I have a table with parent and child ids.
create table if not exists stack (
parent int,
child int
)
Each parent can have multiple children and each child can have multiple children again.
insert into stack (parent, child) values
(1,2),
(2,3),
(3,4),
(4,5),
(5,6),
(6,7),
(7,8),
(8,9),
(9,null),
(1,7),
(7,8),
(8,9),
(9,null);
The data looks like this.
|parent|child|
|------|-----|
|1 |2 |
|2 |3 |
|3 |4 |
|4 |5 |
|5 |6 |
|6 |7 |
|7 |8 |
|8 |9 |
|9 |NULL |
|1 |7 |
|7 |8 |
|8 |9 |
|9 |NULL |
I'd like to find all children. I can use a recursive cte with a UNION ALL.
with recursive cte as (
select
child
from
stack
where
stack.parent = 1
union
select
stack.child
from
cte
left join stack on
cte.child = stack.parent
where
cte.child is not null
)
select * from cte;
This gives me the result I'd like to achieve.
|child|
|-----|
|2 |
|7 |
|3 |
|8 |
|4 |
|9 |
|5 |
|NULL |
|6 |
However I'd like to include the depth / level and also the path for each node. I can do this using a different recursive cte.
with recursive cte as (
select
parent,
child,
0 as level,
array[parent,
child] as path
from
stack
where
stack.parent = 1
union all
select
stack.parent,
stack.child,
cte.level + 1,
cte.path || stack.child
from
cte
left join stack on
cte.child = stack.parent
where
cte.child is not null
)
select * from cte;
That gives me this data.
|parent|child|level|path |
|------|-----|-----|--------------------|
|1 |2 |0 |{1,2} |
|1 |7 |0 |{1,7} |
|2 |3 |1 |{1,2,3} |
|7 |8 |1 |{1,7,8} |
|7 |8 |1 |{1,7,8} |
|3 |4 |2 |{1,2,3,4} |
|8 |9 |2 |{1,7,8,9} |
|8 |9 |2 |{1,7,8,9} |
|8 |9 |2 |{1,7,8,9} |
|8 |9 |2 |{1,7,8,9} |
|4 |5 |3 |{1,2,3,4,5} |
|9 | |3 |{1,7,8,9,} |
|9 | |3 |{1,7,8,9,} |
|9 | |3 |{1,7,8,9,} |
|9 | |3 |{1,7,8,9,} |
|9 | |3 |{1,7,8,9,} |
|9 | |3 |{1,7,8,9,} |
|9 | |3 |{1,7,8,9,} |
|9 | |3 |{1,7,8,9,} |
|5 |6 |4 |{1,2,3,4,5,6} |
|6 |7 |5 |{1,2,3,4,5,6,7} |
|7 |8 |6 |{1,2,3,4,5,6,7,8} |
|7 |8 |6 |{1,2,3,4,5,6,7,8} |
|8 |9 |7 |{1,2,3,4,5,6,7,8,9} |
|8 |9 |7 |{1,2,3,4,5,6,7,8,9} |
|8 |9 |7 |{1,2,3,4,5,6,7,8,9} |
|8 |9 |7 |{1,2,3,4,5,6,7,8,9} |
|9 | |8 |{1,2,3,4,5,6,7,8,9,}|
|9 | |8 |{1,2,3,4,5,6,7,8,9,}|
|9 | |8 |{1,2,3,4,5,6,7,8,9,}|
|9 | |8 |{1,2,3,4,5,6,7,8,9,}|
|9 | |8 |{1,2,3,4,5,6,7,8,9,}|
|9 | |8 |{1,2,3,4,5,6,7,8,9,}|
|9 | |8 |{1,2,3,4,5,6,7,8,9,}|
|9 | |8 |{1,2,3,4,5,6,7,8,9,}|
My problem is that I have a lot of duplicate data. I'd like to get the same result as the UNION query but with the level and the path.
I tried something like
where
cte.child is not null
and stack.parent not in (cte.parent)
or
where
cte.child is not null
and not exists (select parent from cte where cte.parent = stack.parent)
but the first does not change anything and the second returns an error.
ERROR: recursive reference to query "cte" must not appear within a subquery
Any ideas? Thank you very much!
Your problem is inappropriate table data. Your table contains the information that 8 is a direct child to 7 twice for instance. I suggest you remove the duplicate data and implement a unique constraint on the pairs.
