redis-om-python/redis_developer/orm/model.py

685 lines
23 KiB
Python
Raw Normal View History

import abc
import dataclasses
import decimal
import operator
from copy import copy, deepcopy
from dataclasses import dataclass
from enum import Enum
from functools import reduce
from typing import (
AbstractSet,
Any,
Callable,
Dict,
Mapping,
Optional,
Set,
Tuple,
TypeVar,
Union,
2021-09-01 21:56:06 +02:00
Sequence,
no_type_check,
Protocol,
List, Type
)
2021-09-01 01:30:31 +02:00
import uuid
import redis
2021-09-01 21:56:06 +02:00
from pydantic import BaseModel, validator
from pydantic.fields import FieldInfo as PydanticFieldInfo
from pydantic.fields import ModelField, Undefined, UndefinedType
from pydantic.main import ModelMetaclass
2021-09-01 21:56:06 +02:00
from pydantic.typing import NoArgAnyCallable
from pydantic.utils import Representation
from .encoders import jsonable_encoder
model_registry = {}
_T = TypeVar("_T")
class RedisModelError(Exception):
pass
class NotFoundError(Exception):
pass
class Operators(Enum):
EQ = 1
NE = 2
LT = 3
LE = 4
GT = 5
GE = 6
OR = 7
AND = 8
NOT = 9
IN = 10
NOT_IN = 11
LIKE = 12
2021-09-16 21:03:03 +02:00
@dataclass
class NegatedExpression:
expression: 'Expression'
def __invert__(self):
return self.expression
2021-09-21 01:06:04 +02:00
def __and__(self, other):
return Expression(left=self, op=Operators.AND, right=other)
def __or__(self, other):
return Expression(left=self, op=Operators.OR, right=other)
2021-09-16 21:03:03 +02:00
@dataclass
class Expression:
op: Operators
left: Any
right: Any
2021-09-16 21:03:03 +02:00
def __invert__(self):
return NegatedExpression(self)
def __and__(self, other):
return Expression(left=self, op=Operators.AND, right=other)
2021-09-16 23:35:25 +02:00
def __or__(self, other):
return Expression(left=self, op=Operators.OR, right=other)
2021-09-16 21:03:03 +02:00
ExpressionOrNegated = Union[Expression, NegatedExpression]
class QueryNotSupportedError(Exception):
"""The attempted query is not supported."""
class RediSearchFieldTypes(Enum):
TEXT = 'TEXT'
TAG = 'TAG'
NUMERIC = 'NUMERIC'
GEO = 'GEO'
# TODO: How to handle Geo fields?
NUMERIC_TYPES = (float, int, decimal.Decimal)
@dataclass
class FindQuery:
expressions: Sequence[Expression]
expression: Expression = dataclasses.field(init=False)
query: str = dataclasses.field(init=False)
2021-09-16 23:35:25 +02:00
pagination: List[str] = dataclasses.field(init=False)
model: Type['RedisModel']
2021-09-16 23:35:25 +02:00
limit: Optional[int] = None
offset: Optional[int] = None
def __post_init__(self):
self.expression = reduce(operator.and_, self.expressions)
self.query = self.resolve_redisearch_query(self.expression)
2021-09-16 23:35:25 +02:00
self.pagination = self.resolve_redisearch_pagination()
def resolve_field_type(self, field: ModelField) -> RediSearchFieldTypes:
2021-09-21 01:06:04 +02:00
if getattr(field.field_info, 'primary_key', None) is True:
return RediSearchFieldTypes.TAG
2021-09-21 01:06:04 +02:00
elif getattr(field.field_info, 'full_text_search', None) is True:
return RediSearchFieldTypes.TEXT
field_type = field.outer_type_
# TODO: GEO
if any(issubclass(field_type, t) for t in NUMERIC_TYPES):
return RediSearchFieldTypes.NUMERIC
else:
