I've got a scenario where I need to do a join across three tables.
table #1 is a list of users
table #2 contains users who have trait A
table #3 contains users who have trait B
If I want to find all the users who have trait A or trait B (in one simple sql) I think I'm stuck.
If I do a regular join, the people who don't have trait A won't show up in the result set to see if they have trait B (and vice versa).
But if I do an outer join from table 1 to tables 2 and 3, I get all the rows in table 1 regardless of the rest of my where clause specifying a requirement against tables 2 or 3.
Before you come up with multiple sqls and temp tables and whatnot, this program is far more complex, this is just the simple case. It dynamically creates the sql based on lots of external factors, so I'm trying to make it work in one sql.
I expect there are combinations of in or exists that will work, but I was hoping for some thing simple.
But basically the outer join will always yield all results from table 1, yes?
SELECT *
FROM table1
LEFT OUTER
JOIN table2
ON ...
LEFT OUTER
JOIN table3
ON ...
WHERE NOT (table2.pk IS NULL AND table3.pk IS NULL)
or if you want to be sneaky:
WHERE COALESCE(table2.pk, table3.pk) IS NOT NULL
but for you case, i simply suggest:
SELECT *
FROM table1
WHERE table1.pk IN (SELECT fk FROM table2)
OR table1.pk IN (SELECT fk FROM table3)
or the possibly more efficient:
SELECT *
FROM table1
WHERE table1.pk IN (SELECT fk FROM table2 UNION (SELECT fk FROM table3)
If you really just want the list of users that have one trait or the other, then:
SELECT userid FROM users
WHERE userid IN (SELECT userid FROM trait_a UNION SELECT userid FROM trait_b)
Regarding outerjoin specifically, longneck's answer looks like what I was in the midst of writing.
I think you could do a UNION here.
May I suggest:
SELECT columnList FROM Table1 WHERE UserID IN (SELECT UserID FROM Table2)
UNION
SELECT columnList FROM Table1 WHERE UserID IN (SELECT UserID FROM Table3)
Would something like this work? Keep in mind depending on the size of the tables left outer joins can be very expensive with regards to performance.
Select *
from table1
where userid in (Select t.userid
From table1 t
left outer join table2 t2 on t1.userid=t2.userid and t2.AttributeA is not null
left outer join table3 t3 on t1.userid=t3.userid and t3.AttributeB is not null
group by t.userid)
If all you want is the ids of the users then
SELECT UserId From Table2
UNION
SELECT UserId From Table3
is totally sufficient.
If you want some more infos from Table1 on these users, you can join the upper SQL to Table 1:
SELECT <list of columns from Table1>
FROM Table1 Join (
SELECT UserId From Table2
UNION
SELECT UserId From Table3) User on Table1.UserID = Users.UserID
Related
Suppose I have three tables in PostgreSQL:
table1 - id1, a_id, updated_by_id
table2 - id2, a_id, updated_by_id
Users - id, display_name
Suppose I am using the using the following query:
select count(t1.id1) from table1 t1
left join table2 t2 on (t1.a_id=t2.a_id)
full outer join users u1 t1.updated_by_id=u1.id)
full outer join users u2 t2.updated_by_id=u2.id)
where u1.id=100;
I get 50 as count.
Whereas with:
select count(t1.id1) from table1 t1
left join table2 t2 on (t1.a_id=t2.a_id)
full outer join users u1 t1.updated_by_id=u1.id)
full outer join users u2 t2.updated_by_id=u2.id)
where u2.id=100;
I get only 25 as count.
What is my mistake in the second query? What can I do to get the same count?
My requirement is that there is a single user table, referenced by multiple tables. I want to take the complete list of users and get the count of ids from different tables.
But the table on which I have joined alone returns the proper count but rest of them don't return the proper count. Can anybody suggest a way to modify my second query to get the proper count?
To simplify your logic, aggregate first, join later.
