Powered by AppSignal & Oban Pro
Would you like to see your link here? Contact us

GenServers

reading/genservers.livemd

GenServers

Mix.install([
  {:jason, "~> 1.4"},
  {:kino, "~> 0.9", override: true},
  {:youtube, github: "brooklinjazz/youtube"},
  {:hidden_cell, github: "brooklinjazz/hidden_cell"}
])

Navigation

Home Report An Issue Process MailboxGenServer Drills

Review Questions

  • What is the purpose of a GenServer?
  • Explain the lifecycle of sending a message to a GenServer and receiving a response.
  • Why might we use asynchronous vs synchronous messages?
  • What happens when we send a GenServer too many messages? What might happen to the GenServer mailbox?

Generic Server

In Elixir, a Generic Server refers to a programming pattern and behavior implemented using the GenServer module from the OTP (Open Telecom Platform) library. The Generic Server pattern is a way to structure concurrent and fault-tolerant processes that receive and handle messages.

A Generic Server, implemented as a GenServer, is a long-running Elixir process that encapsulates state and behavior. It allows for message-based communication between processes and provides a structured way to handle those messages.

Key characteristics of a Generic Server implemented with GenServer include:

  1. State Management: The GenServer process holds and manages its own internal state. This state can be modified by handling specific messages and updating the state accordingly.

  2. Message Handling: GenServer processes receive messages asynchronously and handle them using pattern matching on the message content. The behavior of the server can be defined based on the message received.

  3. Synchronous and Asynchronous Communication: GenServer processes can communicate synchronously by sending a message and waiting for a response, or asynchronously by sending a message without expecting an immediate reply.

  4. Supervision and Fault Tolerance: GenServer processes are often used within supervision trees, allowing them to be monitored and restarted in the event of failures or crashes. This contributes to the fault-tolerant nature of OTP applications.

  5. Callback Functions: GenServer requires implementing specific callback functions such as handle_call/3, handle_cast/2, and handle_info/2 to define the behavior of the server for different types of messages.

The Generic Server pattern implemented with GenServer is a fundamental building block in Elixir/OTP applications. It provides a structured approach to building concurrent and fault-tolerant systems by encapsulating state, managing message-based communication, and defining behavior through callback functions.

Generic Server Life-Cycle

Generic Servers are a generic wrapper around state and message handling. They are given a callback module that defines callback functions for customizing the behavior of the Generic Server.

The callback module contains the specific message handler callbacks and the callback for initializing state.

Here’s a diagram showing a broad overview of how GenServer’s work with a callback module.

sequenceDiagram
Process ->> Generic Server: start_link/0
Generic Server ->> Callback Module: init/1 
Callback Module ->> Generic Server: initial state
loop send/receive loop
  Generic Server --> Generic Server: receive
  Process ->> Generic Server: send message (e.g.  call/3)
  Generic Server ->> Callback Module: handle message (e.g. handle_call/3)
  Callback Module ->> Generic Server: sends response and new state
  Generic Server --> Generic Server: updates state
  Generic Server ->> Process: sends response
end

GenServer

Message Handler Callbacks

The GenServer module defines all of the boilerplate under the hood, and allows us to conveniently provide callback functions including:

  • init/1 defines the initial state in an {:ok, state} tuple.
  • handle_call/3 handle a synchronous message meant for a GenServer process.
  • handle_cast/2 handle an asynchronous message meant for a GenServer process.
  • handle_info/2 handle a generic asynchronous message.

Starting A GenServer

We commonly use one of the following functions to start a GenServer process.

Sending A GenServer A Message

We can send the GenServer messages with functions such as:

  • GenServer.call/3 send a synchronous message for a GenServer handled by handle_call/3.
  • GenServer.cast/2 send an asynchronous message for a GenServer handled by handle_cast/2.
  • Kernel.send/2 send a generic asynchronous message handled by handle_info/2
  • Process.send/3 send a generic asynchronous message with some additional options handled by handle_info/2.
  • Process.send_after/4 send a generic asynchronous message handled by handle_info/2.

Counter Example

Here’s a simple CounterServer GenServer. It stores a count in state, and we can send it a message to increment the count.

  • GenServer.handle_call/3 returns a synchronous response in a {:reply, response, state} tuple.
  • GenServer.handle_cast/2 asynchronously updates the state in a {:noreply, new_state} tuple.
  • GenServer.init/2 initializes the initial state (0) in an {:ok, state} tuple.
defmodule CounterServer do
  use GenServer

  @impl true
  def init(_init_arg) do
    {:ok, 0}
  end

  @impl true
  def handle_cast(:increment, state) do
    {:noreply, state + 1}
  end

  @impl true
  def handle_call(:get_count, _from, state) do
    # `response` bound for sake of clarity.
    response = state
    {:reply, response, state}
  end
end

Using the GenServer above, here’s an example of sending an asynchronous message with GenServer.cast/2 and an asynchronous message with GenServer.call/3.