2021-09-21 01:06:04 +02:00
# TAG fields are the default field type.
return RediSearchFieldTypes.TAG
@staticmethod
def expand_tag_value(value):
err = RedisModelError(f"Using the IN operator requires passing an iterable of "
"possible values. You passed: {value}")
if isinstance(str, value):
raise err
try:
expanded_value = "|".join(value)
except TypeError:
raise err
return expanded_value
def resolve_value(self, field_name: str, field_type: RediSearchFieldTypes,
op: Operators, value: Any) -> str:
result = ""
if field_type is RediSearchFieldTypes.TEXT:
result = f"@{field_name}:"
if op is Operators.EQ:
2021-09-16 21:03:03 +02:00
result += f'"{value}"'
elif op is Operators.NE:
result = f'-({result}"{value}")'
elif op is Operators.LIKE:
result += value
else:
2021-09-16 21:03:03 +02:00
# TODO: Handling TAG, TEXT switch-offs, etc.
raise QueryNotSupportedError("Only equals (=) and not-equals (!=) comparisons are "
"currently supported for TEXT fields. See docs: TODO")
elif field_type is RediSearchFieldTypes.NUMERIC:
if op is Operators.EQ:
result += f"@{field_name}:[{value} {value}]"
elif op is Operators.NE:
2021-09-21 01:06:04 +02:00
# TODO: Is this enough or do we also need a clause for all values
# ([-inf +inf]) from which we then subtract the undesirable value?
2021-09-21 01:08:24 +02:00
result += f"-(@{field_name}:[{value} {value}])"
elif op is Operators.GT:
result += f"@{field_name}:[({value} +inf]"
elif op is Operators.LT:
result += f"@{field_name}:[-inf ({value}]"
elif op is Operators.GE:
result += f"@{field_name}:[{value} +inf]"
elif op is Operators.LE:
result += f"@{field_name}:[-inf {value}]"
2021-09-21 01:06:04 +02:00
elif field_type is RediSearchFieldTypes.TAG:
if op is Operators.EQ:
result += f"@{field_name}:{{{value}}}"
elif op is Operators.NE:
2021-09-21 01:08:24 +02:00
result += f"-(@{field_name}:{{{value}}})"
2021-09-21 01:06:04 +02:00
elif op is Operators.IN:
expanded_value = self.expand_tag_value(value)
result += f"(@{field_name}:{{{expanded_value}}})"
elif op is Operators.NOT_IN:
expanded_value = self.expand_tag_value(value)
2021-09-21 01:08:24 +02:00
result += f"-(@{field_name}:{{{expanded_value}}})"
2021-09-16 21:03:03 +02:00
return result
2021-09-16 23:35:25 +02:00
def resolve_redisearch_pagination(self):
"""Resolve pagination options for a query."""
if not self.limit and not self.offset:
return []
offset = self.offset or 0
limit = self.limit or 10
return ["LIMIT", offset, limit]
2021-09-16 21:03:03 +02:00
def resolve_redisearch_query(self, expression: ExpressionOrNegated):
"""Resolve an expression to a string RediSearch query."""
field_type = None
field_name = None
2021-09-21 01:06:04 +02:00
encompassing_expression_is_negated = False
result = ""
2021-09-16 21:03:03 +02:00
if isinstance(expression, NegatedExpression):
2021-09-21 01:06:04 +02:00
encompassing_expression_is_negated = True
2021-09-16 21:03:03 +02:00
expression = expression.expression
2021-09-21 01:06:04 +02:00
if isinstance(expression.left, Expression) or \
isinstance(expression.left, NegatedExpression):
result += f"({self.resolve_redisearch_query(expression.left)})"
elif isinstance(expression.left, ModelField):
field_type = self.resolve_field_type(expression.left)
field_name = expression.left.name
else:
2021-09-21 01:06:04 +02:00
import ipdb; ipdb.set_trace()
raise QueryNotSupportedError(f"A query expression should start with either a field "
f"or an expression enclosed in parenthesis. See docs: "
f"TODO")
2021-09-16 21:03:03 +02:00
right = expression.right
right_is_negated = isinstance(right, NegatedExpression)
if isinstance(right, Expression) or right_is_negated:
if expression.op == Operators.AND:
2021-09-16 21:03:03 +02:00
result += " "
elif expression.op == Operators.OR:
2021-09-16 21:03:03 +02:00
result += "| "
else:
raise QueryNotSupportedError("You can only combine two query expressions with"
2021-09-16 21:03:03 +02:00
"AND (&) or OR (|). See docs: TODO")
if right_is_negated:
result += "-"
