Add healthcheck endpoint

* Added healthcheck controller
* Added tests
* Created documentation for this endpoint
* Updated Sparrow dependency

README.md

@@ -499,6 +499,33 @@ If you use dockerized MongoosePush, you need to do the following:
 * Modify the `sys.config` as you see fit (for metrics, see above)
 * Stop MongoosePush docker container and restart it with the modified `sys.config` as volume in `/opt/app/sys.config` (yes, this is not the path we used to copy this file from, this is an override)
 
+### Healthcheck
+
+MongoosePush exposes `/healthcheck` endpoint, from which you can get information about current status of connections in a `JSON` format, grouped by connection pool. An example with 2 pools, one being connected to the service and the other one not would look like this:
+
+```json
+[
+{
+  "pool": "pool_name1",
+  "connection_status":
+    [
+      "connected",
+      "connected"
+    ]
+},
+{
+  "pool": "pool_name2",
+  "connection_status":
+  [
+    "disconnected",
+    "disconnected",
+    "disconnected"
+  ]
+}
+]
+```
+
+Please note that it's not recommended to use this frequently as it puts an extra load on the worker processes.
 
 ### Available metrics
 

lib/mongoose_push_web/controllers/healthcheck_controller.ex

@@ -0,0 +1,58 @@
+defmodule MongoosePushWeb.HealthcheckController do
+  use MongoosePushWeb, :controller
+
+  def send(conn = %Plug.Conn{}, %{}) do
+    stats = :wpool.stats()
+
+    payload =
+      stats
+      |> Enum.map(&extract_connection_info_from_pool/1)
+
+    status = get_status(payload)
+
+    conn
+    |> put_status(status)
+    |> json(payload)
+  end
+
+  defp extract_connection_info_from_pool(pool) do
+    pool_pid = pool[:supervisor]
+    children = Supervisor.which_children(pool_pid)
+    {_, sup_pid, _, _} = List.keyfind(children, [:wpool_process_sup], 3)
+    workers = Supervisor.which_children(sup_pid)
+
+    connections =
+      workers
+      |> Enum.map(fn worker_info ->
+        {_, worker_pid, _, _} = worker_info
+
+        if Sparrow.H2Worker.is_alive_connection(worker_pid) do
+          :connected
+        else
+          :disconnected
+        end
+      end)
+
+    %{
+      pool: pool[:pool],
+      connection_status: connections
+    }
+  end
+
+  defp get_status(connections) do
+    is_everything_disconnected =
+      Enum.all?(connections, fn pool_info ->
+        Enum.all?(pool_info[:connection_status], fn
+          status -> status == :disconnected
+        end)
+      end)
+
+    case is_everything_disconnected do
+      true ->
+        503
+
+      false ->
+        200
+    end
+  end
+end

lib/mongoose_push_web/router.ex

@@ -22,6 +22,7 @@ defmodule MongoosePushWeb.Router do
     pipe_through(:api)
 
     get("/metrics", MongoosePushWeb.PrometheusMetricsController, :send)
+    get("/healthcheck", MongoosePushWeb.HealthcheckController, :send)
   end
 
   scope "/v1", MongoosePushWeb.APIv1 do

mix.exs

@@ -31,7 +31,7 @@ defmodule MongoosePush.Mixfile do
   defp deps do
     [
       {:chatterbox, github: "joedevivo/chatterbox", ref: "1f4ce4f", override: true},
-      {:sparrow, github: "esl/sparrow", ref: "b1896ca"},
+      {:sparrow, github: "esl/sparrow", ref: "80a17bd"},
       {:plug_cowboy, "~> 2.0"},
       {:cowboy, "~> 2.3", override: true},
       {:jason, "~> 1.0"},

