bitwalker · svsool · Sep 10, 2024 · Sep 17, 2024
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -8,6 +8,12 @@
 - Remove calls to deprecated `Logger.warn/2`
 - Correct misspell of 'Empd' -> 'Epmd' in `Cluster.Strategy.LocalEpmd` moduledoc
 
+## 3.4.0
+
+### Added
+
+- Telemetry events added for tracking node connects and disconnects
+
 ### 3.3.0
 
 ### Changed

diff --git a/lib/strategy/strategy.ex b/lib/strategy/strategy.ex
@@ -50,16 +50,36 @@ defmodule Cluster.Strategy do
         fargs = connect_args ++ [n]
         ensure_exported!(connect_mod, connect_fun, length(fargs))
 
+        start = System.monotonic_time()
+
         case apply(connect_mod, connect_fun, fargs) do
           true ->
+            :telemetry.execute(
+              [:libcluster, :connect_node, :ok],
+              %{duration: System.monotonic_time() - start},
+              %{node: n, topology: topology}
+            )
+
             Cluster.Logger.info(topology, "connected to #{inspect(n)}")
             acc
 
           false ->
+            :telemetry.execute(
+              [:libcluster, :connect_node, :error],
+              %{},
+              %{node: n, topology: topology, reason: :unreachable}
+            )
+
             Cluster.Logger.warn(topology, "unable to connect to #{inspect(n)}")
             [{n, false} | acc]
 
           :ignored ->
+            :telemetry.execute(
+              [:libcluster, :connect_node, :error],
+              %{},
+              %{node: n, topology: topology, reason: :not_part_of_network}
+            )
+
             Cluster.Logger.warn(
               topology,
               "unable to connect to #{inspect(n)}: not part of network"
@@ -100,12 +120,26 @@ defmodule Cluster.Strategy do
         fargs = disconnect_args ++ [n]
         ensure_exported!(disconnect_mod, disconnect_fun, length(fargs))
 
+        start = System.monotonic_time()
+
         case apply(disconnect_mod, disconnect_fun, fargs) do
           true ->
+            :telemetry.execute(
+              [:libcluster, :disconnect_node, :ok],
+              %{duration: System.monotonic_time() - start},
+              %{node: n, topology: topology}
+            )
+
             Cluster.Logger.info(topology, "disconnected from #{inspect(n)}")
             acc
 
           false ->
+            :telemetry.execute(
+              [:libcluster, :disconnect_node, :error],
+              %{},
+              %{node: n, topology: topology, reason: :already_disconnected}
+            )
+
             Cluster.Logger.warn(
               topology,
               "disconnect from #{inspect(n)} failed because we're already disconnected"
@@ -114,6 +148,12 @@ defmodule Cluster.Strategy do
             acc
 
           :ignored ->
+            :telemetry.execute(
+              [:libcluster, :disconnect_node, :error],
+              %{},
+              %{node: n, topology: topology, reason: :not_part_of_network}
+            )
+
             Cluster.Logger.warn(
               topology,
               "disconnect from #{inspect(n)} failed because it is not part of the network"
@@ -122,6 +162,12 @@ defmodule Cluster.Strategy do
             acc
 
           reason ->
+            :telemetry.execute(
+              [:libcluster, :disconnect_node, :error],
+              %{},
+              %{node: n, topology: topology, reason: inspect(reason)}
+            )
+
             Cluster.Logger.warn(
               topology,
               "disconnect from #{inspect(n)} failed with: #{inspect(reason)}"

diff --git a/mix.exs b/mix.exs
@@ -1,7 +1,7 @@
 defmodule Cluster.Mixfile do
   use Mix.Project
 
-  @version "3.3.3"
+  @version "3.4.0"
   @source_url "https://github.com/bitwalker/libcluster"
 
   def project do
@@ -40,7 +40,8 @@ defmodule Cluster.Mixfile do
       {:ex_doc, ">= 0.0.0", only: :dev, runtime: false},
       {:dialyxir, "~> 1.0", only: :dev, runtime: false},
       {:exvcr, "~> 0.11", only: :test, runtime: false},
-      {:jason, "~> 1.1"}
+      {:jason, "~> 1.1"},
+      {:telemetry, "~> 1.3"}
     ]
   end
 

diff --git a/mix.lock b/mix.lock
@@ -13,4 +13,5 @@
   "makeup_erlang": {:hex, :makeup_erlang, "1.0.1", "c7f58c120b2b5aa5fd80d540a89fdf866ed42f1f3994e4fe189abebeab610839", [:mix], [{:makeup, "~> 1.0", [hex: :makeup, repo: "hexpm", optional: false]}], "hexpm", "8a89a1eeccc2d798d6ea15496a6e4870b75e014d1af514b1b71fa33134f57814"},
   "meck": {:hex, :meck, "0.9.2", "85ccbab053f1db86c7ca240e9fc718170ee5bda03810a6292b5306bf31bae5f5", [:rebar3], [], "hexpm", "81344f561357dc40a8344afa53767c32669153355b626ea9fcbc8da6b3045826"},
   "nimble_parsec": {:hex, :nimble_parsec, "1.4.0", "51f9b613ea62cfa97b25ccc2c1b4216e81df970acd8e16e8d1bdc58fef21370d", [:mix], [], "hexpm", "9c565862810fb383e9838c1dd2d7d2c437b3d13b267414ba6af33e50d2d1cf28"},
+  "telemetry": {:hex, :telemetry, "1.3.0", "fedebbae410d715cf8e7062c96a1ef32ec22e764197f70cda73d82778d61e7a2", [:rebar3], [], "hexpm", "7015fc8919dbe63764f4b4b87a95b7c0996bd539e0d499be6ec9d7f3875b79e6"},
 }
diff --git a/test/strategy_test.exs b/test/strategy_test.exs
@@ -5,6 +5,7 @@ defmodule Cluster.StrategyTest do
 
   alias Cluster.Strategy
   alias Cluster.Nodes
+  alias Cluster.Telemetry
 
   require Cluster.Nodes
 
@@ -24,36 +25,54 @@ defmodule Cluster.StrategyTest do
       connect = {Nodes, :connect, [self()]}
       list_nodes = {Nodes, :list_nodes, [[Node.self()]]}
 
