在简单的 SQL 查询中可能没有正确使用索引

Question

尺寸：

• 广告系列：3k 行（200 行，campaign.is_active = 1）
• 链接：20k 行（4k 行，links.status = 1 // 500 行，links.status = 1 AND campaign.is_active = 1）
• 点击次数：1000 万行（50k 次创建 > '2020-10-25 00:00:00'）

---

此查询运行 2 秒

SELECT links.id, COUNT(clicks.id)
FROM links 
INNER JOIN campaigns ON campaigns.id = links.campaign_id 
AND campaigns.is_active = 1 
LEFT JOIN clicks ON clicks.link_id = links.id
WHERE links.status = 1 
AND clicks.created > '2020-10-25 00:00:00'
GROUP BY links.id

当我删除以下行时，它只运行 0.13 秒（快 15 倍）

AND campaigns.is_active = 1

campaigns.is_active 上有一个索引。 还尝试在 2 列（campaigns.id + campaign.is_active）上设置索引，但没有帮助。

“campaigns.is_active”只包含 0 或 1。campaigns 表很小，campaigns.is_active 条件实际上减少了行数。所以它应该加快查询速度。

为什么会因为这种情况需要这么长时间？如何解决？

如果我要删除活动的 JOIN，而是将 links.campaign_id 添加到 SELECT 字段，然后在附加查询中查询每个返回的活动 ID，例如“SELECT is_active FROMcampaign WHERE id = ?" 它仍然会更快，因为这样的查询是 0.000x。根据我的经验，当 2 次查询速度更快时，这通常意味着第一次查询没有得到最大程度的优化。

解释-选择

结构

CREATE TABLE `campaigns` (
  `id` int(11) UNSIGNED NOT NULL,
  `is_active` tinyint(4) NOT NULL DEFAULT 0
) ENGINE=InnoDB DEFAULT CHARSET=utf8;


CREATE TABLE `clicks` (
  `id` int(11) UNSIGNED NOT NULL,
  `link_id` int(11) UNSIGNED NOT NULL,
  `created` datetime NOT NULL
) ENGINE=InnoDB DEFAULT CHARSET=utf8;


CREATE TABLE `links` (
  `id` int(11) UNSIGNED NOT NULL,
  `campaign_id` int(8) UNSIGNED NOT NULL,
  `status` tinyint(4) NOT NULL DEFAULT 0
) ENGINE=InnoDB DEFAULT CHARSET=utf8;


ALTER TABLE `campaigns`
  ADD PRIMARY KEY (`id`),
  ADD UNIQUE KEY `id_isactive` (`id`,`is_active`),
  ADD KEY `is_active` (`is_active`)

ALTER TABLE `clicks`
  ADD PRIMARY KEY (`id`),
  ADD KEY `link_id` (`link_id`),
  ADD KEY `created` (`created`)

ALTER TABLE `links`
  ADD PRIMARY KEY (`id`),
  ADD KEY `campaign_id` (`campaign_id`),

Answer 1

这需要多长时间？

SELECT l.id,
       (SELECT COUNT(*)
        FROM clicks cl
        WHERE cl.link_id = l.id AND
              cl.created > '2020-10-25'
       )
FROM links l JOIN
     campaigns ca
     ca.id = l.campaign_id 
WHERE l.status = 1 AND ca.is_active = 1;

编辑：

嗯，order by，你可以试试：

SELECT l.id,
       (SELECT COUNT(*)
        FROM clicks cl
        WHERE cl.link_id = l.id AND
              cl.created > '2020-10-25'
       )
FROM links l 
WHERE EXISTS (SELECT 1
              FROM campaigns ca
              WHERE ca.id = l.campaign_id AND ca.is_active = 1
             )
WHERE l.status = 1 
ORDER BY l.id;

为此，您需要在links(status, id) 和campaigns(campaign_id, is_active) 上建立索引。

Answer 2

问题...如果一个活动当前不活跃，你不希望它有任何输出，对吗？此外，不活动的广告系列不会有任何点击，对吗？那为什么还要检查is_active？

即使我的分析是错误的，在计数完成之前忽略 is_active 可能会更快。

当它不起作用时，请不要使用LEFT。你有一个简单的JOIN。

使用COUNT(*); COUNT(x) 测试 x 不为空。

SELECT  links.id, COUNT(*)
    FROM  links
    JOIN  clicks  ON clicks.link_id = links.id
    WHERE  links.status = 1
      AND  clicks.created > '2020-10-25 00:00:00'
    GROUP BY  links.id

这是多余的：

ADD UNIQUE KEY `id_isactive` (`id`,`is_active`),

因为PRIMARY KEY(id) 声明id 是一个索引并且是唯一的。

Answer 3

我不喜欢和数据库引擎优化器作斗争。

SELECT links.id, campaigns.is_active, COUNT(clicks.id)
FROM links 
INNER JOIN campaigns ON campaigns.id = links.campaign_id 
LEFT JOIN clicks ON clicks.link_id = links.id
WHERE links.status = 1 
AND clicks.created > '2020-10-25 00:00:00'
GROUP BY links.id, campaigns.is_active
HAVING campaigns.is_active = 1;

第二个变种！

-- Second Variant
EXPLAIN
SELECT links.id AS LinksId
      , COUNT(clicks.id) AS ClickCount
FROM links 
LEFT JOIN clicks 
    ON links.id = clicks.link_id
WHERE links.status = 1
AND clicks.created > '2020-10-25 00:00:00'
AND links.campaign_id IN (SELECT campaign_id 
                            FROM campaigns
                            WHERE is_active = 1)
GROUP BY links.id;

第三次是魅力！由于已发布的基数而使用 CTE。

-- Third time is the charm
WITH ActiveCampaigns
AS
    (SELECT *
    FROM campaigns
    WHERE is_active = 1)
SELECT links.id, COUNT(clicks.id)
FROM links 
INNER JOIN ActiveCampaigns 
    ON ActiveCampaigns.id = links.campaign_id 
LEFT JOIN clicks 
    ON clicks.link_id = links.id
WHERE links.status = 1 
AND clicks.created > '2020-10-25 00:00:00'
GROUP BY links.id;