LoadScope scheduler: Sort scopes by number of tests to assign biggest scopes first (#778)

Follow up to # 632. --------- Co-authored-by: Jörg Kohlsdorf <joerg@cornershopapp.com>

Author: cryvate

changelog/632.feature.rst

@@ -0,0 +1 @@
+``--dist=loadscope`` now sorts scopes by number of tests to assign largest scopes early -- in many cases this should improve overall test session running time, as there is less chance of a large scope being left to be processed near the end of the session, leaving other workers idle.

src/xdist/scheduler/loadscope.py

@@ -350,11 +350,18 @@ def schedule(self):
  return
 
  # Determine chunks of work (scopes)
+ unsorted_workqueue = OrderedDict()
  for nodeid in self.collection:
  scope = self._split_scope(nodeid)
- work_unit = self.workqueue.setdefault(scope, default=OrderedDict())
+ work_unit = unsorted_workqueue.setdefault(scope, default=OrderedDict())
  work_unit[nodeid] = False
 
+ # Insert tests scopes into work queue ordered by number of tests.
+ for scope, nodeids in sorted(
+ unsorted_workqueue.items(), key=lambda item: -len(item[1])
+ ):
+ self.workqueue[scope] = nodeids
+
  # Avoid having more workers than work
  extra_nodes = len(self.nodes) - len(self.workqueue)
 

testing/acceptance_test.py

@@ -1232,6 +1232,22 @@ def test(self, i):
  "test_a.py::TestB", result.outlines
  ) in ({"gw0": 10}, {"gw1": 10})
 
+ def test_workqueue_ordered_by_size(self, pytester: pytest.Pytester) -> None:
+ test_file = """
+ import pytest
+ @pytest.mark.parametrize('i', range({}))
+ def test(i):
+ pass
+ """
+ pytester.makepyfile(test_a=test_file.format(10), test_b=test_file.format(20))
+ result = pytester.runpytest("-n2", "--dist=loadscope", "-v")
+ assert get_workers_and_test_count_by_prefix(
+ "test_a.py::test", result.outlines
+ ) == {"gw1": 10}
+ assert get_workers_and_test_count_by_prefix(
+ "test_b.py::test", result.outlines
+ ) == {"gw0": 20}
+
  def test_module_single_start(self, pytester: pytest.Pytester) -> None:
  """Fix test suite never finishing in case all workers start with a single test (#277)."""
  test_file1 = """