Table of contents
- What is SQL
- SQL dialects
- SQL features
- Why learn SQL
- DML, DDL
- SQLite installation
- Tables
- Database file creation
- Creating test schema
- Summary
- First select
- Sqlite prompt configuration
- Select query basics
- Table aliases
- Column aliases
- Rows filtering
- Export data to CSV
- Summary
- Basic sorting
- Sorting by multiple columns
- Sorting’s voodoo techniques
- Summary
- LEFT JOIN
- RIGHT JOIN
- JOIN
- Using joins with more than two tables
- Using subqueries
- More on conditions in joins
- Compound conditions
- You can use any condition you want
- Summary
- UNION, EXCEPT, INTERSECT
- Subqueries
This is the small tutorial about SQL. I hope it will help you to solve problems and make it easy for you to understand SQL.
Top links:
Note: Work in progress. ☜(ﾟヮﾟ☜)
Info for russian-speaking readers
There’s also a book that was generated from one of my previous sites, it’s about SQL in Oracle database.
DownloadWhat is SQL
SQL, or Structure Query Language is the main tool for interaction with relational database management systems(RDBMS).
SQL is used to:
- Retrieve data from DB
- Save data to DB
- Perform changes in existing data
SQL dialects
SQL is standardized, but who cares?
There are many RDBMS, and as a rule, each of them has its differences in SQL implementation - some of them use different syntax for the same things, some of them have or have not some features, but generally, if you know one SQL dialect, it will be not so hard to switch to another dialect when it will be needed.
In this tutorial, we will use SQLite database, since it’s easy to install on most platforms, and also it supports most of the standard SQL.
SQL features
A few words about what is so special in SQL.
Unlike many other programming languages, such as Java, Pascal or JavaScript, in which programming is performed by describing how to do things, in SQL we need to describe what we need to do, or what result we want, but generally, we can’t tell the database how to achieve this result.
Why learn SQL
As told, SQL is the main tool to talk with RDBMS.
When some program or service want to get, save or change data in a database, it does this by using SQL. Even if SQL is not used directly, and for interaction with the database some framework is used(such as Hibernate in Java), under the hood this framework generates SQL, which is then passed to the database.
DML, DDL
SQL commands can be divided into two groups - DML and DDL.
Except for DML and DDL, there are also DCL and TCL groups, but we don’t touch them in this tutorial.
DML stands for Data Manipulation Language. It consists of commands which can change or retrieve already existing data. By changing we mean also insertion of new data and deletion of old data.
So, here are the list of DML commands:
- SELECT
- INSERT
- UPDATE
- DELETE
- MERGE
Note that SELECT command also related to DML.
DDL is stands for Data Definition Language. It consists from commands which are responsible for database objects’ creation or modification.
These commands are related to DDL:
- CREATE
- RENAME
- ALTER
- DROP
- RENAME
- TRUNCATE
- COMMENT
In this tutorial, we will mostly concentrate on DML commands, but we will use DDL for creating our database schema.
SQLite installation
Linux
If you Llinux system, probably SQLite is already installed.
First, try to type into terminal following command:
sqlite3 -version
If you see something like this, SQLite is already installed:
3.31.1 2020-01-27 19:55:54 3bfa9cc97da10598521b342961df8f5f68c7388fa117345eeb516eaa837balt1
To install SQLite, we should search for sqlite package. Since there are so many
different Linux distributives and theirs package managers, it’s hard to tell what exactly
command you need to type, but we will list instructions for most popular linux distrubutives.
Debian-based( Debian, Ubuntu, and others)
sudo apt-get install sqlite3
Fedora
sudo dnf install sqlite
Arch linux
sudo pacman -S sqlite
Windows
You can download precompiled binaries from the official site. Precompiled binaries for Linux also can be downloaded from there.
Tables
In RDMS data is stored in the tables. Tables are the key objects, with which we have to work in SQL.
Tables in the databases don’t differ much from the tables with which you already familiar from school - they consist of columns and rows. Each column in a table has its own name and type. Generally speaking, there are five types of data that we can store in the database:
- Numbers
- Text
- Dates
- Binary data(images, pdf files and so on)
- Nothing
These types are not related to any concrete database engine. Moreover, many
databases have much more types, but those types don’t bring any new kind of
data that we can store - they just split each of our types into more small
subtypes. For example, the Oracle database has a DATE type( accurate to seconds).
Also, it has the TIMESTAMP type, which is also used to store dates,
but with more precision(accurate to parts of second).
WTF is NULL?