mix.lock

@@ -48,7 +48,7 @@
   "pollution": {:hex, :pollution, "0.9.2", "3f67542631071c99f807d2a8f9da799c07cd983c902f5357b9e1569c20a26e76", [:mix], [], "hexpm", "6399fd8ffd97dcc3d9d277f60542a234d644d7bcc0d48c8fda93d6be4801bac2"},
   "quixir": {:hex, :quixir, "0.9.3", "f01c37386b9e1d0526f01a8734a6d7884af294a0ec360f05c24c7171d74632bd", [:mix], [{:pollution, "~> 0.9.2", [hex: :pollution, repo: "hexpm", optional: false]}], "hexpm", "4f3a1fe7c82b767d935b3f7b94cf34b91ef78bb487ef256b303d77417fc7d589"},
   "ranch": {:hex, :ranch, "1.7.1", "6b1fab51b49196860b733a49c07604465a47bdb78aa10c1c16a3d199f7f8c881", [:rebar3], [], "hexpm", "451d8527787df716d99dc36162fca05934915db0b6141bbdac2ea8d3c7afc7d7"},
-  "sparrow": {:git, "https://github.com/esl/sparrow.git", "b1896ca4fb0ca18369dd62de3d4c82f14f5bc66e", [ref: "b1896ca"]},
+  "sparrow": {:git, "https://github.com/esl/sparrow.git", "80a17bdc9a2289229a646eb0bd26a162b39cbf83", [ref: "80a17bd"]},
   "ssl_verify_fun": {:hex, :ssl_verify_fun, "1.1.5", "6eaf7ad16cb568bb01753dbbd7a95ff8b91c7979482b95f38443fe2c8852a79b", [:make, :mix, :rebar3], [], "hexpm", "13104d7897e38ed7f044c4de953a6c28597d1c952075eb2e328bc6d6f2bfc496"},
   "telemetry": {:hex, :telemetry, "0.4.1", "ae2718484892448a24470e6aa341bc847c3277bfb8d4e9289f7474d752c09c7f", [:rebar3], [], "hexpm", "4738382e36a0a9a2b6e25d67c960e40e1a2c95560b9f936d8e29de8cd858480f"},
   "telemetry_metrics": {:hex, :telemetry_metrics, "0.4.2", "1de986fad9aa6bf81f8a33ddfd16e5d8ab0dec6272e624eb517c1a92a44d41a9", [:mix], [{:telemetry, "~> 0.4", [hex: :telemetry, repo: "hexpm", optional: false]}], "hexpm", "e56ffed2dbe293ab6cf7c94980faeb368cb360662c1927f54fc634a4ca55362e"},

test/support/api.ex

@@ -30,7 +30,7 @@ defmodule MongoosePush.Support.API do
 
   def get(path) do
     %Response{status_code: status, headers: headers, body: body} =
-      HTTPoison.get!("https://localhost:8443" <> path)
+      HTTPoison.get!("https://localhost:8443" <> path, [], hackney: [:insecure])
 