If you cannot do so for some reason, make the rows distinct in your query:
with recursive
good_stack as (select distinct * from stack)
,cte as
(
select
parent,
child,
0 as level,
array[parent,
child] as path
from good_stack
where good_stack.parent = 1
union all
select
good_stack.parent,
good_stack.child,
cte.level + 1,
cte.path || good_stack.child
from cte
left join good_stack on cte.child = good_stack.parent
where cte.child is not null and good_stack.child is not null
)
select * from cte;
Demo: https://dbfiddle.uk/?rdbms=postgres_13&fiddle=acb1d7a1a1d26c3fd9caf0e7dedc12b2
(You may also make the columns not nullable. The entries 9|null add no information. If the table were lacking these entries, 9 would still be without a child.)
Is it possible to "group" rows within BigQuery/SQL depending on column values? Let's say I want to assign a string/id for all rows between stream_start_init and stream_start and then do the same for the rows between stream_resume and the last stream_ad.
The amount of stream_ad event can differ hence I can't use a RANK() or ROW() to group them be based on those values.
|id, timestamp, event|
|1 | 1231231 | first_visit|
|2 | 1231232 | login|
|3 | 1231233 | page_view|
|4 | 1231234 | page_view|
|5 | 1231235 | stream_start_init|
|6 | 1231236 | stream_ad|
|7 | 1231237 | stream_ad|
|8 | 1231238 | stream_ad|
|9 | 1231239 | stream_start|
|6 | 1231216 | stream_resume|
|6 | 1231236 | stream_ad|
|7 | 1231217 | stream_ad|
|8 | 1231258 | stream_ad|
|10| 1231240 | page_view|
How I wish the table to be
|id, timestamp, event, group_id|
|1 | 1231231 | first_visit, null|
|2 | 1231232 | login, null|
|3 | 1231233 | page_view, null|
|4 | 1231234 | page_view, null|
|5 | 1231235 | stream_start_init, group_1|
|6 | 1231236 | stream_ad, group_1|
|7 | 1231237 | stream_ad, group_1|
|8 | 1231238 | stream_ad, group_1|
|9 | 1231239 | stream_start, group_1|
|6 | 1231216 | stream_resume, group_2|
|6 | 1231236 | stream_ad, group_2|
|7 | 1231217 | stream_ad, group_2|
|8 | 1231258 | stream_ad, group_2|
|10| 1231240 | page_view, null|
I wouldn't assign a string. I would assign a number. This appears to be a cumulative sum. I think a sum of the number of "stream_start_init" and "stream_resume" does what you want:
select t.*,
countif(event in ('stream_start_init', 'stream_resume')) over (order by timestamp) as group_id
from t;
Note that this produces 0 for the first group -- which seems like a good thing. You can convert that to a NULL using NULLIF().
If you really want strings, you can use CONCAT().
Below is for BigQuery Standard SQL
#standardSQL
SELECT *,
IF(event IN ('stream_start_init', 'stream_start', 'stream_resume', 'stream_ad'),
COUNTIF(event IN ('stream_start_init', 'stream_resume')) OVER(ORDER BY timestamp),
NULL
) AS group_id
FROM `project.dataset.table`
I'm currently trying to create a grid showing worked hours of employees.
Here's what my data look like (simplified) :
|ID |Client |Task |Hours |Date |
------------------------------------------
|1 |ABC |A |3 |09/06/2014|
|2 |ABC |A |5 |09/06/2014|
|3 |DEF |B |8 |10/06/2014|
|4 |DEF |C |8 |11/06/2014|
|5 |ABC |A |8 |12/06/2014|
And here's what the output must look like:
|Client |Task |Sun |Mon |Tue |Wed |Thu |Fri |Sat |
--------------------------------------------------
|ABC |A | |3 | | |8 | | |
|ABC |A | |5 | | | | | |
|DEF |B | | |8 | | | | |
|DEF |C | | | |8 | | | |
My problem is really close to this one. However there's a major diffrence: it's possible in my case to have multiple values for the same combination of Client-Task-Date.
As shown in the desired output, employees will sometime seperate their work hours even if they worked for the same client and on the same task and i can't use aggregate since all the data shown in the grid will be interactive to the end user.
Is there a way to obtain such output using pivot or any other SQL mechanics such as CASE WHEN ?
WITH t AS (
SELECT
Client,
Task,
Hours,
ROW_NUMBER() OVER(PARTITION BY Client,Task,Date ORDER BY Date) rn,
DATEPART(dw,date) DayOfWeek
FROM MyTable
)
SELECT Client, Task, [1] Sun, [2] Mon, [3] Tues, [4] Wed, [5] Thu, [6] Fri, [7] Sat
FROM t
PIVOT(SUM(Hours) FOR DayOfWeek IN ([1],[2],[3],[4],[5],[6],[7])) p