# We're handling the RediSearch operator in this call ("-"), so resolve the
# inner expression instead of the NegatedExpression.
right = right.expression
result += f"({self.resolve_redisearch_query(right)})"
else:
2021-09-16 21:03:03 +02:00
if isinstance(right, ModelField):
raise QueryNotSupportedError("Comparing fields is not supported. See docs: TODO")
else:
2021-09-16 21:03:03 +02:00
result += self.resolve_value(field_name, field_type, expression.op, right)
2021-09-21 01:06:04 +02:00
if encompassing_expression_is_negated:
2021-09-16 21:03:03 +02:00
result = f"-({result})"
return result
def find(self):
2021-09-16 23:35:25 +02:00
args = ["ft.search", self.model.Meta.index_name, self.query]
# TODO: Do we need self.pagination if we're just appending to query anyway?
if self.pagination:
args.extend(self.pagination)
return self.model.db().execute_command(*args)
2021-09-01 01:30:31 +02:00
class PrimaryKeyCreator(Protocol):
2021-09-01 21:56:06 +02:00
def create_pk(self, *args, **kwargs) -> str:
2021-09-01 01:30:31 +02:00
"""Create a new primary key"""
class Uuid4PrimaryKey:
2021-09-01 21:56:06 +02:00
def create_pk(self) -> str:
2021-09-01 01:30:31 +02:00
return str(uuid.uuid4())
class ExpressionProxy:
def __init__(self, field: ModelField):
self.field = field
def __eq__(self, other: Any) -> Expression:
return Expression(left=self.field, op=Operators.EQ, right=other)
def __ne__(self, other: Any) -> Expression:
return Expression(left=self.field, op=Operators.NE, right=other)
def __lt__(self, other: Any) -> Expression:
return Expression(left=self.field, op=Operators.LT, right=other)
def __le__(self, other: Any) -> Expression:
return Expression(left=self.field, op=Operators.LE, right=other)
def __gt__(self, other: Any) -> Expression:
return Expression(left=self.field, op=Operators.GT, right=other)
def __ge__(self, other: Any) -> Expression:
return Expression(left=self.field, op=Operators.GE, right=other)
def __dataclass_transform__(
*,
eq_default: bool = True,
order_default: bool = False,
kw_only_default: bool = False,
field_descriptors: Tuple[Union[type, Callable[..., Any]], ...] = (()),
) -> Callable[[_T], _T]:
return lambda a: a
class FieldInfo(PydanticFieldInfo):
def __init__(self, default: Any = Undefined, **kwargs: Any) -> None:
primary_key = kwargs.pop("primary_key", False)
sortable = kwargs.pop("sortable", Undefined)
index = kwargs.pop("index", Undefined)
full_text_search = kwargs.pop("full_text_search", Undefined)
super().__init__(default=default, **kwargs)
self.primary_key = primary_key
self.sortable = sortable
self.index = index
self.full_text_search = full_text_search
class RelationshipInfo(Representation):
def __init__(
self,
*,
back_populates: Optional[str] = None,
link_model: Optional[Any] = None,
) -> None:
self.back_populates = back_populates
self.link_model = link_model
def Field(
default: Any = Undefined,
*,
default_factory: Optional[NoArgAnyCallable] = None,
alias: str = None,
title: str = None,
description: str = None,
exclude: Union[
AbstractSet[Union[int, str]], Mapping[Union[int, str], Any], Any
] = None,
