优化大数据的“有计数（不同）”查询

Question

我有以下查询，我需要在一个有 1 亿条记录的表上运行，但它非常慢（到目前为止已经运行了 5 个小时） 我不知道如何优化它，将不胜感激任何帮助 该表在 DID 和 week_no 上有一个索引，并包含其他几个未编入索引的列，以及一个已编入索引的主键 (id)

 DELETE FROM test WHERE "DID" IN (SELECT "DID" FROM test GROUP BY "DID" having count(distinct week_no) < 4 );

谢谢！

Answer 1

这将使用DELETE 和JOIN（或USING 用于PostgreSQL）来最有效地编写，以避免必须计算每一行的计数：

对于 PostgreSQL：

DELETE
FROM test t1
USING (SELECT did, COUNT(DISTINCT week_no) AS num_weeks
       FROM test
       GROUP BY did) t2
WHERE t2.did = t1.did AND num_weeks < 4

Demo on dbfiddle

在 MySQL 中：

DELETE t1
FROM test t1
JOIN (SELECT did, COUNT(DISTINCT week_no) AS num_weeks
      FROM test
      GROUP BY did) t2 ON t2.did = t1.did
WHERE num_weeks < 4

Demo on dbfiddle