看看 EXPLAIN EXTENDED 怎么说.

如果它是 DEPENDENT SUBQUERY 或 UNCACHEABLE SUBQUERY，那么每次使用时都会重新评估.

If it says DEPENDENT SUBQUERY or UNCACHEABLE SUBQUERY, then it will be reevaluated each time it's used.

如果子查询使用会话变量或者是相关子查询，就会发生这种情况.

This happens if the subquery uses session variables or is a correlated subquery.

如果没有，它很可能会被缓存.

If it doesn't, it most probably will be cached.

如果您的情况子查询不会被缓存，它将在每个 UNION 的集合中重新评估.

If your case the subquery will not be cached, it will be reevaluated in each UNION'ed set.

不过，您的子查询似乎太复杂了.你为什么不直接使用:

You subquery, though, seems to be too complicated. Why don't you just use:

SELECT id
FROM   playlist_program_map ppm, programs p
WHERE  ppm.playlist_id = 181
       AND p.id = ppm.program_id
       AND submitter_id = 32
       AND feed_id = 2478

如果您在 playlist_program_map (playlist_id) 上有一个索引，那么这个查询应该会很有效.

If you have an index on playlist_program_map (playlist_id), this query should work like a charm.

你能告诉我另外两件事吗:

Could you please tell me two more things:

1. playlist_program_map 中有多少行，DISTINCT playlist_id 值有多少?
   • programs 中有多少行，DISTINCT submitter_id, feed_id 对有多少?

1. How many rows are there in playlist_program_map and how many DISTINCT playlist_id values are there?
   • How many rows are there in programs and how many DISTINCT submitter_id, feed_id pairs are there?

根据您的评论，我可以得出结论，每个 playlist 平均有 10 programs，200 (submitter, feed) 对的 code>programs.这意味着 playlist_program_map 上的索引比 (submitter, feed) 上的索引更具选择性，并且 playlist_program_map 必须在连接中领先.

From your comment I can conclude that there are 10 programs per playlist in average, and 200 programs per (submitter, feed) pair. This means your index on playlist_program_map is more selective than the one on (submitter, feed), and playlist_program_map must be leading in the join.

鉴于您需要加入 2,000,000 中的 10 个程序，因此您的案例中的全文索引似乎也不是很有选择性.

The fulltext index in your case also doesn't seem to be very selective, given that you need to join 10 programs out of 2,000,000.

您***尝试以下方法:

SELECT object_id, programs.created AS created
FROM   playlist_program_map ppm, programs p, comments_programs cp
WHERE  ppm.playlist_id = 181
       AND p.id = ppm.program_id
       AND p.submitter_id = 32
       AND p.feed_id = 2478
       AND cp.object_id = p.id
       AND cp.text REGEXP 'excellent'

，然后对所有三个表重复此操作.

, and repeat this for all three tables.