Basic Aggregate Functions

How to calculate totals, averages, etc. over multiple rows.

You frequently need to get a single piece of information that is derived from multiple rows in a table. For example, when you need to know the total of all invoices for August 2018:

SELECT sum(total)
   FROM invoices
   WHERE invoice_date BETWEEN
         '2018-08-01' AND '2018-08-31';

The aggregate functions are:

• sum : Calculate the total of the values in a column
• avg : Calculate the average (mean) of the values in a column
• min : Determine the mimimum value of a column
• max : Determine the maximum value of a column
• count : Count the number of values (non-null) in a column

All the above are in the SQL standard, most implementations provide others. SUM and AVG can only apply to numeric data, the others can apply to any datatype.

Further examples:

“What is the average length of stay at our hotel?” :

SELECT avg(checkout_date - checkin_date)
  FROM reservations;

“What are the lowest and highest room rates we charge?” :

SELECT min(rate) AS lowest,
       max(rate) AS highest
  FROM rooms;

You can use the count(x) function to count non-null values:

SELECT count(id) AS id_ct, count(postcode) AS post_ct
  FROM customers;

   id_ct | post_ct
  -------+---------
     133 |     126
  (1 row)

Notice that these two results show different values - there are NULL values for postcode but id is mandatory for all rows.

If you just want to count the number of rows, use count(*). This is often used to find how many rows match a WHERE clause:

SELECT count(*) FROM customers WHERE country = 'Belgium';

Grouping Rows for Aggregation

You can calculate aggregates over subsets of rows using the GROUP BY clause:

SELECT count(*) FROM rooms
  GROUP BY room_type;
  count
  -------
      14
      14
       8
      10
       2
  (5 rows)

What do you notice?

The query calculated the counts correctly but we have no idea which room type each value represents. To solve this we are allowed to include the GROUP BY expressions in the list of selected values, as below:

SELECT room_type, count(*) FROM rooms
  GROUP BY room_type;
 room_type    | count
--------------+-------
 PREMIUM      |    14
 PREMIER      |    14
 PREMIER PLUS |     8
 PREMIUM PLUS |    10
 FAMILY       |     2
(5 rows)

Notice the room_type used for GROUP BY is also included in the SELECT list of values.

We can group by multiple expressions, for example:

SELECT trunc(room_no/100) AS floor,
       to_char(checkin_date, 'YYYY-MM') AS month,
       count(*), sum(no_guests), avg(no_guests)
  FROM reservations
  GROUP BY floor, month;

Notice that the GROUP BY is using the column aliases floor and month that have been defined in the select list. This works in many, but not all, SQL implementations. (In those that don’t allow aliases you must use the full expression, for example: trunc(room_no/100) instead of floor)

You can use a WHERE clause to restrict the rows that are included in the aggregate function. For example, if we need the above query for only the 2nd and 3rd floors:

SELECT trunc(room_no/100) AS floor,
       to_char(checkin_date, 'YYYY-MM') AS month,
       count(*), sum(no_guests), avg(no_guests)
  FROM reservations
  WHERE room_no BETWEEN 200 AND 399
  GROUP BY floor, month;

Note that it is NOT usually possible to use column aliases in the where condition.

A WHERE clause is applied before any aggregation, if you need to restrict results using an aggregate function you can’t do that using the WHERE clause.

In the above, to return only results with the number of reservations greater than, say, 4 we use the HAVING clause:

SELECT trunc(room_no/100) AS floor,
       to_char(checkin_date, 'YYYY-MM') AS month,
       count(*), sum(no_guests), avg(no_guests)
   FROM reservations
   GROUP BY floor, month
   HAVING count(*) > 4;    --<< Note the HAVING keyword

The order of clauses in the SELECT statement is:

SELECT ...
   FROM ...
   [WHERE ...]
   [GROUP BY ...
   [HAVING ...] ]
   [ORDER BY ...]

The square brackets indicate optional clauses. Note that HAVING is only relevant when you have a GROUP BY and must follow it in the SELECT statement.

It can be confusing at first knowing whether to use a WHERE clause or a HAVING clause with GROUP BY.

• Use the WHERE clause when values you want to test are available without having to use any aggregate functions (e.g. plain column values).

• Use HAVING when the values you want to test are the results of aggregate functions (e.g. count(*), sum(amount), min(x), etc…).

Inserting data

To add new data to a table use the INSERT command that has the following format:

INSERT INTO table_name (column_name, ...)
       VALUES (value, ...)

