[RTC-454] Healthcheck endpoint (#143)

* Initial healthcheck endpoint implementation
* Add distribution info
* Add test

Dockerfile

@@ -126,7 +126,7 @@ RUN chmod +x docker-entrypoint.sh
 
 ENV HOME=/app
 
-HEALTHCHECK CMD curl --fail -H "authorization: Bearer ${JF_SERVER_API_TOKEN}" http://localhost:${JF_PORT:-8080}/room || exit 1
+HEALTHCHECK CMD curl --fail -H "authorization: Bearer ${JF_SERVER_API_TOKEN}" http://localhost:${JF_PORT:-8080}/health || exit 1
 
 ENTRYPOINT ["./docker-entrypoint.sh"]
 

lib/jellyfish/application.ex

@@ -48,6 +48,8 @@ defmodule Jellyfish.Application do
     # for other strategies and supported options
     opts = [strategy: :one_for_one, name: Jellyfish.Supervisor]
 
+    Application.put_env(:jellyfish, :start_time, System.monotonic_time(:second))
+
     Logger.info("Starting Jellyfish v#{@version}")
     Logger.info("Distribution config: #{inspect(Keyword.delete(dist_config, :cookie))}")
     Logger.info("WebRTC config: #{inspect(webrtc_config)}")

lib/jellyfish_web/api_spec/health_report.ex

@@ -0,0 +1,57 @@
+defmodule JellyfishWeb.ApiSpec.HealthReport do
+  @moduledoc false
+
+  require OpenApiSpex
+  alias OpenApiSpex.Schema
+
+  defmodule Status do
+    @moduledoc false
+
+    require OpenApiSpex
+
+    OpenApiSpex.schema(%{
+      title: "HealthReportStatus",
+      description: "Informs about the status of Jellyfish or a specific service",
+      type: :string,
+      enum: ["UP", "DOWN"],
+      example: "UP"
+    })
+  end
+
+  defmodule Distribution do
+    @moduledoc false
+
+    require OpenApiSpex
+
+    OpenApiSpex.schema(%{
+      title: "HealthReportDistribution",
+      description: "Informs about the status of Jellyfish distribution",
+      type: :object,
+      properties: %{
+        enabled: %Schema{
+          type: :boolean,
+          description: "Whether distribution is enabled on this Jellyfish"
+        },
+        nodeStatus: Status,
+        nodesInCluster: %Schema{
+          type: :integer,
+          description:
+            "Amount of nodes (including this Jellyfish's node) in the distribution cluster"
+        }
+      },
+      required: [:nodeStatus, :nodesInCluster]
+    })
+  end
+
+  OpenApiSpex.schema(%{
+    title: "HealthReport",
+    description: "Describes overall Jellyfish health",
+    type: :object,
+    properties: %{
+      status: Status,
+      uptime: %Schema{type: :integer, description: "Uptime of Jellyfish (in seconds)"},
+      distribution: Distribution
+    },
+    required: [:status, :uptime, :distribution]
+  })
+end

lib/jellyfish_web/api_spec/responses.ex

@@ -25,7 +25,9 @@
   {RoomsListingResponse, "Response containing list of all rooms",
    %OpenApiSpex.Schema{type: :array, items: JellyfishWeb.ApiSpec.Room}},
   {RecordingListResponse, "Response containing list of all recording",
-   %OpenApiSpex.Schema{type: :array, items: %OpenApiSpex.Schema{type: :string}}}
+   %OpenApiSpex.Schema{type: :array, items: %OpenApiSpex.Schema{type: :string}}},
+  {HealthcheckResponse, "Response containing health report of Jellyfish",
+   JellyfishWeb.ApiSpec.HealthReport}
 ]
 |> Enum.map(fn {title, description, schema} ->
   module = Module.concat(JellyfishWeb.ApiSpec, title)

lib/jellyfish_web/controllers/healthcheck_controller.ex

@@ -0,0 +1,47 @@
+defmodule JellyfishWeb.HealthcheckController do
+  use JellyfishWeb, :controller
+  use OpenApiSpex.ControllerSpecs
+
+  alias JellyfishWeb.ApiSpec
+
+  action_fallback JellyfishWeb.FallbackController
+
+  operation :show,
+    operation_id: "healthcheck",
+    summary: "Describes the health of Jellyfish",
+    responses: [
+      ok: ApiSpec.data("Healthy", ApiSpec.HealthcheckResponse),
+      internal_server_error: ApiSpec.data("Unhealthy", ApiSpec.HealthcheckResponse)
+    ]
+
+  def show(conn, _params) do
+    report = get_health_report()
+
+    conn
+    |> put_resp_content_type("application/json")
+    |> render("show.json", report: report)
+  end
+
+  defp get_health_report() do
+    %{
+      status: :up,
+      uptime: get_uptime(),
+      distribution: get_distribution_report()
+    }
+  end
+
+  defp get_uptime() do
+    System.monotonic_time(:second) - Application.fetch_env!(:jellyfish, :start_time)
+  end
+
+  defp get_distribution_report() do
+    alive? = Node.alive?()
+    visible_nodes = Node.list() |> length()
+
+    %{
+      enabled: Application.fetch_env!(:jellyfish, :dist_config)[:enabled],
+      node_status: if(alive?, do: :up, else: :down),
+      nodes_in_cluster: visible_nodes + if(alive?, do: 1, else: 0)
+    }
+  end
+end

