Improve query to join other tables depending on value without union - mysql

I have a table containing the following data:
+-----------------+--------------+----------------------+------------+--------------------+-----------+------+---------+---------+-----+------------------------+---------------------+
| notification_id | from_user_id | from_user_auth_level | to_user_id | to_user_auth_level | status_id | type | subject | message | url | timestamp_inserted_utc | timestamp_read_utc |
+-----------------+--------------+----------------------+------------+--------------------+-----------+------+---------+---------+-----+------------------------+---------------------+
| 1 | NULL | NULL | 1 | 1 | 0 | 2 | test | test | url | 2010-10-10 00:00:00 | 2011-10-10 00:00:00 |
| 2 | 2 | 5 | 1 | 1 | 0 | 2 | test | test | url | 2010-10-10 00:00:00 | 2011-10-10 00:00:00 |
| 3 | 3 | 5 | 1 | 1 | 0 | 2 | test | test | url | 2010-10-10 00:00:00 | 2011-10-10 00:00:00 |
| 4 | 2295 | 4 | 1 | 1 | 0 | 2 | test | test | url | 2010-10-10 00:00:00 | 2011-10-10 00:00:00 |
| 5 | 10 | 1 | 1 | 1 | 0 | 2 | test | test | url | 2010-10-10 00:00:00 | 2011-10-10 00:00:00 |
+-----------------+--------------+----------------------+------------+--------------------+-----------+------+---------+---------+-----+------------------------+---------------------+
And then I have some other tables like 'users', 'companies', 'organizations', ... etc.
I need to be able to get the username, gender and image of every notification (based on the from_user_id and from_user_auth_level).
But the problem resides in the fact, that this info resides in different places, depending on what the user_auth_level is.
For example: if my user is a "regular" user, his auth_level will be 1. And the image will reside in my "users" table, and the gender is applicable.
But if the user has auth_level == 5, it means he is an organization. In this case, gender is not applicable, and the image resides in the "organization" table, this needs to be linked via users to user_roles and then to the organization.
And this goes on for every user type, they all require different joins.
I have created a working query, but this uses UNION's everywhere, and I have read that it is not the best to use for performance reasons, so i'm hoping someone can guide me to improving this query with performance in mind:
SELECT n.*, NULL as username, NULL as gender, NULL as picture
FROM notification as n
WHERE n.from_user_auth_level IS NULL
AND n.to_user_id = $userid
UNION
SELECT n.*, u.username, u.gender as gender, u.profile_picture as picture
FROM notification as n
LEFT JOIN users AS u ON n.from_user_id = u.user_id
WHERE n.from_user_auth_level = 1
AND n.to_user_id = $userid
UNION
SELECT n.*, u.username, NULL as gender, c.logo as picture
FROM notification as n
LEFT JOIN users AS u ON n.from_user_id = u.user_id
LEFT JOIN user_companies AS uc on u.user_id = uc.user_id
LEFT JOIN company as c on uc.company_id = c.company_id
WHERE n.from_user_auth_level = 4
AND n.to_user_id = $userid
UNION
SELECT n.*, u.username, NULL as gender, o.logo as picture
FROM notification as n
LEFT JOIN users AS u ON n.from_user_id = u.user_id
LEFT JOIN user_roles as ur on u.user_id = ur.user_id
LEFT JOIN organization as o on ur.org_id = o.org_id
WHERE n.from_user_auth_level = 5
AND n.to_user_id = $userid
UNION
SELECT n.*, u.username, NULL as gender, o.logo as picture
FROM notification as n
LEFT JOIN users AS u ON n.from_user_id = u.user_id
LEFT JOIN user_roles as ur on u.user_id = ur.user_id
LEFT JOIN organization as o on ur.org_id = o.org_id
WHERE n.from_user_auth_level = 7
AND n.to_user_id = $userid
UNION
SELECT n.*, u.username, NULL as gender, NULL as picture
FROM notification as n
LEFT JOIN users AS u ON n.from_user_id = u.user_id
WHERE n.from_user_auth_level = 9
AND n.to_user_id = $userid"
After I get this result, I use PHP to order the results based on the timestamp_inserted_utc, since it's not possible to get the correct results with the UNION.

