FaaS, or Functions as a Service, is a relatively new offering that you can hardly miss nowadays. Some see it as a successor to Platform as a Service, others see just another way of hosting your application functionalities. The major cloud providers also stepped into this area and offer serverless computing capabilities. As always, your application needs to run somewhere, so it is a nice marketing term for a hosting solution that allows you to make efficient use of available resources by running short-lived pieces of code, which are small, usually stateless, and on demand. OpenFaaS is a project that makes use of Docker containers to convert any process into a serverless function, regardless of the programming language of the code.
Why Functions as a Service?
There are various reasons why Functions as a Service is interesting for many organizations.
First of all, it allows for rapid prototyping of application logic. Development teams only need to write small pieces of code, in the language of their choice, and this can be deployed directly to the cloud. Many cloud providers offer integrations with other services within their platform (such as queues and databases) and this makes it very easy to get up and running in creating a serverless architecture.
Secondly, it provides a way to achieve a highly scalable solution on a highly granular level since individual functions can scale up when the load is heavy, and scale back down when the load is low. Usually, the cloud providers fully manage this scaling so it requires no additional effort.
Lastly, all cloud providers offer a consumption-based pricing model. This means that billing is done by the number of executions combined with the execution time. This can be a very attractive pricing model if the application workload changes aggressively over time (spikes) or when the workload is constantly very low.
An often heard argument against FaaS concerns vendor lock-in. Once you are using a FaaS platform by one of the cloud providers, like Azure Functions, AWS Lambda or Google Functions, it is difficult to move to another provider because each FaaS platform is specific to that provider. The OpenFaaS project is a very interesting alternative for organizations that want to avoid this lock-in.
The OpenFaaS project
A Docker Captain called Alex Ellis1 liked the idea and purpose of serverless functions but wanted to see if there was a way to avoid having a vendor lock-in and use private or hybrid clouds as an alternative. The OpenFaaS project uses Docker and Kubernetes as the underlying technology and this allows OpenFaaS to run functions nearly everywhere. This can range from a single laptop to large-scale cloud systems, so you have plenty of hosting options available.
The OpenFaaS architecture is set up in such a way that you can host any kind of programming language as long as it can talk to standard input (stdin) and output (stdout), the standard streams that are preconnected input and output communication channels between a computer program and its environment. Your code is packaged inside a small Docker container together with a watchdog component. The job of the watchdog is to accept requests from a HTTP API gateway, route the request to the standard input of the application, read the output from the standard output of the process, and return this to the sender. In addition, it enforces timing constraints to the container.
As mentioned already, the OpenFaaS architecture includes an API gateway that provides HTTP and JSON capabilities so you don’t need to write this kind of plumbing anymore.
As Docker can use Prometheus for metric collection, the OpenFaaS project utilizes the same system. It also uses the data to detect whether containers are under a certain load and therefore need scaling. An online dashboard allows you to monitor and set your own alerts.
OpenFaaS also comes with its own command line interface (CLI) which gives you control over the functions and makes it possible to easily create and deploy functions. We will show you how this works in this article.
Taking OpenFaaS for a spin
For the purpose of this article, we will spin up OpenFaaS on a simple Docker Swarm cluster. Even if this is a cluster of only one node and that node is your own machine, it will be fine. Make sure you have installed Docker with Swarm support and that you have initialized a Swarm if you have not done this yet:
docker swarm init
Fetch the actual code using a git clone command to a local folder:
git clone https://github.com/openfaas/faas
This will pull down the required files and when fetching is complete, navigate to the automatically created folder called faas. Next, you will need to run the deployment script.
Use either the .sh if you are running from a Linux bash shell, or the .ps1 when using Windows Powershell:
.\deploy_stack.ps1
This will create the required networks and services (both for running OpenFaaS and for samples) in Docker Stack, and the system is ready to use.
You can now navigate to http://localhost:8080 using your web browser to see the OpenFaas dashboard.
This dashboard shows all the deployed example functions and you can invoke them directly. There is a gallery of functions under Deploy New Function where you can easily add existing functions to your FaaS system.
Create
Let’s see how you can create your own function. For this purpose we will use the CLI and there are various ways of installing this tool as described on the GitHub FaaS-CLI4 site, so pick the one applicable to your operating system. When installed, you can add functions using a new, build and deploy command pattern. The new command requires a function name and a (programming) language template.
faas-cli new helloworld -- lang csharp
After you have created the helloworld function, the following files and folders will have been added:
- A template folder with function templates. OpenFaaS automatically downloads function templates from GitHub when these are not yet available. A more detailed description of these templates is provided in the next section.
