Get count of created entries for each day

问题

Let's say I have a this search query like this:

SELECT COUNT(id), date(created_at)
FROM entries
WHERE date(created_at) >= date(current_date - interval '1 week')
GROUP BY date(created_at)

As you know then for example I get a result back like this:

count | date
  2   |  15.01.2014
  1   |  13.01.2014
  9   |  09.01.2014

But I do not get the days of the week where no entries where created.

How can I get a search result that looks like this, including the days where no entries where created?

count | date
  2   |  15.01.2014
  0   |  14.01.2014
  1   |  13.01.2014
  0   |  12.01.2014
  0   |  11.01.2014
  0   |  10.01.2014
  9   |  09.01.2014

回答1:

SELECT day, COALESCE(ct, 0) AS ct
FROM  (SELECT now()::date - d AS day FROM generate_series (0, 6) d) d  -- 6, not 7
LEFT   JOIN (
   SELECT created_at::date AS day, count(*) AS ct 
   FROM   entries
   WHERE  created_at >= date_trunc('day', now()) - interval '6d'
   GROUP  BY 1
   ) e USING (day);

Use a sargable expression for your WHERE condition, so Postgres can use a plain index on created_at. Far more important for performance than all the rest.
To cover a week (including today), subtract 6 days from the start of "today", not 7.
Assuming that id is defined NOT NULL, count(*) is identical to count(id) here, but slightly faster.
A CTE would be overkill here. It's slower and more verbose.
Aggregate first, join later. That's faster in this case.
now() is the slightly shorter and faster Postgres implementation of the standard SQL CURRENT_TIMESTAMP (which you can use as well).

This should be the shortest and fastest query. Test with EXPLAIN ANALYZE.

Selecting sum and running balance for last 18 months with generate_series
PostgreSQL: running count of rows for a query 'by minute'

回答2:

Try this query:

with a as (select current_date - n as dt from generate_series(0, 6) as t(n)),
     b as (select count(id) cnt, date(created_at) created_at
           from entries
           where date(created_at) >= date(current_date - interval '1 week')
           group by date(created_at))
select coalesce(b.cnt,0), a.dt
from a
left join b on (a.dt = b.created_at)
order by a.dt;

count function will not generate 0 for non-existing rows. So you have to fill the rows for missing dates. With generate_series and simple date arithmetic, you can generate rows for dates of some period (in this case, 1 week). Then you can outer join to generate the final result. coalesce will substitute null to 0.

回答3:

You need to tell SQL to handle NULL. Return 0 if NULL

You can do this by COALESCE

http://www.postgresql.org/docs/devel/static/functions-conditional.html

回答4:

Use generate_series() to create the dates you need and JOIN to this list of dates:

SELECT  COUNT(id), 
    date(gdate)
FROM entries
    RIGHT JOIN generate_series(current_date - interval '1 week', current_date, '1 day') g(gdate) 
    ON date(created_at) = date(gdate) AND date(created_at) >= date(current_date - interval '1 week')
GROUP BY 
    date(gdate)
ORDER BY
    date(gdate) DESC;

来源：https://stackoverflow.com/questions/29362384/get-count-of-created-entries-for-each-day

标签

sql

postgresql

aggregate-functions

generate-series