I'd use the notification table as base and used conditional outer joins as:
select
n.*,t1.gender, t2.orgNo
from
notifications n
left outer join table1 t1 on (n.auth=1 and more join)
left outer join table2 t2 on (n.auth=2 and more..)
You will have more columns but their name would make sense and you may merge at application level.

Suggested indexes (most are "covering"):
n: (from_user_auth_level, to_user_id, from_user_id)
u: (user_id, username, profile_picture, gender)
o: (org_id, logo)
ur: (org_id, user_id)
c: (company_id, logo)
uc: (company_id, user_id)
(They may even speed up Teson's Answer.)

Related

MySQL - Pivot rows to comma separated values [duplicate]

This question already has answers here:
MySQL pivot table query with dynamic columns
(3 answers)
Closed 2 years ago.
I have 3 tables.
1. user
2. user_role
3. user_role_mapping
Sample record,
user
+--------+----------+---------------+
| userid | username | email |
+--------+----------+---------------+
| 1 | user1 |user1#test.com |
| 2 | user2 |user2#test.com |
+--------+----------+---------------+
user_role
+--------+----------+
| roleid | rolename |
+--------+----------+
| 1 | user |
| 2 | manager |
| 3 | director |
| 4 | admin |
+--------+----------+
User_role_mapping
+--------+------+
| roleid |userid|
+--------+------+
| 1 | 1 |
| 2 | 1 |
| 3 | 1 |
+--------+------+
Query
select u.userid, u.username, u.email,
count(case when ur.rolename = 'user' THEN 1 END) user,
count(case when ur.rolename = 'manager' THEN 1 END) manager,
count(case when ur.rolename = 'director' THEN 1 END) director,
count(case when ur.rolename = 'admin' THEN 1 END) admin
from user_role ur
left join userrole_mapping urm
on ur.roleid = urm.roleid
left join user u
on urm.userid = u.userid
group by u.userid, u.username, u.email
Result:-
+--------+----------+---------------+------|---------|----------|-------|
| userid | username | email | user | manager | director | admin |
+--------+----------+---------------+------------------------------------
| 1 | user1 |user1#test.com | 1 | 1 | 1 | 1 |
+--------+----------+---------------+------------------------------------
Here, I am hardcoding the role in the query and the new role can get added in the future and I do not want to do the code change. Is there any better approach? I am fine with comma separated result for role in the result.
For comma separation you can use group_concat:
select u.user_id, u.user_name, u.email,
Group_concat(ur.role_name) roles
from user u
left join user_role_mapping urm
on urm.user_id = u.user_id
left join user_role ur
on ur.role_id = urm.role_id
where u.user_id = ?
group by u.user_id;
Another, maybe better approach w/o hardcoding would be to query the roles separately:
select u.user_id, u.user_name, u.email, ur.role_name
from user u
left join user_role_mapping urm
on urm.user_id = u.user_id
left join user_role ur
on ur.role_id = urm.role_id
where u.user_id = ?;

How to do the join strategy in this case?

This is the situation I'm dealing with:
I have 4 tables:
Users table:
+----+-------+
| id | name |
+----+-------+
| 1 | name1 |
| 2 | name2 |
| 3 | name3 |
+----+-------+
Assignment table:
+----+-----------------+
| id | assignment_name |
+----+-----------------+
| 11 | name1 |
| 12 | name2 |
| 13 | name3 |
+----+-----------------+
Submissions:
+----+---------------+---------+
| id | assignment_id | user_id |
+----+---------------+---------+
| 1 | 11 | 3 |
| 2 | 12 | 1 |
| 3 | 11 | 2 |
+----+---------------+---------+
Group_submissions
+----+----------------+---------+
| id | submission_id | user_id |
+----+----------------+---------+
| 1 | 1 | 1 |
| 2 | 1 | 2 |
| 3 | 3 | 1 |
+----+----------------+---------+
The submission table has an assignment_id to tell in which assignment the submission belongs to.
Also users can submit a group submission, where the one that does the submission goes to the submissions table, while the others go to the group_submissions table. That way it will be counted as one submission instead of being 2,3...N submission based on how many people where in the group.
How can i get the users that have submitted a submission or have participated in a group submission in a given assignment?
The result should return the user or users that have are in the submissions table or in the group_submissions table based on a assignment id
The result should look something likes this:
+----+-------+
| id | name |
+----+-------+
| 1 | name1 |
| 2 | name2 |
+----+-------+
It should basically return the user table.
This is what i have tried so far:
This only gives me the users that aren't in the submissions table but are in the group_submission
select * from users u
right join group_submissions gs on u.id = gs.student_id
right join assignment_submissions ass on gs.submission_id = ass.id
inner join assignments a on a.id = ass.assignment_id
where a.id = number
This only gives me the one user that made the submission (in the submissions table)
select * from users u
right join assignment_submissions ass on u.id= ass.student_id
right join group_submissions gs on ass.id = gs.submission_id
inner join assignments a on a.id = ass.assignment_id
where a.id = number
What should my join strategy be here? Or maybe joins are not the right option here.
NOTE: This is a MySQL database.
You could use exists:
select u.*
from users u
where
exists (
select 1
from submissions s
where s.user_id = u.id and s.assignment_id = ?
)
or exists (
select 1
from group_submissions gs
inner join submissions s on s.id = gs.submission_id
where gs.user_id = u.id and s.assignment_id = ?
)
I am thinking a union could do it. Like this
select * from
(
select u.*, assignment_id, 'assignment_submissions' as type from users u
inner join assignment_submissions ass on u.id= ass.student_id
union
select u.*, assignment_id, 'group_submissions' from users u
inner join group_submissions gs on u.id = gs.student_id
inner join assignments a on a.id = ass.assignment_id
)a
where assignment_id = ?