We need to pay attention to the last item on our list. What does it mean - Nothing? We can think about it as just an empty column in a row. But the math guy who have created relational theory did not like empty columns, and he said that all columns must have a value. But in practice, we need empty columns sometimes, so database developers have created a specific value which means that columns with this value don’t have a value. And this value is NULL. We will talk about nulls later, don’t worry, anything that we need to know for now is that this value means nothing, empty, no value and so on.
Database file creation
Let’s create SQLite database which we will use to keep our database structure for later use and experiments. In terminal, type following command:
sqlite3
Right after this we will see something like this:
SQLite version 3.31.1 2020-01-27 19:55:54
Enter ".help" for usage hints.
Connected to a transient in-memory database.
Use ".open FILENAME" to reopen on a persistent database.
sqlite>
Wee see that now we are connected to in-memory database, what means that if we quit from SQLite command prompt or shutdown the computer, all our data will be lost.
What we want is to save the database to a file so we can
use it later without losing our data. To do this, .save command is used:
sqlite> .save testdb.db
After execution of the above command, file named testdb.db
will be created in the current directory.
To quit SQLite command interpreter, use .quit command.
Creating test schema
Let’s create our tables and fill them with test data. We will use this schema throughout the whole SQL tutorial.
Our database structure represents an imaginary shop service, which allows users to create some orders, put some items into those orders and close them. Each order can be in one of the next states: Closed and Waiting.
create table users(
id number primary key,
login text,
is_active number
);
create table orders(
id number primary key,
user_id number,
order_date text,
order_num number,
status number
);
create table items(
id number,
item_name text
);
create table order_items(
id number,
order_id number,
item_id number,
quantity number
);create table users(
id number primary key,
login text,
is_active number
);
insert into users
values(1, 'JohnDoe', 1);
insert into users
values(2, 'MrWinner', 1);
insert into users
values(3, 'Alex', 1);
insert into users
values(4, 'Barbie', 1);
insert into users
values(5, 'Lisa', 1);
insert into users
values(6, 'Stone', 0);
create table items(
id number primary key,
item_name text,
price number
);
insert into items
values(1, 'Microwave', 300);
insert into items
values(2, 'Mobile phone', 350);
insert into items
values(3, 'Nintendo switch', 400);
insert into items
values(4, 'TV', 1000);
insert into items
values(5, 'Laptop', 1500);
insert into items
values(6, 'Washer', 700);
create table orders(
id number primary key,
user_id number,
order_date text,
order_num number,
status text,
constraint orders_users_fk foreign key(user_id)
references users(id)
);
insert into orders
values(1, 2, '2021-10-11', 32, 'CLOSED');
insert into orders
values(2, 2, '2021-08-15', 230, 'WAITING');
insert into orders
values(3, 4, '2019-04-24', 39, 'CLOSED');
insert into orders
values(4, 5, '2019-04-24', 39, 'WAITING');
create table order_items(
id number,
order_id number,
item_id number,
quantity number,
constraint order_items_order_fk foreign key(order_id)
references orders(id),
constraint order_items_item_fk foreign key(item_id)
references items(id)
);
insert into order_items
values(1, 1, 4, 1);
insert into order_items
values(2, 1, 6, 1);
insert into order_items
values(3, 2, 5, 2);
insert into order_items
values(4, 3, 3, 4);
insert into order_items
values(5, 3, 2, 4);
Save this query to a file, named, for example, schema.sql.
Now, we need to open our database file:
sqlite3 testdb.db
To execute sql script, in SQLite command prompt, we need
to use command read:
sqlite> .read schema.sql
This command executes our script, but to be able to use test data later, we need to save changes back to file:
sqlite> .save
Let’s check which tables are in the database:
sqlite>.tables
Output:
items order_items orders users
So, now we have the database file, called testdb.db,
which contains tables with which we will work later.
In the next chapter we will look at how to retrieve
information from tables and how we can save it to external
file using SQLite command prompt. Now, take some time to relax;
here is a picture to help you do this:
Summary
- SQL is the main language to talk with RDBMS
- SQL consists from DDL, DML, DCL and TCL
- There are many SQL dialects
Nullis specific value that determines empty value- To create new database file, we use command
sqlite3 dbfile.db - To execute a sql script,
.read sqlfile.sqlcommand is used - To save changes back to file,
.savecommand is used .tablescommand shows list of tables in the current database
In this chapter, we gonna look at how to retrieve data from the database.
First select
Let’s connect to our database, which we have created previously:
sqlite3 testdb.db
To select data from database tables, SELECT command is used.
