def test_serialize_env_dict():
# this tests an edge case due to EnvDict's implementation: to enable
# accessing values in the underlying dictionary as attributes, we are
# customizing __getattr__, however, when an object is unserialized,
# Python tries to look for __getstate__ (which triggers calling
# __getattr__), since it cannot find it, it will go to __getitem__
# (given the current implementation of __getattr__). But __getitem__
# uses self.preprocessed. At unserialization time, this attribute does
# not exist yet!, which will cause another call to __getattr__. To avoid
# this recursive loop, we have to prevent special methods to call
# __getitem__ if they do not exist - EnvDict and Env objects are not
# expected to be serialized but we have fix it anyway
env = EnvDict({'a': 1})
assert pickle.loads(pickle.dumps(env))
def test_adds_default_keys_if_they_dont_exist(monkeypatch):
monkeypatch.setattr(getpass, 'getuser', Mock(return_value='User'))
monkeypatch.setattr(os, 'getcwd', Mock(return_value='/some_path'))
mock = Mock(return_value='some_value')
monkeypatch.setattr(default, 'find_root_recursively', mock)
monkeypatch.setattr(expand.default, 'find_root_recursively', mock)
env = EnvDict({'a': 1}, path_to_here='/dir')
assert env.cwd == str(Path('/some_path').resolve())
assert env.here == str(Path('/dir').resolve())
assert env.user == 'User'
assert env.root == 'some_value'
assert env.default_keys == {'cwd', 'here', 'user', 'root'}
def test_expand_raw_dict_nested():
mapping = EnvDict({'key': 'value'})
d = {
'section': {
'some_settting': '{{key}}'
},
'list': ['{{key}}', '{{key}}']
}
assert (expand_raw_dictionary(d, mapping) == {
'section': {
'some_settting': 'value'
},
'list': ['value', 'value']
})
def test_expand_raw_dictionaries_and_extract_tags():
mapping = EnvDict({'key': 'value'})
d = [{'some_setting': '{{key}}'}, {'another_setting': '{{key}}'}]
expanded, tags = expand_raw_dictionaries_and_extract_tags(d, mapping)
assert expanded == (
{
'some_setting': 'value',
},
{
'another_setting': 'value'
},
)
assert tags == {'key'}
def test_dagspec_initialization_from_yaml_and_env(tmp_nbs, monkeypatch):
"""
DAGSpec can be initialized with a path to a spec or a dictionary, but
they have a slightly different behavior. This ensure the cli passes
the path, instead of a dictionary
"""
mock_DAGSpec = Mock(wraps=parsers.DAGSpec)
mock_default_path_to_env = Mock(
wraps=parsers.default.path_to_env_from_spec)
mock_EnvDict = Mock(wraps=parsers.EnvDict)
monkeypatch.setattr(sys, 'argv', ['python'])
monkeypatch.setattr(parsers, 'DAGSpec', mock_DAGSpec)
monkeypatch.setattr(parsers.default, 'path_to_env_from_spec',
mock_default_path_to_env)
monkeypatch.setattr(parsers, 'EnvDict', mock_EnvDict)
# ensure current timestamp does not change
mock = Mock()
mock.datetime.now().isoformat.return_value = 'current-timestamp'
monkeypatch.setattr(expand, "datetime", mock)
parser = CustomParser()
with parser:
pass
dag, args = parser.load_from_entry_point_arg()
# ensure called using the path to the yaml spec
mock_DAGSpec.assert_called_once_with('pipeline.yaml',
env=EnvDict({'sample': False},
path_to_here='.'))
# and EnvDict initialized from env.yaml
mock_EnvDict.assert_called_once_with(str(Path('env.yaml').resolve()),
path_to_here=Path('.'))
def test_default_keys(kwargs, expected):
assert EnvDict(**kwargs).default_keys == expected
def test_default_with_here_absolute(tmp_directory):
here = str(Path(tmp_directory, 'dir').resolve())
env = EnvDict(dict(), path_to_here=here)
assert env.here == here
def test_default_with_here_relative(tmp_directory):
Path('dir').mkdir()
env = EnvDict(dict(), path_to_here='dir')
assert env.here == str(Path(tmp_directory, 'dir').resolve())
def test_expand_raw_dictionary():
mapping = EnvDict({'key': 'value'})
d = {'some_setting': '{{key}}'}
assert expand_raw_dictionary(d, mapping) == {'some_setting': 'value'}
def test_error_when_flatten_key_doesnt_exist():
env = EnvDict({'a': 1})
with pytest.raises(KeyError):
env._replace_flatten_key(2, 'env__b')
def test_replace_nested_flatten_keys_env_dict():
env = EnvDict({'a': {'b': 1, 'c': 1}})
new_env = env._replace_flatten_keys({'env__a__b': 2, 'env__a__c': 2})
assert (new_env.a.b == 2 and new_env.a.c == 2
and env is not new_env) # must return a copy
def test_replace_nested_flatten_key_env_dict():
env = EnvDict({'a': {'b': 1}})
new_env = env._replace_flatten_key(2, 'env__a__b')
assert new_env.a.b == 2 and env is not new_env # must return a copy
def test_replace_flatten_key_env_dict():
env = EnvDict({'a': 1})
new_env = env._replace_flatten_key(2, 'env__a')
assert new_env.a == 2 and env is not new_env # must return a copy
def _init(self, data, env, lazy_import, reload, parent_path,
look_up_project_root_recursively):
