大概的概念
您可以使用qs.union:
- 创建 2 个模型，它们之间没有任何关系。不要忘记使用
class Meta: managed = False
- 从第一个模型中选择，用子查询进行注释并与第二个模型联合：
from django.db import models
from django.db.models import F, OuterRef, Subquery, Value
from django.db.models.functions import Coalesce
# OperationalDevice fields: ip, mac
# AllowedDevice fields: ip, type, owner
USE_EMPTY_STR_AS_DEFAULT = True
null_char_field = models.CharField(null=True)
if USE_EMPTY_STR_AS_DEFAULT:
default_value = ''
else:
default_value = None
# By default Expressions treat strings as "field_name" so if you want to use
# empty string as a second argument for Coalesce, then you should wrap it in
# `Value()`.
# `None` can be used there without wrapping in `Value()`, but in
# `.annotate(type=NoneValue)` it still should be wrapped, so it's easier to
# just "always wrap".
default_value = Value(default_value, output_field=null_char_field)
operational_devices_subquery = OperationalDevice.objects.filter(ip=OuterRef('ip'))
qs1 = (
AllowedDevice.objects
.all()
.annotate(
mac=Coalesce(
Subquery(operational_devices_subquery.values('mac')[:1]),
default_value,
output_field=null_char_field,
),
)
)
qs2 = (
OperationalDevice.objects
.exclude(
ip__in=qs1.values('ip'),
)
.annotate(
type=default_value,
owner=default_value,
)
)
final_qs = qs1.union(qs2)
多个领域的通用方法
可以使用更复杂但“通用”的方法Model._meta.get_fields()。对于“第二个”模型有超过 1 个额外字段的情况（不仅是ip,mac）。示例代码（未测试，但给出一般印象）：
# One more import:
from django.db.models.fields import NOT_PROVIDED
common_field_name = 'ip'
# OperationalDevice fields: ip, mac, some_more_fields ...
# AllowedDevice fields: ip, type, owner
operational_device_fields = OperationalDevice._meta.get_fields()
operational_device_fields_names = {_f.name for _f in operational_device_fields} # or set((_f.name for ...))
allowed_device_fields = AllowedDevice._meta.get_fields()
allowed_device_fields_names = {_f.name for _f in allowed_device_fields} # or set((_f.name for ...))
operational_devices_subquery = OperationalDevice.objects.filter(ip=OuterRef(common_field_name))
left_joined_qs = ( # "Kind-of". Assuming AllowedDevice to be "left" and OperationalDevice to be "right"
AllowedDevice.objects
.all()
.annotate(
**{
_f.name: Coalesce(
Subquery(operational_devices_subquery.values(_f.name)[1]),
Value(_f.get_default()), # Use defaults from model definition
output_field=_f,
)
for _f in operational_device_fields
if _f.name not in allowed_device_fields_names
# NOTE: if fields other than `ip` "overlap", then you might consider
# changing logic here. Current implementation keeps fields from the
# AllowedDevice
}
# Unpacked dict is partially equivalent to this:
# mac=Coalesce(
# Subquery(operational_devices_subquery.values('mac')[:1]),
# default_for_mac_eg_fallback_text_value,
# output_field=null_char_field,
# ),
# other_field = Coalesce(...),
# ...
)
)
lonely_right_rows_qs = (
OperationalDevice.objects
.exclude(
ip__in=AllowedDevice.objects.all().values(common_field_name),
)
.annotate(
**{
_f.name: Value(_f.get_default(), output_field=_f), # Use defaults from model definition
for _f in allowed_device_fields
if _f.name not in operational_device_fields_names
# NOTE: See previous NOTE
}
)
)
final_qs = left_joined_qs.union(lonely_right_rows_qs)
使用 OneToOneField 实现“更好”的 SQL
理论上你可以使用device_info = models.OneToOneField(OperationalDevice, db_column='ip', primary_key=True, related_name='status_info'): in AllowedDevice。在这种情况下，您的第一个 QS 可以在不使用的情况下定义Subquery:
from django.db.models import F
# Now 'ip' is not in field names ('device_info' is there), so add it:
allowed_device_fields_names.add(common_field_name)
# NOTE: I think this approach will result in a more compact SQL query without
# multiple `(SELECT "some_field" FROM device_info_table ... ) as "some-field"`.
# This also might result in better query performance.
honest_join_qs = (
AllowedDevice.objects
.all()
.annotate(
**{
_f.name: F(f'device_info__{_f.name}')
for _f in operational_device_fields
if _f.name not in allowed_device_fields_names
}
)
)
final_qs = honest_join_qs.union(lonely_right_rows_qs)
# or:
# final_qs = honest_join_qs.union(
# OperationalDevice.objects.filter(status_info__isnull=True).annotate(**missing_fields_annotation)
# )
# I'm not sure which approach is better performance-wise...
# Commented one will use something like:
# `SELECT ... FROM "device_info_table" LEFT OUTER JOIN "status_info_table" ON ("device_info_table"."ip" = "status_info_table"."ip") WHERE "status_info_table"."ip" IS NULL
#
# So it might be a little better than first with `union(QS.exclude(ip__in=honest_join_qs.values('ip'))`.
# Because later uses SQL like this:
# `SELECT ... FROM "device_info_table" WHERE NOT ip IN (SELECT ip FROM "status_info_table")`
#
# But it's better to measure timings of both approaches to be sure.
# @GrannyAching, can you compare them and tell in the comments which one is better ?
附：要自动化模型定义，您可以使用manage.py inspectdb
附言或许多表继承与定制OneToOneField(..., parent_link=True)可能比使用对您更有帮助union.
