如何改进这种联合以确保一致的结果？

Question

我一直试图找到一种可靠且经济的方法来计算 7 天的行数，到目前为止，这是我能想到的最好方法：

架构

CREATE TABLE IF NOT EXISTS "data" (
    "id" SERIAL,
    "hash" CHARACTER VARYING(255) NOT NULL,
    "source" CHARACTER VARYING(255) NOT NULL,
    "isFiltered" BOOLEAN NOT NULL,
    "campaignId" INTEGER NOT NULL,
    "data" JSON NOT NULL,
    "meta" JSON NOT NULL,
    "modifiedReason" CHARACTER VARYING(255) NULL DEFAULT NULL,
    "createdAt" TIMESTAMP WITH TIME ZONE NOT NULL,
    "updatedAt" TIMESTAMP WITH TIME ZONE NOT NULL,
    PRIMARY KEY ("id")
);

CREATE TABLE IF NOT EXISTS "campaign" (
    "id" SERIAL,
    "userId" INTEGER NOT NULL,
    "userCompanyId" INTEGER NOT NULL,
    "type" CHARACTER VARYING(255) NULL DEFAULT NULL,
    "title" CHARACTER VARYING(255) NULL DEFAULT NULL,
    "description" CHARACTER VARYING(255) NULL DEFAULT NULL,
    "sources" CHARACTER VARYING(255)[] NULL DEFAULT NULL,
    "configuration" JSON NOT NULL,
    "active" BOOLEAN NULL DEFAULT true,
    "excludedUserNames" CHARACTER VARYING(255)[] NULL DEFAULT NULL,
    "limit" INTEGER NULL DEFAULT 0,
    "startAt" TIMESTAMP WITH TIME ZONE NOT NULL,
    "endAt" TIMESTAMP WITH TIME ZONE NOT NULL,
    "removedAt" TIMESTAMP WITH TIME ZONE NULL DEFAULT NULL,
    "createdAt" TIMESTAMP WITH TIME ZONE NOT NULL,
    "updatedAt" TIMESTAMP WITH TIME ZONE NOT NULL,
    PRIMARY KEY ("id")
);

要测试的数据

INSERT INTO "campaign" ("id", "userId", "userCompanyId", "type", "title", "description", "sources", "configuration", "active", "excludedUserNames", "limit", "startAt", "endAt", "removedAt", "createdAt", "updatedAt") VALUES
    (1, 1, 1, E'test', E'Test', E'Test', E'{a}', E'{"query":{"accounts":[],"hashtags":["GavinTest1234","XYZ"]}}', E'true', NULL, 0, E'2016-01-25 15:06:00+00', E'2016-01-27 23:59:59+00', NULL, E'2016-01-25 15:06:27.474+00', E'2016-01-26 16:48:19.693+00');
INSERT INTO "data" ("id", "hash", "source", "isFiltered", "campaignId", "data", "meta", "modifiedReason", "createdAt", "updatedAt") VALUES
    (1, E'dHdpdHRlci02OTE5MTQ3ODcwNjA1ODQ0NDg=', E'a', E'false', 1, E'{}', E'{"profile":{"url":"xxx","image":"xxx","username":"xxx","name":"xxx","createdAt":"2015-10-05T10:30:11.000Z"},"posts":{"total":32,"perDay":0},"friends":0,"favourites":0,"createdAt":"2016-01-26T09:25:15.000Z","matchedOn":{"hashtags":["GavinTest123"],"accounts":[]}}', NULL, E'2016-01-26 09:25:15.539+00', E'2016-01-26 09:25:15.539+00'),
    (2, E'dHdpdHRlci02OTE5MjAwNDAwNTcyNzAyNzI=', E'a', E'false', 1, E'{}', E'{"profile":{"url":"xxx","image":"xxx","username":"xxx","name":"xxx","createdAt":"2015-10-05T10:30:11.000Z"},"posts":{"total":34,"perDay":0},"friends":0,"favourites":0,"createdAt":"2016-01-26T09:46:07.000Z","matchedOn":{"hashtags":["GavinTest123"],"accounts":[]}}', NULL, E'2016-01-26 09:46:07.942+00', E'2016-01-26 09:46:07.942+00'),
    (3, E'dHdpdHRlci02OTE5NjI4NjM5OTc1NTg3ODQ=', E'a', E'false', 1, E'{}', E'{"profile":{"url":"xxx","image":"xxx","username":"xxx","name":"xxx","createdAt":"2015-10-05T10:30:11.000Z"},"posts":{"total":36,"perDay":0},"friends":0,"favourites":0,"createdAt":"2016-01-26T12:36:17.000Z","matchedOn":{"hashtags":["GavinTest1234"],"accounts":[]}}', NULL, E'2016-01-26 12:36:17.724+00', E'2016-01-26 12:36:17.724+00');

