Question

我有2个MySQL表：

的网页：

id INT,
name VARCHAR,
view_count INT

PAGE_VIEWS：

page_id INT
viewed_at DATETIME,
ip VARCHAR(15),
processed INT(1)

当有人查看某个页面时，会在page_views表中添加一条带有processed = 0标记的新记录。每N分钟未处理的记录应该被提取，计算并添加到view_count表的pages属性中。

没有交易和锁定，它看起来像：

SELECT page_id, COUNT(ip) AS cnt FROM page_views WHERE processed = 0 GROUP BY page_id;
UPDATE page_views SET processed = 1 WHERE processed = 0;
UPDATE pages SET view_count = view_count + ...

但在这种情况下，可以在前2个查询之间将新记录添加到page_views。所以我们应该锁定page_views表进行编写。但是，当第三个查询由于某种原因失败时，还有另外一个案例可能，但processed表已在page_views表上更新，下次不会考虑这些记录。这意味着我们也应该使用交易。

所以算法应该如下：

-- lock `page_views` for writing
-- start transaction
SELECT page_id, COUNT(ip) AS cnt FROM page_views WHERE processed = 0 GROUP BY page_id;
UPDATE page_views SET processed = 1 WHERE processed = 0;
-- unlock `page_views`
UPDATE pages SET view_count = view_count + ...
-- commit transaction

但根据MySQL documentation，LOCK TABLES不是交易安全的，所以我正在寻找处理这种情况的正确方法。

回答评论：

在这种情况下，仅使用事务的原因示例不起作用：

mysql_1> SET TRANSACTION ISOLATION LEVEL REPEATABLE READ;
mysql_1> START TRANSACTION;
mysql_1> SELECT * FROM page_views;
+---------+------------+-----------+-----------+
| page_id | date       | ip        | processed |
+---------+------------+-----------+-----------+
|    2832 | 2015-09-10 | 127.0.0.1 |         1 |
|    2832 | 2015-09-11 | 127.0.0.1 |         1 |
+---------+------------+-----------+-----------+
mysql_2> INSERT INTO page_views SET page_id = 2832, date='2015-09-12', ip='127.0.0.1', processed=0;
mysql_1> UPDATE page_views SET processed = 1 WHERE processed = 0;
mysql_1> COMMIT;
mysql_1> SELECT * FROM page_views;
+---------+------------+-----------+-----------+
| page_id | date       | ip        | processed |
+---------+------------+-----------+-----------+
|    2832 | 2015-09-10 | 127.0.0.1 |         1 |
|    2832 | 2015-09-11 | 127.0.0.1 |         1 |
|    2832 | 2015-09-12 | 127.0.0.1 |         1 | <-- this should be 0
+---------+------------+-----------+-----------+

用于选择和更新的锁定表

0 个答案: