最开始我用子查询实现，巨慢无比：
select id,like_count,favorite_count from message_message where id in(select message_id from message_message_tags where messagetag_id=62101) and (category=9 or category=1) order by (like_count+favorite_count) desc
原因是in子句的数据量大的话，性能就会非常低下。详见：
修改成多表查询：
select t1.sender_id,t1.id,t1.like_count,t1.favorite_count,t1.add_datetime from message_message t1, message_message2tag t2 where t1.id = t2.message_id and t2.tag_id=120560 and t1.category in (9,1) and t1.add_datetime>'2012-06-29' and t1.add_datetime < '2012-07-03' order by (like_count+favorite_count) desc limit 60
性能解决了，但发现有重复记录：
+----------+------------+----------------+-----------+---------------------+
| id | like_count | favorite_count | sender_id | add_datetime |
+----------+------------+----------------+-----------+---------------------+
| 33850491 | 2 | 220 | 471956 | 2012-06-29 20:42:47 |
| 33778187 | 14 | 167 | 122216 | 2012-06-29 12:24:15 |
| 33779339 | 8 | 165 | 122216 | 2012-06-29 12:33:57 |
| 34068745 | 0 | 112 | 3054 | 2012-07-01 15:00:26 |
| 33782239 | 4 | 94 | 116633 | 2012-06-29 12:58:23 |
| 33815718 | 2 | 92 | 4287 | 2012-06-29 17:10:13 |
| 33905850 | 3 | 87 | 85131 | 2012-06-30 10:36:39 |
| 33827272 | 1 | 75 | 171462 | 2012-06-29 18:21:31 |
| 33778798 | 7 | 69 | 122216 | 2012-06-29 12:29:38 |
| 33780283 | 8 | 64 | 122216 | 2012-06-29 12:42:15 |
| 33816638 | 0 | 59 | 4287 | 2012-06-29 17:16:16 |
122216一个会员发了多条记录，看来需要做distinct，直接把distinct加在字段前面是没有效果，可以通过结合子查询来和group来做，最终版：
select * from (select t1.sender_id,t1.id,t1.like_count,t1.favorite_count,t1.add_datetime from message_message t1, message_message2tag t2 where t1.id = t2.message_id and t2.tag_id=120560 and t1.category in (9,1) and t1.add_datetime>'2012-06-29' and t1.add_datetime < '2012-07-03' order by (like_count+favorite_count) desc limit 60) as t group by t.sender_id order by (like_count+favorite_count) desc;
这个人提到说通过
select *, count(distinct name) from table group by name来实现
http://hi.baidu.com/liveinyc/blog/item/facac543a2dc260b9213c658.html