def test_set():
test_map = WhitelistMap.create()
@_whitelist_for_serdes(whitelist_map=test_map)
class HasSets(namedtuple("_HasSets", "reg_set frozen_set")):
def __new__(cls, reg_set, frozen_set):
set_param(reg_set, "reg_set")
inst_param(frozen_set, "frozen_set", frozenset)
return super(HasSets, cls).__new__(cls, reg_set, frozen_set)
foo = HasSets({1, 2, 3, "3"}, frozenset([4, 5, 6, "6"]))
serialized = _serialize_dagster_namedtuple(foo, whitelist_map=test_map)
foo_2 = _deserialize_json(serialized, whitelist_map=test_map)
assert foo == foo_2
# verify that set elements are serialized in a consistent order so that
# equal objects always have a consistent serialization / snapshot ID
big_foo = HasSets(set(string.ascii_lowercase),
frozenset(string.ascii_lowercase))
snap_id = hash_str(
_serialize_dagster_namedtuple(big_foo, whitelist_map=test_map))
roundtrip_snap_id = hash_str(
_serialize_dagster_namedtuple(
_deserialize_json(
_serialize_dagster_namedtuple(big_foo, whitelist_map=test_map),
whitelist_map=test_map,
),
whitelist_map=test_map,
))
assert snap_id == roundtrip_snap_id
def test_descent_path():
class Foo(NamedTuple):
bar: int
with pytest.raises(SerializationError,
match=re.escape("Descent path: <root:dict>.a.b[2].c")):
serialize_value({"a": {"b": [{}, {}, {"c": Foo(1)}]}})
test_map = WhitelistMap.create()
blank_map = WhitelistMap.create()
@_whitelist_for_serdes(whitelist_map=test_map)
class Fizz(NamedTuple):
buzz: int
ser = _serialize_dagster_namedtuple({"a": {
"b": [{}, {}, {
"c": Fizz(1)
}]
}},
whitelist_map=test_map)
with pytest.raises(DeserializationError,
match=re.escape("Descent path: <root:dict>.a.b[2].c")):
_deserialize_json(ser, whitelist_map=blank_map)
def test_from_storage_dict():
old_map = WhitelistMap.create()
@_whitelist_for_serdes(whitelist_map=old_map)
class MyThing(NamedTuple):
orig_name: str
serialized_old = _serialize_dagster_namedtuple(MyThing("old"),
whitelist_map=old_map)
class CompatSerializer(DefaultNamedTupleSerializer):
@classmethod
def value_from_storage_dict(cls, storage_dict, klass, args_for_class,
whitelist_map, descent_path):
# simplified demo of field renaming
return klass(
storage_dict.get("orig_name") or storage_dict.get("new_name"))
new_map = WhitelistMap.create()
@_whitelist_for_serdes(whitelist_map=new_map, serializer=CompatSerializer) # pylint: disable=function-redefined
class MyThing(NamedTuple):
new_name: str
deser_old_val = _deserialize_json(serialized_old, whitelist_map=new_map)
assert deser_old_val.new_name == "old"
serialized_new = _serialize_dagster_namedtuple(MyThing("new"),
whitelist_map=new_map)
deser_new_val = _deserialize_json(serialized_new, whitelist_map=new_map)
assert deser_new_val.new_name == "new"
def test_pipeline_run_dagster_run():
# serialize in current code
test_run = DagsterRun(pipeline_name="test")
test_str = serialize_dagster_namedtuple(test_run)
# deserialize in "legacy" code
legacy_env = WhitelistMap.create()
@_whitelist_for_serdes(legacy_env)
class PipelineRun(
namedtuple(
"_PipelineRun",
"pipeline_name run_id run_config mode solid_selection solids_to_execute "
"step_keys_to_execute status tags root_run_id parent_run_id "
"pipeline_snapshot_id execution_plan_snapshot_id external_pipeline_origin "
"pipeline_code_origin",
)
):
pass
@_whitelist_for_serdes(legacy_env) # pylint: disable=unused-variable
class PipelineRunStatus(Enum):
QUEUED = "QUEUED"
NOT_STARTED = "NOT_STARTED"
result = _deserialize_json(test_str, legacy_env)