lib/jellyfish_web/controllers/healthcheck_json.ex

@@ -0,0 +1,22 @@
+defmodule JellyfishWeb.HealthcheckJSON do
+  @moduledoc false
+
+  def show(%{report: report}) do
+    %{data: data(report)}
+  end
+
+  def data(%{status: status, uptime: uptime, distribution: distribution}) do
+    %{
+      status: status_str(status),
+      uptime: uptime,
+      distribution: %{
+        enabled: distribution.enabled,
+        nodeStatus: status_str(distribution.node_status),
+        nodesInCluster: distribution.nodes_in_cluster
+      }
+    }
+  end
+
+  defp status_str(:up), do: "UP"
+  defp status_str(:down), do: "DOWN"
+end

lib/jellyfish_web/router.ex

@@ -6,6 +6,12 @@ defmodule JellyfishWeb.Router do
     plug :bearer_auth
   end
 
+  scope "/health", JellyfishWeb do
+    pipe_through :api
+
+    get "/", HealthcheckController, :show
+  end
+
   scope "/room", JellyfishWeb do
     pipe_through :api
 

openapi.yaml

@@ -14,6 +14,15 @@ components:
       title: PeerStatus
       type: string
       x-struct: Elixir.JellyfishWeb.ApiSpec.Peer.Status
+    HealthReportStatus:
+      description: Informs about the status of Jellyfish or a specific service
+      enum:
+        - UP
+        - DOWN
+      example: UP
+      title: HealthReportStatus
+      type: string
+      x-struct: Elixir.JellyfishWeb.ApiSpec.HealthReport.Status
     RoomDetailsResponse:
       description: Response containing room details
       properties:
@@ -139,6 +148,16 @@ components:
       title: PeerOptionsWebRTC
       type: object
       x-struct: Elixir.JellyfishWeb.ApiSpec.Peer.WebRTC
+    HealthcheckResponse:
+      description: Response containing health report of Jellyfish
+      properties:
+        data:
+          $ref: '#/components/schemas/HealthReport'
+      required:
+        - data
+      title: HealthcheckResponse
+      type: object
+      x-struct: Elixir.JellyfishWeb.ApiSpec.HealthcheckResponse
     ComponentHLS:
       description: Describes the HLS component
       properties:
@@ -334,6 +353,23 @@ components:
       title: ComponentOptions
       type: object
       x-struct: Elixir.JellyfishWeb.ApiSpec.Component.Options
+    HealthReportDistribution:
+      description: Informs about the status of Jellyfish distribution
+      properties:
+        enabled:
+          description: Whether distribution is enabled on this Jellyfish
+          type: boolean
+        nodeStatus:
+          $ref: '#/components/schemas/HealthReportStatus'
+        nodesInCluster:
+          description: Amount of nodes (including this Jellyfish's node) in the distribution cluster
+          type: integer
+      required:
+        - nodeStatus
+        - nodesInCluster
+      title: HealthReportDistribution
+      type: object
+      x-struct: Elixir.JellyfishWeb.ApiSpec.HealthReport.Distribution
     ComponentPropertiesRTSP:
       description: Properties specific to the RTSP component
       properties:
@@ -368,6 +404,23 @@ components:
       title: PeerOptions
       type: object
       x-struct: Elixir.JellyfishWeb.ApiSpec.Peer.Options
+    HealthReport:
+      description: Describes overall Jellyfish health
+      properties:
+        distribution:
+          $ref: '#/components/schemas/HealthReportDistribution'
+        status:
+          $ref: '#/components/schemas/HealthReportStatus'
+        uptime:
+          description: Uptime of Jellyfish (in seconds)
+          type: integer
+      required:
+        - status
+        - uptime
+        - distribution
+      title: HealthReport
+      type: object
+      x-struct: Elixir.JellyfishWeb.ApiSpec.HealthReport
     Peer:
       description: Describes peer status
       properties:
@@ -528,6 +581,26 @@ info:
   version: 0.2.0
 openapi: 3.0.0
 paths:
+  /health:
+    get:
+      callbacks: {}
+      operationId: healthcheck
+      parameters: []
+      responses:
+        '200':
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/HealthcheckResponse'
+          description: Healthy
+        '500':
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/HealthcheckResponse'
+          description: Unhealthy
+      summary: Describes the health of Jellyfish
+      tags: []
   /hls/{room_id}/subscribe:
     post:
       callbacks: {}

test/jellyfish_web/controllers/healthcheck_controller_test.exs

@@ -0,0 +1,30 @@
+defmodule JellyfishWeb.HealthcheckControllerTest do
+  use JellyfishWeb.ConnCase, async: true
+
+  import OpenApiSpex.TestAssertions
+
+  @schema JellyfishWeb.ApiSpec.spec()
+
+  setup %{conn: conn} do
+    server_api_token = Application.fetch_env!(:jellyfish, :server_api_token)
+    conn = put_req_header(conn, "authorization", "Bearer " <> server_api_token)
+
+    [conn: conn]
+  end
+
+  test "healthcheck without distribution", %{conn: conn} do
+    conn = get(conn, ~p"/health")
+    response = json_response(conn, :ok)
+    assert_response_schema(response, "HealthcheckResponse", @schema)
+
+    assert %{
+             "status" => "UP",
+             "uptime" => _uptime,
+             "distribution" => %{
+               "enabled" => false,
+               "nodeStatus" => "DOWN",
+               "nodesInCluster" => 0
+             }
+           } = response["data"]
+  end
+end