Cannot wait (with timeout) for expectations

[danielemegna]

# inspired from https://hexdocs.pm/phoenix/testing.html
test "GET /", %{conn: conn} do
  bypass = Bypass.open(port: 4204)

  conn = get(conn, "/welcome")

  assert html_response(conn, 200) =~ "Welcome to Phoenix!"
  Bypass.expect_once(bypass, "POST", "/an/external/service", fn conn ->
    Plug.Conn.resp(conn, 201, "")
  end)
end

This is a nice and clean phoenix integration test. It verifies that, when a GET request hits the /welcome route:

• a 200 response is returned
• an html response is returned with the "Welcome to Phoenix!" body
• an external service is contacted once with a POST on the /an/external/service route

Everything is good, until we face a different (quite common) async scenario.

Let's imagine that our Phoenix application, when it receives the /welcome GET request, it spawns a new process to contact the external service and it response synchronously with the 200 response. Since with ExUnit the verify_expectations is executed with hooks automatically at the end of the test:

• if the the test ends after the external service invocation in the async process, the test goes well
• if the the test ends before the external service invocation in the async process, the expectation fails

  ---

Usually popular http mocking frameworks solve this problem adding some waiting capabilities to the expectations (ie. do not fail immediately, do it if expectations are not satisfied in xxxx seconds). We could add some options as keyword list as last optional argument:

Bypass.expect_once(bypass, "POST", "/an/external/service", fn conn ->
  Plug.Conn.resp(conn, 201, "")
end, timeout: 2_000, polling_interval: 400)

---

What do you think? I would be happy to work on it and I have some ideas:

• check polling bypass instance process (with the provided polling_interval) if the expectations are satisfied
• use message passing and take advantage of receive with timeout otp functionalities
• do not change expectation functions, but provide a separated function to sleep (with timeout) at the end of the test that bypass received some requests (eg. Bypass.wait_to_receive(bypass_instance, request_count: 3, timeout: 2_000))

[Odaeus]

This would be great! I have this problem when testing async work triggered by AMQP messages. My current workaround for this goes like:

test "something" do
  test = self()
  Bypass.expect(bypass, "POST", "/endpoint", fn conn ->
    send(test, :request_done)
    # ...response
  end)

  trigger_work()

  # Wait for the message
  assert_received :request_done, :timer.seconds(3)
end

In case that helps anyone.

EDIT: This technique isn't reliable in all circumstances because the test can still end before the process finishes 😕. Which in my cases leads to a failure as the test is supervising and shuts down the process abnormally. send_after could be used as a band-aid.

[ream88]

I just tripped exactly over the same stuff and was looking through the docs on how to solve this. @danielemegna happy to help if you want to takle this issue!

[mauricius]

@Odaeus I use the same technique, which is actually accepted in the community, cosidering that it's also suggested in the documentation of the Mox library.

In my case I usually do the following:

test "something" do
  parent = self()
  ref = make_ref()

  Bypass.expect(bypass, "POST", "/endpoint", fn conn ->
    # ...response
    send(parent, {ref, :sent})
  end)

  trigger_work()

  assert_receive {^ref, :sent}
end

[cschilbe]

When the external request is not the thing you are testing, you may be interested in using the stub/4 function. It was added specifically for this case.

Added here https://github.com/PSPDFKit-labs/bypass/pull/58

Bypass.stub(bypass, "POST", "/1.1/statuses/update.json", fn conn ->
  Agent.get_and_update(AgentModule, fn step_no -> {step_no, step_no + 1} end)
  Plug.Conn.resp(conn, 429, ~s<{"errors": [{"code": 88, "message": "Rate limit exceeded"}]}>)
end)