fix: improve rpc metric; remove ecto metric (#1308)

supabase · Feb 21, 2025 · 51b2b0a · 51b2b0a
1 parent 5192428
commit 51b2b0a
Show file tree

Hide file tree

Showing 5 changed files with 57 additions and 17 deletions.
diff --git a/lib/realtime/monitoring/prom_ex.ex b/lib/realtime/monitoring/prom_ex.ex
@@ -73,8 +73,7 @@ defmodule Realtime.PromEx do
       {OsMon, poll_rate: poll_rate},
       {Tenants, poll_rate: poll_rate},
       {Tenant, poll_rate: poll_rate},
-      {Channels, poll_rate: poll_rate},
-      {PromEx.Plugins.Ecto, otp_app: :realtime, poll_rate: poll_rate, metric_prefix: [:ecto]}
+      {Channels, poll_rate: poll_rate}
     ]
   end
 

diff --git a/lib/realtime/monitoring/prom_ex/plugins/tenants.ex b/lib/realtime/monitoring/prom_ex/plugins/tenants.ex
@@ -18,7 +18,8 @@ defmodule Realtime.PromEx.Plugins.Tenants do
         description: "Latency of rpc calls triggered by a tenant action",
         measurement: :latency,
         unit: {:microsecond, :millisecond},
-        reporter_options: [buckets: [10, 50, 250, 1500, 15_000]]
+        tags: [:success],
+        reporter_options: [buckets: [10, 250, 5000, 15_000]]
       )
     ])
   end

diff --git a/lib/realtime/rpc.ex b/lib/realtime/rpc.ex
@@ -35,17 +35,17 @@ defmodule Realtime.Rpc do
         {:ok, _} ->
           Telemetry.execute(
             [:realtime, :rpc],
-            %{latency: latency, success?: true},
-            %{mod: mod, func: func, target_node: node, origin_node: node()}
+            %{latency: latency},
+            %{mod: mod, func: func, target_node: node, origin_node: node(), success: true}
           )
 
           response
 
         {:error, response} ->
           Telemetry.execute(
             [:realtime, :rpc],
-            %{latency: latency, success?: false},
-            %{mod: mod, func: func, target_node: node, origin_node: node()}
+            %{latency: latency},
+            %{mod: mod, func: func, target_node: node, origin_node: node(), success: false}
           )
 
           {:error, response}
@@ -55,8 +55,8 @@ defmodule Realtime.Rpc do
     kind, reason ->
       Telemetry.execute(
         [:realtime, :rpc],
-        %{latency: 0, success?: false},
-        %{mod: mod, func: func, target_node: node, origin_node: node()}
+        %{latency: 0},
+        %{mod: mod, func: func, target_node: node, origin_node: node(), success: false}
       )
 
       log_error(

diff --git a/mix.exs b/mix.exs
@@ -4,7 +4,7 @@ defmodule Realtime.MixProject do
   def project do
     [
       app: :realtime,
-      version: "2.34.30",
+      version: "2.34.31",
       elixir: "~> 1.17.3",
       elixirc_paths: elixirc_paths(Mix.env()),
       start_permanent: Mix.env() == :prod,

diff --git a/test/realtime/rpc_test.exs b/test/realtime/rpc_test.exs
@@ -11,7 +11,7 @@ defmodule Realtime.RpcTest do
     def test_success, do: {:ok, "success"}
   end
 
-  def handle_telemetry(event, metadata, _, pid: pid), do: send(pid, {event, metadata})
+  def handle_telemetry(event, measurements, metadata, pid: pid), do: send(pid, {event, measurements, metadata})
 
   setup do
     :telemetry.attach(__MODULE__, [:realtime, :rpc], &__MODULE__.handle_telemetry/4, pid: self())
@@ -22,7 +22,14 @@ defmodule Realtime.RpcTest do
   describe "call/5" do
     test "successful RPC call returns exactly what the original function returns" do
       assert {:ok, "success"} = Rpc.call(node(), TestRpc, :test_success, [])
-      assert_receive {[:realtime, :rpc], %{latency: _}}
+
+      assert_receive {[:realtime, :rpc], %{latency: _},
+                      %{
+                        mod: Realtime.RpcTest.TestRpc,
+                        func: :test_success,
+                        origin_node: :nonode@nohost,
+                        target_node: :nonode@nohost
+                      }}
     end
 
     test "raised exceptions are properly caught and logged" do
@@ -35,21 +42,41 @@ defmodule Realtime.RpcTest do
                 ]}}} =
                Rpc.call(node(), TestRpc, :test_raise, [])
 
-      assert_receive {[:realtime, :rpc], %{latency: _}}
+      assert_receive {[:realtime, :rpc], %{latency: _},
+                      %{
+                        mod: Realtime.RpcTest.TestRpc,
+                        func: :test_raise,
+                        origin_node: :nonode@nohost,
+                        target_node: :nonode@nohost
+                      }}
     end
 
     test "timeouts are properly caught and logged" do
       assert {:badrpc, :timeout} =
                Rpc.call(node(), TestRpc, :test_timeout, [], timeout: 100)
 
-      assert_receive {[:realtime, :rpc], %{latency: _}}
+      assert_receive {[:realtime, :rpc], %{latency: _},
+                      %{
+                        mod: Realtime.RpcTest.TestRpc,
+                        func: :test_timeout,
+                        origin_node: :nonode@nohost,
+                        target_node: :nonode@nohost
+                      }}
     end
   end
 
   describe "enhanced_call/5" do
     test "successful RPC call returns exactly what the original function returns" do
       assert {:ok, "success"} = Rpc.enhanced_call(node(), TestRpc, :test_success)
-      assert_receive {[:realtime, :rpc], %{latency: _, success?: true}}
+
+      assert_receive {[:realtime, :rpc], %{latency: _},
+                      %{
+                        mod: Realtime.RpcTest.TestRpc,
+                        func: :test_success,
+                        origin_node: :nonode@nohost,
+                        target_node: :nonode@nohost,
+                        success: true
+                      }}
     end
 
     test "raised exceptions are properly caught and logged" do
@@ -58,7 +85,14 @@ defmodule Realtime.RpcTest do
                         Rpc.enhanced_call(node(), TestRpc, :test_raise)
              end) =~ "ErrorOnRpcCall"
 
-      assert_receive {[:realtime, :rpc], %{latency: _, success?: false}}
+      assert_receive {[:realtime, :rpc], %{latency: _},
+                      %{
+                        mod: Realtime.RpcTest.TestRpc,
+                        func: :test_raise,
+                        origin_node: :nonode@nohost,
+                        target_node: :nonode@nohost,
+                        success: false
+                      }}
     end
 
     test "timeouts are properly caught and logged" do
@@ -67,7 +101,13 @@ defmodule Realtime.RpcTest do
                         Rpc.enhanced_call(node(), TestRpc, :test_timeout, [], timeout: 100)
              end) =~ "ErrorOnRpcCall"
 
-      assert_receive {[:realtime, :rpc], %{latency: 0, success?: false}}
+      assert_receive {[:realtime, :rpc], %{latency: 0},
+                      %{
+                        mod: Realtime.RpcTest.TestRpc,
+                        func: :test_timeout,
+                        origin_node: :nonode@nohost,
+                        target_node: :nonode@nohost
+                      }}
     end
   end
 end