It is probably the most frequently usable command in SQL, and we will use
this command all the time.
Let’s see what data we have in the users table:
sqlite> select *
...> from users;
Result:
1|JohnDoe|1
2|MrWinner|1
3|Alex|1
4|Barbie|1
5|Lisa|1
6|Stone|0
Here we have just ran our first SQL query, which retrieves all data
from the users table. One important thing - a sql query is executed
after the first semicolon.
Ok, we have some results, but they look very ugly. We can’t see table
columns, and here is no any formatting. Before we dive into the
structure of SELECT query, let’s tweak SQLite command prompt, so
that query results will look more sexy.
Sqlite prompt configuration
We gonna add options for displaying column headers and for
aligning columns content. To do this, we need to create
a file named .sqliterc in the home directory with the following content:
.headers on
.mode column
Now, if we run our query again, result will look much better:
id login is_active
---------- ---------- ----------
1 JohnDoe 1
2 MrWinner 1
3 Alex 1
4 Barbie 1
5 Lisa 1
6 Stone 0
Select query basics
We have ran next query:
select *
from users
In this query we have just said: “Database, please,
give me all rows and all columns from the users table.
In select queries, asterisk means “All columns”. If we want to retrieve some specific columns, we need to list them. Let’s get only login names:
select login
from users
If we want to select a few columns, they should be separated by comma:
select login, is_active
from users
Table aliases
It is a good practice to use aliases in SQL queries. By using table aliases, we can say from which table column should be used. Table aliases are separated by space from the table name:
select u.login, u.is_active
from users u
Now it may look like as a redundant thing, but in complex cases, table aliases make queries much, much more readable.
Table names also can be used as aliases, so if the table name is short, it is ok to use it as an alias:
select users.login, users.is_active
from users
Column aliases
Aliases can be given for selected columns, too:
select u.login user_login,
u.is_active is_active_flag
from users
And now our query result looks like there are
users_login and is_active_flag in
the users table:
user_login is_active_flag
---------- --------------
JohnDoe 1
MrWinner 1
Alex 1
Barbie 1
Lisa 1
Stone 0
Rows filtering
Almost always we don’t need all data in a table, we want to get rows that satisfy some conditions. For example, we may want only active users, only inactive users, users which logins start with the symbol ‘A’ and so on.
In SQL, we specify conditions in the WHERE clause:
select u.*
from users u
where u.is_active = 1
id login is_active
---------- ---------- ----------
1 JohnDoe 1
2 MrWinner 1
3 Alex 1
4 Barbie 1
5 Lisa 1
Here we have got only active users.
When using WHERE, only those rows go to the result, which
satisfy all conditions.
In the WHERE clause, we can specify multiple conditions by using
AND and OR keywords.
Let’s get information about users with nicknames “MrWinner” and “Lisa”:
select u.*
from users u
where u.login = 'MrWinner'
or u.login = 'Lisa'
id login is_active
---------- ---------- ----------
2 MrWinner 1
5 Lisa 1
We have used OR to include both rows in the result.
If we use AND, the result will be empty:
select u.*
from users u
where u.login = 'MrWinner'
and u.login = 'Lisa'
That is because there are no rows in the table which have values “MrWinner” and “Lisa” at the same time.
Grouping conditions
Let’s remember school math: things between brackets are calculated first. Same thing with conditions in SQL (actually, it is boolean logic and predicate calculations, but it’s boring):
select u.*
from users u
where (is_active = 0 or login = 'Lisa') and is_active = 1
Result:
id login is_active
---------- ---------- ----------
5 Lisa 1
If we change brackets position, the result will be different:
select u.*
from users u
where is_active = 0 or (login = 'Lisa' and is_active = 1)
Result:
id login is_active
---------- ---------- ----------
5 Lisa 1
6 Stone 0
Export data to CSV
This topic is exclusively about the SQLite database feature. It allows you to save query results into a CSV file.
Here is the how-to:
-
Turn on CSV mode:
sqlite> .mode csv -
Set output file:
sqlite> .output result.csvHere
result.csvis the output file. It can has any other name. -
Execute query:
qlite> select * ...> from users;
Our result csv file now contains:
id,login,is_active
1,JohnDoe,1
2,MrWinner,1
3,Alex,1
4,Barbie,1
5,Lisa,1
6,Stone,0
Additional info about command-line features of SQLite can be found at Official docs.
Summary
- To filter query results, WHERE clause is used
- To change the default calculation order, we can use brackets for grouping conditions
- In SQLite, it is easy to export data into a CSV file
Sometimes we want to get result in specific order,
and order by clause does exactly this.
