SQLAlchemy 无缘无故地发出交叉连接

Question

我在 SQLAlchemy 中设置了一个查询，该查询运行得有点慢，尝试对其进行优化。由于未知的原因，结果使用了隐式交叉连接，这不仅速度慢得多，而且产生完全错误的结果。I\xe2\x80\x99 对表名和参数进行了匿名处理，但没有进行任何更改。有谁知道这是从哪里来的？

\n\n

为了更容易找到：新发出的 SQL 和旧发出的 SQL 的区别在于新发出的 SQL 具有更长的 SELECT，并且在任何 JOIN 之前的 WHERE 中提到了所有三个表。

\n\n

原始代码：

\n\n

cust_name = u'Bob'\nproj_name = u'job1'\nitem_color = u'blue'\nquery = (db.session.query(Item.name)\n                       .join(Project, Customer)\n                       .filter(Customer.name == cust_name,\n                               Project.name == proj_name)\n                       .distinct(Item.name))\n\n# some conditionals determining last filter, resolving to this one:\nquery = query.filter(Item.color == item_color)\n\nresult = query.all()\n

\n\n

由flask_sqlalchemy.get_debug_queries记录的原始发出的SQL：

\n\n

QUERY: SELECT DISTINCT ON (items.name) items.name AS items_name\nFROM items JOIN projects ON projects.id = items._project_id JOIN customers ON customers.id = projects._customer_id\nWHERE customers.name = %(name_1)s AND projects.name = %(name_2)s AND items.color = %(color_1)s\nParameters: `{'name_2': u'job1', 'state_1': u'blue', 'name_1': u'Bob'}\n

\n\n

新代码：

\n\n

cust_name = u'Bob'\nproj_name = u'job1'\nitem_color = u'blue'\nquery = (db.session.query(Item)\n                     .options(Load(Item).load_only('name', 'color'),\n                                joinedload(Item.project, innerjoin=True).load_only('name').\n                                joinedload(Project.customer, innerjoin=True).load_only('name'))\n                     .filter(Customer.name == cust_name,\n                                 Project.name == proj_name)\n                     .distinct(Item.name))\n\n# some conditionals determining last filter, resolving to this one:\nquery = query.filter(Item.color == item_color)\n\nresult = query.all()\n

\n\n

由flask_sqlalchemy.get_debug_queries记录的新发出的SQL：

\n\n

QUERY: SELECT DISTINCT ON (items.nygc_id) items.id AS items_id, items.name AS items_name, items.color AS items_color, items._project_id AS items__project_id, customers_1.id AS customers_1_id, customers_1.name AS customers_1_name, projects_1.id AS projects_1_id, projects_1.name AS projects_1_name\nFROM customers, projects, items JOIN projects AS projects_1 ON projects_1.id = items._project_id JOIN customers AS customers_1 ON customers_1.id = projects_1._customer_id\nWHERE customers.name = %(name_1)s AND projects.name = %(name_2)s AND items.color = %(color_1)s\nParameters: `{'state_1': u'blue', 'name_2': u'job1', 'name_1': u'Bob'}\n

\n\n

如果重要的话，底层数据库是 PostgreSQL。

\n\n

查询的初衷只需要Item.name. 我思考的时间越长，优化尝试看起来就越不可能真正有帮助，但我仍然想知道交叉连接来自哪里，以防它再次发生在添加joinedload、load_only等实际上会有所帮助的地方。

\n

Answer 1

这是因为 ajoinedload与 a 不同join。ed实体joinedload实际上是匿名的，并且您应用的后续过滤器引用同一表的不同实例，因此customers和projects被连接两次。

您应该做的是像以前一样执行 a join，但使用contains_eager使您的连接看起来像joinedload。

query = (session.query(Item)
                .join(Item.project)
                .join(Project.customer)
                .options(Load(Item).load_only('name', 'color'),
                         Load(Item).contains_eager("project").load_only('name'),
                         Load(Item).contains_eager("project").contains_eager("customer").load_only('name'))
                .filter(Customer.name == cust_name,
                        Project.name == proj_name)
                .distinct(Item.name))

这给你查询

SELECT DISTINCT ON (items.name) customers.id AS customers_id, customers.name AS customers_name, projects.id AS projects_id, projects.name AS projects_name, items.id AS items_id, items.name AS items_name, items.color AS items_color 
FROM items JOIN projects ON projects.id = items._project_id JOIN customers ON customers.id = projects._customer_id 
WHERE customers.name = %(name_1)s AND projects.name = %(name_2)s AND items.color = %(color_1)s