change supervisor strategy

Author: magnetised

packages/sync-service/lib/electric/shape_cache/shape_status/shape_db/supervisor.ex

@@ -65,12 +65,16 @@ defmodule Electric.ShapeCache.ShapeStatus.ShapeDb.Supervisor do
              lazy: not exclusive_mode},
             id: {:pool, :write}
           ),
-          # write buffer for batching SQLite writes to avoid timeout cascades
+          # Write buffer for batching SQLite writes to avoid timeout cascades.
           {ShapeDb.WriteBuffer, opts},
           {Task, fn -> ShapeDb.Statistics.initialize(stack_id) end}
         ]
       ])
 
-    Supervisor.init(children, strategy: :one_for_one)
+    # Because the full state of the system is split between the actual db, the
+    # writeBuffer and the ShapeStatus ets caches, we are not safe to adopt a
+    # one_for_one strategy and need to propagate an exit in the children of
+    # this supervisor to the parent
+    Supervisor.init(children, strategy: :one_for_all, max_restarts: 0)
   end
 end

packages/sync-service/lib/electric/shape_cache/shape_status/shape_db/write_buffer.ex

@@ -274,8 +274,6 @@ defmodule Electric.ShapeCache.ShapeStatus.ShapeDb.WriteBuffer do
 
   @impl GenServer
   def init(opts) do
-    Process.flag(:trap_exit, true)
-
     stack_id = Keyword.fetch!(opts, :stack_id)
     manual_flush_only = Keyword.get(opts, :manual_flush_only, false)
 

packages/sync-service/test/electric/shape_cache/shape_status/shape_db_test.exs

@@ -657,6 +657,17 @@ defmodule Electric.ShapeCache.ShapeStatus.ShapeDbTest do
         end)
       end
     end
+
+    test "crashing WriteBuffer restarts entire supervision tree", ctx do
+      assert supervisor_pid = GenServer.whereis(ShapeDb.Supervisor.name(ctx.stack_id))
+      super_ref = Process.monitor(supervisor_pid)
+      assert write_buffer_pid = GenServer.whereis(ShapeDb.WriteBuffer.name(ctx.stack_id))
+
+      buffer_ref = Process.monitor(write_buffer_pid)
+      Process.exit(write_buffer_pid, :some_reason)
+      assert_receive {:DOWN, ^buffer_ref, :process, ^write_buffer_pid, :some_reason}
+      assert_receive {:DOWN, ^super_ref, :process, ^supervisor_pid, _}
+    end
   end
 
   describe "pool scaling" do