Basic sorting
Let’s look at the data in the orders table:
select user_id, order_date, order_num
from orders
user_id order_date order_num
---------- ---------- ----------
2 2021-10-11 32
2 2021-08-15 230
4 2019-04-24 39
5 2019-04-24 39
We can retrieve orders info sorted by the order_num column:
select user_id, order_date, order_num
from orders
order by order_num asc
Result:
user_id order_date order_num
---------- ---------- ----------
2 2021-10-11 32
4 2019-04-24 39
5 2019-04-24 39
2 2021-08-15 230
As we can see, now orders are sorted by order number in
ascending order. This is because we have set ascending
order in the order by by adding the asc keyword.
It’s unnecessary to add asc if we want to sort rows in
ascending order - it is the default. Next query
is identical to the previous one:
select user_id, order_date, order_num
from orders
order by order_num
To sort rows in descending order, we should add
desc to order by clause:
select user_id, order_date, order_num
from orders
order by order_num desc
user_id order_date order_num
---------- ---------- ----------
2 2021-08-15 230
4 2019-04-24 39
5 2019-04-24 39
2 2021-10-11 32
Sorting by multiple columns
To sort rows by multiple columns, those columns should be separated by commas:
select user_id, order_date, order_num
from orders
order by user_id desc, order_date
user_id order_date order_num
---------- ---------- ----------
5 2019-04-24 39
4 2019-04-24 39
2 2021-08-15 230
2 2021-10-11 32
Here rows come sorted by userId in descending order
first, and by order date in ascending order second.
It is observed by the fact that rows with the same
user_id(where it equals to 2) sorted by order_date
in ascending order.
Sorting’s voodoo techniques
It is unnecessary for a column to be included into a select query - we can sort by column, but don’t select it at all:
select order_date, order_num, status
from orders
order by user_id desc
order_date order_num status
---------- ---------- ----------
2019-04-24 39 WAITING
2019-04-24 39 CLOSED
2021-10-11 32 CLOSED
2021-08-15 230 WAITING
Rows are sorted by the user_id column, as in the
previous example, but the user_id column itself does not
exist in the result set.
We can sort columns by their order in select clause, not just
by their name:
select user_id, order_date, order_num, status
from orders
order by 1, 2 desc
user_id order_date order_num status
---------- ---------- ---------- ----------
2 2021-10-11 32 CLOSED
2 2021-08-15 230 WAITING
4 2019-04-24 39 CLOSED
5 2019-04-24 39 WAITING
The above query is the same as the next one:
select user_id, order_date, order_num, status
from orders
order by user_id, order_date desc
Such a feature can look attractive, but in practice you should avoid sorting by columns’ order, because such queries may become invalid after changing column order. Let’s demonstrate this.
Suppose we have a query:
select order_date, order_num, status
from orders
order by 1, 2 desc
order_date order_num status
---------- ---------- ----------
2019-04-24 39 CLOSED
2019-04-24 39 WAITING
2021-08-15 230 WAITING
2021-10-11 32 CLOSED
But some time later we decide to retrieve the user_id
column as well:
select user_id, order_date, order_num, status
from orders
order by 1, 2 desc
user_id order_date order_num status
---------- ---------- ---------- ----------
2 2021-10-11 32 CLOSED
2 2021-08-15 230 WAITING
4 2019-04-24 39 CLOSED
5 2019-04-24 39 WAITING
Now rows come in an absolutely different order, because
the user_id column is first and the order_date is
second.
Summary
-
Order byclause is used for result sorting. -
There is no need to add
asckeyword to sort rows in ascending order - this value is assumed to be the default. -
It is the bad pattern to order rows by specifying columns order - it is better to list columns by their names.
Joins are some of the most essential concepts of SQL. They are used for querying data from multiple tables. In this article we will look at
JOINandLEFT/RIGHT JOIN- these are all what you need to know to freely work with multiple tables.
LEFT JOIN
Let’s forget about tables and databases for a while, and imagine that we have two bags with balls of various colors. Also, let’s imagine that one bag is placed on the left side and another bag is placed on the right side relating to us.
Now, let’s play the very simple game - we will take one random ball from the left bag and then we will search for all balls with the same colour in this bag. Then we search for balls with this colour in the right bag. We can write down this operation like “Find all balls of color C in bag A, and then find all balls of color C in bag B”.
In SQL, we could write a query for this:
-- Search for Red balls
select ball, colour
from A
left join B on B.colour = A.colour
where A.colur = 'Red'
We will dig deeper into examples from our test database a little bit later.
