编写两次引用同一个表的 SQL 查询

Question

考虑以下表定义

class MCastSession(Base):
  __tablename__ = 'mcast_session'
  id = Column(Integer, primary_key=True)
  ip = Column(Integer)
  port = Column(Integer)
  __table_args__ = ( UniqueConstraint('ip', 'port'), )

class Topic(Base):
  __tablename__ = 'topic'
  id = Column(Integer, primary_key=True)
  name = Column(String, unique=True)
  mcast_session_id = Column(Integer, ForeignKey('mcast_session.id'))
  mcast_session = relationship('MCastSession')

class Host(Base):
  __tablename__ = 'host'
  id = Column(Integer, primary_key=True)
  name = Column(String, unique=True)

class Subscriber(Base):
  __tablename__ = 'subscriber'
  id = Column(Integer, primary_key=True)
  topic_id = Column(Integer, ForeignKey('topic.id'))
  topic = relationship('Topic')
  host_id = Column(Integer, ForeignKey('host.id'))
  host = relationship('Host')
  __table_args__ = ( UniqueConstraint('topic_id', 'host_id'), )

Example data:
Topic Session
T1    IP1:port1
T2    IP1:port2
T3    IP1:port2
T4    IP2:port1

Topic Host
T1    H1
T2    H1
T4    H2

我想编写一个查询来获取订阅多播 ip 但不处理该 ip 的所有主题的所有主机。在上面的例子中。 H1 有 T1，因此订阅了 IP1，但没有 T3，T3 也有相同的 IP1。所以查询应该返回 H1。 H2 为其订阅的 ips (T4) 处理所有主题(T4)，因此 H2 不应出现在结果中。如何编写上述查询？

Answer 1

下面的查询将得到目标：

select distinct host1.name host_name
  from Subscriber Subscriber1
 inner join host host1
    on host1.id = Subscriber1.Host_Id
 inner join topic topic1
    on topic1.id = Subscriber1.Topic_Id
 inner join mcast_session mcast_session1
    on mcast_session1.id = topic1.mcast_session_id
 where (select count(*)
          from mcast_session
         where mcast_session.ip = mcast_session1.ip) !=
       (select count(*)
          from topic
         inner join mcast_session
            on topic.mcast_session_id = mcast_session.id
         where mcast_session.ip = mcast_session1.ip)

为了解释逻辑，查询可能会有所帮助：

select  host1.name host_name,
       topic1.name topic_name,
       mcast_session1.ip,
       mcast_session1.port,
(select count(*)
          from mcast_session
         where mcast_session.ip = mcast_session1.ip) host_to_topic_registeration,
       (select count(*)
          from topic
         inner join mcast_session
            on topic.mcast_session_id = mcast_session.id
         where mcast_session.ip = mcast_session1.ip
           ) ip_topic_count
  from Subscriber Subscriber1
 inner join host host1
    on host1.id = Subscriber1.Host_Id
 inner join topic topic1
    on topic1.id = Subscriber1.Topic_Id
 inner join mcast_session mcast_session1
    on mcast_session1.id = topic1.mcast_session_id

sqlfiddle sample：

Answer 2

查看所需结果的另一种方法是应用以下逻辑：

1. 为每个IP地址计算Topics广播的总数
2. 对于每个Host，计算按IP 地址分组的Topics 广播总数
3. Desires Host 是指每个IP 地址的Topics 数量不等于（实际上更少）IP 地址总数的那些。

以下 SA 代码应为您提供所需的 Host 实例：

# subquery to get number of topics per IP
subq_ip_topics = (session.query(
        MCastSession.ip.label("mcast_session_ip"),
        func.count(Topic.id).label("num_topics")
    )
    .join(Topic)
    .group_by(MCastSession.ip)
    ).subquery().alias("ip_topics")

# subquery to get number of topics per host per ip
subq_host_ip_topics = (session.query(
        Host.id.label("host_id"),
        MCastSession.ip.label("mcast_session_ip"),
        func.count(Topic.id).label("num_topics")
    )
    .join(Subscriber)
    .join(Topic)
    .join(MCastSession)
    .group_by(Host.id, MCastSession.ip)
    ).subquery().alias("host_ip_topics")

# final query: get those Hosts where results on both sub-queries do not match
query = (session.query(Host)
        .join(subq_host_ip_topics, Host.id == subq_host_ip_topics.c.host_id)
        .join(subq_ip_topics, and_(
            subq_host_ip_topics.c.mcast_session_ip == subq_ip_topics.c.mcast_session_ip,
            subq_host_ip_topics.c.num_topics != subq_ip_topics.c.num_topics
            ))
        )

产生下面的SQL 代码（对于SQLite）：

SELECT  host.id AS host_id, host.name AS host_name

FROM    host

JOIN   (SELECT  host.id AS host_id,
                mcast_session.ip AS mcast_session_ip,
                count(topic.id) AS num_topics
        FROM    host
        JOIN    subscriber
            ON  host.id = subscriber.host_id
        JOIN    topic
            ON  topic.id = subscriber.topic_id
        JOIN    mcast_session
            ON  mcast_session.id = topic.mcast_session_id
        GROUP BY host.id, mcast_session.ip
       ) AS host_ip_topics
    ON  host.id = host_ip_topics.host_id

JOIN   (SELECT  mcast_session.ip AS mcast_session_ip,
                count(topic.id) AS num_topics
        FROM    mcast_session
        JOIN    topic
            ON  mcast_session.id = topic.mcast_session_id
        GROUP BY mcast_session.ip
       ) AS ip_topics
    ON  host_ip_topics.mcast_session_ip = ip_topics.mcast_session_ip
    AND host_ip_topics.num_topics != ip_topics.num_topics

---

现在，如果您想在查询中多次使用同一个表，您可以使用aliased。下面的代码将返回一个元组列表(MCaseSession, NNN)，其中NNN是具有相同IP的MCastSession对象的数量：

aliased_MCastSession = aliased(MCastSession, name="MCastSession2")
qry = session.query(\
    MCastSession, \
    func.count(aliased_MCastSession.id).label("number_with_same_ip")).\
filter(MCastSession.ip == aliased_MCastSession.ip).\
group_by(MCastSession)

但我不需要为提出的解决方案执行此操作，因为我为此使用了子查询。