PostgreSQL 在一个查询中获取事件发生的每日、每周和每月平均值

Question

目前我有这个相当大的查询

1. 通过获取count()按事件名称和日期分组的事件，将每日、每周、每月计数聚合到中间表中。
2. 通过avg()按事件分组来选择每个中间表的平均计数，对结果进行联合，并且因为我想为每天、每周、每月设置一个单独的列，将填充值 0 放入空列中。
3. 然后我对所有列求和，0 基本上充当空操作，这给我每个事件的单个值。

---

查询虽然很大，但我觉得我正在做很多重复的工作。有什么办法可以更好地执行此查询或使其更小吗？我以前没有真正做过这样的查询，所以我不太确定。

WITH monthly_counts as (
  SELECT
    event,
    count(*) as count
  FROM tracking_stuff
  WHERE
    event = 'thing'
    OR event = 'thing2'
    OR event = 'thing3'
  GROUP BY event, date_trunc('month', created_at)
),
weekly_counts as (
  SELECT
    event,
    count(*) as count
  FROM tracking_stuff
  WHERE
    event = 'thing'
    OR event = 'thing2'
    OR event = 'thing3'
  GROUP BY event, date_trunc('week', created_at)
),
daily_counts as (
  SELECT
    event,
    count(*) as count
  FROM tracking_stuff
  WHERE
    event = 'thing'
    OR event = 'thing2'
    OR event = 'thing3'
  GROUP BY event, date_trunc('day', created_at)
),
query as (
  SELECT
    event,
    0 as daily_avg,
    0 as weekly_avg,
    avg(count) as monthly_avg
  FROM monthly_counts
  GROUP BY event
  UNION
  SELECT
    event,
    0 as daily_avg,
    avg(count) as weekly_avg,
    0 as monthly_avg
  FROM weekly_counts
  GROUP BY event
  UNION
  SELECT
    event,
    avg(count) as daily_avg,
    0 as weekly_avg,
    0 as monthly_avg
  FROM daily_counts
  GROUP BY event
)
SELECT
  event,
  sum(daily_avg) as daily_avg,
  sum(weekly_avg) as weekly_avg,
  sum(monthly_avg) as monthly_avg
FROM query
GROUP BY event;

Answer 1

我会以这样的方式编写查询：

select event, daily_avg, weekly_avg, monthly_avg
from (
    select event, avg(count) monthly_avg
    from (
        select event, count(*)
        from tracking_stuff
        where event in ('thing1', 'thing2', 'thing3')
        group by event, date_trunc('month', created_at)
    ) s
    group by 1
) monthly
join (
    select event, avg(count) weekly_avg
    from (
        select event, count(*)
        from tracking_stuff
        where event in ('thing1', 'thing2', 'thing3')
        group by event, date_trunc('week', created_at)
    ) s
    group by 1
) weekly using(event)
join (
    select event, avg(count) daily_avg
    from (
        select event, count(*)
        from tracking_stuff
        where event in ('thing1', 'thing2', 'thing3')
        group by event, date_trunc('day', created_at)
    ) s
    group by 1
) daily using(event)
order by 1;

如果where条件消除了很大一部分数据（比如超过一半），则使用cte可以稍微加快查询执行速度：

with the_data as (
    select event, created_at
    from tracking_stuff
    where event in ('thing1', 'thing2', 'thing3')
    )

select event, daily_avg, weekly_avg, monthly_avg
from (
    select event, avg(count) monthly_avg
    from (
        select event, count(*)
        from the_data
        group by event, date_trunc('month', created_at)
    ) s
    group by 1
) monthly
--  etc ... 

出于好奇，我对数据进行了测试：

create table tracking_stuff (event text, created_at timestamp);
insert into tracking_stuff
    select 'thing' || random_int(9), '2016-01-01'::date+ random_int(365)
    from generate_series(1, 1000000);

在每个查询我已经更换了thing同thing1，所以查询消除对行的2/3。

10 个测试的平均执行时间：

Original query          1106 ms
My query without cte    1077 ms
My query with cte        902 ms
Clodoaldo's query       5187 ms

Answer 2

在 9.5+ 中使用 grouping sets

FROM 和 WHERE 子句选择的数据按每个指定的分组集分别分组，像简单的 GROUP BY 子句一样为每个组计算聚合，然后返回结果

select event,
    avg(total) filter (where day is not null) as avg_day,
    avg(total) filter (where week is not null) as avg_week,
    avg(total) filter (where month is not null) as avg_month    
from (
    select
        event,
        date_trunc('day', created_at) as day,
        date_trunc('week', created_at) as week,
        date_trunc('month', created_at) as month,
        count(*) as total
    from tracking_stuff
    where event in ('thing','thing2','thing3')
    group by grouping sets ((event, 2), (event, 3), (event, 4))
) s
group by event