assert isinstance(result, PipelineRun)
assert result.pipeline_name == test_run.pipeline_name
def test_forward_compat_serdes_new_field_with_default():
test_map = WhitelistMap.create()
@_whitelist_for_serdes(whitelist_map=test_map)
class Quux(namedtuple("_Quux", "foo bar")):
def __new__(cls, foo, bar):
return super(Quux, cls).__new__(cls, foo, bar) # pylint: disable=bad-super-call
assert test_map.has_tuple_entry("Quux")
klass, _, _ = test_map.get_tuple_entry("Quux")
assert klass is Quux
quux = Quux("zip", "zow")
serialized = _serialize_dagster_namedtuple(quux, whitelist_map=test_map)
# pylint: disable=function-redefined
@_whitelist_for_serdes(whitelist_map=test_map)
class Quux(namedtuple("_Quux", "foo bar baz")): # pylint: disable=bad-super-call
def __new__(cls, foo, bar, baz=None):
return super(Quux, cls).__new__(cls, foo, bar, baz=baz)
assert test_map.has_tuple_entry("Quux")
klass, _, _ = test_map.get_tuple_entry("Quux")
assert klass is Quux
deserialized = _deserialize_json(serialized, whitelist_map=test_map)
assert deserialized != quux
assert deserialized.foo == quux.foo
assert deserialized.bar == quux.bar
assert deserialized.baz is None
def test_from_unpacked():
test_map = WhitelistMap.create()
class CompatSerializer(DefaultNamedTupleSerializer):
@classmethod
def value_from_unpacked(cls, unpacked_dict, klass):
return DeprecatedAlphabet.legacy_load(unpacked_dict)
@_whitelist_for_serdes(whitelist_map=test_map, serializer=CompatSerializer)
class DeprecatedAlphabet(namedtuple("_DeprecatedAlphabet", "a b c")):
def __new__(cls, a, b, c):
raise Exception("DeprecatedAlphabet is deprecated")
@classmethod
def legacy_load(cls, storage_dict):
# instead of the DeprecatedAlphabet, directly invoke the namedtuple constructor
return super().__new__(
cls,
storage_dict.get("a"),
storage_dict.get("b"),
storage_dict.get("c"),
)
serialized = '{"__class__": "DeprecatedAlphabet", "a": "A", "b": "B", "c": "C"}'
nt = _deserialize_json(serialized, whitelist_map=test_map)
assert isinstance(nt, DeprecatedAlphabet)
def test_to_storage_value():
test_map = WhitelistMap.create()
class MySerializer(DefaultNamedTupleSerializer):
@classmethod
def value_to_storage_dict(cls, value, whitelist_map, descent_path):
return DefaultNamedTupleSerializer.value_to_storage_dict(
SubstituteAlphabet(value.a, value.b, value.c),
test_map,
descent_path,
)
@_whitelist_for_serdes(whitelist_map=test_map, serializer=MySerializer)
class DeprecatedAlphabet(namedtuple("_DeprecatedAlphabet", "a b c")):
def __new__(cls, a, b, c):
return super(DeprecatedAlphabet, cls).__new__(cls, a, b, c)
@_whitelist_for_serdes(whitelist_map=test_map)
class SubstituteAlphabet(namedtuple("_SubstituteAlphabet", "a b c")):
def __new__(cls, a, b, c):
return super(SubstituteAlphabet, cls).__new__(cls, a, b, c)
nested = DeprecatedAlphabet(None, None, "_C")
deprecated = DeprecatedAlphabet("A", "B", nested)
serialized = _serialize_dagster_namedtuple(deprecated,
whitelist_map=test_map)
alphabet = _deserialize_json(serialized, whitelist_map=test_map)
assert not isinstance(alphabet, DeprecatedAlphabet)
assert isinstance(alphabet, SubstituteAlphabet)
assert not isinstance(alphabet.c, DeprecatedAlphabet)
assert isinstance(alphabet.c, SubstituteAlphabet)
def test_forward_compat():
old_map = WhitelistMap.create()
@_whitelist_for_serdes(whitelist_map=old_map)
class Quux(namedtuple("_Quux", "bar baz")):
def __new__(cls, bar, baz):
return super().__new__(cls, bar, baz)
# new version has a new field with a new type
new_map = WhitelistMap.create()