include: Union[
AbstractSet[Union[int, str]], Mapping[Union[int, str], Any], Any
] = None,
const: bool = None,
gt: float = None,
ge: float = None,
lt: float = None,
le: float = None,
multiple_of: float = None,
min_items: int = None,
max_items: int = None,
min_length: int = None,
max_length: int = None,
allow_mutation: bool = True,
regex: str = None,
primary_key: bool = False,
sortable: Union[bool, UndefinedType] = Undefined,
index: Union[bool, UndefinedType] = Undefined,
full_text_search: Union[bool, UndefinedType] = Undefined,
schema_extra: Optional[Dict[str, Any]] = None,
) -> Any:
current_schema_extra = schema_extra or {}
field_info = FieldInfo(
default,
default_factory=default_factory,
alias=alias,
title=title,
description=description,
exclude=exclude,
include=include,
const=const,
gt=gt,
ge=ge,
lt=lt,
le=le,
multiple_of=multiple_of,
min_items=min_items,
max_items=max_items,
min_length=min_length,
max_length=max_length,
allow_mutation=allow_mutation,
regex=regex,
primary_key=primary_key,
sortable=sortable,
index=index,
full_text_search=full_text_search,
**current_schema_extra,
)
field_info._validate()
return field_info
@dataclass
class PrimaryKey:
name: str
field: ModelField
class DefaultMeta:
global_key_prefix: Optional[str] = None
model_key_prefix: Optional[str] = None
primary_key_pattern: Optional[str] = None
database: Optional[redis.Redis] = None
primary_key: Optional[PrimaryKey] = None
2021-09-01 21:56:06 +02:00
primary_key_creator_cls: Type[PrimaryKeyCreator] = None
index_name: str = None
abstract: bool = False
class ModelMeta(ModelMetaclass):
def __new__(cls, name, bases, attrs, **kwargs): # noqa C901
meta = attrs.pop('Meta', None)
new_class = super().__new__(cls, name, bases, attrs, **kwargs)
meta = meta or getattr(new_class, 'Meta', None)
base_meta = getattr(new_class, '_meta', None)
if meta and meta != DefaultMeta and meta != base_meta:
new_class.Meta = meta
new_class._meta = meta
elif base_meta:
new_class._meta = deepcopy(base_meta)
new_class.Meta = new_class._meta
# Unset inherited values we don't want to reuse (typically based on the model name).
new_class._meta.abstract = False
new_class._meta.model_key_prefix = None
new_class._meta.index_name = None
else:
new_class._meta = deepcopy(DefaultMeta)
new_class.Meta = new_class._meta
# Not an abstract model class
if abc.ABC not in bases:
key = f"{new_class.__module__}.{new_class.__name__}"
model_registry[key] = new_class
# Create proxies for each model field so that we can use the field
# in queries, like Model.get(Model.field_name == 1)
for field_name, field in new_class.__fields__.items():
setattr(new_class, field_name, ExpressionProxy(field))
# Check if this is our FieldInfo version with extended ORM metadata.
if isinstance(field.field_info, FieldInfo):
if field.field_info.primary_key:
new_class._meta.primary_key = PrimaryKey(name=field_name, field=field)
if not getattr(new_class._meta, 'global_key_prefix', None):
new_class._meta.global_key_prefix = getattr(base_meta, "global_key_prefix", "")
if not getattr(new_class._meta, 'model_key_prefix', None):
