From 9e45222cba67a13002dae4ff2ee9969aa48bcd07 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=C5=81ukasz=20Jan=20Niemier?= <lukasz@supabase.io>
Date: Wed, 8 Jan 2025 12:39:54 +0100
Subject: [PATCH] feat: use Telemetry in MetricsCleaner

---
 lib/supavisor/metrics_cleaner.ex | 68 ++++++++++++++++++--------------
 1 file changed, 39 insertions(+), 29 deletions(-)

diff --git a/lib/supavisor/metrics_cleaner.ex b/lib/supavisor/metrics_cleaner.ex
index 27f61cc8..1888191d 100644
--- a/lib/supavisor/metrics_cleaner.ex
+++ b/lib/supavisor/metrics_cleaner.ex
@@ -11,18 +11,32 @@ defmodule Supavisor.MetricsCleaner do
 
   def init(_args) do
     Logger.info("Starting MetricsCleaner")
+
+    :telemetry.attach(
+      {__MODULE__, :report},
+      [:supavisor, :metrics_cleaner, :stop],
+      &__MODULE__.__report_long_cleanups__/4,
+      []
+    )
+
     {:ok, %{check_ref: check()}}
   end
 
-  def handle_info(:check, state) do
-    Process.cancel_timer(state.check_ref)
-
-    start = System.monotonic_time(:millisecond)
-    loop_and_cleanup_metrics_table()
-    exec_time = System.monotonic_time(:millisecond) - start
+  @doc false
+  def __report_long_cleanups__(_event_name, %{duration: duration}, _metadata, _config) do
+    exec_time = :erlang.convert_time_unit(duration, :native, :millisecond)
 
     if exec_time > :timer.seconds(5),
       do: Logger.warning("Metrics check took: #{exec_time} ms")
+  end
+
+  def handle_info(:check, state) do
+    Process.cancel_timer(state.check_ref)
+
+    :telemetry.span([:supavisor, :metrics_cleaner], %{}, fn ->
+      count = loop_and_cleanup_metrics_table()
+      {[], %{orphaned_metrics: count}, %{}}
+    end)
 
     {:noreply, %{state | check_ref: check()}}
   end
@@ -38,32 +52,28 @@ defmodule Supavisor.MetricsCleaner do
     metrics_table = Supavisor.Monitoring.PromEx.Metrics
     tenant_registry_table = :syn_registry_by_name_tenants
 
-    func = fn
-      {{_,
-        %{
-          type: type,
-          mode: mode,
-          user: user,
-          tenant: tenant,
-          db_name: db,
-          search_path: search_path
-        }} = key, _},
-      _ ->
-        case :ets.lookup(tenant_registry_table, {{type, tenant}, user, mode, db, search_path}) do
-          [] ->
-            Logger.warning("Found orphaned metric: #{inspect(key)}")
-            :ets.delete(metrics_table, key)
-
-          _ ->
-            nil
-        end
-
-      _, acc ->
-        acc
+    func = fn elem, acc ->
+      with {{_,
+             %{
+               type: type,
+               mode: mode,
+               user: user,
+               tenant: tenant,
+               db_name: db,
+               search_path: search_path
+             }} = key, _} <- elem,
+           [] <- :ets.lookup(tenant_registry_table, {{type, tenant}, user, mode, db, search_path}) do
+        Logger.warning("Found orphaned metric: #{inspect(key)}")
+        :ets.delete(metrics_table, key)
+
+        acc + 1
+      else
+        _ -> acc
+      end
     end
 
     {_, tids} = Peep.Persistent.storage(Supavisor.Monitoring.PromEx.Metrics)
 
-    Enum.each(List.wrap(tids), &:ets.foldl(func, nil, &1))
+    Enum.each(List.wrap(tids), &:ets.foldl(func, 0, &1))
   end
 end