账号备份优化

pull/8873/head
feng626 2 years ago
parent 7ae395f7e8
commit 757e688ab3

@ -49,16 +49,23 @@ class BaseAccountHandler:
return header_fields
@classmethod
def create_row(cls, account, serializer_cls, header_fields=None):
serializer = serializer_cls(account)
if not header_fields:
header_fields = cls.get_header_fields(serializer)
data = cls.unpack_data(serializer.data)
def create_row(cls, data, header_fields):
data = cls.unpack_data(data)
row_dict = {}
for field, header_name in header_fields.items():
row_dict[header_name] = str(data[field])
row_dict[header_name] = str(data.get(field, field))
return row_dict
@classmethod
def add_rows(cls, data, header_fields, sheet):
data_map = defaultdict(list)
for i in data:
row = cls.create_row(i, header_fields)
if sheet not in data_map:
data_map[sheet].append(list(row.keys()))
data_map[sheet].append(list(row.values()))
return data_map
class AssetAccountHandler(BaseAccountHandler):
@staticmethod
@ -73,22 +80,25 @@ class AssetAccountHandler(BaseAccountHandler):
data_map = defaultdict(list)
# TODO 可以优化一下查询 在账号上做type的缓存 避免数据量大时连表操作
accounts = Account.objects.filter(
qs = Account.objects.filter(
asset__platform__type__in=types
).annotate(type=F('asset__platform__type'))
if not accounts.first():
if not qs.exists():
return data_map
type_dict = dict(Type.CHOICES)
header_fields = cls.get_header_fields(AccountSecretSerializer(accounts.first()))
for account in accounts:
sheet_name = type_dict[account.type]
row = cls.create_row(account, AccountSecretSerializer, header_fields)
if sheet_name not in data_map:
data_map[sheet_name].append(list(row.keys()))
data_map[sheet_name].append(list(row.values()))
logger.info('\n\033[33m- 共收集 {} 条账号\033[0m'.format(accounts.count()))
header_fields = cls.get_header_fields(AccountSecretSerializer(qs.first()))
account_type_map = defaultdict(list)
for account in qs:
account_type_map[account.type].append(account)
data_map = {}
for tp, accounts in account_type_map.items():
sheet_name = type_dict[tp]
data = AccountSecretSerializer(accounts, many=True).data
data_map.update(cls.add_rows(data, header_fields, sheet_name))
logger.info('\n\033[33m- 共收集 {} 条账号\033[0m'.format(qs.count()))
return data_map

Loading…
Cancel
Save