Guessing missing details, this query would give you the exact count, how many times each user was referenced in table1 and table2 respectively for all users:
SELECT *
FROM users u
LEFT JOIN (
SELECT updated_by_id AS id, count(*) AS t1_ct
FROM table1
GROUP BY 1
) t1 USING (id)
LEFT JOIN (
SELECT updated_by_id AS id, count(*) AS t2_ct
FROM table2
GROUP BY 1
) t2 USING (id);
In particular, avoid multiple 1-n relationships multiplying each other when joined together:
Two SQL LEFT JOINS produce incorrect result
To retrieve a single or few users only, LATERAL joins will be faster (Postgres 9.3+):
SELECT *
FROM users u
LEFT JOIN LATERAL (
SELECT count(*) AS t1_ct
FROM table1
WHERE updated_by_id = u.id
) ON true
LEFT JOIN LATERAL (
SELECT count(*) AS t2_ct
FROM table2
WHERE updated_by_id = u.id
) ON true
WHERE u.id = 100;
What is the difference between LATERAL JOIN and a subquery in PostgreSQL?
Explain perceived difference
The particular mismatch you report is due to the specifics of a FULL OUTER JOIN:
First, an inner join is performed. Then, for each row in T1 that does
not satisfy the join condition with any row in T2, a joined row is
added with null values in columns of T2. Also, for each row of T2 that
does not satisfy the join condition with any row in T1, a joined row
with null values in the columns of T1 is added.
So you get NULL values appended on the respective other side for missing matches. count() does not count NULL values. So you can get a different result depending on whether you filter on u1.id=100 or u2.id=100.
This is just to explain, you don't need a FULL JOIN here. Use the presented alternatives instead.
I have 3 tables of accounts that all contain the same fields. Table1 contains all accounts while Table2 and Table3 contain subsets of the accounts. I'm trying to select records in Table1 that do no exist in Table2 or Table3.
Let's say the table layout is like this and is the same for all 3 tables:
|AcctNum|Name|State|
I know how to do this if it was just Table1 and Table2, using a left join and Is Null, but the 3rd table is throwing me. Is this possible to do in one query? Can you combine left joins? I should point out I'm using Access 2010.
Yes you can combine left joins and with the odd syntax Access uses the query should look like this:
SELECT T1.AcctNum
FROM (Table1 AS T1 LEFT JOIN Table2 AS T2 ON T1.AcctNum = T2.AcctNum)
LEFT JOIN Table3 AS T3 ON T1.AcctNum = T3.AcctNum
WHERE (((T2.AcctNum) Is Null) AND ((T3.AcctNum) Is Null));
You can use Access to create a view called TableCombined that is a union of both Table2 and Table3.
At that point, you can use your left join and Is Null query and join TableCombined to Table1.
Hope this helps!
You can also do a NOT EXISTS statement which makes sense logically for what you are trying to achieve.
For example:
SELECT ACCTNUM
FROM TABLE1
WHERE NOT EXISTS (SELECT TABLE2.ACCTNUM FROM TABLE2 INNER JOIN TABLE3 WHERE TABLE2.ACCTNUM IS NULL AND TABLE3.ACCTNUM IS NULL)
I have two tables with binding primary key in database and I desire to find a disjoint set between them. For example,
Table1 has columns (ID, Name) and sample data: (1 ,John), (2, Peter), (3, Mary)
Table2 has columns (ID, Address) and sample data: (1, address2), (2, address2)
So how do I create a SQL query so I can fetch the row with ID from table1 that is not in table2. In this case, (3, Mary) should be returned?
PS: The ID is the primary key for those two tables.
Try this
SELECT ID, Name
FROM Table1
WHERE ID NOT IN (SELECT ID FROM Table2)
Use LEFT JOIN
SELECT a.*
FROM table1 a
LEFT JOIN table2 b
on a.ID = b.ID
WHERE b.id IS NULL
There are basically 3 approaches to that: not exists, not in and left join / is null.
LEFT JOIN with IS NULL
SELECT l.*
FROM t_left l
LEFT JOIN
t_right r
ON r.value = l.value
WHERE r.value IS NULL
NOT IN
SELECT l.*
FROM t_left l
WHERE l.value NOT IN
(
SELECT value
FROM t_right r
)
NOT EXISTS
SELECT l.*
FROM t_left l
WHERE NOT EXISTS
(
SELECT NULL
FROM t_right r
WHERE r.value = l.value
)
Which one is better? The answer to this question might be better to be broken down to major specific RDBMS vendors. Generally speaking, one should avoid using select ... where ... in (select...) when the magnitude of number of records in the sub-query is unknown. Some vendors might limit the size. Oracle, for example, has a limit of 1,000. Best thing to do is to try all three and show the execution plan.