For example:

INSERT INTO customers (name, email, address, city, postcode, country)
  VALUES ('John Smith','j.smith@johnsmith.org',
          '11 New Road','Liverpool','L10 2AB','UK');

Note:

1. You do not need to supply the value for the automatically generated id column, it is populated from a sequence generator object.
2. The order of values in the VALUES (...) clause must correspond to the columns in the column name list. The first value is stored in the first named column, the second value in the second named column and so forth.

Updating Existing Data

When you need to change values in a table, use the UPDATE command. The general construction to update a row is:

UPDATE table
  SET column1 = value1,
      column2 = value2
  WHERE condition;

Note that UPDATE usually requires a WHERE clause to specify the row or rows to be updated. As with SELECT, if you don’t specify a condition to restrict the rows, the command applies to all the rows in the table.

For example, to update the name and country of the customer with ID 3:

UPDATE customers
  SET name='Bob Marley',
      country='Jamaica'
  WHERE id=3;

Deleting a row

The syntax to delete a row is:

DELETE FROM table WHERE condition;

For example, to delete the booking with ID 4:

DELETE FROM reservations WHERE id=4;

Introduction

So far we’ve only looked at one table in any query. Many problems require data from several tables - how do we do that?

For example, if I want to phone or email customers who have not yet paid their invoices, which tables do I need to look at?

Use joins to combine data from more than one table. Joins use column values to match rows in one table to rows in another.

The join columns are usually referred to as foreign keys and primary keys.

Foreign and Primary Keys

Each table should have a Primary Key. This is one or more columns whose values, which cannot be NULL, are combined to provide a unique identifying value for each row. Natural primary keys are often difficult to find so many tables use an arbitrary integer whose value is automatically generated when the row is created. When joining tables we need to match a single row to one or more other rows, usually in another table - for example, matching a customer to her/his reservations. The single row (customer) is usually identified by its primary key value.

Foreign Keys are the columns in a table that reference corresponding columns in another table (although self-referencing foreign keys can reference the same table). For example, the res_id column in the invoices table references the id column in the reservations table (see diagram above).

The referenced column is almost always the primary key of the referenced table because a foreign key must always reference exactly one row in the referenced table (primary keys guarantee that).

Using JOIN in SQL

To join reservations and invoices in SQL:

SELECT r.cust_id, r.room_no, i.invoice_date, i.total
  FROM reservations r JOIN
       invoices i ON (r.id = i.res_id);

Notice:

• The new keyword JOIN with ON (predicate)
• Table aliases (r and i) used to qualify columns

The new syntax follows the following pattern:

SELECT ...
  FROM ... [JOIN ... ON (...)]...
  [WHERE ...]
  [GROUP BY ... [HAVING ...] ]
  [ORDER BY ...]

Use the JOIN to define the combined row source then you can use WHERE, DISTINCT, GROUP BY, ORDER BY, etc… as with single-table queries. For example:

    SELECT r.cust_id, r.room_no, i.invoice_date, i.total
      FROM reservations r JOIN
           invoices i ON (i.res_id = r.id)
      WHERE r.checkin_date > '2018-07-01'
        AND i.total < 500
      ORDER BY i.invoice_date DESC, r.cust_id;

There is no theoretical limit to the number of tables that can be joined in a query, although practical considerations like
complexity and performance must be considered. It is quite common, though, to find up to seven or eight tables joined in a query.

Multi-table joins just extend the syntax to add more tables, as below:

SELECT c.name, c.phone, c.email, i.invoice_date, i.total
  FROM customers c JOIN
       reservations r ON (r.cust_id = c.id) JOIN
       invoices i ON (r.id = i.res_id)
  WHERE i.invoice_date < current_date - interval '1 month'
    AND i.paid = FALSE
  ORDER BY i.invoice_date DESC, c.id;

Note
You have just learned about what is called the INNER JOIN, which is the most common kind of join. Indeed, you can use the keyword INNER in the JOIN syntax, as follows:

SELECT c.name, c.phone, c.email, i.invoice_date, i.total
  FROM customers c INNER JOIN
       reservations r ON (r.cust_id = c.id) INNER JOIN
       invoices i ON (r.id = i.res_id)
  WHERE i.invoice_date < current_date - interval '1 month'
    AND i.paid = FALSE
  ORDER BY i.invoice_date DESC, c.id;

The INNER keyword is not required (it’s the default) but some organisations might require it for the sake of coding standards.

There are other kinds of JOIN, specifically the OUTER JOIN and the CROSS JOIN but these are less frequently used in applications.
If you want to find out about these kinds of JOIN refer to the PostgreSQL documentation.

Exercise 5

1. Try and understand each of the queries above in your psql prompt
2. Which customers occupied room 111 and what are their details?
3. List the customer name, room details (room number, type and rate), nights stay and departure dates for all UK customers.
4. List name, phone and email along with all reservations and invoices for customer Mary Saveley.