# pylint: disable=function-redefined
@_whitelist_for_serdes(whitelist_map=new_map)
class Quux(namedtuple("_Quux", "foo bar baz")):
def __new__(cls, foo, bar, baz):
return super().__new__(cls, foo, bar, baz)
@_whitelist_for_serdes(whitelist_map=new_map)
class Foo(namedtuple("_Foo", "wow")):
def __new__(cls, wow):
return super().__new__(cls, wow)
new_quux = Quux(foo=Foo("wow"), bar="bar", baz="baz")
# write from new
serialized = _serialize_dagster_namedtuple(new_quux, whitelist_map=new_map)
# read from old, foo ignored
deserialized = _deserialize_json(serialized, whitelist_map=old_map)
assert deserialized.bar == "bar"
assert deserialized.baz == "baz"
def test_skip_when_empty():
test_map = WhitelistMap.create()
@_whitelist_for_serdes(whitelist_map=test_map)
class SameSnapshotTuple(namedtuple("_Tuple", "foo")):
def __new__(cls, foo):
return super(SameSnapshotTuple, cls).__new__(cls, foo) # pylint: disable=bad-super-call
old_tuple = SameSnapshotTuple(foo="A")
old_serialized = _serialize_dagster_namedtuple(old_tuple,
whitelist_map=test_map)
old_snapshot = hash_str(old_serialized)
# Without setting skip_when_empty, the ID changes
@_whitelist_for_serdes(whitelist_map=test_map) # pylint: disable=function-redefined
class SameSnapshotTuple(namedtuple("_Tuple", "foo bar")):
def __new__(cls, foo, bar=None):
return super(SameSnapshotTuple, cls).__new__( # pylint: disable=bad-super-call
cls, foo, bar)
new_tuple_without_serializer = SameSnapshotTuple(foo="A")
new_snapshot_without_serializer = hash_str(
_serialize_dagster_namedtuple(new_tuple_without_serializer,
whitelist_map=test_map))
assert new_snapshot_without_serializer != old_snapshot
# By setting a custom serializer and skip_when_empty, the snapshot stays the same
# as long as the new field is None
class SkipWhenEmptySerializer(DefaultNamedTupleSerializer):
@classmethod
def skip_when_empty(cls) -> Set[str]:
return {"bar"}
@_whitelist_for_serdes(whitelist_map=test_map,
serializer=SkipWhenEmptySerializer) # pylint: disable=function-redefined
class SameSnapshotTuple(namedtuple("_Tuple", "foo bar")):
def __new__(cls, foo, bar=None):
return super(SameSnapshotTuple, cls).__new__( # pylint: disable=bad-super-call
cls, foo, bar)
for bar_val in [None, [], {}, set()]:
new_tuple = SameSnapshotTuple(foo="A", bar=bar_val)
new_snapshot = hash_str(
_serialize_dagster_namedtuple(new_tuple, whitelist_map=test_map))
assert old_snapshot == new_snapshot
rehydrated_tuple = _deserialize_json(old_serialized,
whitelist_map=test_map)
assert rehydrated_tuple.foo == "A"
assert rehydrated_tuple.bar is None
new_tuple_with_bar = SameSnapshotTuple(foo="A", bar="B")
assert new_tuple_with_bar.foo == "A"
assert new_tuple_with_bar.bar == "B"
def test_namedtuple_name_map():
wmap = WhitelistMap.create()
@_whitelist_for_serdes(whitelist_map=wmap)
class Thing(NamedTuple):
name: str
wmap.register_serialized_name("Thing", "SerializedThing")
thing = Thing("foo")
thing_serialized = _serialize_dagster_namedtuple(thing, wmap)
assert seven.json.loads(thing_serialized)["__class__"] == "SerializedThing"
with pytest.raises(DeserializationError):
_deserialize_json(thing_serialized, wmap)
wmap.register_deserialized_name("SerializedThing", "Thing")
assert _deserialize_json(thing_serialized, wmap) == thing
def test_long_int():
test_map = WhitelistMap.create()
@_whitelist_for_serdes(whitelist_map=test_map)
class NumHolder(NamedTuple):
num: int
x = NumHolder(98765432109876543210)
