refactoring folders

lib/elixir_ai/application.ex

@@ -12,6 +12,8 @@ defmodule ElixirAi.Application do
        [Application.get_env(:libcluster, :topologies, []), [name: ElixirAi.ClusterSupervisor]]},
       {Phoenix.PubSub, name: ElixirAi.PubSub},
       {ElixirAi.LiveViewPG, []},
+      {ElixirAi.RunnerPG, []},
+      {ElixirAi.SingletonPG, []},
       {ElixirAi.PageToolsPG, []},
       {ElixirAi.AudioProcessingPG, []},
       {DynamicSupervisor, name: ElixirAi.AudioWorkerSupervisor, strategy: :one_for_one},

lib/elixir_ai/chat_runner/chat_runner.ex

@@ -68,6 +68,7 @@ defmodule ElixirAi.ChatRunner do
 
   def init(name) do
     Phoenix.PubSub.subscribe(ElixirAi.PubSub, conversation_message_topic(name))
+    :pg.join(ElixirAi.RunnerPG, {:runner, name}, self())
 
     messages =
       case Conversation.find_id(name) do

lib/elixir_ai/chat_runner/runner_pg.ex

@@ -0,0 +1,16 @@
+defmodule ElixirAi.RunnerPG do
+  @moduledoc """
+  Named :pg scope for tracking ChatRunner processes across the cluster.
+  Each ChatRunner joins {:runner, name} on init; :pg syncs membership
+  automatically and removes dead processes without any additional cleanup.
+  """
+
+  def child_spec(_opts) do
+    %{
+      id: __MODULE__,
+      start: {:pg, :start_link, [__MODULE__]},
+      type: :worker,
+      restart: :permanent
+    }
+  end
+end

lib/elixir_ai/cluster_singleton/singleton_pg.ex

@@ -0,0 +1,16 @@
+defmodule ElixirAi.SingletonPG do
+  @moduledoc """
+  Named :pg scope for tracking cluster singleton processes across the cluster.
+  Each singleton joins {:singleton, __MODULE__} on init; :pg syncs membership
+  automatically and removes dead processes without any additional cleanup.
+  """
+
+  def child_spec(_opts) do
+    %{
+      id: __MODULE__,
+      start: {:pg, :start_link, [__MODULE__]},
+      type: :worker,
+      restart: :permanent
+    }
+  end
+end

lib/elixir_ai/conversation_manager.ex

@@ -20,6 +20,7 @@ defmodule ElixirAi.ConversationManager do
 
   def init(_) do
     Logger.info("ConversationManager initializing...")
+    :pg.join(ElixirAi.SingletonPG, {:singleton, __MODULE__}, self())
     send(self(), :load_conversations)
     {:ok, %{conversations: :loading, subscriptions: MapSet.new(), runners: %{}}}
   end

lib/elixir_ai_web/features/admin/admin_live.ex

@@ -1,41 +1,64 @@
 defmodule ElixirAiWeb.AdminLive do
+  import ElixirAi.PubsubTopics
   use ElixirAiWeb, :live_view
   require Logger
 
-  @refresh_ms 1_000
-
   def mount(_params, _session, socket) do
+    socket =
       if connected?(socket) do
         :net_kernel.monitor_nodes(true)
+        # Join before monitoring so our own join doesn't trigger a spurious refresh.
         :pg.join(ElixirAi.LiveViewPG, {:liveview, __MODULE__}, self())
-      schedule_refresh()
+        {pg_ref, _} = :pg.monitor_scope(ElixirAi.LiveViewPG)
+        {runner_pg_ref, _} = :pg.monitor_scope(ElixirAi.RunnerPG)
+        {singleton_pg_ref, _} = :pg.monitor_scope(ElixirAi.SingletonPG)
+
+        socket
+        |> assign(pg_ref: pg_ref)
+        |> assign(runner_pg_ref: runner_pg_ref)
+        |> assign(singleton_pg_ref: singleton_pg_ref)
+      else
+        assign(socket, pg_ref: nil, runner_pg_ref: nil, singleton_pg_ref: nil)
       end
 