- A helloworld folder with the Function.csproj and the FunctionHandler.cs files.
- A helloworld.yml file. Similar to docker-compose or Kubernetes, the definition of an OpenFaaS function is described in a yaml file. This makes it easier to build and deploy sets of functions and keep your server in the desired state.
Templates
Now that we’ve created a new function let’s have a look at the template on which it is built. Each language template contains a dockerfile and an entry point application file. The dockerfile contains a multi-stage definition; the first part builds the function code, the second part uses the output, includes the watchdog component, and defines the entry function to be the watchdog itself. The dockerfile for the C# template (located at template\csharp\Dockerfile) is shown below.
The csharp template function is located at template\csharp\Function\FunctionHandler.cs.
Because OpenFaaS uses the standard input and output streams your C# function is actually a small console application, as can be seen below.
Build
When you have completed writing your helloworld function, you need to build this before it can be deployed. Use the CLI to invoke the build command which instructs Docker to build the dockerfile that contains the instructions for both building and packaging.
faas-cli build -f .\helloworld.yml
The generated build folder contains the project file which you can open with, for example, Visual Studio (Code). This allows you to run, debug and edit with intellisense before it gets packaged inside a docker container. While running the console application, keep in mind that a control-Z will end your input to your function and allows the function to return something to the console.
After building you need to execute the deploy command:
faas-cli deploy -f .\helloworld.yml
This applies the functions defined in the yaml file to the configured provider. The above-defined function will live under the http://localhost:8080/function/helloworld endpoint. A curl POST request to this endpoint returns the defined string:
curl -d "myname" http://localhost:8080/function/ helloworld Hi there - your input was: myname
The dashboard and the faas-cli are other ways to invoke this function. All this time OpenFaaS is using Docker to deploy and host your containers. You can still use the docker commands to see your functions running on the nodes:
docker service ps helloworld
If you have a cluster, the container will be hosted on one of the nodes and kept available when nodes crash. Calling the functions a number of times will produce metrics which are collected inside Prometheus. With the Prometheus portal on port 9090 you can build all kinds of dashboard and monitor metrics such as the state, number of invocations, and execution durations.
Prometheus has predefined rules to alert on services that are down as well as high invocation counts. OpenFaaS uses these metrics and alerts to scale the functions up or down.
When to choose OpenFaas instead of a vendor-specific FaaS
OpenFaaS or vendor-specific systems such as Azure Functions or AWS Lambda are all valid platforms to develop and host functions. However, you do have to realize that a vendoragnostic solution comes at a price.
Although you can choose your own host(s) with OpenFaaS and avoid vendor-lock-in, OpenFaaS lacks (cloud provider specific) integrations which can limit the rapid prototyping capabilities. For instance, Azure Functions provides seamless integration (triggers) with other Azure services such as CosmosDB, Storage Queues, Blob Storage, ServiceBus, EventGrid etc. With OpenFaaS, you only have HTTP triggered functions and you have to develop other integrations yourself.
There is a learning curve for each FaaS platform, but the learning curve for OpenFaaS is a bit steeper if you don’t have some container technology experience. That being said, containers are conquering the world so we highly recommend upgrading your skills on that part.
A possible benefit of OpenFaaS is that it is open source. So if you want to change the way it autoscales or want support for a new language you can develop that yourself. This level of flexibility is usually not provided by a cloud vendor. However, open source has its drawbacks in the form of support, lifecycle management, and documentation, which might not align with your requirements.
Use the OpenFaaS platform when:
- you don’t require easy integration with cloud-specific services
- your development team is comfortable with using container technology
- you require full control of a FaaS platform and your development team also has the capability to maintain it.
Summarizing
OpenFaas allows you to run your own functions anywhere thanks to container technology. It supports many programming languages out of the box and since the platform is open source, you can extend or change it to your needs. It does, however, lack the integration options that cloud-specific platforms provide. As with any FaaS platform you host yourself, you need to make sure you secure it by using a gateway solution such as Treafik, Kong or API Management before you bring it into production.
If you like a DIY challenge while learning .NET Core, OpenFaas & Kubernetes, we can highly recommend this post by Scott Hanselman.
This article is part of our latest magazine; XPRT.#6 Download it here or get your free copy.