Query to select from one table and then from the other, based on results

I have been trying to select all things from one table and then for each result select from another table only one newest result using MySQL.
The first table is a standard one, with AI id and text name
users
+----+-------+
| id | name |
+----+-------+
| 1 | John |
| 2 | Peter |
+----+-------+
then there is the second one with AI id, int user_id, text action and datetime date
actions
+----+---------+--------+------------+-----+
| id | user_id | action | date | str |
+----+---------+--------+------------+-----+
| 7 | 2 | drink | 2019-01-10 | 5 |
| 6 | 1 | sleep | 2019-02-14 | -2 |
| 5 | 2 | walk | 2019-04-24 | 4 |
| 4 | 1 | jump | 2019-03-14 | 3 |
| 3 | 2 | talk | 2019-04-30 | -8 |
| 2 | 2 | train | 2019-04-14 | -1 |
| 1 | 1 | drive | 2019-04-01 | 1 |
+----+---------+--------+------------+-----+
So now I want to select all from table_users and for each found row search table_actions and find only newest one, either based on id or date.
So it would look either like this (by id)
[0] => ['user_id'] = 1
['name'] = John
['action'] = sleep
['date'] = 2019-02-14
['str'] = -2
[1] => ['user_id'] = 2
['name'] = Peter
['action'] = drink
['date'] = 2019-01-10
['str'] = 5
or like this
[0] => ['id'] = 1
['name'] = John
['table_actions'] => ['id'] = 6
['user_id'] = 1
['action'] = sleep
['date'] = 2019-02-14
['str'] = -2
this sounds easy, but I tried few things and nothing worked like this. The closes one was with something like (dont have exact version on my hand, just remembering from on top of my head):
SELECT users.*, actions.*
FROM users
LEFT JOIN actions ON users.id = (
SELECT user_id FROM actions
WHERE users.id = actions.user_id
LIMIT 1
)
GROUP BY actions.user_id
with that I would get all results from users and then for each get one result from actions, but that result from actions would not be the newest one, apparently it groups as it likes to, I tried MAX(actions.id), but I have got no luck with that.
Does anyone know the solution ? for now I have to take all from users and for each result take another query in my php code and I feel there is an elegant and faster way to do that.
With this query:
select
user_id,
max(date) date
from actions
group by user_id
you get the latest date for each user and then you must join to it the 2 tables:
select
u.id, u.name, a.id, a.action, a.date, a.str
from users u
inner join actions a on a.user_id = u.id
inner join (
select
user_id,
max(date) date
from actions
group by user_id
) g on g.user_id = a.user_id and g.date = a.date
If you want the latest results by id and not by date:
select
u.id, u.name, a.id, a.action, a.date, a.str
from users u
inner join actions a on a.user_id = u.id
inner join (
select
user_id,
max(id) id
from actions
group by user_id
) g on g.user_id = a.user_id and g.id = a.id

