Writing Functions

This guide covers writing functions using the Functions Framework for Ruby. For more information about the Framework, see the Overview Guide.

About functions

Functions are Ruby blocks that are run when an input is received. Those inputs can be HTTP requests or events in a recognized format. Functions that receive HTTP requests return an HTTP response, but event functions have no return value.

When you define a function, you must provide an identifying name. The Functions Framework allows you to use any string as a function name; however, many deployment environments restrict the characters that can be used in a name. For maximum portability, it is recommended that you use names that are allowed for Ruby methods, i.e. beginning with a letter, and containing only letters, numbers, and underscores.

Defining an HTTP function

An HTTP function is a simple web service that takes an HTTP request and returns an HTTP response. The following example defines an HTTP function named "hello" that returns a simple message in the HTTP response body:

require "functions_framework"

FunctionsFramework.http "hello" do |request|
  # Return the response body.
  "Hello, world!\n"
end

HTTP functions take a Rack Request object and return an HTTP response. We'll now cover these in a bit more detail.

Using the Request object

An HTTP function is passed a request, which is an object of type Rack::Request. This object provides methods for obtaining request information such as the method, path, query parameters, body content, and headers. You can also obtain the raw Rack environment using the env method. The following example includes some request information in the response:

require "functions_framework"

FunctionsFramework.http "request_info_example" do |request|
  # Include some request info in the response body.
  "Received #{request.request_method} from #{request.url}!\n"
end

The Functions Framework sets up a logger in the Rack environment, so you can use the logger method on the request object if you want to emit logs. These logs will be written to the standard error stream, and will appear in the Google Cloud Logs if your function is running on a Google Cloud serverless hosting environment.

require "functions_framework"

FunctionsFramework.http "logging_example" do |request|
  # Log some request info.
  request.logger.info "I received #{request.request_method} from #{request.url}!"
  # A simple response body.
  "ok"
end

Response types

The above examples return simple strings as the response body. Often, however, you will need to return more complex responses such as JSON, binary data, or even rendered HTML. The Functions Framework recognizes a variety of return types from an HTTP function:

  • String : If you return a string, the framework will use it as the response body in with a 200 (success) HTTP status code. It will set the Content-Type header to text/plain.
  • Array : If you return an array, the framework will assume it is a standard three-element Rack response array, as defined in the Rack spec.
  • Rack::Response : You can return a Rack::Response object. The Framework will call #finish on this object and retrieve the contents.
  • Hash : If you return a Hash, the Framework will attempt to encode it as JSON, and return it in the response body with a 200 (success) HTTP status code. The Content-Type will be set to application/json.
  • StandardError : If you return an exception object, the Framework will return a 500 (server error) response. See the section below on Error Handling.

Using Sinatra

The Functions Framework, and the functions-as-a-service (FaaS) solutions it targets, are optimized for relatively simple HTTP requests such as webhooks and simple APIs. If you want to deploy a large application or use a monolithic framework such as Ruby on Rails, you may want to consider a solution such as Google Cloud Run that is tailored to larger applications. However, a lightweight framework such as Sinatra is sometimes useful when writing HTTP functions.

It is easy to connect an HTTP function to a Sinatra app. First, declare the dependency on Sinatra in your Gemfile:

source "https://rubygems.org"
gem "functions_framework", "~> 0.9"
gem "sinatra", "~> 2.0"

Write the Sinatra app using the "modular" Sinatra interface (i.e. subclass Sinatra::Base), and then run the Sinatra app directly as a Rack handler from the function. Here is a basic example:

require "functions_framework"
require "sinatra/base"

class App < Sinatra::Base
  get "/hello/:name" do
    "Hello, #{params[:name]}!"
  end
end

FunctionsFramework.http "sinatra_example" do |request|
  App.call request.env
end

This technique gives you access to pretty much any feature of the Sinatra web framework, including routes, templates, and even custom middleware.

Defining an Event function

An event function is a handler for a standard cloud event. It can receive industry-standard CloudEvents, as well as events sent by Google Cloud services such as Pub/Sub and Storage. Event functions do not have a return value.

The following is a simple event handler that receives an event and logs some information about it:

require "functions_framework"

FunctionsFramework.cloud_event "hello" do |event|
  FunctionsFramework.logger.info "I received an event of type #{event.type}!"
end

The event parameter will be either a CloudEvents V0.3 Event object (see spec) or a CloudEvents V1.0 Event object (see spec).

Some Google Cloud services send events in a legacy event format that was defined prior to CloudEvents. The Functions Framework will convert these legacy events to an equivalent CloudEvents V1 type, so your function will always receive a CloudEvent object when it is sent an event from Google Cloud. The precise mapping between legacy events and CloudEvents is not specified in detail here, but in general, the data from the legacy event will be mapped to the data field in the CloudEvent, and the context from the legacy event will be mapped to equivalent CloudEvent attributes.