     {status, headers, body}
   end

test/unit/healthcheck_test.exs

@@ -0,0 +1,159 @@
+defmodule MongoosePushWeb.HealthcheckTest do
+  alias MongoosePush.Support.API
+  use ExUnit.Case, async: false
+  import Mox
+
+  setup :verify_on_exit!
+
+  setup do
+    TestHelper.reload_app()
+  end
+
+  test "Successful connection to services" do
+    {200, _, body} = API.get("/healthcheck")
+    pools = Jason.decode!(body)
+
+    fcm_pools =
+      :mongoose_push
+      |> Application.get_env(:fcm)
+      |> Enum.map(fn {name, info} ->
+        {name, Keyword.get(info, :pool_size)}
+      end)
+
+    apns_pools =
+      :mongoose_push
+      |> Application.get_env(:apns)
+      |> Enum.map(fn {name, info} ->
+        {name, Keyword.get(info, :pool_size)}
+      end)
+
+    for {pool_name, worker_count} <- fcm_pools ++ apns_pools do
+      pool_info = %{
+        "pool" => Atom.to_string(pool_name),
+        "connection_status" => List.duplicate("connected", worker_count)
+      }
+
+      assert true == Enum.member?(pools, pool_info)
+    end
+  end
+
+  describe "Unsuccessful FCM connection" do
+    setup do
+      old_config = Application.fetch_env!(:mongoose_push, :fcm)
+
+      new_config =
+        old_config
+        |> Enum.map(fn {pool_name, pool_info} ->
+          # We simulate broken FCM connection by changing the port
+          {pool_name, Keyword.update(pool_info, :port, 4444, &(&1 + 1))}
+        end)
+
+      Application.stop(:mongoose_push)
+      Application.load(:mongoose_push)
+      Application.put_env(:mongoose_push, :fcm, new_config)
+      Application.start(:mongoose_push)
+    end
+
+    test "is reflected in heathcheck endpoint" do
+      {200, _, body} = API.get("/healthcheck")
+      pools = Jason.decode!(body)
+
+      fcm_pools =
+        :mongoose_push
+        |> Application.get_env(:fcm)
+        |> Enum.map(fn {name, info} ->
+          {name, Keyword.get(info, :pool_size)}
+        end)
+
+      apns_pools =
+        :mongoose_push
+        |> Application.get_env(:apns)
+        |> Enum.map(fn {name, info} ->
+          {name, Keyword.get(info, :pool_size)}
+        end)
+
+      for {pool_name, worker_count} <- apns_pools do
+        pool_info = %{
+          "pool" => Atom.to_string(pool_name),
+          "connection_status" => List.duplicate("connected", worker_count)
+        }
+
+        assert true == Enum.member?(pools, pool_info)
+      end
+
+      for {pool_name, worker_count} <- fcm_pools do
+        pool_info = %{
+          "pool" => Atom.to_string(pool_name),
+          "connection_status" => List.duplicate("disconnected", worker_count)
+        }
+
+        assert true == Enum.member?(pools, pool_info)
+      end
+    end
+  end
+
+  describe "Unsuccessful APNS and FCM connection" do
+    setup do
+      old_fcm_config = Application.fetch_env!(:mongoose_push, :fcm)
+
+      new_fcm_config =
+        old_fcm_config
+        |> Enum.map(fn {pool_name, pool_info} ->
+          # We simulate broken FCM connection by changing the port
+          {pool_name, Keyword.update(pool_info, :port, 4444, &(&1 + 1))}
+        end)
+
+      old_apns_config = Application.fetch_env!(:mongoose_push, :apns)
+
+      new_apns_config =
+        old_apns_config
+        |> Enum.map(fn {pool_name, pool_info} ->
+          # We simulate broken APNS connection by changing the port
+          {pool_name, Keyword.update(pool_info, :use_2197, false, &(!&1))}
+        end)
+
+      Application.stop(:mongoose_push)
+      Application.load(:mongoose_push)
+      Application.put_env(:mongoose_push, :fcm, new_fcm_config)
+      Application.put_env(:mongoose_push, :apns, new_apns_config)
+      Application.start(:mongoose_push)
+    end
+
+    test "is reflected in heathcheck endpoint" do
+      {503, _, body} = API.get("/healthcheck")
+      pools = Jason.decode!(body)
+
+      fcm_pools =
+        :mongoose_push
+        |> Application.get_env(:fcm)
+        |> Enum.map(fn {name, info} ->
+          {name, Keyword.get(info, :pool_size)}
+        end)
+
+      apns_pools =
+        :mongoose_push
+        |> Application.get_env(:apns)
+        |> Enum.map(fn {name, info} ->
+          {name, Keyword.get(info, :pool_size)}
+        end)
+
+      for {pool_name, worker_count} <- apns_pools do
+        pool_info = %{
+          "pool" => Atom.to_string(pool_name),
+          "connection_status" => List.duplicate("disconnected", worker_count)
+        }
+
+        assert true == Enum.member?(pools, pool_info)
+      end
+
+      for {pool_name, worker_count} <- fcm_pools do
+        pool_info = %{
+          "pool" => Atom.to_string(pool_name),
+          "connection_status" => List.duplicate("disconnected", worker_count)
+        }
+
+        assert true == Enum.member?(pools, pool_info)
+      end
+    end
+  end
+end