- + - + - + - + A x

Playwright and Phoenix Ecto SQL Sandbox

06 Nov 2021

Playwright is a library for browser automation from Microsoft. There’s an unofficial playwright-elixir library for using it from Elixir. You can think of it as an alternative to Wallaby or Hound, but using different technology underneath. The Elixir client library is not complete yet and is still under development, but let’s make it work with the Phoenix Ecto sandbox!

Why Playwright

Playwright docs have a page on that topic. I think Playwright makes a lot of sense from design perspective. It uses snappier websockets compared to WebDriver’s HTTP based API. It started as a fork of puppeteer but is tailored more for testing. And because the Playwright team treats it as a complete solution for testing, it’s not limited by cross-company politics, so they have more opportunities to optimize both the technicalities and overall user experience.

Playwright supports testing against Firefox and WebKit (closest thing to Safari we can run on CI) while unifying the behavior of testing commands. WebDriver protocol really is underspecified and/or the respective implementations from browser makers are not their priority. Playwright’s solution is to bundle compatible and tested versions of browsers with the library. Thank you Microsoft for the effort to make this work!


We’ll be using a new, freshly generated Phoenix 1.6.2 app, and Playwright 1.16. After generating a new project with mix phx.new hello and using mix phx.gen.auth (to have some database data to show, in this case a listing of users), we can install Playwright using npm (this also downloads the browsers compatible with this version of Playwright and ffmpeg for video recording):

$ cd assets/ && npm i --save-dev playwright@1.16 && cd -

> playwright@1.16.3 install /home/michal/projects/hello/assets/node_modules/playwright
> node install.js

Downloading Playwright build of chromium v930007 - 127.5 Mb
Downloading Playwright build of ffmpeg v1006 - 2.6 Mb
Downloading Playwright build of firefox v1297 - 72.1 Mb
Downloading Playwright build of webkit v1564 - 79.3 Mb

+ playwright@1.16.3
added 46 packages from 85 contributors and audited 46 packages in 34.17s

Next, let’s add playwright-elixir to our dependencies:

  def deps do
      {:playwright, "~> 0.1.16-preview-1", only: :test}

To confirm that the sandbox is working, we can add a listing of all users to the index page:

1 2 3 4 5 6 7 8 9 10 11 12
<section id="users" class="row"> <article class="column"> <h2>Users</h2> <ul> <%= for user <- Hello.Repo.all(Hello.Accounts.User) do %> <li> <%= user.email %> created at <%= inspect(user.inserted_at) %> </li> <% end %> </ul> </article> </section>

And to confirm that playwright-elixir is correctly set up, we can add a very simple test that accesses our app:

1 2 3 4 5 6 7 8 9 10 11 12 13
defmodule HelloWeb.Integration.ATest do use Hello.DataCase, async: false use PlaywrightTest.Case, headless: false test "navigating to our app", %{page: page} do text = page |> Playwright.Page.goto("http://localhost:4002") |> Playwright.Page.text_content("#users h2") assert text == "Users" end end

To access the server in tests we need to change the Endpoint config in config/test.exs to say server: true. Then, the test should pass, and it will cause the browser to shortly show up (thanks to the headless: false option we passed in):

$ mix test test/hello_web/integration/

Finished in 1.7 seconds (0.00s async, 1.7s sync)
1 test, 0 failures

Testing registration

To see that sandboxing works as intended, we can write a test that registers a user, waits a few seconds, refreshes the page and confirms there is only one user registered.

test "registering a user", %{page: page} do
  email = Hello.AccountsFixtures.unique_user_email()
  password = Hello.AccountsFixtures.valid_user_password()

  |> Playwright.Page.goto("http://localhost:4002/users/register")
  |> Playwright.Page.fill("#user_email", email)
  |> Playwright.Page.fill("#user_password", password)
  |> Playwright.Page.click("button[type='submit']")

  Playwright.Page.goto(page, "http://localhost:4002")


  users = page |> Playwright.Page.query_selector_all("#users li")

  assert Enum.count(users) == 1

Let’s save that test in test/hello_web/integration/a_test.exs, but also create a second test file at test/hello_web/integration/b_test.exs, copy the registration test there, and mark both of them as async:true, so that they can execute in parallel.

defmodule HelloWeb.Integration.BTest do
  use Hello.DataCase, async: true
  use PlaywrightTest.Case, headless: false

When two such tests execute at the same time, we want the assertion to pass, and we want to see two browser windows with a different list of users each. But currently that’s not what’s happening:

$ mix test test/hello_web/integration/
14:46:54.855 [error] #PID<0.631.0> running HelloWeb.Endpoint (connection #PID<0.625.0>, stream id 3) terminated
Server: localhost:4002 (http)
Request: POST /users/register
** (exit) an exception was raised:
    ** (DBConnection.OwnershipError) cannot find ownership process for #PID<0.631.0>.
Finished in 9.2 seconds (9.2s async, 0.00s sync)
2 tests, 2 failures

We can fix that using the Ecto Sandbox!