{:ok, pid} = GenServer.start_link(CounterServer, [])
GenServer.cast(pid, :increment)
GenServer.call(pid, :get_count)

In Livebook, we can use Kino.Process.render_seq_trace/2 to visualize the CAST and CALL messages sent to the GenServer.

Kino.Process.render_seq_trace(fn ->
  {:ok, pid} = GenServer.start_link(CounterServer, [])
  GenServer.cast(pid, :increment)
  GenServer.call(pid, :get_count)
end)

The INFO: ack message is an acknowledgement message sent back to the parent process and received under the hood by GenServer.start_link/3.

Messages With Params

We send messages as a collection (typically a tuple) to add additional data.

def handle_cast({:increment_by, increment_by}, state)
  {:noreply, state + increment_by}
end

Your Turn

In the CounterServer above, add a handler for a :decrement message.

Client API

Typically we don’t use GenServer functions directly, and instead create a Client API. The Client API makes our code more reusable, readable, and easy to change. This is purely for code organization.

Client API functions typically use __MODULE__ when referencing the current module to make it easier to rename the module in the future.

Callback functions are often referred to as the Server API.

defmodule ClientServerExample do
  use GenServer
  # Client API

  def start_link(_opts) do
    GenServer.start_link(__MODULE__, [])
  end

  def increment(pid) do
    GenServer.cast(pid, :increment)
  end

  def get_count(pid) do
    GenServer.call(pid, :get_count)
  end

  # Server API

  def init(_init_arg) do
    {:ok, 0}
  end

  def handle_cast(:increment, state) do
    {:noreply, state + 1}
  end

  def handle_call(:get_count, _from, state) do
    response = state
    {:reply, response, state}
  end
end

Some projects will split the client and the server module when they get large enough. If we split the modules then we can’t use __MODULE__ to refer to the server module from the client module.

defmodule ClientExample do
  def start_link(_opts) do
    GenServer.start_link(ServerExample, [])
  end

  def increment(pid) do
    GenServer.cast(pid, :increment)
  end

  def get_count(pid) do
    GenServer.call(pid, :get_count)
  end
end

defmodule ServerExample do
  use GenServer

  def init(_init_arg) do
    {:ok, 0}
  end

  def handle_cast(:increment, state) do
    {:noreply, state + 1}
  end

  def handle_call(:get_count, _from, state) do
    response = state
    {:reply, response, state}
  end
end

Synchronous Vs Asynchronous

Here are some key characteristics of synchronous messages vs asynchronous messages.

  • Synchronous messages:

    • Block the calling process until a response is received.
    • Offer reliability and predictability.
    • Provide a straightforward and ordered flow of communication.
    • Can be slower in terms of performance.
  • Asynchronous messages:

    • Do not block the calling process.
    • Are faster in terms of performance.
    • May introduce unexpected behavior and timing issues.
    • Can result in a less predictable flow of communication.
    • Often referred to as fire-and-forget as they do not care about a response.

    Here’s an example SlowCounter server that takes a second to increment the count. We’re going to use it to demonstrate the difference between synchronous and asynchronous code.

defmodule SlowCounter do
  use GenServer

  def init(_init_arg) do
    {:ok, 0}
  end

  def handle_call(:slow_increment, _from, state) do
    Process.sleep(2000)
    IO.puts("Finished Synch Increment")
    {:reply, state + 1, state + 1}
  end

  def handle_cast(:slow_increment, state) do
    Process.sleep(2000)
    IO.puts("Finished Asynch Increment")
    {:noreply, state + 1}
  end
end

Synchronous messages block the caller process. Evaluate the cell below and notice it takes 2 seconds to finish.

{:ok, pid} = GenServer.start_link(SlowCounter, [])
GenServer.call(pid, :slow_increment)
IO.puts("Evaluated")

Asynchronous messages do not block the caller process. Evaluate the cell below and you’ll notice it evaluates immediately, but prints the “Finished Asynch Increment” message 2 seconds later.

{:ok, pid} = GenServer.start_link(SlowCounter, [])
GenServer.cast(pid, :slow_increment)
IO.puts("Evaluated")

Synchronous Mailbox

Whether synchronous or asynchronous, GenServers handle messages in the order they are received. These messages are stored in a process mailbox until they are handled.

The synchronous mailbox improves the predictability of a GenServer. It’s a major reason for why Elixir is so powerful and simple when building concurrent systems compared to other languages.

Notice that when we call cast/2twice below it takes 4 seconds for the second cast/2 handler to finish. They are not handled concurrently.