If we modify the rules a little and will take all balls from the left bag and search for all balls from the right bag with the same colour as the balls from the left bag, then we should remove the filter condition from our query:
select ball, colour
from A
left join B on B.colour = A.colour
What if we will not find any balls in the right bag, which colours
match with the balls from the left bag? In this case, our result row
set will contain only balls from the left bag. This is the
only thing that makes JOIN and LEFT or RIGHT JOIN different.
For a concrete example, suppose we are gonna fetch info about all orders.
We have the orders table, which contains info about orders,
but we also want to see what items were in each order, and this
info is stored in the order_items table.
First, let’s see what data lies in each of these tables:
Orders:
select *
from orders
id user_id order_date order_num status
---------- ---------- ---------- ---------- ----------
1 2 2021-10-11 32 CLOSED
2 2 2021-08-15 230 WAITING
3 4 2019-04-24 39 CLOSED
4 5 2019-04-24 39 WAITING
Order_items:
select *
from order_items
id order_id item_id quantity
---------- ---------- ---------- ----------
1 1 4 1
2 1 6 1
3 2 5 2
4 3 3 4
5 3 2 4
To query data from both tables, we need to pick a condition by which these tables will be joined( like balls colors).
Here we can(and need, actually) to use order id -
order_id column in the order_items table
is linked to the id column in the orders table.
Such relation has a corresponding specific object
in databases, called “foreign key”.
So, here is our query:
select o.id, o.order_date, o.order_num, i.quantity
from orders o
left join order_items i on i.order_id = o.id
id order_date order_num quantity
---------- ---------- ---------- ----------
1 2021-10-11 32 1
1 2021-10-11 32 1
2 2021-08-15 230 2
3 2019-04-24 39 4
3 2019-04-24 39 4
4 2019-04-24 39
Now let’s talk about this query in more detail.
We take all rows from the orders table, and then,
for every row, we take all rows from the order_items
table with the same value in the order_id column.
For the order with id = 4, we did not find any rows in the
order_items table, so there is no value in the quantity
field for this row.
RIGHT JOIN
If we will take balls from the right bag first and then search
for similar colours in the left bag, this will be called RIGHT JOIN
operation.
JOIN
As was said, JOIN is the operation that leaves only those rows which exist in both tables. Let’s get all user orders:
select u.id, u.login, o.order_date, o.status
from users u
join orders o on o.user_id = u.id
Result:
id login order_date status
---------- ---------- ---------- ----------
2 MrWinner 2021-10-11 CLOSED
2 MrWinner 2021-08-15 WAITING
4 Barbie 2019-04-24 CLOSED
5 Lisa 2019-04-24 WAITING
Note that users with id = 1 and 2 did not appear in a result, because they do not have any orders.
Or, let’s rewrite the example from
the LEFT JOIN part, but this time we will
use JOIN:
select o.id, o.order_date, o.order_num, i.quantity
from orders o
join order_items i on i.order_id = o.id
Result:
id order_date order_num quantity
---------- ---------- ---------- ----------
1 2021-10-11 32 1
1 2021-10-11 32 1
2 2021-08-15 230 2
3 2019-04-24 39 4
3 2019-04-24 39 4
Row with empty quantity disappeared, because there is no
such row in the order_items table, where order_id is equal 4.
Using joins with more than two tables
When we want to join more than two tables, we just apply join rules for the first two tables first, then we join result of this join with third table then result is joined with the fourth table and so on.
select u.id,
u.login,
o.order_date,
i.item_name,
oi.quantity
from users u
join orders o on o.user_id = u.id
left join order_items oi on oi.order_id = o.id
left join items i on i.id = oi.item_id
where u.is_active = 1
Result:
id login order_date item_name quantity
---------- ---------- ---------- ---------- ----------
2 MrWinner 2021-10-11 TV 1
2 MrWinner 2021-10-11 Washer 1
2 MrWinner 2021-08-15 Laptop 2
4 Barbie 2019-04-24 Mobile pho 4
4 Barbie 2019-04-24 Nintendo s 4
5 Lisa 2019-04-24
Here, the users table first joined with
the orders table first.
We can see what we have after this join:
select u.id, u.login, o.order_date
from users u
join orders o on o.user_id = u.id
id login order_date
---------- ---------- ----------
2 MrWinner 2021-10-11
2 MrWinner 2021-08-15
4 Barbie 2019-04-24
5 Lisa 2019-04-24
We see, that MrWinner has two orders (different dates, and also different order_id values), Barbie and Lisa both have one order.