Error handling

If your function encounters an error, it can raise an exception. The Functions Framework will catch StandardError exceptions and handle them appropriately.

If you raise an exception in an HTTP function, the Functions Framework will return a 500 (server error) response. You can control whether the exception details (e.g. exception type, message, and backtrace) are sent with the response by setting the detailed-errors configuration in the server. The Framework will also log the error for you.

If you need more control over the error response, you can also construct the HTTP response yourself. For example:

require "functions_framework"

FunctionsFramework.http "error_reporter" do |request|
  begin
    raise "whoops!"
  rescue RuntimeError => e
    [500, {}, ["Uh, oh, got an error message: #{e.message}."]]
  end
end

The runtime environment

A serverless environment may be somewhat different from server-based runtime environments you might be used to. Serverless runtimes often provide a simpler programming model, transparent scaling, and cost savings, but they do so by controlling how your code is managed and executed. The Functions Framework is designed around a "functions-as-a-service" (FaaS) paradigm, which runs self-contained stateless functions that have an input and a return value. It's important to understand what that means for your Ruby code in order to get the most out of a cloud serverless product.

For example, multithreading is a core element of the Functions Framework. When you write functions, you should assume that multiple executions may be taking place concurrently in different threads, and thus you should avoid operations that can cause concurrency issues or race conditions. The easiest way to do this is to make your functions self-contained and stateless. Avoid global variables and don't share mutable data between different function executions.

Additionally, a serverless runtime may throttle the CPU whenever no actual function executions are taking place. This lets it reduce the CPU resources used (and therefore the cost to you), while keeping your application warmed up and ready to respond to new requests quickly. An important implication, though, is that you should avoid starting up background threads or processes. They may not get any CPU time during periods when your Ruby application is not actually executing a function.

In the sections below, we'll discuss a few techniques and features of the Functions Framework to help you write Ruby code that fits well into a serverless paradigm.

Startup tasks

It is sometimes useful to perform one-time initialization that applies to many function executions, for example to warm up caches, perform precomputation, or establish shared remote connections. To run code during initialization, use FunctionsFramework.on_startup to define a startup task.

require "functions_framework"

FunctionsFramework.on_startup do |function|
  # Perform initialization here.
  require "my_cache"
  MyCache.warmup
end

FunctionsFramework.http "hello" do |request|
  # Initialization will be done by the time a normal function is called.
end

Startup tasks are run once per Ruby instance during cold start -- that is, after the Ruby VM boots up but before the framework starts receiving requests and executing functions. You can define multiple startup tasks, and they will run in order, and are guaranteed to complete before any function is executed.

The block is optionally passed the FunctionsFramework::Function representing the function that will be run. You code can, for example, perform different initialization depending on the FunctionsFramework::Function#name or FunctionsFramework::Function#type.

In most cases, initialization code should live in an on_startup block instead of at the "top level" of your Ruby file. This is because some serverless runtimes may load your Ruby code at build or deployment time (for example, to verify that it properly defines the requested function), and this will execute any code present at the top level of the Ruby file. If top-level code is long-running or depends on runtime resources or environment variables, this could cause the deployment to fail. By performing initialization in an on_startup block instead, you ensure it will run only when an actual runtime server is starting up, not at build/deployment time.

require "functions_framework"

# DO NOT perform initialization here because this could get run at build time.
#   require "my_cache"
#   MyCache.warmup

# Instead initialize in an on_startup block, which is executed only when a
# runtime server is starting up.
FunctionsFramework.on_startup do
  # Perform initialization here.
  require "my_cache"
  MyCache.warmup
end

# ...

Because startup tasks run during cold start, they could have an impact on your function's startup latency. To mitigate this issue, it is possible to run parts of your initialization lazily, as described below in the section below on lazy initialization.

The execution context and global data

When your function block executes, the object context (i.e. self) is set to an instance of FunctionsFramework::Function::Callable. Each function invocation (including functions that might be running concurrently in separate threads) runs within a different instance, to help you avoid having functions interfere with each other.

The object context also defines a few methods that may be useful when writing your function.

First, you can obtain the logger by calling the FunctionsFramework::Function::Callable#logger convenience method. This is the same logger that is provided by the HTTP request object or by the FunctionsFramework.logger global method.

Second, you can access global shared data by passing a key to FunctionsFramework::Function::Callable#global. Global shared data is a set of key-value pairs that are available to every function invocation. By default, two keys are available to all functions:

  • :function_name whose String value is the name of the running function.
  • :function_type whose value is either :http or :cloud_event depending on the type of the running function.

Following is a simple example using the logger and global methods of the context object:

require "functions_framework"

FunctionsFramework.cloud_event "hello" do |event|
  logger.info "Now running the function called #{global(:function_name)}"
end

To avoid concurrency issues, global shared data is immutable when executing a function. You cannot add or delete keys or change the value of existing keys. However, the global data is settable during startup tasks, because startup tasks never run concurrently. You can use this feature to initialize shared resources, as described below.