Specifically form PostgreSQL, execution plan of NOT EXISTS and LEFT JOIN / IS NULL are the same. I personally prefer the NOT EXISTS option because it shows better the intent. After all the semantic is that you want to find records in A that its pk do not exist in B.
Old but still gold, specific to PostgreSQL though: https://explainextended.com/2009/09/16/not-in-vs-not-exists-vs-left-join-is-null-postgresql/
Fast Alternative
I ran some tests (on postgres 9.5) using two tables with ~2M rows each. This query below performed at least 5* better than the other queries proposed:
-- Count
SELECT count(*) FROM (
(SELECT id FROM table1) EXCEPT (SELECT id FROM table2)
) t1_not_in_t2;
-- Get full row
SELECT table1.* FROM (
(SELECT id FROM table1) EXCEPT (SELECT id FROM table2)
) t1_not_in_t2 JOIN table1 ON t1_not_in_t2.id=table1.id;
Keeping in mind the points made in #John Woo's comment/link above, this is how I typically would handle it:
SELECT t1.ID, t1.Name
FROM Table1 t1
WHERE NOT EXISTS (
SELECT TOP 1 NULL
FROM Table2 t2
WHERE t1.ID = t2.ID
)
SELECT COUNT(ID) FROM tblA a
WHERE a.ID NOT IN (SELECT b.ID FROM tblB b) --For count
SELECT ID FROM tblA a
WHERE a.ID NOT IN (SELECT b.ID FROM tblB b) --For results
If I run the two statements in batch will they return one table to two to my sqlcommand object with the data merged. What I am trying to do is optimize a search by searching twice, the first time on one set of data and then a second on another. They have the same fields and I’d like to have all the records from both tables show and be added to each other. I need this so that I can sort the data between both sets of data but short of writing a stored procedure I can’t think of a way of doing this.
Eg. Table 1 has columns A and B, Table 2 has these same columns but different data source. I then wan to merge them so that if a only exists in one column it is added to the result set and if both exist it eh tables the column B will be summed between the two.
Please note that this is not the same as a full outer join operation as that does not merge the data.
[EDIT]
Here's what the code looks like:
Select * From
(Select ID,COUNT(*) AS Count From [Table1]) as T1
full outer join
(Select ID,COUNT(*) AS Count From [Table2]) as T2
on t1.ID = T2.ID
Perhaps you're looking for UNION?
IE:
SELECT A, B FROM Table1
UNION
SELECT A, B FROM Table2
Possibly:
select table1.a, table1.b
from table1
where table1.a not in (select a from table2)
union all
select table1.a, table1.b+table2.b as b
from table1
inner join table2 on table1.a = table2.a
edit: perhaps you would benefit from unioning the tables before counting. e.g.
select id, count() as count from
(select id from table1
union all
select id from table2)
I'm not sure if I understand completely but you seem to be asking about a UNION
SELECT A,B
FROM tableX
UNION ALL
SELECT A,B
FROM tableY
To do it, you would go:
SELECT * INTO TABLE3 FROM TABLE1
UNION
SELECT * FROM TABLE2
Provided both tables have the same columns
I think what you are looking for is this, but I am not sure I am understanding your language correctly.
select id, sum(count) as count
from (
select id, count() as count
from table1
union all
select id, count() as count
from table2
) a
group by id
Here is my situation:
Table one contains a set of data that uses an id for an unique identifier. This table has a one to many relationship with about 6 other tables such that.
Given Table 1 with Id of 001:
Table 2 might have 3 rows with foreign key: 001
Table 3 might have 12 rows with foreign key: 001
Table 4 might have 0 rows with foreign key: 001
Table 5 might have 28 rows with foreign key: 001
I need to write a report that lists all of the rows from Table 1 for a specified time frame followed by all of the data contained in the handful of tables that reference it.