# Don't look at the base class for this.
new_class._meta.model_key_prefix = f"{new_class.__module__}.{new_class.__name__}"
if not getattr(new_class._meta, 'primary_key_pattern', None):
new_class._meta.primary_key_pattern = getattr(base_meta, "primary_key_pattern",
"{pk}")
if not getattr(new_class._meta, 'database', None):
new_class._meta.database = getattr(base_meta, "database",
redis.Redis(decode_responses=True))
if not getattr(new_class._meta, 'primary_key_creator_cls', None):
new_class._meta.primary_key_creator_cls = getattr(base_meta, "primary_key_creator_cls",
Uuid4PrimaryKey)
if not getattr(new_class._meta, 'index_name', None):
new_class._meta.index_name = f"{new_class._meta.global_key_prefix}:" \
f"{new_class._meta.model_key_prefix}:index"
return new_class
class RedisModel(BaseModel, abc.ABC, metaclass=ModelMeta):
pk: Optional[str] = Field(default=None, primary_key=True)
Meta = DefaultMeta
class Config:
orm_mode = True
arbitrary_types_allowed = True
extra = 'allow'
def __init__(__pydantic_self__, **data: Any) -> None:
2021-09-01 01:30:31 +02:00
super().__init__(**data)
__pydantic_self__.validate_primary_key()
def __lt__(self, other):
my_keys = set(self.__fields__.keys())
other_keys = set(other.__fields__.keys())
shared_keys = list(my_keys & other_keys)
lt = [getattr(self, k) < getattr(other, k) for k in shared_keys]
return len(lt) > len(shared_keys) / 2
2021-09-01 21:56:06 +02:00
@validator("pk", always=True)
def validate_pk(cls, v):
if not v:
v = cls._meta.primary_key_creator_cls().create_pk()
2021-09-01 21:56:06 +02:00
return v
@classmethod
def validate_primary_key(cls):
"""Check for a primary key. We need one (and only one)."""
primary_keys = 0
for name, field in cls.__fields__.items():
2021-09-01 01:30:31 +02:00
if getattr(field.field_info, 'primary_key', None):
primary_keys += 1
if primary_keys == 0:
raise RedisModelError("You must define a primary key for the model")
elif primary_keys > 1:
raise RedisModelError("You must define only one primary key for a model")
@classmethod
2021-08-31 21:03:53 +02:00
def make_key(cls, part: str):
global_prefix = getattr(cls._meta, 'global_key_prefix', '').strip(":")
model_prefix = getattr(cls._meta, 'model_key_prefix', '').strip(":")
2021-09-01 21:56:06 +02:00
return f"{global_prefix}:{model_prefix}:{part}"
2021-08-31 21:03:53 +02:00
@classmethod
2021-09-01 01:30:31 +02:00
def make_primary_key(cls, pk: Any):
2021-08-31 21:03:53 +02:00
"""Return the Redis key for this model."""
return cls.make_key(cls._meta.primary_key_pattern.format(pk=pk))
2021-08-31 21:03:53 +02:00
def key(self):
"""Return the Redis key for this model."""
pk = getattr(self, self._meta.primary_key.field.name)
2021-08-31 21:03:53 +02:00
return self.make_primary_key(pk)
@classmethod
def db(cls):
return cls._meta.database
2021-09-16 21:03:03 +02:00
@classmethod
def from_redis(cls, res: Any):
import six
from six.moves import xrange, zip as izip
2021-09-16 21:03:03 +02:00
def to_string(s):
if isinstance(s, six.string_types):
return s
elif isinstance(s, six.binary_type):
return s.decode('utf-8','ignore')
else:
return s # Not a string we care about
2021-09-16 21:03:03 +02:00
docs = []
step = 2 # Because the result has content
offset = 1
for i in xrange(1, len(res), step):
fields_offset = offset
fields = dict(
dict(izip(map(to_string, res[i + fields_offset][::2]),
map(to_string, res[i + fields_offset][1::2])))
)
try:
del fields['id']
except KeyError:
pass
doc = cls(**fields)
docs.append(doc)
return docs
@classmethod
def find(cls, *expressions: Expression):
query = FindQuery(expressions=expressions, model=cls)
raw_result = query.find()
return cls.from_redis(raw_result)
@classmethod
2021-09-16 23:35:25 +02:00
def find_one(cls, *expressions: Expression):
query = FindQuery(expressions=expressions, model=cls, limit=1, offset=0)
raw_result = query.find()
return cls.from_redis(raw_result)[0]
@classmethod
def add(cls, models: Sequence['RedisModel']) -> Sequence['RedisModel']:
return [model.save() for model in models]
@classmethod
def update(cls, **field_values):
return cls
@classmethod
def values(cls):
"""Return raw values from Redis instead of model instances."""