Using the global data mechanism is generally preferred over actual Ruby global variables, because the Functions Framework can help you avoid concurrent edits. Additionally, the framework will isolate the sets of global data associated with different sets of functions, which lets you run functions in isolation during unit tests. If you are testing multiple functions, they will not interfere with each other as they might if they used global variables.

Sharing resources

Although functions should generally be self-contained and stateless, it is sometimes useful to share certain kinds of resources across multiple function invocations that run on the same Ruby instance. For example, you might establish a single connection to a remote database or other service, and share it across function invocations to avoid incurring the overhead of re-establishing it for every function invocation.

The best practice for sharing a resource across function invocations is to initialize it in a FunctionsFramework.on_startup block, and reference it from global shared data. (As discussed above, the best practice is to initialize shared resources in a startup task rather than at the top level of a Ruby file, and to use the Functions Framework's global data mechanism rather than Ruby's global variables.)

Here is a simple example:

require "functions_framework"

# Use an on_startup block to initialize a shared client and store it in
# the global shared data.
FunctionsFramework.on_startup do
  require "google/cloud/storage"
  set_global :storage_client, Google::Cloud::Storage.new
end

# The shared storage_client can be accessed by all function invocations
# via the global shared data.
FunctionsFramework.http "storage_example" do |request|
  bucket = global(:storage_client).bucket "my-bucket"
  file = bucket.file "path/to/my-file.txt"
  file.download.to_s
end

Importantly, if you do share a resource across function invocations, make sure the resource is thread-safe, so that separate functions running concurrently in different threads can access them safely. The API clients provided by Google, for example, are thread-safe and can be used concurrently.

Also of note: There is no guaranteed cleanup hook. The Functions Framework does not provide a way to register a cleanup task, and we recommend against using resources that require explicit "cleanup". This is because serverless runtimes may perform CPU throttling, and therefore there may not be an opportunity for cleanup tasks to run. (For example, you could register a Kernel.at_exit task, but the Ruby VM may still terminate without calling it.)

Lazy initialization

Because startup tasks run during cold start, they could have an impact on your function's startup latency. You can mitigate this by initializing some globals lazily. When setting a global, instead of computing and setting the value directly (e.g. constructing a shared API client object directly), you can provide a block that describes how to construct it on demand.

Here is an example using the storage client we saw above.

require "functions_framework"

# This startup block describes _how_ to initialize a shared client, but
# does not construct it immediately.
FunctionsFramework.on_startup do
  require "google/cloud/storage"
  set_global :storage_client do
    Google::Cloud::Storage.new
  end
end

# The first time this function is invoked, it will call the above block
# to construct the storage client. Subsequent invocations will not need
# to construct it again, but will reuse the same shared object.
FunctionsFramework.http "storage_example" do |request|
  bucket = global(:storage_client).bucket "my-bucket"
  file = bucket.file "path/to/my-file.txt"
  file.download.to_s
end

The block will not be called until a function actually attempts to access the global. From that point, subsequent accesses of the global will return that same shared value; the block will be called at most once. This is true even if multiple functions are run concurrently in different threads.

Lazy initialization is particularly useful if you define several different functions that may use different sets of shared resources. Instead of initializing all resources eagerly up front, you could initialize them lazily and run only the code needed by the function that is actually invoked.

Structuring a project

A Functions Framework based "project" or "application" is a typical Ruby application. It should include a Gemfile that specifies the gem dependencies (including the functions_framework gem itself), and any other dependencies needed by the function. It must include at least one Ruby source file that defines functions, and can also include additional Ruby files defining classes and methods that assist in the function implementation.

By convention, the "main" Ruby file that defines functions should be called app.rb and be located at the root of the project. The path to this file is sometimes known as the function source. The Functions Framework allows you to specify an arbitrary source, but some hosting environments (such as Google Cloud Functions) require it to be ./app.rb.

A source file can define any number of functions (with distinct names). Each of the names is known as a function target.

Following is a typical layout for a Functions Framework based project.

(project directory)
|
+- Gemfile
|
+- app.rb
|
+- lib/
|  |
|  +- hello.rb
|
+- test/
   |
   ...

# Gemfile
source "https://rubygems.org"
gem "functions_framework", "~> 0.9"

# app.rb
require "functions_framework"

FunctionsFramework.on_startup do
  require_relative "lib/hello"
end

FunctionsFramework.http "hello" do |request|
  Hello.new(request).build_response
end

# lib/hello.rb
class Hello
  def initialize request
    @request = request
  end

  def build_response
    "Received request: #{@request.request_method} #{@request.url}\n"
  end
end

Next steps

To learn about writing unit tests for functions, see Testing Functions.

To learn how to run your functions in a server, see Running a Functions Server.

To learn how to deploy your functions to Google Cloud Functions or Google Cloud Run, see Deploying Functions.