+      Telemetry.setup_telemetry([:libcluster, :connect_node, :ok])
+
       assert capture_log(fn ->
                assert :ok =
                         Strategy.connect_nodes(__MODULE__, connect, list_nodes, [:"foo@some.host"])
              end) =~ "connected to :\"foo@some.host\""
 
       assert_receive {:connect, :"foo@some.host"}
+
+      assert_receive {:telemetry_event,
+        {[:libcluster, :connect_node, :ok], %{duration: _},
+          %{node: :"foo@some.host", topology: _}, _}}
     end
 
     test "handles connect failure" do
       connect = {Nodes, :connect, [self(), false]}
       list_nodes = {Nodes, :list_nodes, [[Node.self()]]}
 
+      Telemetry.setup_telemetry([:libcluster, :connect_node, :error])
+
       assert capture_log(fn ->
                assert {:error, ["foo@some.host": false]} =
                         Strategy.connect_nodes(__MODULE__, connect, list_nodes, [:"foo@some.host"])
              end) =~ "unable to connect to :\"foo@some.host\""
 
       assert_receive {:connect, :"foo@some.host"}
+
+      assert_receive {:telemetry_event,
+        {[:libcluster, :connect_node, :error], %{},
+          %{node: :"foo@some.host", topology: _, reason: :unreachable}, _}}
     end
 
     test "handles connect ignore" do
       connect = {Nodes, :connect, [self(), :ignored]}
       list_nodes = {Nodes, :list_nodes, [[Node.self()]]}
 
+      Telemetry.setup_telemetry([:libcluster, :connect_node, :error])
+
       assert capture_log(fn ->
                assert {:error, ["foo@some.host": :ignored]} =
                         Strategy.connect_nodes(__MODULE__, connect, list_nodes, [:"foo@some.host"])
              end) =~ "unable to connect to :\"foo@some.host\""
 
       assert_receive {:connect, :"foo@some.host"}
+
+      assert_receive {:telemetry_event,
+        {[:libcluster, :connect_node, :error], %{},
+          %{node: :"foo@some.host", topology: _, reason: :not_part_of_network,}, _}}
     end
   end
 
@@ -71,6 +90,8 @@ defmodule Cluster.StrategyTest do
       disconnect = {Nodes, :disconnect, [self()]}
       list_nodes = {Nodes, :list_nodes, [[:"foo@some.host"]]}
 
+      Telemetry.setup_telemetry([:libcluster, :disconnect_node, :ok])
+
       assert capture_log(fn ->
                assert :ok =
                         Strategy.disconnect_nodes(__MODULE__, disconnect, list_nodes, [
@@ -79,12 +100,18 @@ defmodule Cluster.StrategyTest do
              end) =~ "disconnected from :\"foo@some.host\""
 
       assert_receive {:disconnect, :"foo@some.host"}
+
+      assert_receive {:telemetry_event,
+        {[:libcluster, :disconnect_node, :ok], %{duration: _},
+          %{node: :"foo@some.host", topology: _}, _}}
     end
 
     test "handles disconnect error" do
       disconnect = {Nodes, :disconnect, [self(), :failed]}
       list_nodes = {Nodes, :list_nodes, [[:"foo@some.host"]]}
 
+      Telemetry.setup_telemetry([:libcluster, :disconnect_node, :error])
+
       assert capture_log(fn ->
                assert {:error, ["foo@some.host": :failed]} =
                         Strategy.disconnect_nodes(__MODULE__, disconnect, list_nodes, [
@@ -94,6 +121,10 @@ defmodule Cluster.StrategyTest do
                "disconnect from :\"foo@some.host\" failed with: :failed"
 
       assert_receive {:disconnect, :"foo@some.host"}
+
+      assert_receive {:telemetry_event,
+        {[:libcluster, :disconnect_node, :error], %{},
+          %{node: :"foo@some.host", topology: _, reason: ":failed"}, _}}
     end
   end
 end
diff --git a/test/support/telemetry.ex b/test/support/telemetry.ex
@@ -0,0 +1,26 @@
+defmodule Cluster.Telemetry do
+  @moduledoc false
+
+  use ExUnit.Case
+
+  def setup_telemetry(event) do
+    telemetry_handle_id = "test-telemetry-handler-#{inspect(self())}"
+
+    :ok = :telemetry.attach_many(
+      telemetry_handle_id,
+      [
+        event,
+      ],
+      &send_to_pid/4,
+      nil
+    )
+
+    :ok = on_exit(fn -> :telemetry.detach(telemetry_handle_id) end)
+  end
+
+  defp send_to_pid(event, measurements, metadata, config) do
+    pid = config[:pid] || self()
+
+    send(pid, {:telemetry_event, {event, measurements, metadata, config}})
+  end
+end