My current approach in pseudo code would look like this:
select * from table 1
foreach(result) {
print result;
select * from table 2 where id = result.id;
foreach(result2) {
print result2;
}
select * from table 3 where id = result.id
foreach(result3) {
print result3;
}
//continued for each table
}
This means that the single report can run in the neighbor hood of 1000 queries. I know this is excessive however my sql-fu is a little weak and I could use some help.
LEFT OUTER JOIN Tables2-N on Table1
SELECT Table1.*, Table2.*, Table3.*, Table4.*, Table5.*
FROM Table1
LEFT OUTER JOIN Table2 ON Table1.ID = Table2.ID
LEFT OUTER JOIN Table3 ON Table1.ID = Table3.ID
LEFT OUTER JOIN Table4 ON Table1.ID = Table4.ID
LEFT OUTER JOIN Table5 ON Table1.ID = Table5.ID
WHERE (CRITERIA)
Join doesn't do it for me. I hate having to de-tangle the data on the client side. All those nulls from left-joining.
Here's a set-based solution that doesn't use Joins.
INSERT INTO #LocalCollection (theKey)
SELECT id
FROM Table1
WHERE ...
SELECT * FROM Table1 WHERE id in (SELECT theKey FROM #LocalCollection)
SELECT * FROM Table2 WHERE id in (SELECT theKey FROM #LocalCollection)
SELECT * FROM Table3 WHERE id in (SELECT theKey FROM #LocalCollection)
SELECT * FROM Table4 WHERE id in (SELECT theKey FROM #LocalCollection)
SELECT * FROM Table5 WHERE id in (SELECT theKey FROM #LocalCollection)
Ah! Procedural! My SQL would look like this, if you needed to order the results from the other tables after the results from the first table.
Insert Into #rows Select id from Table1 where date between '12/30' and '12/31'
Select * from Table1 t join #rows r on t.id = r.id
Select * from Table2 t join #rows r on t.id = r.id
--etc
If you wanted to group the results by the initial ID, use a Left Outer Join, as mentioned previously.
You may be best off to use a reporting tool like Crystal or Jasper, or even XSL-FO if you are feeling bold. They have things built in to handle specifically this. This is not something the would work well in raw SQL.
If the format of all of the rows (the headers as well as all of the details) is the same, it would also be pretty easy to do it as a stored procedure.
What I would do: Do it as a join, so you will have the header data on every row, then use a reporting tool to do the grouping.
SELECT * FROM table1 t1
INNER JOIN table2 t2 ON t1.id = t2.resultid -- this could be a left join if the table is not guaranteed to have entries for t1.id
INNER JOIN table2 t3 ON t1.id = t3.resultid -- etc
OR if the data is all in the same format you could do.
SELECT cola,colb FROM table1 WHERE id = #id
UNION ALL
SELECT cola,colb FROM table2 WHERE resultid = #id
UNION ALL
SELECT cola,colb FROM table3 WHERE resultid = #id
It really depends on the format you require the data in for output to the report.
If you can give a sample of how you would like the output I could probably help more.
Join all of the tables together.
select * from table_1 left join table_2 using(id) left join table_3 using(id);
Then, you'll want to roll up the columns in code to format your report as you see fit.
What I would do is open up cursors on the following queries:
SELECT * from table1 order by id
SELECT * from table1 r, table2 t where t.table1_id = r.id order by r.id
SELECT * from table1 r, table3 t where t.table1_id = r.id order by r.id
And then I would walk those cursors in parallel, printing your results. You can do this because all appear in the same order. (Note that I would suggest that while the primary ID for table1 might be named id, it won't have that name in the other tables.)
Do all the tables have the same format? If not, then if you have to have a report that can display the n different types of rows. If you are only interested in the same columns then it is easier.
Most databases have some form of dynamic SQL. In that case you can do the following:
create temporary table from
select * from table1 where rows within time frame
x integer
sql varchar(something)
x = 1
while x <= numresults {
sql = 'SELECT * from table' + CAST(X as varchar) + ' where id in (select id from temporary table'
execute sql
x = x + 1
}
But I mean basically here you are running one query on your main table to get the rows that you need, then running one query for each sub table to get rows that match your main table.
If the report requires the same 2 or 3 columns for each table you could change the select * from tablex to be an insert into and get a single result set at the end...