Creating a self-reflexive cross join - mysql

Question

Hi I'm looking to see if there is alternative syntax for the following self-reflexive cross-join. The objective is a sort of row-filler for a table - dates should have entries for every cdn. I am using MySQL
select
d.labelDate,
n.cdn,
networks.sites
from (
select
distinct labelDate
from
cdn_trend
) as d
cross join (
select
distinct cdn
from cdn_trend
) as n
left join cdn_trend as networks
on networks.labelDate = d.labelDate
and networks.cdn = n.cdn
order by
labelDate,
cdn
I've tried recasting the cross-join using simple aliases but that gives me column errors in the join. Is it possible to do so or should I consider using views instead?
As a cross join should simply return the Cartesian product of two tables it should be the same as simply selecting both without a join. However, the following raises an "unknown column d.labelDate in on clause" exception
select distinct d.labelDate, n.cdn, networks.sites
from
cdn_trend as d,
cdn_trend as n
left join cdn_trend as networks ON
(n.labelDate = networks.labelDate
and d.cdn = networks.cdn)
order by labelDate, cdn
Error Code: 1054. Unknown column 'd.cdn' in 'on clause'
Because the length of dand n are relatively small the size of the query is fast enough.

Answer 1

I think you were close in your original intent... For every date, you want the results of every network node status. If you list multiple tables in the WHERE clause without a join condition, it by default will create a Cartesian... From that, join to your detail table...
select
d.labelDate,
n.cdn,
networks.sites
from
( select d.LabelDate, n.cdn
from
( select distinct labelDate
from cdn_trend ) as d,
( select distinct cdn
from cdn_trend ) as n ) as CrossResults
LEFT JOIN cdn_trend as networks
on CrossResults.labelDate = networks.labelDate
and CrossResults.cdn = networks.cdn
order by
networks.labelDate,
networks.cdn

Answer 2

Reading the comments and the extra info you need a pivot with y - lableDate and x - cdn and values - sites assuming the values for cdn are (a,b,c) and that sites is a number you can try this:
SELECT
labelDate,
SUM(IF(cdn = 'a',sites,0)) as cdn_a,
SUM(IF(cdn = 'b',sites,0)) as cdn_b,
SUM(IF(cdn = 'c',sites,0)) as cdn_c
FROM
cdn_trend
GROUP BY
labelDate
And the output should be something like this (i used the sample data from you) :
labelDate cdn_a cdn_b cdn_c
2013-04 NULL 5 4
2013-05 6 NULL NULL
....

Answer 3

After some playing around this is the best I could come up with. It seems that parametrising the table name would be possible but would involve yet another layer of statement generation that I fortunately don't need for this project.
-- --------------------------------------------------------------------------------
-- Routine DDL
-- Note: comments before and after the routine body will not be stored by the server
-- --------------------------------------------------------------------------------
DELIMITER $$
CREATE DEFINER="root"#"localhost" PROCEDURE "cdn_pivot"(
IN slice varchar(64),
IN start date,
IN stop date)
BEGIN
SET ##group_concat_max_len = 32000;
SET #sql = NULL;
SELECT
GROUP_CONCAT(DISTINCT
CONCAT(
' sum(IF(cdn = ''',
cdn,
''', sites,NULL)) "'
,cdn, '"'
)
) INTO #sql
FROM cdns ORDER BY sites;
SET #stmt = CONCAT('SELECT labelDate, ',
#sql,
' from cdns
WHERE slice = ''',
slice,
''' AND ( labelDate between''',
start,
''' AND ''',
stop,
'''
)
GROUP BY labelDate');
prepare stmt from #stmt;
execute stmt;
deallocate prepare stmt;
SET ##group_concat_max_len = 1024;
END
This can then simply be called e.,g.
call cdn_pivot('Top100', '2013-01-01', 2013-02-01')
Given the problems associated with testing this code and keeping it with any client-side code it's very tempting to generate the dynamic part of the head on the client and, at least for this kind of use case, the performance penalty of the additional query shouldn't be too high. The key thing is obviously understanding how to generate the columns dynamically.