from __future__ import annotations import contextlib import contextvars import copy import dataclasses import itertools from collections import Counter from collections.abc import Callable from typing import ( TYPE_CHECKING, Any, Optional, TypeVar, cast, ) from django.db import models from django.db.models import Prefetch from django.db.models.constants import LOOKUP_SEP from django.db.models.expressions import BaseExpression, Combinable from django.db.models.fields.reverse_related import ( ManyToManyRel, ManyToOneRel, OneToOneRel, ) from django.db.models.manager import BaseManager from django.db.models.query import QuerySet from graphql import ( FieldNode, GraphQLInterfaceType, GraphQLObjectType, GraphQLOutputType, GraphQLWrappingType, get_argument_values, ) from graphql.execution.collect_fields import collect_sub_fields from graphql.language.ast import OperationType from graphql.type.definition import GraphQLResolveInfo, get_named_type from strawberry import relay from strawberry.extensions import SchemaExtension from strawberry.relay.utils import SliceMetadata from strawberry.schema.schema import Schema from strawberry.schema.schema_converter import get_arguments from strawberry.types import get_object_definition, has_object_definition from strawberry.types.base import StrawberryContainer from strawberry.types.info import Info from strawberry.types.object_type import StrawberryObjectDefinition from typing_extensions import assert_never, assert_type from strawberry_django.fields.types import resolve_model_field_name from strawberry_django.pagination import OffsetPaginated, apply_window_pagination from strawberry_django.queryset import get_queryset_config, run_type_get_queryset from strawberry_django.relay.list_connection import DjangoListConnection from strawberry_django.resolvers import django_fetch from .descriptors import ModelProperty from .utils.inspect import ( PrefetchInspector, get_model_field, get_model_fields, get_possible_concrete_types, get_possible_type_definitions, is_inheritance_manager, is_inheritance_qs, is_polymorphic_model, ) from .utils.typing import ( AnnotateCallable, AnnotateType, PrefetchCallable, PrefetchType, TypeOrMapping, TypeOrSequence, WithStrawberryDjangoObjectDefinition, get_django_definition, has_django_definition, unwrap_type, ) if TYPE_CHECKING: from collections.abc import Generator from django.contrib.contenttypes.fields import GenericRelation from strawberry.types.execution import ExecutionContext from strawberry.types.field import StrawberryField from strawberry.utils.await_maybe import AwaitableOrValue __all__ = [ "DjangoOptimizerExtension", "OptimizerConfig", "OptimizerStore", "PrefetchType", "optimize", ] _M = TypeVar("_M", bound=models.Model) _sentinel = object() _annotate_placeholder = "__annotated_placeholder__" @dataclasses.dataclass class OptimizerConfig: """Django optimization configuration. Attributes ---------- enable_only: Enable `QuerySet.only` optimizations enable_select_related: Enable `QuerySet.select_related` optimizations enable_prefetch_related: Enable `QuerySet.prefetch_related` optimizations enable_annotate: Enable `QuerySet.annotate` optimizations enable_nested_relations_prefetch: Enable prefetch of nested relations optimizations. prefetch_custom_queryset: Use custom instead of _base_manager for prefetch querysets """ enable_only: bool = dataclasses.field(default=True) enable_select_related: bool = dataclasses.field(default=True) enable_prefetch_related: bool = dataclasses.field(default=True) enable_annotate: bool = dataclasses.field(default=True) enable_nested_relations_prefetch: bool = dataclasses.field(default=True) prefetch_custom_queryset: bool = dataclasses.field(default=False) @dataclasses.dataclass class OptimizerStore: """Django optimization store. Attributes ---------- only: Set of values to optimize using `QuerySet.only` selected: Set of values to optimize using `QuerySet.select_related` prefetch_related: Set of values to optimize using `QuerySet.prefetch_related` annotate: Dict of values to use in `QuerySet.annotate` """ only: list[str] = dataclasses.field(default_factory=list) select_related: list[str] = dataclasses.field(default_factory=list) prefetch_related: list[PrefetchType] = dataclasses.field(default_factory=list) annotate: dict[str, AnnotateType] = dataclasses.field(default_factory=dict) def __bool__(self): return any( [self.only, self.select_related, self.prefetch_related, self.annotate], ) def __ior__(self, other: OptimizerStore): self.only.extend(other.only) self.select_related.extend(other.select_related) self.prefetch_related.extend(other.prefetch_related) self.annotate.update(other.annotate) return self def __or__(self, other: OptimizerStore): new = self.copy() new |= other return new def copy(self): """Create a shallow copy of the store.""" return self.