The Vexing Question of NULL

A column can be assigned a NULL value to indicate it has no value. This can happen when the data for this column is unknown at the time the row is created, for example, employee leaving date, order shipment date, etc… It can also be used when the data is optional.

Be careful with expressions - any expression that includes a NULL value results in NULL as the expression value.

Because NULL is ’no value’ it cannot be compared to anything else. For example, you will never get any results from:

SELECT * FROM customers WHERE postcode = NULL;

nor will you get any from:

SELECT * FROM customers WHERE postcode != NULL;

Instead you must use:

  ... WHERE postcode IS NULL

or

  ... WHERE postcode IS NOT NULL

This behaviour has some impacts on operations like JOIN, where NULL values won’t match. You could work around this, but see the warning below, by using:

  ... ON (a.col = b.col OR
          a.col IS NULL AND b.col IS NULL)

WARNING:
However, be aware that this is not a sensible situation - join columns containing NULL should be expected to not match or should be disallowed (see Primary Keys above)

You can explicitly provide NULL as a value in INSERT and UPDATE statements, for example:

    INSERT INTO rooms (room_no, rate, room_type, no_guests)
      VALUES (213, 95.00, NULL, 2);

    UPDATE rooms SET room_type = NULL, no_guests = NULL
      WHERE room_no = 204;

In INSERT statements if you omit a column from the column list (following the table name) then that column will be given either:

• an autogenerated value (if it has datatype SERIAL)
• a default value if one has been specified in the CREATE TABLE command
• NULL if neither of the above apply

Functions to Handle NULL

There are some functions that can operate on NULL values, especially the coalesce(x, y) function. This function looks at the first argument x and if it is NULL returns the value of the second argument y otherwise it returns the value of x. For example:

SELECT room_no, rate, coalesce(room_type, 'None') type
  FROM rooms
  WHERE no_guests IS NULL;

Notes:

• The coalesce function can take more than two arguments and returns the first of these (from left to right) that is not null.
• This feature is provided by most SQL vendors but goes by different names, e.g. ifnull(x, y) in MySQL, nvl(x, y) in Oracle, etc…

Exercise 6

1. Which customers have not yet provided a phone number?
2. Update room 304 such that it does not have a room_type.
3. List customers (name and city) qand their reservations replacing the room number with ‘Not Assigned’ if it is NULL.

Use the CREATE TABLE command, which in the simplest case has the general form:

CREATE TABLE <tablename> <column definition>, <column definition>, ...);

To create an inventory table for our hotel we might need:

CREATE TABLE inventory (
  id            SERIAL PRIMARY KEY,
  description   VARCHAR(30) NOT NULL,
  cost          NUMERIC(6,2)
);

Note: you may never need to do this. Database design is a task that requires specialist skills and considerable experience.

Naming Tables and Columns

In the CREATE TABLE command you must give the name of the table (e.g. inventory) and the names of each of the columns (in the parenthesised column definitions) (e.g. id, description, cost).

Names of tables and columns (and any other objects in the database) must start with a letter, can contain letters, digits and the underscore symbol (_) up to 64 bytes (in PostgreSQL). Names are not case-sensitive so that NAME, name and NaMe are all the same.

Data Types of Columns

In the above example:

The id column uses SERIAL as its data type, making it an autoincrementing integer that increments by 1, starting from 1, for each time a new row is inserted into the table. For this to work, the id column must be omitted from the INSERT command. id is also designated as the PRIMARY KEY of the table (note that SERIAL doesn’t make the column the primary key). PRIMARY KEY also implies that the column cannot be set to NULL.

The description column is a variable length character value (VARCHAR) that can hold up to a maximum of 30 characters. The NOT NULL constraint means the value cannot be left empty, each row must have a description.

The cost column is NUMERIC(6,2), a number that can accurately store up to 6 digits, two of which are the fractional part. For example, it can hold 1234.56, -23.45 or 0.01. Note that the NUMERIC data type stores and computes values using decimal values and does not lose accuracy in the same was as, say, floating point values. NUMERIC values take longer in calculations because they don’t use simple binary values - user either integer or floating point for speed with compute-heavy numbers.

NEVER use floating point for financial values.

Other Common Data Types

There are several more standard data types (plus a few non-standard ones), including:

You can read more about data types in the PostgreSQL documentation. Refer to https://www.postgresql.org/docs/12/datatype.html
for more information.

Changing a Table Definition

Using the ALTER TABLE command to add and remove columns:

ALTER TABLE inventory ADD COLUMN room_no INTEGER;

ALTER TABLE customers DROP COLUMN phone;

There are some constraints on adding and removing columns, for example, you cannot add a NOT NULL column to a table that already contains some rows.