-    {:ok, assign(socket, cluster_info: gather_info())}
+    {:ok,
+     socket
+     |> assign(nodes: gather_node_statuses())
+     |> assign(singleton_locations: gather_singleton_locations())
+     |> assign(chat_runners: gather_chat_runners())
+     |> assign(liveviews: gather_liveviews())}
   end
 
   def handle_info({:nodeup, _node}, socket) do
-    {:noreply, assign(socket, cluster_info: gather_info())}
+    {:noreply, assign(socket, nodes: gather_node_statuses())}
   end
 
   def handle_info({:nodedown, _node}, socket) do
-    {:noreply, assign(socket, cluster_info: gather_info())}
+    {:noreply, assign(socket, nodes: gather_node_statuses())}
   end
 
-  def handle_info(:refresh, socket) do
+  def handle_info(:refresh_singletons, socket) do
-    schedule_refresh()
+    {:noreply, assign(socket, singleton_locations: gather_singleton_locations())}
-    {:noreply, assign(socket, cluster_info: gather_info())}
   end
 
-  defp schedule_refresh, do: Process.send_after(self(), :refresh, @refresh_ms)
+  def handle_info({ref, change, _group, _pids}, %{assigns: %{singleton_pg_ref: ref}} = socket)
+      when is_reference(ref) and change in [:join, :leave] do
+    {:noreply, assign(socket, singleton_locations: gather_singleton_locations())}
+  end
 
-  defp gather_info do
+  def handle_info({ref, change, _group, _pids}, %{assigns: %{pg_ref: ref}} = socket)
-    import ElixirAi.PubsubTopics
+      when is_reference(ref) and change in [:join, :leave] do
+    {:noreply, assign(socket, liveviews: gather_liveviews())}
+  end
 
+  def handle_info({ref, change, _group, _pids}, %{assigns: %{runner_pg_ref: ref}} = socket)
+      when is_reference(ref) and change in [:join, :leave] do
+    {:noreply, assign(socket, chat_runners: gather_chat_runners())}
+  end
+
+  defp gather_node_statuses do
     all_nodes = [Node.self() | Node.list()]
-    configured = ElixirAi.ClusterSingleton.configured_singletons()
 
-    node_statuses =
     Enum.map(all_nodes, fn node ->
       status =
         if node == Node.self() do
@@ -53,39 +76,45 @@ defmodule ElixirAiWeb.AdminLive do
 
       {node, status}
     end)
-
-    singleton_locations =
-      Enum.map(configured, fn module ->
-        location =
-          case Horde.Registry.lookup(ElixirAi.ChatRegistry, module) do
-            [{pid, _}] -> node(pid)
-            _ -> nil
   end
 
-        {module, location}
+  defp gather_singleton_locations do
-      end)
+    running =
-
+      :pg.which_groups(ElixirAi.SingletonPG)
-    # All ChatRunner entries in the distributed registry, keyed by conversation name.
-    # Each entry is a {name, node, pid, supervisor_node} tuple.
-    chat_runners =
-      Horde.DynamicSupervisor.which_children(ElixirAi.ChatRunnerSupervisor)
       |> Enum.flat_map(fn
-        {_, pid, _, _} when is_pid(pid) ->
+        {:singleton, module} ->
-          case Horde.Registry.select(ElixirAi.ChatRegistry, [
+          case :pg.get_members(ElixirAi.SingletonPG, {:singleton, module}) do
-                 {{:"$1", pid, :"$2"}, [], [{{:"$1", pid, :"$2"}}]}
+            [pid | _] -> [{module, node(pid)}]
-               ]) do
-            [{name, ^pid, _}] when is_binary(name) -> [{name, node(pid), pid}]
             _ -> []
           end
 