__class__( only=self.only[:], select_related=self.select_related[:], prefetch_related=self.prefetch_related[:], annotate=self.annotate.copy(), ) @classmethod def with_hints( cls, *, only: TypeOrSequence[str] | None = None, select_related: TypeOrSequence[str] | None = None, prefetch_related: TypeOrSequence[PrefetchType] | None = None, annotate: TypeOrMapping[AnnotateType] | None = None, ): """Create a new store with the given hints.""" return cls( only=[only] if isinstance(only, str) else list(only or []), select_related=( [select_related] if isinstance(select_related, str) else list(select_related or []) ), prefetch_related=( [prefetch_related] if isinstance(prefetch_related, (str, Prefetch, Callable)) else list(prefetch_related or []) ), annotate=( # placeholder here, # because field name is evaluated later on .annotate call: {_annotate_placeholder: annotate} if isinstance(annotate, (BaseExpression, Combinable, Callable)) else dict(annotate or {}) ), ) def with_resolved_callables(self, info: GraphQLResolveInfo): """Resolve any prefetch/annotate callables using the provided info and return a new store. This is used to resolve callables using the correct info object, scoped to their respective fields. """ if not any(callable(p) for p in self.prefetch_related) and not any( callable(a) for a in self.annotate.values() ): return self prefetch_related: list[PrefetchType] = [ p(info) if callable(p) else p for p in self.prefetch_related ] annotate: dict[str, AnnotateType] = { label: annotation(info) if callable(annotation) else annotation for label, annotation in self.annotate.items() } return self.__class__( only=self.only, select_related=self.select_related, prefetch_related=prefetch_related, annotate=annotate, ) def with_prefix(self, prefix: str, *, info: GraphQLResolveInfo): """Create a copy of this store with the given prefix. This is useful when we need to apply the same store to a nested field. `prefix` will be prepended to all fields in the store. Any callables will be resolved, just like with_resolved_callables, to apply the prefix to their results. """ prefetch_related = [] for p in self.prefetch_related: if isinstance(p, Callable): assert_type(p, PrefetchCallable) p = p(info) # noqa: PLW2901 if isinstance(p, str): prefetch_related.append(f"{prefix}{LOOKUP_SEP}{p}") elif isinstance(p, Prefetch): # add_prefix modifies the field's prefetch object, so we copy it before p_copy = copy.copy(p) p_copy.add_prefix(prefix) prefetch_related.append(p_copy) else: # pragma:nocover assert_never(p) annotate = {} for k, v in self.annotate.items(): if isinstance(v, Callable): assert_type(v, AnnotateCallable) v = v(info) # noqa: PLW2901 annotate[f"{prefix}{LOOKUP_SEP}{k}"] = v return self.__class__( only=[f"{prefix}{LOOKUP_SEP}{i}" for i in self.only], select_related=[f"{prefix}{LOOKUP_SEP}{i}" for i in self.select_related], prefetch_related=prefetch_related, annotate=annotate, ) def apply( self, qs: QuerySet[_M], *, info: GraphQLResolveInfo, config: OptimizerConfig | None = None, ) -> QuerySet[_M]: """Apply this store optimizations to the given queryset.""" config = config or OptimizerConfig() qs = self._apply_prefetch_related( qs, info=info, config=config, ) qs, extra_only_set = self._apply_select_related( qs, info=info, config=config, ) qs = self._apply_only( qs, info=info, config=config, extra_only_set=extra_only_set, ) qs = self._apply_annotate( qs, info=info, config=config, ) return qs # noqa: RET504 def _apply_prefetch_related( self, qs: QuerySet[_M], *, info: GraphQLResolveInfo, config: OptimizerConfig, ) -> QuerySet[_M]: if not config.enable_prefetch_related or not self.prefetch_related: return qs abort_only = set() prefetch_lists = [ qs._prefetch_related_lookups, # type: ignore self.prefetch_related, ] # Add all str at the same time to make it easier to handle Prefetch below to_prefetch: dict[str, str | Prefetch] = { p: p for p in itertools.chain(*prefetch_lists) if isinstance(p, str) } # Merge already existing prefetches together for p in itertools.chain(*prefetch_lists): # Already added above if isinstance(p, str): continue if isinstance(p, Callable): assert_type(p, PrefetchCallable) p = p(info) # noqa: PLW2901 path = p.prefetch_to existing = to_prefetch.get(path) # The simplest case. The prefetch doesn't exist or is a string. # In this case, just replace it. if not existing or isinstance(existing, str): to_prefetch[path] = p if isinstance(existing, str): abort_only.add(path) continue p1 = PrefetchInspector(existing) p2 = PrefetchInspector(p) if getattr(existing, "_optimizer_sentinel", None) is _sentinel: ret = p1.merge(p2, allow_unsafe_ops=True) elif getattr(p, "_optimizer_sentinel", None) is _sentinel: ret = p2.merge(p1, allow_unsafe_ops=True) else: # The order here doesn't matter ret = p1.merge(p2) to_prefetch[path] = ret.prefetch # Abort only optimization if one prefetch related was made for everything for ao in abort_only: # cast is safe as the loop above only adds Prefetch instances as abort_only prefetch = cast("Prefetch", to_prefetch[ao]) if prefetch.queryset is not None: # type: ignore - queryset can be None prefetch.queryset.query.deferred_loading = ( set(), True, ) # First prefetch_related(None) to clear all existing prefetches, and then # add ours, which also contains them. This is to avoid the # "lookup was already seen with a different queryset" error return qs.prefetch_related(None).prefetch_related(*to_prefetch.values()) def _apply_select_related( self, qs: QuerySet[_M], *, info: GraphQLResolveInfo, config: OptimizerConfig, ) -> tuple[QuerySet[_M], set[str]]: only_set = set(self.only) extra_only_set = set() select_related_set = set(self.select_related) # inspect the queryset to find any existing select_related fields def get_related_fields_with_prefix( queryset_select_related: dict[str, Any], prefix: str = "", ): for parent, nested in queryset_select_related.items(): current_path = f"{prefix}{parent}" yield current_path if nested: # If there are nested relations, dive deeper yield from get_related_fields_with_prefix( nested, prefix=f"{current_path}{LOOKUP_SEP}", ) if isinstance(qs.query.select_related, dict): select_related_set.update( get_related_fields_with_prefix(qs.query.select_related) ) if config.enable_select_related and