Exercise 7

1. Create a table for charge points. This must record the hotel shops, bars, cafes and restaurants that a customer can use during their stay.
2. Include an auto-incrementing primary key
3. Include the charge point name, a description and maximum customer credit value
4. Insert charge points for ‘Shop’, ‘Pool Bar’, ‘Elysium Restaurant’ and ‘Room Service’ with credit limits of £1000 for each.
5. Create a table for charge items that records amounts charged to rooms by customers using our shop, bars, restaurants, etc. This must include the room number of the room charged, the charge point used, the amount, the date and time of the charge and any customer comments.

Defining Primary and Foreign Keys

Defining Primary Keys

Use the following templates to define a Primary Key.

For a single-column PK use:

CREATE TABLE <table name> (
  ...
  <column name>   <data type>   PRIMARY KEY,
  ...
)

For example:

CREATE TABLE rooms (
  room_no       INTEGER   PRIMARY KEY,
  ...
);

To define a multi-column primary key you must define a constraint separate from the column definitions, as below:

CREATE TABLE <table name> (
  ...
  <pk col 1>     <data type>,
  <pk col 2>     <data type>,
  ... ,
  PRIMARY KEY (<pk col 1>, <pk col 2>),
  ...
);

For example:

CREATE TABLE invoice_items (
  inv_id        INTEGER REFERENCES invoices(id),
  item_no       INTEGER,
  ... ,
  PRIMARY KEY (inv_id, item_no),
  ...
);

There can be only one primary key in a table definition. The PRIMARY KEY definition implies NOT NULL so no column in a table’s PK can be set to NULL.

Note: a partial primary key can be a foreign key as well.

Defining Foreign Keys

To define foreign keys use either:

For a single-column foreign key:

  <column name>   <data type>   REFERENCES <table name> (<column name>);

where the <column name> in the REFERENCES clause is the column name in the referenced table, not the one being defined at this point. For example, in the reservations table:

  ...
  cust_id         INTEGER NOT NULL   REFERENCES customers (id),
  ...

For multi-column foreign keys we must again use a separate constraint definition, as shown:

CREATE TABLE customer_challenges (
  id           SERIAL PRIMARY KEY,
  inv_id       INTEGER,
  item_no      INTEGER,
  ...
  FOREIGN KEY (inv_id, item_no) REFERENCES invoice_items (inv_id, item_no),
  ...
);

Exercise 8

1. Try to delete the customer Mary Saveley. What happens and why?
2. Insert a new room, number 313 as room type ‘SUPER PREMIER’.
3. Define primary and foreign keys required by the charge_items table
4. Insert some rows into the charge_items table. You can invent the details.

Introduction to node-postgres

“node-postgres is a collection of node.js modules for interfacing with your PostgreSQL database.” - https://node-postgres.com/

In the following, we will use node-postgres to…

1. Connect to a database
2. Send SQL queries to the database and get results

Loading data from a database with a GET endpoint

Let’s build a brand new NodeJS application with a single GET endpoint to load the list of customers that you already have in the customers table of the cyf_hotels database.

First, create a new NodeJS application that we will call cyf-hotels-api (enter server.js when asked for the entry point):

mkdir cyf-hotels-api
cd cyf-hotels-api
npm init

As before, we will use the Express library to build our API, and the node-postgres library to connect with our database:

npm install --save express
npm install --save pg

Create a server.js file, import express, initialise the server and start listening for requests:

const express = require("express");
const app = express();

app.listen(3000, function () {
  console.log("Server is listening on port 3000. Ready to accept requests!");
});

Import pg library and create a new GET endpoint to load the list of customers:

const { Pool } = require("pg");

const db = new Pool({
  user: "keith", // replace with you username
  host: "localhost",
  database: "cyf_hotels",
  password: "",
  port: 5432,
});

app.get("/customers", function (req, res) {
  db.query("SELECT id, name, city, phone FROM customers")
    .then((result) => {
      res.json(result.rows);
    })
    .catch((error) => {
      console.log(error);
    });
});

In the code above:

• We first import the Pool class from the pg library, which is used to connect to a database
• We create a new connection (db) where we specify the credentials to connect to the cyf_hotel database
• We then create a new /customers endpoint where we use the method query() to send a SQL query to load all the customers from the table customers and return the results with result.rows. The query method returns a Promise: so we can access the returned rows using a .then block. You can write any valid SQL query that you learned in the query() method!

Note: There is no semicolon to end the SQL statement.

Start your server with node server.js and try to reach the /customers endpoint to see the list of customers currently available in your cyf_hotels database. You can try to create/update/delete customers using psql to verify that your API always returns what is stored in your database.