查询

SELECT   q."date", 
         q."hashtag", 
         Max(q."count") AS "count" 
FROM     ( 
     SELECT     To_char(d::date, 'DD/MM/YYYY') AS "date", 
                0                              AS "count", 
                c_h.hashtag::text              AS "hashtag" 
     FROM       generate_series('2016-01-20', '2016-01-26', '1 day'::interval) d 
     INNER JOIN "campaign" c 
     ON         ( 
                          c."id" = 1) 
     INNER JOIN Json_array_elements(c.configuration->'query'->'hashtags') c_h(hashtag) 
     ON         true 
     UNION ALL 
     SELECT     To_char(cdi."createdAt"::date, 'DD/MM/YYYY') AS "date", 
                Count(cdi_h.hashtag::text)                   AS "count", 
                cdi_h.hashtag::text                          AS "hashtag" 
     FROM       "data" cdi 
     INNER JOIN "campaign" c 
     ON         ( 
                          c."id" = cdi."campaignId") 
     INNER JOIN json_array_elements(c.configuration->'query'->'hashtags') c_h(hashtag) 
     ON         true 
     INNER JOIN json_array_elements(cdi.meta->'matchedOn'->'hashtags') cdi_h(hashtag) 
     ON         ( 
                          c_h.hashtag::text = cdi_h.hashtag::text) 
     WHERE      c."id" = 1 
     AND        ( 
                          cdi."createdAt"::date >= '2016-01-20' 
               AND        cdi."createdAt"::date <= '2016-01-26') 
     GROUP BY   to_char(cdi."createdAt"::   date, 'DD/MM/YYYY'), 
               cdi_h.hashtag::text 
     ORDER BY  "date" ASC, 
               "hashtag" ASC, 
               "count" ASC ) q 
GROUP BY q."date", 
         q."hashtag";

结果

因此，正如您从查询中看到的那样，我首先返回一个结果集，其中包含每个日期和主题标签的行。

然后，我想查询我的数据表并返回每天有多少匹配项的计数，覆盖默认值。

查询有效，而且成本并不太高，但是它太丑了，我相信有更好的方法让我看不见。

除了这个，还有没有人有更好的建议，或者我的替代方法是只返回我们在 7 天内拥有的结果（分组和计数）并在服务器端做一些处理。

- 编辑 -

抱歉，SQL Fiddle 只有 9.3 并且在我尝试为大家创建一个 fiddle 时崩溃了。

Answer 1

您的查询可以在多个方面得到简化：

SELECT to_char(day, 'DD/MM/YYYY') AS date
     , hashtag
     , count(d.*)::int AS count
FROM  (
       campaign c
CROSS  JOIN json_array_elements_text(c.configuration#>'{query,hashtags}') ch(hashtag)
CROSS  JOIN (SELECT g::date AS day
             FROM generate_series(timestamp '2016-01-20', '2016-01-26', interval '1 day') g) day 
      )
NATURAL LEFT JOIN (
   SELECT "createdAt"::date AS day, dh.hashtag
   FROM   data, json_array_elements_text(meta#>'{matchedOn,hashtags}') dh(hashtag)
   WHERE  "campaignId" = 1
   AND    "createdAt" >= '2016-01-20'
   AND    "createdAt" <  '2016-01-27'
   ) d
WHERE  c.id = 1
GROUP  BY day, hashtag
ORDER  BY day, hashtag, count;

由于多种原因，这应该更快。尤其是它可以使用多列索引data("campaignId", "createdAt")- 除非您拥有它，否则您应该创建它。

LEFT JOIN是您需要的核心功能。然后对列进行计数，仅非空值计数...

顺便说一句，这是 a 有用的极少数情况之一NATURAL JOIN。不过你不需要它。这只是一个小语法快捷方式。

而是使用timestamp输入generate_series()：

• 在 PostgreSQL 中生成两个日期之间的时间序列