Fix flaky PublicationManager relation tracker restart test (#4635)

Fixes #4606.

## Problem

The `PublicationManagerTest` "handles relation tracker restart" test is
flaky. It stops the `RelationTracker` (the supervisor then restarts it)
and relies on `assert_pub_tables` as a "wait for restart" barrier — but
it isn't one.

The relation is already in the publication and is **not** removed during
the restart, so `assert_pub_tables(ctx, [ctx.relation], …)` passes
immediately, potentially before the supervisor has restarted and
re-registered the process. The subsequent
`PublicationManager.remove_shape/2` call (→
`GenServer.call(name(stack_id), …)`) then races with a "no process"
exit.

## Fix

Wait for the `RelationTracker` to be re-registered, then for it to
finish restoring its filters from `ShapeStatus`, before issuing further
calls:

```elixir
assert wait_until(fn -> is_pid(GenServer.whereis(relation_tracker_name)) end, 2_000)
:ok = PublicationManager.wait_for_restore(ctx.stack_id, timeout: 2_000)
```

Rather than add a bespoke helper, this promotes the existing private
`wait_until` poller (previously in `shape_cache_test.exs`) into the
shared `Support.TestUtils` as `wait_until/2`, and migrates its original
call site to the shared version.

## Verification

15/15 seeded runs of the target test pass; the full
`publication_manager_test.exs` (26 tests) and the migrated
`shape_cache_test.exs` test are green. Test-only change, so no
changeset.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Oleksii Sholik <oleksii@sholik.dev>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: 3 people

.changeset/fix-relation-tracker-restart-test-race.md

@@ -0,0 +1,5 @@
+---
+'@core/sync-service': patch
+---
+
+Fix a race condition in the PublicationManager relation tracker restart test that caused intermittent failures.

packages/sync-service/test/electric/replication/publication_manager_test.exs

@@ -652,6 +652,18 @@ defmodule Electric.Replication.PublicationManagerTest do
       relation_tracker_name = PublicationManager.RelationTracker.name(ctx.stack_id)
       GenServer.stop(relation_tracker_name)
 
+      # Wait for the supervisor to restart and re-register the RelationTracker,
+      # then for it to finish restoring filters from ShapeStatus.
+      #
+      # The assert_pub_tables below is NOT a sufficient barrier on its own: the
+      # relation is already in the publication and is not removed during the
+      # restart, so the assertion passes immediately - potentially before the
+      # restarted process has re-registered. Calling remove_shape in that window
+      # races with a "no process" exit, which is the flakiness this guards
+      # against.
+      assert wait_until(fn -> is_pid(GenServer.whereis(relation_tracker_name)) end, 2_000)
+      :ok = PublicationManager.wait_for_restore(ctx.stack_id, timeout: 2_000)
+
       # After restart, the publication manager should repopulate from ShapeStatus.
       # The publication should still have the relation.
       assert_pub_tables(ctx, [ctx.relation], 2_000)

packages/sync-service/test/electric/shape_cache_test.exs

@@ -22,7 +22,8 @@ defmodule Electric.ShapeCacheTest do
       assert_shape_cleanup: 1,
       patch_shape_status: 1,
       patch_shape_cache: 1,
-      complete_txn_fragment: 3
+      complete_txn_fragment: 3,
+      wait_until: 2
     ]
 
   @stub_inspector Support.StubInspector.new(
@@ -1338,9 +1339,10 @@ defmodule Electric.ShapeCacheTest do
       index_after = SubqueryIndex.for_stack(ctx.stack_id)
       assert index_after != nil
 
-      assert wait_until(200, fn ->
-               SubqueryIndex.has_positions?(index_after, shape_handle)
-             end)
+      assert wait_until(
+               fn -> SubqueryIndex.has_positions?(index_after, shape_handle) end,
+               200
+             )
     end
 
     test "restores shapes with subqueries and their materializers when backup missing", ctx do
@@ -1389,22 +1391,6 @@ defmodule Electric.ShapeCacheTest do
         :ok = stop_supervised(name)
       end
     end
-
-    defp wait_until(timeout_ms, fun, started_at \\ System.monotonic_time(:millisecond))
-
-    defp wait_until(timeout_ms, fun, started_at) do
-      cond do
-        fun.() ->
-          true
-
-        System.monotonic_time(:millisecond) - started_at >= timeout_ms ->
-          false
-
-        true ->
-          Process.sleep(10)
-          wait_until(timeout_ms, fun, started_at)
-      end
-    end
   end
 
   describe "wait_for_restore eager subquery consumer start" do

packages/sync-service/test/support/test_utils.ex

@@ -138,6 +138,30 @@ defmodule Support.TestUtils do
     Electric.StatusMonitor.mark_pg_lock_as_errored(stack_id, error_message)
   end
 
+  @doc """
+  Poll `fun` every 10ms until it returns a truthy value or `timeout_ms` elapses.
+
+  Returns `true` if the condition was met within the timeout, `false` otherwise.
+  """
+  def wait_until(fun, timeout_ms \\ 500)
+      when is_function(fun, 0) and is_integer(timeout_ms) do
+    do_wait_until(fun, timeout_ms, System.monotonic_time(:millisecond))
+  end
+
+  defp do_wait_until(fun, timeout_ms, started_at) do
+    cond do
+      fun.() ->
+        true
+
+      System.monotonic_time(:millisecond) - started_at >= timeout_ms ->
+        false
+
+      true ->
+        Process.sleep(10)
+        do_wait_until(fun, timeout_ms, started_at)
+    end
+  end
+
   def generate_shape(relation, where_clause \\ nil, selected_columns \\ nil) do
     all_columns = Enum.uniq(["id", "value", "foo_enum"] ++ (selected_columns || []))
     selected_columns = selected_columns || all_columns