Important notice: when joins are performed,
all columns from both tables are available
for later use in a query. They don’t have
to appear in the selected columns list. In our query,
for example, we used the is_active column from the users
table in the where clause, but the column itself did
not appear in a result set.
Next, we are
jeft joining this result with the
order_items table:
select u.id, u.login, o.order_date, oi.quantity
from users u
join orders o on o.user_id = u.id
left join order_items oi on oi.order_id = o.id
id login order_date quantity
---------- ---------- ---------- ----------
2 MrWinner 2021-10-11 1
2 MrWinner 2021-10-11 1
2 MrWinner 2021-08-15 2
4 Barbie 2019-04-24 4
4 Barbie 2019-04-24 4
5 Lisa 2019-04-24
Take a look at the row with user Lisa - it does not
have a value in the quantity column. This is because there are
no items in her order. And since we use left join, rows
that have been produced by previous joins did not disappear.
Just to see the difference, we can rewrite this query
using join instead of left join:
select u.id, u.login, o.order_date, oi.quantity
from users u
join orders o on o.user_id = u.id
join order_items oi on oi.order_id = o.id
id login order_date quantity
---------- ---------- ---------- ----------
2 MrWinner 2021-10-11 1
2 MrWinner 2021-10-11 1
2 MrWinner 2021-08-15 2
4 Barbie 2019-04-24 4
4 Barbie 2019-04-24 4
Here, rows, which do not have order items, have been removed from the result set.
But let’s go back to our main example. The last
join is with the items table, from which we select
item names.
The final part is not join, but filtering
(where u.is_active = 1) - we keep
only those rows, which have “1” as value of
the users.is_active column (actually,
no rows were removed at this step, because all
users in result set is active).
Using subqueries
Instead tables, we can use subqueries in any JOIN clause (It applies not only to joins - in SQL you can use subqueries almost everywhere where tables can be used).
select u.login,
ord.order_date,
ord.order_num,
ord.item_name
from users u
left join (
select o.user_id,
o.order_date,
o.order_num,
oi.quantity,
i.item_name,
i.price
from orders o
join order_items oi on oi.order_id = o.id
left join items i on i.id = oi.item_id
where o.status = 'CLOSED'
) ord on ord.user_id = u.id
Result:
login order_date order_num item_name
---------- ---------- ---------- ----------
JohnDoe
MrWinner 2021-10-11 32 TV
MrWinner 2021-10-11 32 Washer
Alex
Barbie 2019-04-24 39 Mobile pho
Barbie 2019-04-24 39 Nintendo s
Lisa
Stone
Subqueries did not explained in this tutorial yet, but
here it should be clear - we wrap some query in parens,
assign alias name to it(ord in our case), and use it
like it is a table.
Tip: Select only those columns in subqueries, which
will be used later. Here, we did not use item price column
(i.price),
so it must be removed. Unnecessary columns, joins and conditions
in subqueries make queries more complicated and harder
to read later, especially for other people.
More on conditions in joins
Compound conditions
We can use compound conditions by which we are going to join tables:
select u.login, o.order_num, o.status
from users u
left join orders o on o.user_id = u.id and o.status = 'CLOSED'
Result:
login order_num status
---------- ---------- ----------
JohnDoe
MrWinner 32 CLOSED
Alex
Barbie 39 CLOSED
Lisa
Stone
Here, only those rows from the orders table are
attached to rows from the
users table, where not only user_id has matched rows, but
also rows that have status ‘CLOSED’.
Be careful here - it is the join condition, and it affects only on joined rows - not on a whole result. Compare query above with this one and see the difference:
select u.login, o.order_num, o.status
from users u
left join orders o on o.user_id = u.id
where o.status = 'CLOSED'
Result:
login order_num status
---------- ---------- ----------
MrWinner 32 CLOSED
Barbie 39 CLOSED
Conditions in a where clause are applied to
a whole dataset which is result of all joins
in a query.
You can use any condition you want
You don’t have to join by id columns or
something like that. Yes, usually, joins
are performed by linking rows in one set to rows
in another set by matching some columns, but it is not
necessary. Join conditions are like all other conditions
in SQL - they are just a rule, or set of rules,
which can be evaluated to true or false, and nothing more.