ser_x = _serialize_dagster_namedtuple(x, test_map)
roundtrip_x = _deserialize_json(ser_x, test_map)
assert x.num == roundtrip_x.num
def test_enum_backcompat():
test_env = WhitelistMap.create()
class MyEnumSerializer(EnumSerializer):
@classmethod
def value_from_storage_str(cls, storage_str, klass):
return getattr(klass, storage_str)
@classmethod
def value_to_storage_str(cls, value, whitelist_map, descent_path):
val_as_str = str(value)
actual_enum_val = val_as_str.split(".")[1:]
backcompat_name = (
"OldEnum" # Simulate changing the storage name to some legacy backcompat name
)
return ".".join([backcompat_name, *actual_enum_val])
@_whitelist_for_serdes(test_env, serializer=MyEnumSerializer)
class MyEnum(Enum):
RED = "color.red"
BLUE = "color.red"
# Ensure that serdes roundtrip preserves value
register_serdes_enum_fallbacks({"OldEnum": MyEnum}, whitelist_map=test_env)
my_enum = MyEnum("color.red")
enum_json = serialize_value(my_enum, whitelist_map=test_env)
result = _deserialize_json(enum_json, test_env)
assert result == my_enum
# ensure that "legacy" environment can correctly interpret enum stored under legacy name.
legacy_env = WhitelistMap.create()
@_whitelist_for_serdes(legacy_env)
class OldEnum(Enum):
RED = "color.red"
BLUE = "color.blue"
result = _deserialize_json(enum_json, legacy_env)
old_enum = OldEnum("color.red")
assert old_enum == result
def test_persistent_tuple():
test_map = WhitelistMap.create()
@_whitelist_for_serdes(whitelist_map=test_map)
class Alphabet(namedtuple("_Alphabet", "a b c")):
def __new__(cls, a, b, c):
return super(Alphabet, cls).__new__(cls, a, b, c)
foo = Alphabet(a="A", b="B", c="C")
serialized = _serialize_dagster_namedtuple(foo, whitelist_map=test_map)
foo_2 = _deserialize_json(serialized, whitelist_map=test_map)
assert foo == foo_2
def test_solid_handle_node_handle():
# serialize in current code
test_handle = NodeHandle("test", None)
test_str = serialize_dagster_namedtuple(test_handle)
# deserialize in "legacy" code
legacy_env = WhitelistMap.create()
@_whitelist_for_serdes(legacy_env)
class SolidHandle(namedtuple("_SolidHandle", "name parent")):
pass
result = _deserialize_json(test_str, legacy_env)
assert isinstance(result, SolidHandle)
assert result.name == test_handle.name
def test_pipeline_run_status_dagster_run_status():
# serialize in current code
test_status = DagsterRunStatus("QUEUED")
test_str = serialize_value(test_status)
# deserialize in "legacy" code
legacy_env = WhitelistMap.create()
@_whitelist_for_serdes(legacy_env)
class PipelineRunStatus(Enum):
QUEUED = "QUEUED"
result = _deserialize_json(test_str, legacy_env)
assert isinstance(result, PipelineRunStatus)
assert result.value == test_status.value
def test_backward_compat_serdes():
test_map = WhitelistMap.create()
@_whitelist_for_serdes(whitelist_map=test_map)
class Quux(namedtuple("_Quux", "foo bar baz")):
def __new__(cls, foo, bar, baz):
return super(Quux, cls).__new__(cls, foo, bar, baz) # pylint: disable=bad-super-call
quux = Quux("zip", "zow", "whoopie")
serialized = _serialize_dagster_namedtuple(quux, whitelist_map=test_map)
# pylint: disable=function-redefined
@_whitelist_for_serdes(whitelist_map=test_map)
class Quux(namedtuple("_Quux", "foo bar")): # pylint: disable=bad-super-call
def __new__(cls, foo, bar):
return super(Quux, cls).__new__(cls, foo, bar)
deserialized = _deserialize_json(serialized, whitelist_map=test_map)
assert deserialized != quux
assert deserialized.foo == quux.foo
assert deserialized.bar == quux.bar
