Sort observations before returning

The order of the observations are important when doing updates
as it influences the pertubations. Now this property is used for
updates, which means the order matters.

src/ert/storage/local_experiment.py

@@ -207,7 +207,7 @@ def update_parameters(self) -> List[str]:
 
  @cached_property
  def observations(self) -> Dict[str, xr.Dataset]:
- observations = list(self.mount_point.glob("observations/*"))
+ observations = sorted(list(self.mount_point.glob("observations/*")))
  return {
  observation.name: xr.open_dataset(observation, engine="scipy")
  for observation in observations

tests/unit_tests/analysis/test_es_update.py

@@ -594,3 +594,15 @@ def test_temporary_parameter_storage_with_inactive_fields(
  ensemble._path / f"realization-{iens}" / f"{param_group}.nc", engine="scipy"
  )
  np.testing.assert_array_equal(ds["values"].values[0], fields[iens]["values"])
+
+
+def test_that_observations_keep_sorting(snake_oil_case_storage, snake_oil_storage):
+ """
+ The order of the observations influence the update as it affects the
+ perturbations, so we make sure we maintain the order throughout.
+ """
+ ert_config = snake_oil_case_storage
+ prior_ens = snake_oil_storage.get_ensemble_by_name("default_0")
+ assert list(ert_config.observations.keys()) == list(
+ prior_ens.experiment.observations.keys()
+ )