select_related_set: qs = qs.select_related(*select_related_set) # Update our extra_select_related_only_set with the fields that were # selected by select_related to make sure they actually get selected for select_related in select_related_set: if select_related in only_set: continue if not any(only.startswith(select_related) for only in only_set): extra_only_set.add(select_related) return qs, extra_only_set def _apply_only( self, qs: QuerySet[_M], *, info: GraphQLResolveInfo, config: OptimizerConfig, extra_only_set: set[str], ) -> QuerySet[_M]: only_set = set(self.only) | extra_only_set if config.enable_only and only_set: qs = qs.only(*only_set) return qs def _apply_annotate( self, qs: QuerySet[_M], *, info: GraphQLResolveInfo, config: OptimizerConfig, ) -> QuerySet[_M]: if not config.enable_annotate or not self.annotate: return qs to_annotate = {} for k, v in self.annotate.items(): if isinstance(v, Callable): assert_type(v, AnnotateCallable) v = v(info) # noqa: PLW2901 to_annotate[k] = v return qs.annotate(**to_annotate) def _get_django_type( field: StrawberryField, ) -> type[WithStrawberryDjangoObjectDefinition] | None: f_type = unwrap_type(field.type) return f_type if has_django_definition(f_type) else None def _get_prefetch_queryset( remote_model: type[models.Model], schema: Schema, field: StrawberryField, parent_type: GraphQLObjectType | GraphQLInterfaceType, field_node: FieldNode, *, config: OptimizerConfig | None, info: GraphQLResolveInfo, related_field_id: str | None = None, ) -> QuerySet: # We usually want to use the `_base_manager` for prefetching, as it is what django # itself states we should be using: # https://docs.djangoproject.com/en/5.0/topics/db/managers/#base-managers # But in case prefetch_custom_queryset is enabled, we use the custom queryset # from _default_manager instead. if config and config.prefetch_custom_queryset: qs = remote_model._default_manager.all() else: qs = remote_model._base_manager.all() # type: ignore if f_type := _get_django_type(field): qs = run_type_get_queryset( qs, f_type, info=Info( _raw_info=info, _field=field, ), ) return _optimize_prefetch_queryset( qs, schema, field, parent_type, field_node, config=config, info=info, related_field_id=related_field_id, ) def _optimize_prefetch_queryset( qs: QuerySet[_M], schema: Schema, field: StrawberryField, parent_type: GraphQLObjectType | GraphQLInterfaceType, field_node: FieldNode, *, config: OptimizerConfig | None, info: GraphQLResolveInfo, related_field_id: str | None = None, ) -> QuerySet[_M]: from strawberry_django.fields.field import ( StrawberryDjangoConnectionExtension, StrawberryDjangoField, ) from strawberry_django.relay.cursor_connection import ( DjangoCursorConnection, apply_cursor_pagination, ) if ( not config or not config.enable_nested_relations_prefetch or related_field_id is None or not isinstance(field, StrawberryDjangoField) or is_optimized_by_prefetching(qs) ): return qs mark_optimized = True strawberry_schema = cast("Schema", info.schema._strawberry_schema) # type: ignore field_name = strawberry_schema.config.name_converter.from_field(field) field_info = Info( _raw_info=info, _field=field, ) _field_args, field_kwargs = get_arguments( field=field, source=None, info=field_info, kwargs=get_argument_values( parent_type.fields[field_name], field_node, info.variable_values, ), config=strawberry_schema.config, scalar_registry=strawberry_schema.schema_converter.scalar_registry, ) field_kwargs.pop("info", None) # Disable the optimizer to avoid doing double optimization while running get_queryset with DjangoOptimizerExtension.disabled(): qs = field.get_queryset( qs, field_info, _strawberry_related_field_id=related_field_id, **field_kwargs, ) connection_extension = next( ( e for e in field.extensions if isinstance(e, StrawberryDjangoConnectionExtension) ), None, ) if connection_extension is not None: connection_type_def = get_object_definition( connection_extension.connection_type, strict=True, ) connection_type = ( connection_type_def.concrete_of and connection_type_def.concrete_of.origin ) if ( connection_type is relay.ListConnection or connection_type is DjangoListConnection ): slice_metadata = SliceMetadata.from_arguments( Info(_raw_info=info, _field=field), first=field_kwargs.get("first"), last=field_kwargs.get("last"), before=field_kwargs.get("before"), after=field_kwargs.get("after"), ) qs = apply_window_pagination( qs, related_field_id=related_field_id, offset=slice_metadata.start, limit=slice_metadata.end - slice_metadata.start, max_results=connection_extension.max_results, ) elif connection_type is DjangoCursorConnection: qs, _ = apply_cursor_pagination( qs, related_field_id=related_field_id, info=Info(_raw_info=info, _field=field), first=field_kwargs.get("first"), last=field_kwargs.get("last"), before=field_kwargs.get("before"), after=field_kwargs.get("after"), max_results=connection_extension.max_results, ) else: mark_optimized = False if isinstance(field.type, type) and issubclass(field.type, OffsetPaginated): pagination = field_kwargs.get("pagination") qs = apply_window_pagination( qs, related_field_id=related_field_id, offset=pagination.offset if pagination else 0, limit=pagination.limit if pagination else -1, ) if mark_optimized: qs = mark_optimized_by_prefetching(qs) return qs def _get_selections( info: GraphQLResolveInfo, parent_type: GraphQLObjectType | GraphQLInterfaceType, ) -> dict[str, list[FieldNode]]: return collect_sub_fields( info.schema, info.fragments, info.variable_values, cast("GraphQLObjectType", parent_type), info.field_nodes, ) def _generate_selection_resolve_info( info: GraphQLResolveInfo, field_nodes: list[FieldNode], return_type: GraphQLOutputType, parent_type: GraphQLObjectType | GraphQLInterfaceType, ): field_node = field_nodes[0] return GraphQLResolveInfo( field_name=field_node.name.value, field_nodes=field_nodes, return_type=return_type, parent_type=cast("GraphQLObjectType", parent_type), path=info.path.add_key(0).add_key(field_node.name.value, parent_type.name), schema=info.schema, fragments=info.fragments, root_value=info.root_value, operation=info.operation, variable_values=info.variable_values, context=info.context, is_awaitable=info.is_awaitable, ) def _get_field_data( selections: list[FieldNode], object_definition: StrawberryObjectDefinition, schema: Schema, *, parent_type: GraphQLObjectType | GraphQLInterfaceType, info: GraphQLResolveInfo, ) -> tuple[StrawberryField, GraphQLObjectType, FieldNode, GraphQLResolveInfo] | None: selection = selections[0] field_name = selection.name.value for field in object_definition.fields: if schema.config.name_converter.get_graphql_name(field) == field_name: break else: return None # Do not optimize the field if the user asked not to if getattr(field, "disable_optimization", False): return None definition = parent_type.fields[selection.name.value].type while isinstance(definition, GraphQLWrappingType): definition = definition.of_type field_info = _generate_selection_resolve_info( info, selections, definition, parent_type, ) return field, definition, selection, field_info def _get_hints_from_field( field: StrawberryField, *, f_info: GraphQLResolveInfo, prefix: str = "", ) -> OptimizerStore | None: if not ( field_store := cast("Optional[OptimizerStore]", getattr(field, "store", None)) ): return None if len(field_store.annotate) == 1 and _annotate_placeholder in field_store.annotate: # This is a special case where we need to update the field name, # because when field_store was created on __init__, # the field name wasn't available. # This allows for annotate expressions to be declared as: # total: int = gql.django.field(annotate=Sum("price")) # noqa: ERA001 # Instead of the more redundant: # total: int = gql.django.field(annotate={"total": Sum("price")}) # noqa: ERA001 field_store.annotate = { field.name: field_store.annotate[_annotate_placeholder], } # with_prefix also resolves callables, so we only need one or the other return ( field_store.with_prefix(prefix, info=f_info) if prefix else field_store.with_resolved_callables(f_info) ) def _get_hints_from_model_property( field: StrawberryField, model: type[models.Model], *, f_info: GraphQLResolveInfo, prefix: str = "", ) -> OptimizerStore | None: model_attr = getattr(model, field.python_name, None) if ( model_attr is not None and isinstance(model_attr, ModelProperty) and model_attr.store ): attr_store = model_attr.store # with_prefix also resolves callables, so we only need one or the other store = ( attr_store.with_prefix(prefix, info=f_info) if prefix else attr_store.with_resolved_callables(f_info) ) else: store = None return store def _must_use_prefetch_related( config: OptimizerConfig, field: StrawberryField, model_field: models.ForeignKey | OneToOneRel, ) -> bool: f_type = _get_django_type(field) # - If the field has a get_queryset method, use Prefetch so it will be respected # - If the model is using django-polymorphic, # use Prefetch so its custom queryset will be used, returning polymorphic models return ( (f_type and hasattr(f_type, "get_queryset")) or is_polymorphic_model(model_field.related_model) or is_inheritance_manager( model_field.related_model._default_manager if config.prefetch_custom_queryset else model_field.related_model._base_manager # type: ignore ) ) def _get_hints_from_django_foreign_key( field: StrawberryField, field_definition: GraphQLObjectType, field_selection: FieldNode, model_field: models.ForeignKey | OneToOneRel, model_fieldname: str, schema: Schema, *, config: OptimizerConfig, parent_type: GraphQLObjectType | GraphQLInterfaceType, field_info: GraphQLResolveInfo, path: str, cache: dict[type[models.Model], list[tuple[int, OptimizerStore]]], level: int = 0, ) -> OptimizerStore: if _must_use_prefetch_related(config, field, model_field): store = _get_hints_from_django_relation( field, field_selection=field_selection, model_field=model_field, model_fieldname=model_fieldname, schema=schema, config=config, parent_type=parent_type, field_info=field_info, path=path, cache=cache, level=level, ) store.only.append(path) return store store = OptimizerStore.with_hints( only=[path], select_related=[path], ) # If adding a reverse relation, make sure to select its pointer to us, # or else this might causa a refetch from the database if isinstance(model_field, OneToOneRel): remote_field = model_field.remote_field store.only.append( f"{path}{LOOKUP_SEP}{resolve_model_field_name(remote_field)}", ) for f_type_def in get_possible_type_definitions(field.type): f_model = model_field.related_model f_store = _get_model_hints( f_model, schema, f_type_def, parent_type=field_definition, info=field_info, config=config, cache=cache, level=level + 1, ) if f_store is not None: cache.setdefault(f_model, []).append((level, f_store)) store |= f_store.with_prefix(path, info=field_info) return store def _get_hints_from_django_relation( field: StrawberryField, field_selection: FieldNode, model_field: ( models.ManyToManyField | ManyToManyRel | ManyToOneRel | GenericRelation | OneToOneRel | models.ForeignKey ), model_fieldname: str, schema: Schema, *, config: OptimizerConfig, parent_type: GraphQLObjectType | GraphQLInterfaceType, field_info: GraphQLResolveInfo, path: str, cache: dict[type[models.Model], list[tuple[int, OptimizerStore]]], level: int = 0, ) -> OptimizerStore: try: from django.contrib.contenttypes.fields import GenericRelation except (ImportError, RuntimeError): # pragma: no cover GenericRelation = None # noqa: N806 store = OptimizerStore() f_types = list(get_possible_type_definitions(field.type)) if len(f_types) > 1: # This might be a generic foreign key. # In this case, just prefetch it store.prefetch_related.append(model_fieldname) return store field_store = getattr(field, "store", None) if field_store and field_store.prefetch_related: # Skip optimization if 'prefetch_related' is present in the field's store. # This is necessary because 'prefetch_related' likely modifies the queryset # with filtering or annotating, making the optimization redundant and # potentially causing an extra unused query. return store remote_field = model_field.remote_field remote_model = remote_field.model field_store = None f_type = f_types[0] subclasses = [] for concrete_field_type in get_possible_concrete_types( remote_model, schema, f_type ): django_definition = get_django_definition(concrete_field_type.origin) if ( django_definition and django_definition.model != remote_model and not django_definition.model._meta.abstract and issubclass(django_definition.model, remote_model) ): subclasses.append(django_definition.model) concrete_store = _get_model_hints( remote_model, schema, concrete_field_type, parent_type=_get_gql_definition(schema, concrete_field_type), info=field_info, config=config, cache=cache, level=level + 1, ) if concrete_store is not None: field_store = ( concrete_store if field_store is None else field_store | concrete_store ) if field_store is None: return store related_field_id = getattr(remote_field, "attname", None) or getattr( remote_field, "name", None ) if ( config.enable_only and field_store.only and not isinstance(remote_field, ManyToManyRel) ): # If adding a reverse relation, make sure to select its # pointer to us, or else this might causa a refetch from # the database if GenericRelation is not None and isinstance( model_field, GenericRelation, ): field_store.only.append(model_field.object_id_field_name) field_store.only.append(model_field.content_type_field_name) elif related_field_id is not None: field_store.only.append(related_field_id) path_lookup = f"{path}{LOOKUP_SEP}" if store.only and field_store.only: extra_only = [o for o in store.only or [] if o.startswith(path_lookup)] store.only = [o for o in store.only if o not in extra_only] field_store.only.extend(o[len(path_lookup) :] for o in extra_only) if store.select_related and field_store.select_related: extra_sr = [o for o in store.select_related or [] if o.startswith(path_lookup)] store.select_related = [o for o in store.select_related if o not in extra_sr] field_store.select_related.extend(o[len(path_lookup) :] for o in extra_sr) cache.setdefault(remote_model, []).append((level, field_store)) base_qs = _get_prefetch_queryset( remote_model, schema, field, parent_type, field_selection, config=config, info=field_info, related_field_id=related_field_id, ) if is_inheritance_qs(base_qs): base_qs = base_qs.select_subclasses(*subclasses) field_qs = field_store.apply(base_qs, info=field_info, config=config) field_prefetch = Prefetch(path, queryset=field_qs) field_prefetch._optimizer_sentinel = _sentinel # type: ignore store.prefetch_related.append(field_prefetch) return store def _get_hints_from_django_field( field: StrawberryField, field_definition: GraphQLObjectType, field_selection: FieldNode, model: type[models.Model], schema: Schema, *, config: OptimizerConfig, parent_type: GraphQLObjectType | GraphQLInterfaceType, field_info: GraphQLResolveInfo, prefix: str = "", cache: dict[type[models.Model], list[tuple[int, OptimizerStore]]], level: int = 0, ) -> OptimizerStore | None: try: from django.contrib.contenttypes.fields import ( GenericForeignKey, GenericRelation, ) except (ImportError, RuntimeError): # pragma: no cover GenericForeignKey = None # noqa: N806 GenericRelation = None # noqa: N806 relation_fields = (models.ManyToManyField, ManyToManyRel, ManyToOneRel) else: relation_fields = ( models.ManyToManyField, ManyToManyRel, ManyToOneRel, GenericRelation, ) # If the field has a base resolver, don't try to optimize it. The user should # be defining custom hints in this case, which should already be in the store # GlobalID and special cases setting `can_optimize` are ok though, as those resolvers # are auto generated by us if ( field.base_resolver is not None and field.type != relay.GlobalID and not getattr(field.base_resolver.wrapped_func, "can_optimize", False) ): return None model_fieldname: str = getattr(field, "django_name", None) or field.python_name if (model_field := get_model_field(model, model_fieldname)) is None: return None lookup_prefix = prefix + LOOKUP_SEP if prefix else "" path = f"{lookup_prefix}{model_fieldname}" if isinstance(model_field, (models.ForeignKey, OneToOneRel)): store = _get_hints_from_django_foreign_key( field, field_definition=field_definition, field_selection=field_selection, model_field=model_field, model_fieldname=model_fieldname, schema=schema, config=config, parent_type=parent_type, field_info=field_info, path=path, cache=cache, level=level, ) elif GenericForeignKey and isinstance(model_field, GenericForeignKey): # There's not much we can do to optimize generic foreign keys regarding # only/select_related because they can be anything. # Just prefetch_related them store = OptimizerStore.with_hints(prefetch_related=[model_fieldname]) elif isinstance(model_field, relation_fields): store = _get_hints_from_django_relation( field, field_selection=field_selection, model_field=model_field, model_fieldname=model_fieldname, schema=schema, config=config, parent_type=parent_type, field_info=field_info, path=path, cache=cache, level=level, ) else: store = OptimizerStore.with_hints(only=[path]) return store def _get_model_hints( model: type[models.Model], schema: Schema, object_definition: StrawberryObjectDefinition, *, parent_type: GraphQLObjectType | GraphQLInterfaceType, info: GraphQLResolveInfo, config: OptimizerConfig | None = None, prefix: str = "", cache: dict[type[models.Model], list[tuple[int, OptimizerStore]]] | None = None, level: int = 0, subclass_collection: set[type[models.Model]] | None = None, ) -> OptimizerStore | None: cache = cache or {} # In case this is a relay field, the selected fields are inside edges -> node selection if issubclass(object_definition.origin, relay.Connection): return _get_model_hints_from_connection( model, schema, object_definition, parent_type=parent_type, info=info, config=config, prefix=prefix, cache=cache, level=level, subclass_collection=subclass_collection, ) # In case this is a Paginated field, the selected fields are inside results selection if issubclass(object_definition.origin, OffsetPaginated): return _get_model_hints_from_paginated( model, schema, object_definition, parent_type=parent_type, info=info, config=config, prefix=prefix, cache=cache, level=level, subclass_collection=subclass_collection, ) store = OptimizerStore() config = config or OptimizerConfig() dj_definition = get_django_definition(object_definition.origin) if dj_definition is None or dj_definition.disable_optimization: return None if not issubclass(model, dj_definition.model): # If this is a PolymorphicModel, also try to optimize fields in subclasses # of the current model. if not dj_definition.model._meta.abstract and issubclass( dj_definition.model, model ): if subclass_collection is not None: subclass_collection.add(dj_definition.model) if is_polymorphic_model(model): # These must be prefixed with app_label__ModelName___ (note three underscores) # This is a special syntax for django-polymorphic: # https://django-polymorphic.readthedocs.io/en/stable/advanced.html#polymorphic-filtering-for-fields-in-inherited-classes # "prefix" however is written in terms of not including the final LOOKUP_SEP (i.e. "__") # So we don't include the final __ here. return _get_model_hints( dj_definition.model, schema, object_definition, parent_type=parent_type, info=info, config=config, prefix=f"{prefix}{dj_definition.model._meta.app_label}__{dj_definition.model._meta.model_name}_", ) if is_inheritance_manager(model._default_manager) and ( path_from_parent := dj_definition.model._meta.get_path_from_parent( model ) ): prefix = LOOKUP_SEP.join( p.join_field.get_accessor_name() for p in path_from_parent ) return _get_model_hints( dj_definition.model, schema, object_definition, parent_type=parent_type, info=info, config=config, prefix=prefix, ) return None dj_type_store = getattr(dj_definition, "store", None) if dj_type_store: store |= dj_type_store lookup_prefix = prefix + LOOKUP_SEP if prefix else "" # Make sure that the model's pk is always selected when using only pk = model._meta.pk if pk is not None: store.only.append(lookup_prefix + pk.attname) # If this is a polymorphic Model, make sure to select its content type if is_polymorphic_model(model): store.only.extend( lookup_prefix + f for f in model.polymorphic_internal_model_fields ) selections = [ field_data for f_selection in _get_selections(info, parent_type).values() if ( field_data := _get_field_data( f_selection, object_definition, schema, parent_type=parent_type, info=info, ) ) is not None ] fields_counter = Counter(field_data[0] for field_data in selections) for field, f_definition, f_selection, f_info in selections: # If a field is selected more than once in the query, that means it is being # aliased. In this case, optimizing it would make one query to affect the other, # resulting in wrong results for both. if fields_counter[field] > 1: continue # Add annotations from the field if they exist if field_store := _get_hints_from_field(field, f_info=f_info, prefix=prefix): store |= field_store # Then from the model property if one is defined if model_property_store := _get_hints_from_model_property( field, model, f_info=f_info, prefix=prefix, ): store |= model_property_store # Lastly, from the django field itself if model_field_store := _get_hints_from_django_field( field, f_definition, f_selection, model, schema, config=config, parent_type=parent_type, field_info=f_info, prefix=prefix, cache=cache, level=level, ): store |= model_field_store # Django keeps track of known fields. That means that if one model select_related or # prefetch_related another one, and later another one select_related or # prefetch_related the model again, if the used fields there where not optimized in # this call django would have to fetch those again. By mergint those with us we are # making sure to avoid that for inner_level, inner_store in cache.get(model, []): if inner_level > level and inner_store: # We only want the only/select_related from this. prefetch_related is # something else store.only.extend(inner_store.only) store.select_related.extend(inner_store.select_related) # In case we skipped optimization