Snowflake-使用timediff更新相关子查询

Question

I am running this query on Snowflake Database:

UPDATE "click" c
SET "Registration_score" =
(SELECT COUNT(*) FROM "trackingpoint" t
WHERE 1=1
AND c."CookieID" = t."CookieID"
AND t."page" ilike '%Registration complete'
AND TIMEDIFF(minute,c."Timestamp",t."Timestamp") < 4320
AND TIMEDIFF(second,c."Timestamp",t."Timestamp") > 0);

The Database returns Unsupported subquery type cannot be evaluated . However, if I run it without the last two conditions (with TIMEDIFF), it works without problem. I confirmed that the actual TIMEDIFF statements are alright with these queries:

select count(*) from "trackingpoint"
where TIMEDIFF(minute, '2018-01-01', "Timestamp") > 604233;
select count(*) from "click"
where TIMEDIFF(minute, '2018-01-01', "Timestamp") > 604233;

and these work without problem. I don't see a reason why the TIMEDIFF condition shoud prevent the database from returning the result. Any idea what should I alter to make it work?

Answer 1

so using the following setup

create table click (id number, 
   timestamp timestamp_ntz,
   cookieid number,
   Registration_score number);
create table trackingpoint(id number, 
   timestamp timestamp_ntz, 
   cookieid number, 
   page text );


insert into click values (1,'2018-03-20', 101, 0),
    (2,'2019-03-20', 102, 0);
insert into trackingpoint values (1,'2018-03-20 00:00:10', 101, 'user reg comp'),
    (2,'2018-03-20 00:00:11', 102, 'user reg comp'),
    (3,'2018-03-20 00:00:13', 102, 'pet reg comp'),
    (4,'2018-03-20 00:00:15', 102, 'happy dance');

you can see we get the rows we expect

select c.*, t.*
from click c
join trackingpoint t 
    on c.cookieid = t.cookieid ;

now there are two ways to get your count, the first as you have it, which is good if your counting only one thing, as all the rules are join filtering:

select c.id,
  count(1) as new_score
from click c
join trackingpoint t 
    on c.cookieid = t.cookieid
    and t.page ilike '%reg comp'
    and TIMEDIFF(minute, c.timestamp, t.timestamp) < 4320
group by 1;

or you can (in snowflake syntax) move the count to the aggregate/select side,and thus get more than one answer if that's what you need (this is the place I find myself more, thus why I present it):

select c.id,
    sum(iff(t.page ilike '%reg comp' AND TIMEDIFF(minute, c.timestamp, t.timestamp) < 4320, 1, 0)) as new_score
from click c
join trackingpoint t 
    on c.cookieid = t.cookieid
group by 1;

thus plugging this into the UPDATE pattern (see last example in the doc's) https://docs.snowflake.net/manuals/sql-reference/sql/update.html

you can move to a single subselect instead of a corolated subquery which snowflake doesn't support, which is the error message you are getting.

UPDATE click c
SET Registration_score = s.new_score
from (
    select ic.id,
        count(*) as new_score
    from click ic
    join trackingpoint it 
        on ic.cookieid = it.cookieid
        and it.page ilike '%reg comp'
        and TIMEDIFF(minute, ic.timestamp, it.timestamp) < 4320
    group by 1) as s
WHERE c.id = s.id; 

The reason add the TIMEDIFF turns your query into a correlated sub-query, is each row of the UPDATE, now relates to the sub-query results, the correlation. The work around is to make "big but simpler" sub-query and join to that.