mirror of https://github.com/jumpserver/jumpserver
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
225 lines
7.0 KiB
225 lines
7.0 KiB
# -*- coding: utf-8 -*-
|
|
#
|
|
from collections import defaultdict
|
|
from itertools import chain
|
|
from typing import Callable
|
|
|
|
from django.db import models
|
|
from django.db.models.signals import m2m_changed
|
|
from rest_framework.response import Response
|
|
from rest_framework.settings import api_settings
|
|
|
|
from common.drf.filters import (
|
|
IDSpmFilterBackend, CustomFilterBackend, IDInFilterBackend,
|
|
IDNotFilterBackend, NotOrRelFilterBackend, LabelFilterBackend
|
|
)
|
|
from common.utils import get_logger, lazyproperty
|
|
from .action import RenderToJsonMixin
|
|
from .serializer import SerializerMixin
|
|
|
|
__all__ = [
|
|
'CommonApiMixin', 'PaginatedResponseMixin', 'RelationMixin',
|
|
'ExtraFilterFieldsMixin',
|
|
]
|
|
|
|
logger = get_logger(__name__)
|
|
|
|
|
|
class PaginatedResponseMixin:
|
|
|
|
def get_paginated_response_from_queryset(self, queryset):
|
|
page = self.paginate_queryset(queryset)
|
|
if page is not None:
|
|
serializer = self.get_serializer(page, many=True)
|
|
return self.get_paginated_response(serializer.data)
|
|
|
|
serializer = self.get_serializer(queryset, many=True)
|
|
return Response(serializer.data)
|
|
|
|
|
|
class RelationMixin:
|
|
m2m_field = None
|
|
from_field = None
|
|
to_field = None
|
|
to_model = None
|
|
|
|
def __init__(self, *args, **kwargs):
|
|
super().__init__(*args, **kwargs)
|
|
|
|
assert self.m2m_field is not None, '''
|
|
`m2m_field` should not be `None`
|
|
'''
|
|
|
|
self.from_field = self.m2m_field.m2m_field_name()
|
|
self.to_field = self.m2m_field.m2m_reverse_field_name()
|
|
self.to_model = self.m2m_field.related_model
|
|
self.through = getattr(self.m2m_field.model, self.m2m_field.attname).through
|
|
|
|
def get_queryset(self):
|
|
# 注意,此处拦截了 `get_queryset` 没有 `super`
|
|
queryset = self.through.objects.all()
|
|
return queryset
|
|
|
|
def send_m2m_changed_signal(self, instances, action):
|
|
if not isinstance(instances, list):
|
|
instances = [instances]
|
|
|
|
from_to_mapper = defaultdict(list)
|
|
|
|
for i in instances:
|
|
to_id = getattr(i, self.to_field).id
|
|
# TODO 优化,不应该每次都查询数据库
|
|
from_obj = getattr(i, self.from_field)
|
|
from_to_mapper[from_obj].append(to_id)
|
|
|
|
for from_obj, to_ids in from_to_mapper.items():
|
|
m2m_changed.send(
|
|
sender=self.through, instance=from_obj, action=action,
|
|
reverse=False, model=self.to_model, pk_set=to_ids
|
|
)
|
|
|
|
def perform_create(self, serializer):
|
|
instance = serializer.save()
|
|
self.send_m2m_changed_signal(instance, 'post_add')
|
|
|
|
def perform_destroy(self, instance):
|
|
instance.delete()
|
|
self.send_m2m_changed_signal(instance, 'post_remove')
|
|
|
|
|
|
class QuerySetMixin:
|
|
action: str
|
|
get_serializer_class: Callable
|
|
get_queryset: Callable
|
|
|
|
def get_queryset(self):
|
|
queryset = super().get_queryset()
|
|
if not hasattr(self, 'action'):
|
|
return queryset
|
|
if self.action == 'metadata':
|
|
queryset = queryset.none()
|
|
return queryset
|
|
|
|
def paginate_queryset(self, queryset):
|
|
page = super().paginate_queryset(queryset)
|
|
serializer_class = self.get_serializer_class()
|
|
if page and serializer_class and hasattr(serializer_class, 'setup_eager_loading'):
