From 55873d32ce605bee41e1fc86d302c038dae5bf68 Mon Sep 17 00:00:00 2001
From: Isaac Whitfield <iw@whitfin.io>
Date: Fri, 20 Sep 2024 06:55:54 -0700
Subject: [PATCH] Optimize Janitor and enable provisioning

---
 lib/cachex/services/janitor.ex | 85 ++++++++++++++++++++++++++++------
 1 file changed, 72 insertions(+), 13 deletions(-)

diff --git a/lib/cachex/services/janitor.ex b/lib/cachex/services/janitor.ex
index 0122c48..f836d7f 100644
--- a/lib/cachex/services/janitor.ex
+++ b/lib/cachex/services/janitor.ex
@@ -12,15 +12,16 @@ defmodule Cachex.Services.Janitor do
   scans, so it should be expected that this can take a while to execute.
   """
   use GenServer
+  use Cachex.Provision
 
   # import parent macros
   import Cachex.Errors
   import Cachex.Spec
 
   # add some aliases
+  alias Cachex.Query
   alias Cachex.Services
   alias Services.Informant
-  alias Services.Overseer
 
   ##############
   # Public API #
@@ -88,14 +89,19 @@ defmodule Cachex.Services.Janitor do
   #
   # This will create the structure used to store metadata about
   # the run cycles of the Janitor, and schedule the first run.
-  def init(cache),
-    do: {:ok, {schedule_check(cache), %{}}}
+  def init(_),
+    do: {:ok, {nil, false, nil}}
+
+  @doc false
+  # Defines provisions required by this service.
+  def provisions,
+    do: [:cache]
 
   @doc false
   # Returns metadata about the last run of this Janitor.
   #
   # The returned information should be treated as non-guaranteed.
-  def handle_call(:last, _ctx, {_cache, last} = state),
+  def handle_call(:last, _ctx, {_cache, _active, last} = state),
     do: {:reply, {:ok, last}, state}
 
   @doc false
@@ -103,36 +109,89 @@ defmodule Cachex.Services.Janitor do
   #
   # This will drop to the ETS level and use a select to match documents which
   # need to be removed; they are then deleted by ETS at very high speeds.
-  def handle_info(:purge, {cache(name: name), _last}) do
-    start_time = now()
-    new_caches = Overseer.retrieve(name)
+  def handle_info(:purge, {cache, false, _last}) do
+    started = now()
+
+    {duration, active} =
+      :timer.tc(fn ->
+        query =
+          Query.build(
+            where: {:not, {:==, :expiration, nil}},
+            output: true,
+            batch_size: 1
+          )
+
+        cache
+        |> Cachex.stream!(query, const(:local) ++ const(:notify_false))
+        |> Enum.empty?()
+      end)
+
+    last = %{
+      count: 0,
+      started: started,
+      duration: duration
+    }
+
+    handle_activation(active, {cache, false, last})
+  end
+
+  @doc false
+  # Executes an expiration cleanup against a cache table.
+  #
+  # This will drop to the ETS level and use a select to match documents which
+  # need to be removed; they are then deleted by ETS at very high speeds.
+  def handle_info(:purge, {cache, true, _last}) do
+    started = now()
 
     {duration, {:ok, count} = result} =
       :timer.tc(fn ->
-        Cachex.purge(new_caches, const(:local))
+        Cachex.purge(cache, const(:local))
       end)
 
     case count do
       0 -> nil
-      _ -> Informant.broadcast(new_caches, const(:purge_override_call), result)
+      _ -> Informant.broadcast(cache, const(:purge_override_call), result)
     end
 
     last = %{
       count: count,
-      duration: duration,
-      started: start_time
+      started: started,
+      duration: duration
     }
 
-    {:noreply, {schedule_check(new_caches), last}}
+    {:noreply, {schedule(cache), true, last}}
   end
 
+  @doc false
+  # Receives a provisioned cache instance.
+  #
+  # The provided cache is then stored in the state and used for cache calls going
+  # forwards, in order to skip the lookups inside the cache overseer for performance.
+  def handle_provision({:cache, cache}, {nil, active, last}),
+    do: {:ok, {schedule(cache), active, last}}
+
+  def handle_provision({:cache, cache}, {_cache, active, last}),
+    do: {:ok, {cache, active, last}}
+
   ###############
   # Private API #
   ###############
 
+  @doc false
+  # Handles lazy Janitor activation paradigms.
+  #
+  # If the first argument is true, it reflects that there are no records with
+  # an expiration defined so we re-schedule and skip.  In the case there are
+  # records with expiration defined, we continue to the main purge cycle.
+  defp handle_activation(true, {cache, _active, last}),
+    do: {:noreply, {schedule(cache), false, last}}
+
+  defp handle_activation(false, {cache, _active, last}),
+    do: handle_info(:purge, {cache, true, last})
+
   # Schedules a check to occur after the designated interval. Once scheduled,
   # returns the state - this is just sugar for pipelining with a state.
-  defp schedule_check(cache(expiration: expiration(interval: interval)) = cache) do
+  defp schedule(cache(expiration: expiration(interval: interval)) = cache) do
     :erlang.send_after(interval, self(), :purge)
     cache
   end