Presto SQL 查询在算术时间间隔内计算满足一组条件的行数

Question

I want to figure out how the count of rows, that satisfy a set of conditions, has changed over time. To this end, I would like to count the number of rows that satified the conditions at a starting date and then perform the same calculation for each day following up until present day. My desired output table would something like the below table (excluding the unnamed column):

|------------+------------+----------------------------|
|            |       date | rows_staisfying_conditions |
|------------+------------+----------------------------|
| start date | 2021-09-01 |                       2367 |
|            | 2021-09-02 |                       2784 |
|            | 2021-09-03 |                       3011 |
|            | 2021-09-04 |                       3601 |
| today      | 2021-09-05 |                       4155 |
|------------+------------+----------------------------|

A naiive approach to generating the above table is to have one CTE for each day, and then join the CTEs (see below code). The problem is that this is verbose and does not scale.

WITH day0 AS (
    SELECT count(*) AS day0
    FROM (
        SELECT DISTINCT account_id
        FROM default_table
        WHERE
            secret_column = 'secret value'
            AND lower(device_os) LIKE '%android%'
            AND from_iso8601_timestamp(timestamp) < from_iso8601_timestamp('2021-09-01T00:00:00.0000000Z') + interval '0' day
    )
),
day1 AS (
    SELECT count(*) AS day1
    FROM (
        SELECT DISTINCT account_id
        FROM default_table
        WHERE
            secret_column = 'secret value'
            AND lower(device_os) LIKE '%android%'
            AND from_iso8601_timestamp(timestamp) < from_iso8601_timestamp('2021-09-01T00:00:00.0000000Z') + interval '1' day
    )
),
⋮
day4 AS (
    SELECT count(*) AS day1
    FROM (
        SELECT DISTINCT account_id
        FROM default_table
        WHERE
            secret_column = 'secret value'
            AND lower(device_os) LIKE '%android%'
            AND from_iso8601_timestamp(timestamp) < from_iso8601_timestamp('2021-09-01T00:00:00.0000000Z') + interval '4' day
    )
),
SELECT *
FROM
    day0
    FULL JOIN day1 ON TRUE
    FULL JOIN day2 ON TRUE
    FULL JOIN day3 ON TRUE
    FULL JOIN day4 ON TRUE

Does anyone have a suggestion for how I could compute the above table in a scalable manner?

Answer 1

You can use sum window finction over grouped by date and ordered sum using next frame :

WITH dataset (date, condition) AS
(
  VALUES
  (date '2021-09-01', true),
  (date '2021-09-01', true),
  (date '2021-09-01', false),
  (date '2021-09-02', true),
  (date '2021-09-03', true)
)

SELECT date, sum(cnt) over (order by date range between unbounded preceding and current row) rows_staisfying_conditions
FROM (
         SELECT date, sum(case when condition then 1 else 0 end) cnt
         FROM dataset
         GROUP BY date
     )

Output:

date	rows_staisfying_conditions
2021-09-01	2
2021-09-02	3
2021-09-03	4