Count rows from one tables of users in another table

I want to create a query for project listings that would give the number of registered applications, excluding the ones for which the user does not exist.
In this case, considering user 10 does not exist, I should have the query results as folows:
RESULTS
+----+------------+--------------+
| id | project | applications |
+----+------------+--------------+
| 1 | MyProject1 | 3 |
| 2 | MyProject2 | 0 |
| 3 | MyProject3 | 0 |
+----+------------+--------------+
TABLES
Projects
+----+------------+
| id | name |
+----+------------+
| 1 | MyProject1 |
| 2 | MyProject2 |
| 3 | MyProject3 |
+----+------------+
applications
+----+------+------------+
| id | user | project_id |
+----+------+------------+
| 1 | 3 | 1 |
| 2 | 4 | 1 |
| 3 | 5 | 1 |
| 4 | 10 | 1 |
+----+------+------------+
users
+----+---------+
| id | Name |
+----+---------+
| 1 | Smith |
| 2 | John |
| 3 | Paul |
| 4 | Chris |
| 5 | Gabriel |
+----+---------+
The below query is not excluding the non-existing users:
SELECT `projects` . * , (
SELECT COUNT( * )
FROM `applications`
WHERE `applications`.`project_id` = `projects`.`id`
AND EXISTS (
SELECT `applications`.`id`
FROM `applications` , `users`,`project`
WHERE `application`.`user` = `users`.`id` AND `applications`.`project_id` = `project`.`id`
)
) AS `applications`
FROM `projects` ORDER BY `id` DESC LIMIT 30
I think you want left join and group by:
select p.id, p.name, count(u.id)
from projects p left join
applications a
on p.id = a.project_id left join
users u
on a.user_id = u.id
group by p.id, p.name;
However, you might want to think about fixing the data. It seems like there should be foreign key relationships between applications and projects and applications and users. The ability to have an invalid user means that there is no valid foreign key relationship to users.
Your query looks overly complicated. This should do:
select
id,
name as project,
(
select count(*)
from applications a
where a.project_id = p.id
and a.user in (select id from users)
) as applications
from projects p;
Based on previous solution
select p.id, p.name, count(u.id)
from projects p left join
applications a
on p.id = a.project_id left join
users u
on a.user = u.id
where u.id is not null
group by p.id, p.name;
When you do a left join, if the search value doesn't exists, it returns null. Then filtering by excluding null users, will give you the result.
Please find a sqlfiddle to illustrate it : http://www.sqlfiddle.com/#!9/cbfec6/3
But easiest solution would be
select p.id, p.name, count(u.id)
from projects p,applications a, users u
where a.user = u.id
and p.id = a.project_id
group by p.id, p.name;

Joining tables when certain one table has no values?

I have these tables:
USER TABLE
uid | name | role
| |
1 | bob | package 1
2 | jill | package 2
3 | pam | package 1
NODE TABLE
nid | uid | type
| |
1 | 1 | car
2 | 1 | car
3 | 1 | car
4 | 2 | page
5 | 1 | car
6 | 3 | car
If I do:
select u.uid, u.name, count(nid) as totalNodes from USER as u left join NODE on n.uid = u.uid where n.type = 'car' group by u.uid
I end up with:
uid | name | totalNodes
| |
1 | bob | 4
3 | pam | 1
In other words, Jill is excluded. Why? And how can I avoid this? I.e. I want Jill to also appear in the list, but with totalNodes as 0 (or even NULL).
You need to perform your aggregate before attempting to join the tables as what you are currently doing is left joining, then restricting the data (at which point Jill is excluded) then grouping. If you do the count and restriction in a subquery you can then left join these results to the user table for the output you want:
SELECT u.uid, u.name, IFNULL(c.nodecount,0) AS `count`
FROM USER u LEFT JOIN (
SELECT uid, `type` , COUNT(nid) AS nodecount
FROM node
WHERE TYPE = 'car'
GROUP BY uid, type
) AS c ON u.uid = c.uid
use RIGHT JOIN instead left,
try :
select u.uid, u.name, count(nid) as totalNodes from USER as u
right join NODE on n.uid = u.uid where n.type IS NULL or n.type = 'car' group by n.uid
see this excellent post a visual explanation of joins :
http://www.codinghorror.com/blog/2007/10/a-visual-explanation-of-sql-joins.html
mysql syntax of join :
http://dev.mysql.com/doc/refman/5.0/en/join.html