TIL GenServer's `handle_continue/2`

15 Feb 2019 · by Sophie DeBenedetto

What happens when starting up your GenServer requires executing a long-running process? We don’t want the execution of that process to block the GenServer from completing start-up. We also don’t want to execute that process asynchronously in a way that creates a race condition between the running of the process and other messages arriving in our GenServer’s inbox. In this post, we’ll take a closer look at these two problems and understand how OTP 21’s GenServer.handle_continue/2 is the perfect solution.

Starting GenServers without Blocking

Let’s say that we are building a GenServer for a shopping list fulfillment application. Our GenServer will hold state describing a grocery shopping list and be aware of the available inventory as it relates to that shopping list. When our GenServer starts up, it will receive a shopping list and put it in state. But wait! Our GenServer initialization process will then need to take that shopping list and retrieve relevant inventory from another source.

Our first attempt at solving this problem could look something like this:

defmodule ShoppingListFulfillment do
  use GenServer

  def start_link(shopping_list) do
    GenServer.start_link(__MODULE__, shopping_list)
  end

  def init(shopping_list) do
    state = %{
      shopping_list: shopping_list
      inventory: get_inventory_for(shopping_list)
    }
    {:ok, state}
  end

  defp get_inventory_for(shopping_list) do
    # something that could be time consuming!
    # like a web request or a database call
    # returns some inventory info for each item on the shopping list
  end
end

Here, we’re calling our “get inventory for shopping list items” code in the init/1 callback that gets triggered when start_link/1 is called.

This problem with this approach is that start_link/1 will block until init/1 returns {:ok, state}. We won’t return from init/1 until after the inventory-fetching code runs. That could be time consuming. We don’t want our GenServer blocked by this.

Let’s explore an asynchronous approach.

Asynchronous Callbacks and Race Conditions

We can use Kernel.send/2 in our init callback to kick off some asynchronous work without blocking GenServer.start_link/1. When we use send/2 and give it a first argument of self, i.e. the PID of our GenServer, our GenServer will handle that message with a handle_info/2 function that matches the message we sent.

defmodule ShoppingListFulfillment do
  use GenServer

  def start_link(shopping_list) do
    GenServer.start_link(__MODULE__, shopping_list)
  end

  def init(shopping_list) do
    state = %{
      shopping_list: shopping_list
      inventory: []
    }
    send(self, :get_inventory)
    {:ok, state}
  end

  def handle_info(:get_inventory, %{shopping_list: shopping_list}) do
    inventory = get_inventory_for(shopping_list)
    state = %{
      shopping_list: shopping_list,
      inventory: inventory
    }
    {:noreply, state}
  end

  defp get_inventory_for(shopping_list) do
    # something that could be time consuming!
    # like a web request or a database call
    # returns some inventory info for each item on the shopping list
  end
end

This approach un-blocks GenServer.start_link/1. It no longer needs to wait for the work of getting inventory. Now that happens asynchronously and updates state once we have finished fetching inventory info.

There is a downside to this approach though. Just because we send the :get_inventory message inside the init function, doesn’t mean that :get_inventory is the first message our GenServer will receive and process. This could lead to a race condition!

What happens if our GenServer receives a message asking for the availability of an item on the shopping list before it receives and finishes processing the message to get the inventory? That could cause a false negative! We would see that inventory in state is empty, and tell the sender that their item is not available. Oh no!

If only there was some way to fetch inventory asynchronously, without blocking start_link/1, while still ensuring that it executes before any other messages received by our GenServer are responded to…

Using handle_continue/2

The release of OTP 21 a few months back gives us a way to solve this problem. The GenServer.handle_continue/2 callback is called by a GenServer process whenever a previous callback returns {:continue, :message}.

handle_continue/2 is invoked immediately after the previous callback, which makes it useful for performing work after initialization or for splitting the work in a callback in multiple steps, updating the process state along the way. *

This approach ensures that our GenServer won’t handle any other messages until handle_continue/2 is finished. No more race conditions!

Let’s take a look:

defmodule ShoppingListFulfillment do
  use GenServer

  def start_link(shopping_list) do
    GenServer.start_link(__MODULE__, shopping_list)
  end

  def init(shopping_list) do
    state = %{
      shopping_list: shopping_list
      inventory: []
    }
    {:ok, state, {:continue, :get_inventory}}
  end

  def handle_continue(:get_inventory, %{shopping_list: shopping_list}) do
    inventory = get_inventory_for(shopping_list)
    state = %{
      shopping_list: shopping_list,
      inventory: inventory
    }
    {:noreply, state}
  end

  defp get_inventory_for(shopping_list) do
    # something that could be time consuming!
    # like a web request or a database call
    # returns some inventory info for each item on the shopping list
  end
end

Now, init/2 returns {:ok, state, {:continue, :get_inventory}}. This immediately triggers the callback handle_continue(:get_inventory, state). This callback is guaranteed to finish running before our GenServer moves on to processing any other messages.

Conclusion

OTP 21’s handle_continue/2 callback allows us to handle expensive GenServer initialization work in a non-blocking, asynchronous manner that avoids race conditions. If you’re building a GenServer that needs to handle an initialization callback, consider reaching for handle_continue/2.

Sophie DeBenedetto

Sophie is an engineer and teacher at The Flatiron School. She loves teaching and learning and finding the Elixir School community felt like the perfect fit!