For example:
select u.id, u.login, o.user_id, o.order_num
from users u
join orders o on (1 = 1)
Boom:
id login user_id order_num
---------- ---------- ---------- ----------
1 JohnDoe 2 32
1 JohnDoe 2 230
1 JohnDoe 4 39
1 JohnDoe 5 39
2 MrWinner 2 32
2 MrWinner 2 230
2 MrWinner 4 39
2 MrWinner 5 39
3 Alex 2 32
3 Alex 2 230
3 Alex 4 39
3 Alex 5 39
4 Barbie 2 32
4 Barbie 2 230
4 Barbie 4 39
4 Barbie 5 39
5 Lisa 2 32
5 Lisa 2 230
5 Lisa 4 39
5 Lisa 5 39
6 Stone 2 32
6 Stone 2 230
6 Stone 4 39
6 Stone 5 39
This is called “Decart multiplication”, or “cartesian product”.
Here, for each row in the users table, join
condition is evaluated to True for all rows
in the orders table. In SQL, there is special
syntax for such kind of joins - CROSS JOIN. Next
query is identical to the previous one:
select u.id, u.login, o.user_id, o.order_num
from users u
cross join orders o
Summary
- Joins allow us to query data from multiple tables
- We can join subqueries
- Join conditions can be complex
UNION, EXCEPT, INTERSECT
In this chapter, we will look at so-called Set operators, which work with the datasets returned by two queries.
To make examples more understandable, we will create the
guest_orders table, which will store orders that were
placed by guest users.
Our first step is to open our testdb.db file:
sqlite3 testdb.db
Then, we need to run this query:
create table guest_orders(
login text,
order_date text,
order_num,
status text
);
insert into guest_orders
values('Anonymous', '2021-11-04', '238', 'CLOSED');
insert into guest_orders
values('JohnDoe', '2021-11-04', '239', 'CLOSED');
insert into guest_orders
values('JohnDoe', '2021-11-19', '254', 'WAITING');
insert into guest_orders
values('Alex', '2021-12-14', '283', 'WAITING');
insert into guest_orders
values('Stone', '2021-12-14', '290', 'CLOSED');
That is it. No further actions are required. When we open
the testdb.db file again, it will already contain this
table.
UNION
Let’s say we want to do some analytic research, and for this, we need the logins of all users who have placed the orders (without taking into account order statuses).
We already can get all logins of guest users:
select login
from guest_orders;
login
---------
Anonymous
JohnDoe
JohnDoe
Alex
Stone
As all logins of the registered users that
specified in the orders table:
select u.login
from orders o
join users u on u.id = o.user_id;
login
--------
MrWinner
MrWinner
Barbie
Lisa
The UNION operator combines results from two queries and
removes any duplicate rows so that each row in the result
set is unique:
select login
from guest_orders
UNION
select u.login
from users u
join orders o on o.user_id = u.id
Result:
login
---------
Alex
Anonymous
Barbie
JohnDoe
Lisa
MrWinner
Stone
It will be easier to understand how the UNION operator works
if we run each query separately:
First query:
select login
from guest_orders
login
---------
Anonymous
JohnDoe
JohnDoe
Alex
Stone
Second query:
select u.login
from users u
join orders o on o.user_id = u.id
login
--------
MrWinner
MrWinner
Barbie
Lisa
If we mix all logins, we will get this:
Anonymous
JohnDoe
JohnDoe
Alex
Stone
MrWinner
MrWinner
Barbie
Lisa
Now, if we remove all duplicate rows, we will
get the original result. This is how the UNION
operator works - getting all data and removing all duplicates,
nothing complicated.
Now, get prepared for the main rule of all set operators: each query must return the same number of columns:
select item_name, id
from items
UNION
select item_id
from order_items;
We will get the error message from SQLite:
Error: SELECTs to the left and right of UNION do not have the same number of result columns.
There is no need for any explanations, I think.
UNION ALL
The UNION ALL operator works like the UNION operator,
except that it doesn’t remove duplicates:
select login
from guest_orders
UNION ALL
select u.login
from users u
join orders o on o.user_id = u.id
Result:
Anonymous
JohnDoe
JohnDoe
Alex
Stone
MrWinner
MrWinner
Barbie
Lisa
Info: If you know that there are no
duplicates in both queries, use the UNION ALL operator,
so that a database will not spend time removing them.
EXCEPT
Let’s get the items that don’t appear in any order:
select id
from items
except
select item_id
from order_items;
Result:
id
--
1
Yeah, just one item. Anyway, the EXCEPT operator returns
unique rows from the first query, which don’t exist in
the second query.
The Except operator works with two datasets, which are select queries.
They can be both simple queries, as in our example, and very
complex ones, but they are always just the select queries, nothing more.