self._lazy_import = lazy_import
# initialized with a path to a yaml file...
if isinstance(data, (str, Path)):
# TODO: test this
if parent_path is not None:
raise ValueError('parent_path must be None when '
f'initializing {type(self).__name__} with '
'a path to a YAML spec')
# resolve the parent path to make sources and products unambiguous
# even if the current working directory changes
self._path = Path(data).resolve()
self._parent_path = str(self._path.parent)
if not Path(data).is_file():
raise FileNotFoundError(
'Error initializing DAGSpec with argument '
f'{data!r}: Expected it to be a path to a YAML file, but '
'such file does not exist')
content = Path(data).read_text()
try:
data = yaml.safe_load(content)
except (yaml.parser.ParserError,
yaml.constructor.ConstructorError) as e:
error = e
else:
error = None
if error:
if '{{' in content or '}}' in content:
raise DAGSpecInitializationError(
'Failed to initialize spec. It looks like '
'you\'re using placeholders (i.e. {{placeholder}}). '
'Make sure values are enclosed in parentheses '
'(e.g. key: "{{placeholder}}"). Original '
'parser error:\n\n'
f'{error}')
else:
raise error
# initialized with a dictionary...
else:
self._path = None
# FIXME: add test cases, some of those features wont work if
# _parent_path is None. We should make sure that we either raise
# an error if _parent_path is needed or use the current working
# directory if it's appropriate - this is mostly to make relative
# paths consistent: they should be relative to the file that
# contains them
self._parent_path = (None if not parent_path else str(
Path(parent_path).resolve()))
self.data = data
if isinstance(self.data, list):
self.data = {'tasks': self.data}
# validate keys defined at the top (nested keys are not validated here)
self._validate_top_keys(self.data, self._path)
logger.debug('DAGSpec enviroment:\n%s', pp.pformat(env))
env = env or dict()
path_to_defaults = default.path_to_env_from_spec(
path_to_spec=self._path)
if path_to_defaults:
defaults = yaml.safe_load(Path(path_to_defaults).read_text())
self.env = EnvDict(env,
path_to_here=self._parent_path,
defaults=defaults)
else:
self.env = EnvDict(env, path_to_here=self._parent_path)
self.data, tags = expand_raw_dictionary_and_extract_tags(
self.data, self.env)
logger.debug('Expanded DAGSpec:\n%s', pp.pformat(data))
# if there is a "location" top key, we don't have to do anything else
# as we will just load the dotted path when .to_dag() is called
if 'location' not in self.data:
Meta.initialize_inplace(self.data)
import_tasks_from = self.data['meta']['import_tasks_from']
if import_tasks_from is not None:
# when using a relative path in "import_tasks_from", we must
# make it absolute...
if not Path(import_tasks_from).is_absolute():
# use _parent_path if there is one
if self._parent_path:
self.data['meta']['import_tasks_from'] = str(
Path(self._parent_path, import_tasks_from))
# otherwise just make it absolute
else:
self.data['meta']['import_tasks_from'] = str(
Path(import_tasks_from).resolve())
imported = yaml.safe_load(
Path(self.data['meta']['import_tasks_from']).read_text())
if self.env is not None:
(imported,
tags_other) = expand_raw_dictionaries_and_extract_tags(
imported, self.env)
tags = tags | tags_other
# relative paths here are relative to the file where they
# are declared
base_path = Path(self.data['meta']['import_tasks_from']).parent
for task in imported:
add_base_path_to_source_if_relative(task,
base_path=base_path)
self.data['tasks'].extend(imported)
# check if there are any params declared in env, not used in
# in the pipeline
extra = set(self.env) - self.env.default_keys - tags
if extra:
warnings.warn('The following placeholders are declared in the '
'environment but '
f'unused in the spec: {extra}')
self.data['tasks'] = [
normalize_task(task) for task in self.data['tasks']
]
# NOTE: for simple projects, project root is the parent folder
# of pipeline.yaml, for package projects is the parent folder
# of setup.py
if look_up_project_root_recursively:
project_root = (
None if not self._parent_path else
default.find_root_recursively(
starting_dir=self._parent_path,
filename=None if not self._path else self._path.name))
else:
project_root = self._parent_path
# make sure the folder where the pipeline is located is in sys.path
# otherwise dynamic imports needed by TaskSpec will fail
with add_to_sys_path(self._parent_path, chdir=False):
self.data['tasks'] = [
TaskSpec(t,
self.data['meta'],
project_root=project_root,
lazy_import=lazy_import,
reload=reload) for t in self.data['tasks']
]
else:
self.data['meta'] = Meta.empty()
def test_add_cli_args_from_env_dict_keys():
parser = ArgumentParser()
_add_cli_args_from_env_dict_keys(parser, EnvDict({'a': 1}))
assert {action.dest for action in parser._actions} == {'env__a', 'help'}