--dist=loadgroup mix tests of different groups between same workers.

goal:
To segregate tests to nodes by marker such as test of a different marker are grouped in the separate nodes.

each test file include tests cases with one of 4 different xdist_group:

@pytest.mark.xdist_group("setup_one")
def test_smonethig()
    ...


@pytest.mark.xdist_group("setup_two")
def test_smonethig__setup_two()
    ...

@pytest.mark.xdist_group("setup_three")
def test_smonethig__setup_three()
    ...

@pytest.mark.xdist_group("setup_four")

when run : -n 4 --dist=loadgroup -vv
i can see same groups were run on gw0, gw1, gw2, and gw3.

Q1: shouldnt they run each group on separate node?

currently i have added markers manually.

Q2: ps dynamically adding xdist_group marker on pytest_collection_modifyitems():

def pytest_collection_modifyitems(config: Config, items: list[nodes.Item]):
    ...
    test_case.add_marker(pytest.mark.xdist_group(name=test_case_app_setup)) 
   #or
   test_case.nodeid = f"test_case@test_case_app_setup"

doesnt work. (in verbose mode nodeid finish on test name with @group_name when marked manually but still mixed between nodes, dynamically marking doesnt work).
is it becasue xdist override this method?
or to add marker dynamically should implement it in other hook?

I can create dict(app_setup: list(nodesId) and assigned it to config attribute.
But still cant cant figure how to pass it to xdist "Scheduler"..
Many thanks in advance

Env:

platform linux -- Python 3.13.7, 
pytest-8.4.1, 
pluggy-1.6.0
configfile: pyproject.toml
plugins: respx-0.22.0, 
cov-6.2.1, 
mock-3.14.1, 
xdist-3.8.0, 
requests-mock-1.12.1, 
split-0.10.0, 
anyio-4.10.0

Reproduce the behavior with -n 4 --dist=loadgroup -vv and the four xdist_group markers described. Start at pytest_collection_modifyitems() and the xdist Scheduler entry points to determine how collection and group assignment are transferred to workers. Done means the intended group-to-worker behavior and the supported approach for dynamically assigned groups are clarified or corrected.