for a relation, we might end up with a new QuerySet # which would not select its parent relation field on `.only()`, causing n+1 issues. # Make sure that in this case we also select it. if level == 0 and store.only and info.path.prev: own_fk_fields = [ field for field in get_model_fields(model).values() if isinstance(field, models.ForeignKey) ] path = info.path while path := path.prev: type_ = schema.get_type_by_name(path.typename) if not isinstance(type_, StrawberryObjectDefinition): continue if not (strawberry_django_type := get_django_definition(type_.origin)): continue for field in own_fk_fields: if field.related_model is strawberry_django_type.model: store.only.append(field.attname) break return store def _get_gql_definition( schema: Schema, definition: StrawberryObjectDefinition, ) -> GraphQLInterfaceType | GraphQLObjectType: if definition.is_interface: return schema.schema_converter.from_interface(definition) return schema.schema_converter.from_object(definition) def _get_model_hints_from_connection( model: type[models.Model], schema: Schema, object_definition: StrawberryObjectDefinition, *, parent_type: GraphQLObjectType | GraphQLInterfaceType, info: GraphQLResolveInfo, config: OptimizerConfig | None = None, prefix: str = "", cache: dict[type[models.Model], list[tuple[int, OptimizerStore]]] | None = None, level: int = 0, subclass_collection: set[type[models.Model]] | None = None, ) -> OptimizerStore | None: store = None n_type = object_definition.type_var_map.get("NodeType") if n_type is None: specialized_type_var_map = object_definition.specialized_type_var_map or {} n_type = specialized_type_var_map["NodeType"] n_type = unwrap_type(n_type) n_definition = get_object_definition(n_type, strict=True) for edges in _get_selections(info, parent_type).values(): edge = edges[0] if edge.name.value != "edges": continue e_field = object_definition.get_field("edges") if e_field is None: break e_definition = e_field.type while isinstance(e_definition, StrawberryContainer): e_definition = e_definition.of_type if has_object_definition(e_definition): e_definition = get_object_definition(e_definition, strict=True) assert isinstance(e_definition, StrawberryObjectDefinition) e_gql_definition = _get_gql_definition( schema, e_definition, ) assert isinstance(e_gql_definition, (GraphQLObjectType, GraphQLInterfaceType)) e_info = _generate_selection_resolve_info( info, edges, e_gql_definition, parent_type, ) for nodes in _get_selections(e_info, e_gql_definition).values(): node = nodes[0] if node.name.value != "node": continue for concrete_n_type in get_possible_concrete_types( model, schema, n_definition ): n_gql_definition = _get_gql_definition(schema, concrete_n_type) assert isinstance( n_gql_definition, (GraphQLObjectType, GraphQLInterfaceType), ) n_info = _generate_selection_resolve_info( info, nodes, n_gql_definition, e_gql_definition, ) concrete_store = _get_model_hints( model=model, schema=schema, object_definition=concrete_n_type, parent_type=n_gql_definition, info=n_info, config=config, prefix=prefix, cache=cache, level=level, subclass_collection=subclass_collection, ) if concrete_store is not None: store = concrete_store if store is None else store | concrete_store return store def _get_model_hints_from_paginated( model: type[models.Model], schema: Schema, object_definition: StrawberryObjectDefinition, *, parent_type: GraphQLObjectType | GraphQLInterfaceType, info: GraphQLResolveInfo, config: OptimizerConfig | None = None, prefix: str = "", cache: dict[type[models.Model], list[tuple[int, OptimizerStore]]] | None = None, level: int = 0, subclass_collection: set[type[models.Model]] | None = None, ) -> OptimizerStore | None: store = None n_type = unwrap_type(object_definition.type_var_map.get("NodeType")) n_definition = get_object_definition(n_type, strict=True) for selections in _get_selections(info, parent_type).values(): selection = selections[0] if selection.name.value != "results": continue for concrete_n_type in get_possible_concrete_types(model, schema, n_definition): n_gql_definition = _get_gql_definition( schema, concrete_n_type, ) assert isinstance( n_gql_definition, (GraphQLObjectType, GraphQLInterfaceType) ) n_info = _generate_selection_resolve_info( info, selections, n_gql_definition, n_gql_definition, ) concrete_store = _get_model_hints( model=model, schema=schema, object_definition=concrete_n_type, parent_type=n_gql_definition, info=n_info, config=config, prefix=prefix, cache=cache, level=level, subclass_collection=subclass_collection, ) if concrete_store is not None: store = concrete_store if store is None else store | concrete_store return store def optimize( qs: QuerySet[_M] | BaseManager[_M], info: GraphQLResolveInfo | Info, *, config: OptimizerConfig | None = None, store: OptimizerStore | None = None, ) -> QuerySet[_M]: """Optimize the given queryset considering the gql info. This will look through the gql selections, fields and model hints and apply `only`, `select_related`, `prefetch_related` and `annotate` optimizations according those on the `QuerySet`_. Note: ---- This do not execute the queryset, it only optimizes it for when it is actually executed. It will also avoid doing any extra optimization if the queryset already has cached results in it, to avoid triggering extra queries later. Args: ---- qs: The queryset to be optimized info: The current field execution info config: Optional config to use when doing the optimization store: Optional initial store to use for the optimization Returns: ------- The optimized queryset .. _QuerySet: https://docs.djangoproject.com/en/dev/ref/models/querysets/ """ if isinstance(qs, BaseManager): qs = cast("QuerySet[_M]", qs.all()) if isinstance(qs, list): # return sliced queryset as-is return qs # Avoid optimizing twice and also modify an already resolved queryset if is_optimized(qs) or qs._result_cache is not None: # type: ignore return qs if isinstance(info, Info): info = info._raw_info config = config or OptimizerConfig() store = store or OptimizerStore() schema = cast("Schema", info.schema._strawberry_schema) # type: ignore gql_type = get_named_type(info.return_type) strawberry_type = schema.get_type_by_name(gql_type.name) if strawberry_type is None: return qs inheritance_qs = is_inheritance_qs(qs) subclasses = set() if inheritance_qs else None for inner_object_definition in get_possible_concrete_types( qs.model, schema, strawberry_type ): parent_type = _get_gql_definition(schema, inner_object_definition) new_store = _get_model_hints( qs.model, schema, inner_object_definition, parent_type=parent_type, info=info, config=config, subclass_collection=subclasses, ) if new_store is not None: store |= new_store if store: if inheritance_qs and subclasses: qs = qs.select_subclasses(*subclasses) qs = store.apply(qs, info=info, config=config) qs_config = get_queryset_config(qs) qs_config.optimized = True return qs def is_optimized(qs: QuerySet) -> bool: config = get_queryset_config(qs) return config.optimized or config.optimized_by_prefetching def mark_optimized_by_prefetching(qs: QuerySet[_M]) -> QuerySet[_M]: get_queryset_config(qs).optimized_by_prefetching = True return qs def is_optimized_by_prefetching(qs: QuerySet) -> bool: return get_queryset_config(qs).optimized_by_prefetching optimizer: contextvars.ContextVar[DjangoOptimizerExtension | None] = ( contextvars.ContextVar( "optimizer_ctx", default=None, ) ) class DjangoOptimizerExtension(SchemaExtension): """Automatically optimize returned querysets from internal resolvers. Attributes ---------- enable_only_optimization: Enable `QuerySet.only` optimizations enable_select_related_optimization: Enable `QuerySet.select_related` optimizations enable_prefetch_related_optimization: Enable `QuerySet.prefetch_related` optimizations enable_nested_relations_prefetch: Enable prefetch of nested relations. This will allow for nested relations to be prefetched even when using filters/ordering/pagination. Note however that for connections, it will only work when for the `ListConnection` and `DjangoListConnection` types, as this optimization is not safe to be applied automatically for custom connections. enable_annotate_optimization: Enable `QuerySet.annotate` optimizations Examples -------- Add the following to your schema configuration. >>> import strawberry >>> from strawberry_django_plus.optimizer import DjangoOptimizerExtension ... >>> schema = strawberry.Schema( ... Query, ... extensions=[ ... DjangoOptimizerExtension(), ... ] ... ) """ enabled: contextvars.ContextVar[bool] = contextvars.ContextVar( "optimizer_enabled_ctx", default=True, ) def __init__( self, *, enable_only_optimization: bool = True, enable_select_related_optimization: bool = True, enable_prefetch_related_optimization: bool = True, enable_annotate_optimization: bool = True, enable_nested_relations_prefetch: bool = True, execution_context: ExecutionContext | None = None, prefetch_custom_queryset: bool = False, ): super().__init__(execution_context=execution_context) self.enable_only = enable_only_optimization self.enable_select_related = enable_select_related_optimization self.enable_prefetch_related = enable_prefetch_related_optimization self.enable_annotate_optimization = enable_annotate_optimization self.enable_nested_relations_prefetch = enable_nested_relations_prefetch self.prefetch_custom_queryset = prefetch_custom_queryset if enable_nested_relations_prefetch: from strawberry_django.utils.patches import apply_pagination_fix apply_pagination_fix() def on_execute(self) -> Generator[None]: token = optimizer.set(self) try: yield finally: optimizer.reset(token) def resolve( self, next_: Callable, root: Any, info: GraphQLResolveInfo, *args, **kwargs, ) -> AwaitableOrValue[Any]: ret = next_(root, info, *args, **kwargs) if not self.enabled.get(): return ret if isinstance(ret, BaseManager): ret = ret.all() if isinstance(ret, QuerySet) and ret._result_cache is None: # type: ignore config = OptimizerConfig( enable_only=( self.enable_only and info.operation.operation == OperationType.QUERY ), enable_select_related=self.enable_select_related, enable_prefetch_related=self.enable_prefetch_related, enable_annotate=self.enable_annotate_optimization, prefetch_custom_queryset=self.prefetch_custom_queryset, enable_nested_relations_prefetch=self.enable_nested_relations_prefetch, ) ret = django_fetch(optimize(qs=ret, info=info, config=config)) return ret @classmethod @contextlib.contextmanager def disabled(cls): token = cls.enabled.set(False) try: yield finally: cls.enabled.reset(token) def optimize( self, qs: QuerySet[_M] | BaseManager[_M], info: GraphQLResolveInfo | Info, *, store: OptimizerStore | None = None, ) -> QuerySet[_M]: if not self.enabled.get(): return qs config = OptimizerConfig( enable_only=self.enable_only and info.operation.operation == OperationType.QUERY, enable_select_related=self.enable_select_related, enable_prefetch_related=self.enable_prefetch_related, enable_annotate=self.enable_annotate_optimization, prefetch_custom_queryset=self.prefetch_custom_queryset, ) return optimize(qs, info, config=config, store=store)