assert not hasattr(deserialized, "baz")
def test_legacy_event_log_load():
# ensure EventLogEntry 0.14.3+ can still be loaded by older dagster versions
# to avoid downgrades etc from creating operational issues
legacy_env = WhitelistMap.create()
# snapshot of EventLogEntry pre commit ea19544
@_whitelist_for_serdes(
whitelist_map=legacy_env,
storage_name="EventLogEntry", # use this to avoid collision with current EventLogEntry
)
class OldEventLogEntry( # pylint: disable=unused-variable
NamedTuple(
"_OldEventLogEntry",
[
("error_info", Optional[SerializableErrorInfo]),
("message", str),
("level", Union[str, int]),
("user_message", str),
("run_id", str),
("timestamp", float),
("step_key", Optional[str]),
("pipeline_name", Optional[str]),
("dagster_event", Optional[DagsterEvent]),
],
)
):
def __new__(
cls,
error_info,
message,
level,
user_message,
run_id,
timestamp,
step_key=None,
pipeline_name=None,
dagster_event=None,
job_name=None,
):
pipeline_name = pipeline_name or job_name
return super().__new__(
cls,
check.opt_inst_param(error_info, "error_info", SerializableErrorInfo),
check.str_param(message, "message"),
level, # coerce_valid_log_level call omitted
check.str_param(user_message, "user_message"),
check.str_param(run_id, "run_id"),
check.float_param(timestamp, "timestamp"),
check.opt_str_param(step_key, "step_key"),
check.opt_str_param(pipeline_name, "pipeline_name"),
check.opt_inst_param(dagster_event, "dagster_event", DagsterEvent),
)
# current event log entry
new_event = EventLogEntry(
user_message="test 1 2 3",
error_info=None,
level="debug",
run_id="fake_run_id",
timestamp=time.time(),
)
storage_str = serialize_dagster_namedtuple(new_event)
result = _deserialize_json(storage_str, legacy_env)
assert result.message is not None
def test_namedtuple_backcompat():
old_map = WhitelistMap.create()
@_whitelist_for_serdes(whitelist_map=old_map)
class OldThing(NamedTuple):
old_name: str
def get_id(self):
json_rep = _serialize_dagster_namedtuple(self,
whitelist_map=old_map)
return hash_str(json_rep)
# create the old things
old_thing = OldThing("thing")
old_thing_id = old_thing.get_id()
old_thing_serialized = _serialize_dagster_namedtuple(old_thing, old_map)
new_map = WhitelistMap.create()
class ThingSerializer(DefaultNamedTupleSerializer):
@classmethod
def value_from_storage_dict(cls, storage_dict, klass, args_for_class,
whitelist_map, descent_path):
raw_dict = {
key: unpack_inner_value(value, whitelist_map,
f"{descent_path}.{key}")
for key, value in storage_dict.items()
}
# typical pattern is to use the same serialization format from an old field and passing
# it in as a new field
return klass(
**{
key: value
for key, value in raw_dict.items() if key in args_for_class
},
new_name=raw_dict.get("old_name"),
)
@classmethod
def value_to_storage_dict(
cls,
value,
whitelist_map,
descent_path,
):
storage = super().value_to_storage_dict(value, whitelist_map,
descent_path)
name = storage.get("new_name") or storage.get("old_name")
if "new_name" in storage:
del storage["new_name"]
# typical pattern is to use the same serialization format
# it in as a new field
storage["old_name"] = name
storage["__class__"] = "OldThing" # persist using old class name
return storage
@_whitelist_for_serdes(whitelist_map=new_map, serializer=ThingSerializer)
class NewThing(NamedTuple):
new_name: str
def get_id(self):
json_rep = _serialize_dagster_namedtuple(self,
whitelist_map=new_map)
return hash_str(json_rep)