{:ok, pid} = GenServer.start_link(SlowCounter, [])
GenServer.cast(pid, :slow_increment)
GenServer.cast(pid, :slow_increment)

Concurrency

To have concurrent code, we simply need two GenServers. Notice both messages below finish at the same time because each is being processes concurrently.

> If you have fewer CPU cores than processes, concurrent code does not run in parallel, so while code gains the benefits of task-switching typically giving the illusion of running in parallel, they will not actually process in parallel at the same time.

{:ok, counter1} = GenServer.start_link(SlowCounter, [])
{:ok, counter2} = GenServer.start_link(SlowCounter, [])

GenServer.cast(counter1, :slow_increment)
GenServer.cast(counter2, :slow_increment)

Regular Messages

You might wonder why we have both handle_cast/2 and handle_info/2 if they both handle asynchronous messages. Here are some key characteristics of each.

  • GenServer.handle_cast/2
    • Non-blocking message handling.
    • Typically only used to update GenServer state.
  • GenServer.handle_info/2
    • Non-blocking message handling
    • Used for a wider variety of messages such as system level behavior, or handling messages that are sent to many different types of processes.
    • Can receive messages sent after an amount of time with Process.send_after/4.

See HexDocs: Receiving Regular Messages for more information.

GenServers Sending Themselves A Message

A GenServer cannot synchronously send itself a message using call/3 because the current message blocks the process mailbox.

defmodule SendingSelfExample do
  use GenServer

  def init(_init_arg) do
    {:ok, nil}
  end

  def handle_call(:talking_to_myself, _from, state) do
    Process.sleep(1000)
    IO.puts("Talking to myself")
    GenServer.call(self(), :talking_to_myself)
    {:reply, "response", state}
  end

  def handle_cast(:talking_to_myself, state) do
    Process.sleep(1000)
    IO.puts("Talking to myself")
    GenServer.cast(self(), :talking_to_myself)
    {:noreply, state}
  end
end

Run the code below and you’ll notice the process crashes with the message process attempted to call itself.

{:ok, pid} = GenServer.start(SendingSelfExample, [])
GenServer.call(pid, :talking_to_myself)

Debugging

We can use :sys.get_state/2 to view the state of a GenServer. This should only be used for debugging purposes. Generally speaking, do not use :sys.get_state/2 to expose the state of a GenServer or for testing.

{:ok, pid} = GenServer.start_link(CounterServer, [])

:sys.get_state(pid)

Named Processes

GenServer.start_link/3 takes additional options as the third argument. We can provide a :name option to name the process. Names are typically atoms or module names (which are just atoms under the hood.)

Named processes are unique, there cannot be two processes with the same name. Named processes are also easy to reference as you can use the name of the process to send them a message.

defmodule NamedCounter do
  def start_link(_opts) do
    GenServer.start_link(NamedCounter, [], name: NamedCounter)
  end

  def init(_init_arg) do
    {:ok, 0}
  end

  def handle_cast(:increment, state) do
    {:noreply, state + 1}
  end
end

Run the cell below multiple times.

Starting an already started named process returns {:error, {:already_started, #PID<0.613.0>}}.

NamedCounter.start_link([])

Most GenServer related functions accept the named process name instead of a pid.

GenServer.cast(NamedCounter, :increment)
:sys.get_state(NamedCounter)

However some do not. of needed we can use Process.whereis/1 to find the pid of a named process.

Process.whereis(NamedCounter)

Configuration

Often we make GenServer modules configurable for different situations.

The start_link/1 function in a GenServer module typically accepts a list of options. The init/1 callback also accepts an argument, typically (but not necessarily) used to configure the initial state.

Here’s an example. There are many ways to make a GenServer configurable depending on your use case.

defmodule ConfigurableServer do
  def start_link(opts) do
    name = Keyword.get(opts, :name)
    initial_state = Keyword.get(opts, :state, 0)

    GenServer.start_link(__MODULE__, initial_state, name: name)
  end

  def init(init_arg) do
    {:ok, init_arg}
  end
end

ConfigurableServer.start_link(name: :example_name, state: 10)

Further Reading

Consider the following resource(s) to deepen your understanding of the topic.

Commit Your Progress

DockYard Academy now recommends you use the latest Release rather than forking or cloning our repository.

Run git status to ensure there are no undesirable changes. Then run the following in your command line from the curriculum folder to commit your progress.

$ git add .
$ git commit -m "finish GenServers reading"
$ git push

We’re proud to offer our open-source curriculum free of charge for anyone to learn from at their own pace.

We also offer a paid course where you can learn from an instructor alongside a cohort of your peers. We will accept applications for the June-August 2023 cohort soon.

Navigation

Home Report An Issue Process MailboxGenServer Drills