What a developer must care about is the columns’ order. It may turn out that in some query he implied one meaning for the column, while in the second query he did a mistake and specified a column with a completely different sense for the same position. Such a situation is worse than an exception, because it’s easy to miss wrong results, and it’s just the time bomb.
Let’s demonstrate such a situation. Suppose we want to get all users that haven’t created any order, and we write this query:
select id, login
from users
except
select o.id, u.login
from orders o
join users u on u.id = o.user_id;
Result:
id login
-- -------
1 JohnDoe
3 Alex
4 Barbie
5 Lisa
6 Stone
Looks good, but it is the incorrect result. The problem here is that in the second query we fetch order id, while we need to get user id. Since both are numbers, it is easy to interpret this result as the right one.
This is the corrected query:
select id, login
from users
except
select o.user_id, u.login
from orders o
join users u on u.id = o.user_id;
Result:
id login
-- -------
1 JohnDoe
3 Alex
6 Stone
INTERSECT
The INTERSECT operator is reversed EXCEPT:
It returns unique (as all set operators) rows that both exist in the first and
the second query.
The following SQL statement returns the logins that appear both in the
guest_orders and users tables:
select login
from guest_orders
intersect
select login
from users;
Result:
login
-------
Alex
JohnDoe
Stone
Subqueries
Subqueries are just the SQL queries that are parts of another SQL query.
Subqueries in a select clause
We can wrap any query in parentheses and put it
in a select list alongside with ordinary columns:
select id,
order_date,
(select sum(quantity) from order_items) items_cnt
from orders o;
Note that subqueries that are placed in a select clause should return only one row and one column or return no data at all. The next query is illegal:
select id,
order_date,
(select oi.order_item, oi.id from order_items oi) items_cnt
from orders o;
Correlated subqueries
Correlated subqueries are the queries that use data from external (relating to them) queries:
select id,
order_date,
(select sum(i.quantity) from order_items i where i.order_id = o.id) items_cnt
from orders o;
Result:
id order_date items_cnt
-- ---------- ---------
1 2021-10-11 2
2 2021-08-15 2
3 2019-04-24 8
4 2019-04-24
For each fetched row from the orders table, the subquery
is executed, and every time it will use the value
of o.id column, which it takes from the “outer” query.
Subqueries as tables
We can use subqueries anywhere in a SELECT query where
we can use tables. For example:
select o.id,
i.item_name,
ord_items.quantity
from orders o
join (
select oi.order_id,
oi.item_id,
oi.quantity
from order_items oi
where quantity >= 2
) ord_items on ord_items.order_id = o.id
join items i on i.id = ord_items.item_id
Result:
id item_name quantity
-- --------------- --------
2 Laptop 2
3 Nintendo switch 4
3 Mobile phone 4
In the example above, we have joined the orders table with
the subquery instead of a table. Then, we treat its data like
if it’s contained in a some table with the ord_items alias.
Here is another example, where we use a subquery as a single source of data:
select a.*
from (
select *
from items
) a
where a.price < 500
Result:
id item_name price
-- --------------- -----
1 Microwave 300
2 Mobile phone 350
3 Nintendo switch 400
Applying all above rules we can nest one subquery into another:
select id,
order_date,
status,
login,
qty
from (
select ord_info.id,
ord_info.order_date,
ord_info.status,
ord_info.login,
(
select sum(oi.quantity)
from order_items oi
where oi.order_id = ord_info.id
) qty
from (
select o.id, o.order_date, o.status, u.login
from orders o
join users u on u.id = o.user_id
) ord_info
) summary;
Result:
id order_date status login qty
-- ---------- ------- -------- ---
1 2021-10-11 CLOSED MrWinner 2
2 2021-08-15 WAITING MrWinner 2
3 2019-04-24 CLOSED Barbie 8
4 2019-04-24 WAITING Lisa
Here we have used the correlated subquery to get items quantity for each order, while data about orders has been obtained by joining two tables - orders and users (and this join has been performed in the subquery). Then, we have wrapped the whole query into one single subquery and have used it like it’s a table. Logically, we can imagine the query above like the next one:
select id,
order_date,
status,
login,
qty
from summary;
Where the summary table is our big outer subquery:
select ord_info.id,
ord_info.order_date,
ord_info.status,
ord_info.login,
(
select sum(oi.quantity)
from order_items oi
where oi.order_id = ord_info.id
) qty
from orders_info ord_info
While the orders_info table can be disassembled into this:
select o.id, o.order_date, o.status, u.login
from orders o
join users u on u.id = o.user_id
Thus, subqueries can be used as build blocks for a more complex query, making it easy to write and understand.