return cls
2021-08-31 21:03:53 +02:00
def delete(self):
return self.db().delete(self.key())
def save(self, *args, **kwargs) -> 'RedisModel':
2021-09-01 21:56:06 +02:00
raise NotImplementedError
@classmethod
def schema(cls):
2021-09-01 21:56:06 +02:00
raise NotImplementedError
class HashModel(RedisModel, abc.ABC):
2021-09-01 21:56:06 +02:00
def __init_subclass__(cls, **kwargs):
super().__init_subclass__(**kwargs)
for name, field in cls.__fields__.items():
if issubclass(field.outer_type_, RedisModel):
raise RedisModelError(f"HashModels cannot have embedded model "
f"fields. Field: {name}")
2021-09-01 21:56:06 +02:00
for typ in (Set, Mapping, List):
if issubclass(field.outer_type_, typ):
raise RedisModelError(f"HashModels cannot have set, list,"
f" or mapping fields. Field: {name}")
2021-09-01 21:56:06 +02:00
def save(self, *args, **kwargs) -> 'HashModel':
document = jsonable_encoder(self.dict())
2021-08-31 21:03:53 +02:00
success = self.db().hset(self.key(), mapping=document)
2021-09-01 21:56:06 +02:00
return success
@classmethod
def get(cls, pk: Any) -> 'HashModel':
document = cls.db().hgetall(cls.make_primary_key(pk))
if not document:
raise NotFoundError
return cls.parse_obj(document)
@classmethod
@no_type_check
def _get_value(cls, *args, **kwargs) -> Any:
"""
Always send None as an empty string.
TODO: We do this because redis-py's hset() method requires non-null
values. Is there a better way?
"""
val = super()._get_value(*args, **kwargs)
if val is None:
return ""
return val
@classmethod
def schema_for_type(cls, name, typ: Type, field_info: FieldInfo):
if any(issubclass(typ, t) for t in NUMERIC_TYPES):
return f"{name} NUMERIC"
elif issubclass(typ, str):
if getattr(field_info, 'full_text_search', False) is True:
return f"{name} TAG {name}_fts TEXT"
else:
return f"{name} TAG"
else:
return f"{name} TAG"
2021-09-16 21:03:03 +02:00
@classmethod
def schema(cls):
hash_prefix = cls.make_key(cls._meta.primary_key_pattern.format(pk=""))
schema_prefix = f"ON HASH PREFIX 1 {hash_prefix} SCHEMA"
schema_parts = [schema_prefix]
for name, field in cls.__fields__.items():
_type = field.outer_type_
if getattr(field.field_info, 'primary_key', None):
if issubclass(_type, str):
redisearch_field = f"{name} TAG"
else:
redisearch_field = cls.schema_for_type(name, _type, field.field_info)
schema_parts.append(redisearch_field)
elif getattr(field.field_info, 'index', None) is True:
schema_parts.append(cls.schema_for_type(name, _type, field.field_info))
if getattr(field.field_info, 'sortable', False) is True:
schema_parts.append("SORTABLE")
return " ".join(schema_parts)
class JsonModel(RedisModel, abc.ABC):
2021-09-01 21:56:06 +02:00
def save(self, *args, **kwargs) -> 'JsonModel':
success = self.db().execute_command('JSON.SET', self.key(), ".", self.json())
return success
2021-09-01 21:56:06 +02:00
@classmethod
def get(cls, pk: Any) -> 'JsonModel':
document = cls.db().execute_command("JSON.GET", cls.make_primary_key(pk))
if not document:
raise NotFoundError
return cls.parse_raw(document)