peewee update和save性能分析

背景

python項目中使用了peewee這款orm框架，在對數據庫更新時有兩種語法，分別是save和update方法。有同事說從peewee的日誌來看，update比save更快，於是做了一個簡單的比較實驗，看看真實情況如何。

基礎環境：
python: 3.8.10
peewee: 3.16.2
數據庫：sqlite

準備

插入1w條數據

import datetime
from peewee import AutoField, DateTimeField, Model, SqliteDatabase, TextField, IntegerField

class BaseModel(Model):
    """A base model that will use our Sqlite database."""
    id = AutoField()
    update_time = DateTimeField(default=datetime.datetime.now)

    class Meta:
        database = db
		
		
class User(BaseModel):
    name = TextField()
    age = IntegerField()

    class Meta:
        table_name = "user"
		
		
if __name__ == "__main__":
    User.truncate_table()
    db.connect()
    db.create_tables([User])
	
	data = []
    for i in range(10000):
        data.append({"name": f"person_P{i}", "age": i})
        print(i)

    User.insert_many(data).execute()

update 更新

if __name__ == "__main__":
    import logging
    import time

    logger = logging.getLogger('peewee')
    logger.propagate = False
    logger.addHandler(logging.StreamHandler())
    logger.setLevel(logging.DEBUG)
	
    start = time.time()
    for user in users:
        User.update(age=6012).where(User.id==user.id).execute()
        print(user.id, user.name, user.age)
	end = time.time()
    print(f"total: {end-start}")
	

peewee：日誌

('UPDATE "user" SET "update_time" = ?, "name" = ?, "age" = ? WHERE ("user"."id" = ?)', [datetime.datetime(2023, 8, 29, 17, 30, 36, 719081), 'person_P9996', 12341, 10024])
('UPDATE "user" SET "update_time" = ?, "name" = ?, "age" = ? WHERE ("user"."id" = ?)', [datetime.datetime(2023, 8, 29, 17, 30, 36, 719088), 'person_P9997', 12341, 10025])
('UPDATE "user" SET "update_time" = ?, "name" = ?, "age" = ? WHERE ("user"."id" = ?)', [datetime.datetime(2023, 8, 29, 17, 30, 36, 719096), 'person_P9998', 12341, 10026])
('UPDATE "user" SET "update_time" = ?, "name" = ?, "age" = ? WHERE ("user"."id" = ?)', [datetime.datetime(2023, 8, 29, 17, 30, 36, 719103), 'person_P9999', 12341, 10027])

結果：67.96582674980164 s

save更新

if __name__ == "__main__":
    import logging
    import time

    logger = logging.getLogger('peewee')
    logger.propagate = False
    logger.addHandler(logging.StreamHandler())
    logger.setLevel(logging.DEBUG)
	
    start = time.time()
    for user in users:
       user.age = 12341
        user.save()
        print(user.id, user.name, user.age)
	end = time.time()
    print(f"total: {end-start}")

peewee日誌：

('UPDATE "user" SET "age" = ? WHERE ("user"."id" = ?)', [2000, 10025])
10025 person_P9997 3000
('UPDATE "user" SET "age" = ? WHERE ("user"."id" = ?)', [2000, 10026])
10026 person_P9998 3000
('UPDATE "user" SET "age" = ? WHERE ("user"."id" = ?)', [2000, 10027])

結果：67.52418804168701 s

結果分析

從三個方面來分析：

• 從打印的日誌來看，save會更新記錄全部字段，update只會更新指定的字段。
• 從結果來看，1w行以內的更新操作兩者沒有性能的差別。
• 從方法實現來看，update方法是底層方法，save方法調用了update方法或insert方法實現更新操作。

所以理論上來說update 比 save 更底層，效率略高。實際使用中save寫法較爲方便，個人更喜歡save方法。

update 方法

    def __sql__(self, ctx):
        super(Update, self).__sql__(ctx)

        with ctx.scope_values(subquery=True):
            ctx.literal('UPDATE ')

            expressions = []
            for k, v in sorted(self._update.items(), key=ctx.column_sort_key):
                if not isinstance(v, Node):
                    if isinstance(k, Field):
                        v = k.to_value(v)
                    else:
                        v = Value(v, unpack=False)
                elif isinstance(v, Model) and isinstance(k, ForeignKeyField):
                    # NB: we want to ensure that when passed a model instance
                    # in the context of a foreign-key, we apply the fk-specific
                    # adaptation of the model.
                    v = k.to_value(v)

                if not isinstance(v, Value):
                    v = qualify_names(v)

                expressions.append(NodeList((k, SQL('='), v)))

            (ctx
             .sql(self.table)
             .literal(' SET ')
             .sql(CommaNodeList(expressions)))

            if self._from:
                with ctx.scope_source(parentheses=False):
                    ctx.literal(' FROM ').sql(CommaNodeList(self._from))

            if self._where:
                with ctx.scope_normal():
                    ctx.literal(' WHERE ').sql(self._where)
            self._apply_ordering(ctx)
            return self.apply_returning(ctx)

update方法可以看到是拼接出一個sql語句，update xx set xx=xx where

save 方法

    def save(self, force_insert=False, only=None):
        field_dict = self.__data__.copy()
        if self._meta.primary_key is not False:
            pk_field = self._meta.primary_key
            pk_value = self._pk
        else:
            pk_field = pk_value = None
        if only is not None:
            field_dict = self._prune_fields(field_dict, only)
        elif self._meta.only_save_dirty and not force_insert:
            field_dict = self._prune_fields(field_dict, self.dirty_fields)
            if not field_dict:
                self._dirty.clear()
                return False

        self._populate_unsaved_relations(field_dict)
        rows = 1

        if self._meta.auto_increment and pk_value is None:
            field_dict.pop(pk_field.name, None)

        if pk_value is not None and not force_insert:
            if self._meta.composite_key:
                for pk_part_name in pk_field.field_names:
                    field_dict.pop(pk_part_name, None)
            else:
                field_dict.pop(pk_field.name, None)
            if not field_dict:
                raise ValueError('no data to save!')
            rows = self.update(**field_dict).where(self._pk_expr()).execute()
        elif pk_field is not None:
            pk = self.insert(**field_dict).execute()
            if pk is not None and (self._meta.auto_increment or
                                   pk_value is None):
                self._pk = pk
                # Although we set the primary-key, do not mark it as dirty.
                self._dirty.discard(pk_field.name)
        else:
            self.insert(**field_dict).execute()

        self._dirty -= set(field_dict)  # Remove any fields we saved.
        return rows

save方法是調用update方法或insert方法間接實現更新。