# exercising the old serialization format
register_serdes_tuple_fallbacks({"OldThing": NewThing},
whitelist_map=new_map)
new_thing = NewThing("thing")
new_thing_id = new_thing.get_id()
new_thing_serialized = _serialize_dagster_namedtuple(new_thing, new_map)
assert new_thing_id == old_thing_id
assert new_thing_serialized == old_thing_serialized
# ensure that the new serializer can correctly interpret old serialized data
old_thing_deserialized = _deserialize_json(old_thing_serialized, new_map)
assert isinstance(old_thing_deserialized, NewThing)
assert old_thing_deserialized.get_id() == new_thing_id
# ensure that the new things serialized can still be read by old code
new_thing_deserialized = _deserialize_json(new_thing_serialized, old_map)
assert isinstance(new_thing_deserialized, OldThing)
assert new_thing_deserialized.get_id() == old_thing_id
def test_external_job_origin_instigator_origin():
def build_legacy_whitelist_map():
legacy_env = WhitelistMap.create()
@_whitelist_for_serdes(legacy_env)
class ExternalJobOrigin(
namedtuple("_ExternalJobOrigin",
"external_repository_origin job_name")):
def get_id(self):
return create_snapshot_id(self)
@_whitelist_for_serdes(legacy_env)
class ExternalRepositoryOrigin(
namedtuple("_ExternalRepositoryOrigin",
"repository_location_origin repository_name")):
def get_id(self):
return create_snapshot_id(self)
class GrpcServerOriginSerializer(DefaultNamedTupleSerializer):
@classmethod
def skip_when_empty(cls):
return {"use_ssl"}
@_whitelist_for_serdes(whitelist_map=legacy_env,
serializer=GrpcServerOriginSerializer)
class GrpcServerRepositoryLocationOrigin(
namedtuple(
"_GrpcServerRepositoryLocationOrigin",
"host port socket location_name use_ssl",
), ):
def __new__(cls,
host,
port=None,
socket=None,
location_name=None,
use_ssl=None):
return super(GrpcServerRepositoryLocationOrigin,
cls).__new__(cls, host, port, socket,
location_name, use_ssl)
return (
legacy_env,
ExternalJobOrigin,
ExternalRepositoryOrigin,
GrpcServerRepositoryLocationOrigin,
)
legacy_env, klass, repo_klass, location_klass = build_legacy_whitelist_map(
)
from dagster.core.host_representation.origin import (
ExternalInstigatorOrigin,
ExternalRepositoryOrigin,
GrpcServerRepositoryLocationOrigin,
)
# serialize from current code, compare against old code
instigator_origin = ExternalInstigatorOrigin(
external_repository_origin=ExternalRepositoryOrigin(
repository_location_origin=GrpcServerRepositoryLocationOrigin(
host="localhost", port=1234, location_name="test_location"),
repository_name="the_repo",
),
instigator_name="simple_schedule",
)
instigator_origin_str = serialize_dagster_namedtuple(instigator_origin)
instigator_to_job = _deserialize_json(instigator_origin_str, legacy_env)
assert isinstance(instigator_to_job, klass)
# ensure that the origin id is stable
assert instigator_to_job.get_id() == instigator_origin.get_id()
# # serialize from old code, compare against current code
job_origin = klass(
external_repository_origin=repo_klass(
repository_location_origin=location_klass(
host="localhost", port=1234, location_name="test_location"),
repository_name="the_repo",
),
job_name="simple_schedule",
)
job_origin_str = serialize_value(job_origin, legacy_env)
from dagster.serdes.serdes import _WHITELIST_MAP
job_to_instigator = deserialize_json_to_dagster_namedtuple(job_origin_str)
assert isinstance(job_to_instigator, ExternalInstigatorOrigin)
# ensure that the origin id is stable
assert job_to_instigator.get_id() == job_origin.get_id()
