使用date_trunc的滚动平均值

Question

I'm using date_trunc to aggregate over a timestamp field. I'm trying to count the number of unique values in a given field, over a month. If I choose month as my resolution :

SELECT date_trunc('month', timestamp), COUNT(DISTINCT(foo)) FROM ...

then all resulting entries are at the beginning of each month, and so “2017-01-01" would be counting the unique entries where the timestamp field is anywhere in January.

Is there a way to specify an offset such that I might have an entry at 2017-01-15, comprising entries up to 2017-02-15 ?

Thanks.

Answer 1

SELECT 
    date_trunc('month', timestamp + interval '15 day'),
    COUNT(DISTINCT(foo)) 
FROM ...
group by 1

Answer 2

This kind of queries are never very efficient. If speed is an issue, perhaps it would be better to do this outside of the database through looping in the application language

In the subquery, for every record, we generate a series of dates which will include it in their aggregations, then we aggregate by the generated dates.

SELECT
    mydate
  , COUNT(DISTINCT foo)
FROM (SELECT GENERATE_SERIES( DATE("timestamp") - INTERVAL '30 DAYS'
                            , DATE("timestamp"), '1 DAY') mydate
           , foo
      FROM mytable) expanded
GROUP BY 1

Also, please try NOT to use timestamp as a column name as its a postgresql data type.