{
  "generatedBy": "Microsoft.NET.Sdk.Functions-1.0.14",
  "configurationSource": "attributes",
  "bindings": [
    {
      "type": "timerTrigger",
      "schedule": "0 */5 * * * *",
      "useMonitor": true,
      "runOnStartup": false,
      "name": "myTimer"
    }
  ],
  "disabled": false,
  "scriptFile": "../bin/FunctionApp1.dll",
  "entryPoint": "FunctionApp1.Function2.Run"
}

Each function app can either have a DLL function or a function created by portal. But not both.

Now, lets see what are the other components inside a binding:

  • Listener (IListener): This one has a method named StartAsync, which basically start listening for an event to trigger the function. For Azure Blob trigger, it scans the logs for new or modified files. For Queue trigger, it check for new messages in the specified queue.
  • Value Provider (IValueProvider): Once the listener has the event that should trigger the function, it uses the value provider to fill all input parameters. If the trigger was a blog trigger, then the value will be the file, the metadata, the name and content of the file itself. For a Queue, it will be content of the message and so on and so forth.
  • Value Converter (IAsyncConverter): This will be used to convert the values found in the value provider to all the input parameters configured for the function. In the previous example, StorageQueueMessageToStringConverter will be used to convert the CloudQueueMessage instance to the string parameter named myQueueItem. The same will be used to write data to output parameters, the framework will choose the best converter that matches the Output parameter type (Stream) and the type of variable that will be used to write to stream, ex: String or any Stream Writer.
  • Executor (ITriggerExecutor): Now we know that the function should be triggered, we have the data from input binding and we bound all input parameters, now we need to call the function body. This takes place using an implementation of ITriggerExecutor which can use reflection to call the function or in a very special case, it may get executed as a web hook like the case with HTTP trigger.

In the next post, I will dig more into Azure Function App to explore the extensions and host.

Azure Functions 2.0 anatomy – Part 1

In these series of posts, I will explain the internals of Azure Functions and how it works. By the end of it, you should be able to understand what happens once an Azure Function is deployed till the moment it is triggered and invocation completes.

Is it a Web App?

Yes, every Azure Function App that uses .Net as a Language runs as as.net core web app. When the app starts, it loads all your functions and proxies and set up the routes and listeners for each function.

Web App Structure

Just like any other Azure Web App, it consists of the following folders:

  • data: This folder contains another folder named “functions” which has another folder called extensions. The extensions folder has a list of all extensions installed to your Functions App. Each file has a reference to a nuget package that has the extension implementation.
  • Log Files
  • site: it has the wwwroot which hosts the asp.net core web app and all the functions
Data -> Functions folder structure

But for Azure Functions App, the following folders are added:

.nuget

Contains all nuget packages installed when you install a new extension. A reference for each extension installed exist in a file under: data\functions\extensions. The sample extension file below specify the package :Microsoft.Azure.WebJobs.Extensions.ServiceBus

In Azure Functions 1.0, it was running on .Net 4.6 and the runtime already included all the supported triggers and bindings. But, with version 2.0, the runtime only includes http and timer binding and all the other bindings can be developed and installed separately using extensions. This allows you to develop your own extensions and upload it without the need to wait for a new version of Azure Functions.

{
  "Id": "d3a33dd0-3941-4acf-a303-72773995901d",
  "Status": 1,
  "StartTime": "2019-01-02T02:49:23.5568341+00:00",
  "EndTime": "2019-01-02T02:50:50.1029788+00:00",
  "Error": null,
  "Properties": {
    "id": "Microsoft.Azure.WebJobs.Extensions.ServiceBus",
    "version": "3.0.0"
  }
}

ASP.NET

This folder contains all data protection keys used in the web app

To have a look at this folder structure and explore it yourself, use the following URL: https://%5BFUNCTION_APP_NAME%5D.scm.azurewebsites.net/DebugConsole

wwwroot folder structure

wwwroot folder structure

As you can see, there are 3 folders. Each folder represent a single function. All DLLs for each function exist in the bin folder.

The host.json file by default only has the version number for the function

{
  "version": "2.0"
}

Inside each function, there is a function.json file that has the configuration for this function.

{
  "generatedBy": "Microsoft.NET.Sdk.Functions-1.0.14",
  "configurationSource": "attributes",
  "bindings": [
    {
      "type": "timerTrigger",
      "schedule": "0 */5 * * * *",
      "useMonitor": true,
      "runOnStartup": false,
      "name": "myTimer"
    }
  ],
  "disabled": false,
  "scriptFile": "../bin/FunctionApp1.dll",
  "entryPoint": "FunctionApp1.Function2.Run"
}

The Azure Functions run-time use this file to load the metadata for each function during application startup. The most important properties are the scriptFile which points to the DLL that has the code and the entryPoint that has the exact c# method name that will be executed when the function triggers.
The configurationSource specifies where to read the function configuration such as connection strings for the blob storage or service bus. attributes means it will be retrieved from code.
The binding key lists all bindings for this function. For this sample, it is a timer and it is the trigger, so it has a single binding. For others, it can be bindings that represents input, output and a trigger.

In the next post, I will explain the internals of each function, and how the bindings and triggers works.

How Azure Functions Blob Trigger works

Introduction

Azure functions enable you to quickly build a piece of functionality that can be triggered by an external system such as Azure Blob Storage, Storage Queue, CosmoDB, Event Hub and the list goes on. In this post, I will explain how Azure Blob Trigger works.

Sample Function

If you created a new Azure Function using Visual Studio, you will end up with the following code:

 public static class Function1
    {
        [FunctionName("Function1")]
        public static void Run([BlobTrigger("samples-workitems/{name}", Connection = "")]Stream myBlob, string name, TraceWriter log)
        {
            log.Info($"C# Blob trigger function Processed blob\n Name:{name} \n Size: {myBlob.Length} Bytes");
        }
    }

The secret lies in the attribute [BlobTrigger]. It accepts the path to the storage container which Azure Web Jobs will monitor. As you know, Azure Functions is based on Azure Web Jobs, so the same triggers are used.

How Azure Web Job knows about a file that is being added, updated or removed to a container?

Run the sample app you just created in visual studio, it should work on the local storage emulator. Now, you should be able to invoke the function by creating a blob container called samples-workitems in your development storage account and you should end up with the following structure.

blob container

You can see the container “samples-workitems” which Azure Web Jobs will monitor and invoke the function whenever a file is changed there. But there is also another container named “azure-webjobs-hosts” which is the secret for how Azure monitor the files in that container.

If you opened that container, you will find a folder called “blobreceipts” which has a folder for the function name.

Inside the folder with the function name, there are some other folders with strange names like below

etags

These strange names are the ETag of each blob file added, edited or removed. So, when you change a file in any container, Web Jobs will create a new folder here with the new version ETag for that file and inside this folder you will find the same structure of the file being edited. In our case, there should be a folder called samples-workitems and inside this folder the file that was modified.

When a new file is added, Azure will check if  its ETag exist in the azure-webjobs-hosts folder and if not, then it will call the Azure Function. This way it will prevent duplicate calls for the same file. This pattern is called blob receipt.

Note that this process depends on Azure Blob Storage Logging which can take up to 10 minutes to write to the container azure-webjobs-hosts to improve performance. If you need your function to trigger faster then consider using Storage Queue trigger instead.