Phoenix Ecto Sandbox configuration

Phoenix needs some information to identify the test that browser requests originate from. We need to make sure that Playwright passes that information. This is usually done in the user-agent header, but we’ll use a separate x-phoenix-ecto-sandbox header.

Let’s create our own Hello.PlaywrightCase module. We can copy and modify the PlaywrightTest.Case module that playwright-elixir provides. Please have in mind that the playwright-elixir libary is still in early stages of development, so this will look cleaner in the future (what you see below is mostly a copy-pasta of library code with a few changes):

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96
defmodule Hello.PlaywrightCase do use ExUnit.CaseTemplate defmacro __using__(options \\ %{}) do case_using = super(options) quote do unquote(case_using) alias Playwright.Runner.Config setup_all do inline_options = unquote(options) |> Enum.into(%{}) launch_options = Map.merge(Config.launch_options(), inline_options) runner_options = Map.merge(Config.playwright_test(), inline_options) Application.put_env(:playwright, LaunchOptions, launch_options) {:ok, _} = Application.ensure_all_started(:playwright) case runner_options.transport do :driver -> {connection, browser} = Playwright.BrowserType.launch() [ connection: connection, browser: browser, transport: :driver ] :websocket -> options = Config.connect_options() {connection, browser} = Playwright.BrowserType.connect(options.ws_endpoint) [ connection: connection, browser: browser, transport: :websocket ] end end setup ctx do pid = Ecto.Adapters.SQL.Sandbox.start_owner!(Hello.Repo, shared: not ctx[:async]) metadata = Phoenix.Ecto.SQL.Sandbox.metadata_for(Hello.Repo, pid) context = Hello.PlaywrightCase.new_context(ctx[:browser], metadata) page = Hello.PlaywrightCase.new_page(context) on_exit(:ok, fn -> Playwright.Page.close(page) Ecto.Adapters.SQL.Sandbox.stop_owner(pid) end) [page: page] end end end # Copied and modified from Playwright.Browser.new_context # to pass sandbox metadata in an extra header with each browser request def new_context(%Playwright.Browser{connection: connection} = subject, metadata) do sandbox_header = Phoenix.Ecto.SQL.Sandbox.encode_metadata(metadata) context = Playwright.Runner.Channel.send(subject, "newContext", %{ noDefaultViewport: false, sdkLanguage: "elixir", extraHTTPHeaders: [%{name: "x-phoenix-ecto-sandbox", value: sandbox_header}] }) case context do %Playwright.BrowserContext{} -> Playwright.Runner.Channel.patch(connection, context.guid, %{browser: subject}) _other -> raise( "expected new_context to return a Playwright.BrowserContext, received: #{inspect(context)}" ) end end # Copied and modified from Playwright.Browser.new_page # so that we don't start a new context for each page def new_page(%{connection: connection} = context) do page = Playwright.BrowserContext.new_page(context) Playwright.Runner.Channel.patch(connection, context.guid, %{owner_page: page}) case page do %Playwright.Page{} -> Playwright.Runner.Channel.patch(connection, page.guid, %{owned_context: context}) _other -> raise("expected new_page to return a Playwright.Page, received: #{inspect(page)}") end end end

Having that set up, we can modify our tests to use our Case module:

defmodule HelloWeb.Integration.ATest do
  use Hello.PlaywrightCase, async: true, headless: false

We add appropriate config to config/test.exs:

config :hello, sql_sandbox: true

And conditionally add the Phoenix.Ecto.SQL.Sandbox plug in lib/hello_web/endpoint.ex:

  if Application.get_env(:hello, :sql_sandbox) do
    plug Phoenix.Ecto.SQL.Sandbox, header: "x-phoenix-ecto-sandbox"

For applications that use Phoenix Channels or LiveViews, we’d need to pass the metadata header in Phoenix socket assigns, and call Sandbox.allow when initializing them. Follow Phoenix.Ecto.SQL.Sandbox docs to get that set up. For this example with only one “dead view”, we’ll only need the plug.

Now when executing the tests we see that they pass:

$ mix test test/hello_web/integration/

Finished in 9.2 seconds (9.2s async, 0.00s sync)
2 tests, 0 failures

Both tests created a separate user, and listing all users did not show the user created in the other test.

Closing thoughts

Overall, I am very excited to use Playwright-based browser automation for tests instead of WebDriver-based solutions. Playwright is snappier and more reliable - it’s just overall better! I hope to find the time to contribute to the Elixir client a bit more.