fix: prevent deadlocks by detecting existing transactions

lib/membership/member.ex

@@ -763,7 +763,7 @@ defmodule Mv.Membership.Member do
   end
 
   # Regenerates cycles with new type/amount
-  # CycleGenerator uses its own transaction with advisory lock
+  # CycleGenerator detects if already in transaction and uses advisory lock accordingly
   defp regenerate_cycles(member_id) do
     case Mv.MembershipFees.CycleGenerator.generate_cycles_for_member(member_id) do
       {:ok, _cycles} -> :ok

lib/mv/membership_fees/cycle_generator.ex

@@ -188,6 +188,26 @@ defmodule Mv.MembershipFees.CycleGenerator do
     # Convert UUID to integer for advisory lock (use hash)
     lock_key = :erlang.phash2(member_id)
 
+    # Check if we're already in a transaction (e.g., called from Ash action)
+    if Repo.in_transaction?() do
+      # Already in transaction: use advisory lock directly without starting new transaction
+      # This prevents nested transactions which can cause deadlocks
+      Ecto.Adapters.SQL.query!(Repo, "SELECT pg_advisory_xact_lock($1)", [lock_key])
+
+      case fun.() do
+        {:ok, result, notifications} when is_list(notifications) ->
+          # Notifications will be sent after the outer transaction commits
+          # Return in same format as non-transaction case for consistency
+          {:ok, result}
+
+        {:ok, result} ->
+          {:ok, result}
+
+        {:error, reason} ->
+          {:error, reason}
+      end
+    else
+      # Not in transaction: start new transaction with advisory lock
       result =
         Repo.transaction(fn ->
           # Acquire advisory lock for this transaction
@@ -220,6 +240,7 @@ defmodule Mv.MembershipFees.CycleGenerator do
           {:error, reason}
       end
     end
+  end
 
   defp do_generate_cycles(member, today) do
     # Reload member with relationships to ensure fresh data