         _ ->
           []
       end)
+      |> Map.new()
+
+    ElixirAi.ClusterSingleton.configured_singletons()
+    |> Enum.map(fn module -> {module, Map.get(running, module)} end)
+  end
+
+  # All ChatRunner entries via :pg membership, keyed by conversation name.
+  # Each entry is a {name, node, pid} tuple.
+  defp gather_chat_runners do
+    :pg.which_groups(ElixirAi.RunnerPG)
+    |> Enum.flat_map(fn
+      {:runner, name} ->
+        :pg.get_members(ElixirAi.RunnerPG, {:runner, name})
+        |> Enum.map(fn pid -> {name, node(pid), pid} end)
+
+      _ ->
+        []
+    end)
     |> Enum.sort_by(&elem(&1, 0))
+  end
 
   # :pg is cluster-wide — one local call returns members from all nodes.
   # Processes are automatically removed from their group when they die.
-    liveviews =
+  defp gather_liveviews do
     :pg.which_groups(ElixirAi.LiveViewPG)
     |> Enum.flat_map(fn
       {:liveview, view} ->
@@ -95,14 +124,6 @@ defmodule ElixirAiWeb.AdminLive do
       _ ->
         []
     end)
-
-    %{
-      nodes: node_statuses,
-      configured_singletons: configured,
-      singleton_locations: singleton_locations,
-      chat_runners: chat_runners,
-      liveviews: liveviews
-    }
   end
 
   def render(assigns) do
@@ -111,13 +132,13 @@ defmodule ElixirAiWeb.AdminLive do
       <h1 class="text-lg font-semibold text-seafoam-200 tracking-wide">Cluster Admin</h1>
 
       <div class="grid gap-4 grid-cols-1 lg:grid-cols-2 xl:grid-cols-3">
-        <%= for {node, status} <- @cluster_info.nodes do %>
+        <%= for {node, status} <- @nodes do %>
           <% node_singletons =
-            Enum.filter(@cluster_info.singleton_locations, fn {_, loc} -> loc == node end) %>
+            Enum.filter(@singleton_locations, fn {_, loc} -> loc == node end) %>
           <% node_runners =
-            Enum.filter(@cluster_info.chat_runners, fn {_, rnode, _} -> rnode == node end) %>
+            Enum.filter(@chat_runners, fn {_, rnode, _} -> rnode == node end) %>
           <% node_liveviews =
-            @cluster_info.liveviews
+            @liveviews
             |> Enum.filter(fn {_, n} -> n == node end)
             |> Enum.group_by(fn {view, _} -> view end) %>
 
@@ -126,7 +147,9 @@ defmodule ElixirAiWeb.AdminLive do
               <div class="flex items-center gap-2">
                 <span class="font-mono text-sm font-semibold text-seafoam-200">{node}</span>
                 <%= if node == Node.self() do %>
-                  <span class="text-xs bg-seafoam-800/50 text-seafoam-400 px-1.5 py-0.5 rounded">self</span>
+                  <span class="text-xs bg-seafoam-800/50 text-seafoam-400 px-1.5 py-0.5 rounded">
+                    self
+                  </span>
                 <% end %>
               </div>
               <.status_badge status={status} />
@@ -191,7 +214,7 @@ defmodule ElixirAiWeb.AdminLive do
       </div>
 
       <% unlocated =
-        Enum.filter(@cluster_info.singleton_locations, fn {_, loc} -> is_nil(loc) end) %>
+        Enum.filter(@singleton_locations, fn {_, loc} -> is_nil(loc) end) %>
       <%= if unlocated != [] do %>
         <section>
           <h2 class="text-xs font-semibold uppercase tracking-widest text-red-500 mb-2">
@@ -207,7 +230,9 @@ defmodule ElixirAiWeb.AdminLive do
         </section>
       <% end %>
 
-      <p class="text-xs text-seafoam-800">Refreshes every 1s or on node events.</p>
+      <p class="text-xs text-seafoam-800">
+        Nodes, singletons, liveviews &amp; runners all refresh on membership changes.
+      </p>
     </div>
     """
   end