|
|
ids = [str(obj.id) for obj in page]
|
|
page = self.get_queryset().filter(id__in=ids)
|
|
page = serializer_class.setup_eager_loading(page)
|
|
page_mapper = {str(obj.id): obj for obj in page}
|
|
page = [page_mapper.get(_id) for _id in ids if _id in page_mapper]
|
|
return page
|
|
|
|
|
|
class ExtraFilterFieldsMixin:
|
|
"""
|
|
额外的 api filter
|
|
"""
|
|
default_added_filters = (
|
|
CustomFilterBackend, IDSpmFilterBackend, IDInFilterBackend,
|
|
IDNotFilterBackend, LabelFilterBackend
|
|
)
|
|
filter_backends = api_settings.DEFAULT_FILTER_BACKENDS
|
|
extra_filter_fields = []
|
|
extra_filter_backends = []
|
|
|
|
def set_compatible_fields(self):
|
|
"""
|
|
兼容老的 filter_fields
|
|
"""
|
|
if not hasattr(self, 'filter_fields') and hasattr(self, 'filterset_fields'):
|
|
self.filter_fields = self.filterset_fields
|
|
|
|
def get_filter_backends(self):
|
|
self.set_compatible_fields()
|
|
if self.filter_backends != self.__class__.filter_backends:
|
|
return self.filter_backends
|
|
backends = list(chain(
|
|
self.filter_backends,
|
|
self.default_added_filters,
|
|
self.extra_filter_backends,
|
|
))
|
|
# 这个要放在最后
|
|
backends.append(NotOrRelFilterBackend)
|
|
return backends
|
|
|
|
def filter_queryset(self, queryset):
|
|
for backend in self.get_filter_backends():
|
|
queryset = backend().filter_queryset(self.request, queryset, self)
|
|
return queryset
|
|
|
|
|
|
class OrderingFielderFieldsMixin:
|
|
"""
|
|
额外的 api ordering
|
|
"""
|
|
ordering_fields = None
|
|
extra_ordering_fields = []
|
|
|
|
@lazyproperty
|
|
def ordering_fields(self):
|
|
return self._get_ordering_fields()
|
|
|
|
def _get_ordering_fields(self):
|
|
if isinstance(self.__class__.ordering_fields, (list, tuple)):
|
|
return self.__class__.ordering_fields
|
|
|
|
try:
|
|
valid_fields = self.get_valid_ordering_fields()
|
|
except Exception as e:
|
|
logger.debug('get_valid_ordering_fields error: %s, pass' % e)
|
|
# 这里千万不要这么用,会让 logging 重复,至于为什么,我也不知道
|
|
# logging.debug('get_valid_ordering_fields error: %s' % e)
|
|
valid_fields = []
|
|
|
|
fields = list(chain(
|
|
valid_fields,
|
|
self.extra_ordering_fields
|
|
))
|
|
return fields
|
|
|
|
def get_valid_ordering_fields(self):
|
|
if getattr(self, 'model', None):
|
|
model = self.model
|
|
elif getattr(self, 'queryset', None):
|
|
model = self.queryset.model
|
|
else:
|
|
queryset = self.get_queryset()
|
|
if isinstance(queryset, list):
|
|
model = None
|
|
else:
|
|
model = queryset.model
|
|
|
|
if not model:
|
|
return []
|
|
|
|
excludes_fields = (
|
|
models.UUIDField, models.Model, models.ForeignKey,
|
|
models.FileField, models.JSONField, models.ManyToManyField,
|
|
models.DurationField,
|
|
)
|
|
valid_fields = []
|
|
for field in model._meta.fields:
|
|
if isinstance(field, excludes_fields):
|
|
continue
|
|
valid_fields.append(field.name)
|
|
return valid_fields
|
|
|
|
|
|
class CommonApiMixin(
|
|
SerializerMixin, ExtraFilterFieldsMixin, OrderingFielderFieldsMixin,
|
|
QuerySetMixin, RenderToJsonMixin, PaginatedResponseMixin
|
|
):
|
|
def is_swagger_request(self):
|
|
return getattr(self, 'swagger_fake_view', False) or \
|
|
getattr(self, 'raw_action', '') == 'metadata'
|
|
|
|
def get_queryset(self):
|
|
queryset = super().get_queryset()
|
|
|
|
if 'name' in self.ordering_fields:
|
|
self.ordering = ('name',)
|
|
|
|
return queryset
|