left join与inner join结果之和与总体不等的思考

测试环境是mysql，但实际上跟数据库环境没有关系，是逻辑上的问题。
1、建表及测试数据构造

drop tables if exists xxxxx_iqy0708;
create table xxxxx_iqy0708(ip varchar(100));
drop table if exists xxxxx_chusou0708;
create table xxxxx_chusou0708(appkey varchar(100), ip varchar(100), identify varchar(100));
insert into xxxxx_iqy0708
VALUES('xx1.xx2.xx3.xx4'),('xx2.xx2.xx3.xx4'),('xx3.xx2.xx3.xx4');
insert into xxxxx_chusou0708(appkey,ip,identify)
VALUES
('CSAndroid','xx1.xx2.xx3.xx4','id1'),
('CSAndroid','xx2.xx2.xx3.xx4','id2'),
('CSAndroid','xx5.xx2.xx3.xx4','id8'),
('CSAndroid','xx6.xx2.xx3.xx4','id3'),
('CSIos','xx1.xx2.xx3.xx4','id2'),
('CSIos','xx2.xx2.xx3.xx4','id1'),
('CSIos','xx3.xx2.xx3.xx4','id3'),
('CSIos','xx2.xx2.xx3.xx4','id3'),
('CSIos','xx5.xx2.xx3.xx4','id3');

2、查询数据
2.1、用ip关联取两者差集

select appkey,count(distinct identify)
from xxxxx_chusou0708 a1
left join xxxxx_iqy0708 a2 on a1.ip=a2.ip
where a2.ip is null
group by appkey;

--注：根据ip关联，则其对应的identify可能已经被关系上了，但也可能未被关联上；就是说一个identify对应多个ip

appkey	count(distinct identify)
CSAndroid	2
CSIos	1

2.2、用ip关联取两者交集

select appkey,count(distinct identify)
from xxxxx_chusou0708 a1
inner join xxxxx_iqy0708 a2 on a1.ip=a2.ip
group by appkey;

-- ip关联，对应的是关联上的identify；同上，就是说一个identify对应多个ip

appkey	count(distinct identify)
CSAndroid	2
CSIos	3

2.3、总体计算identify数据

select appkey, count(distinct identify)
from xxxxx_chusou0708
group by appkey;

-- 所有的去重，理论上，这个会小于等于2.1+2.3之和。

appkey	count(distinct identify)
CSAndroid	4
CSIos	3

3、说明
原始数据的查看：

select * from xxxxx_chusou0708;
select * from xxxxx_iqy0708;

如果严格关联，2.3=2.1+2.2；但本文所述情况例外，有时挺绕的，不仔细想想，搞不明白。