Paulo Phagula

Musings and Scribbles on Software Development

A few MySQL subtleties and how to go about them

MySQL is the most widely used database in the world. It is in the LAMP stack commonly used by Web Developers and supports many software bundles like WordPress and Drupal which in turn support most of the sites in the internet.

Its users range from the one-man-band-man hard-core developers writing code in their basements, to the super-mega enterprises like Facebook, Github, Google, and Wikipedia just to name a few.

(Almost everyone uses it — that’s what I want to say)

Now, despite its very large user base, there are still some misconceptions/subtleties about how to use it and properly setup, that usually go unnoticed until its too late, or that are initially unusual/unexpected by people (like me) used to other database products. At large, this is due to ignorance on developers (which commonly have to act as DBAs), but mostly to MySQL itself which acts in a very uncommon way — read insecure, non-ANSI compliant, shit-show like style — by default.

The fact that it’s so easy to setup and use, allows almost anyone, beginner or not, to just fire it and rock on.

This post is not a rant about MySQL nor a promotion of <insert-your-favorite-db-here> — I’ll do my best to control my emotions, I promise!. Its about sharing details on a set of subtleties I found in my experience (coming from SQL Server and Oracle), and how you can go about them so you’re not caught off guard as I was.

Division by zero equals NULL

Here we go

mysql> SELECT 1/0;
+------+
| 1/0  |
+------+
| NULL |
+------+
1 row in set (0.00 sec)

OMG. Imagine that is somewhere deep in the middle of your code. Incorrect reports, incorrect business decisions taken, kittens dying.

How could you avoid this? Set your MySQL mode to ERROR_FOR_DIVISION_BY_ZERO and STRICT_ALL_TABLES. Now, whenever you do that an error will be produced instead. Let’s try it.

mysql> SET SESSION sql_mode = 'ERROR_FOR_DIVISION_BY_ZERO,STRICT_ALL_TABLES';
Query OK, 0 rows affected, 1 warning (0.00 sec)

mysql> SELECT @@sql_mode;
+----------------------------------------------+
| @@sql_mode                                   |
+----------------------------------------------+
| STRICT_ALL_TABLES,ERROR_FOR_DIVISION_BY_ZERO |
+----------------------------------------------+
1 row in set (0.00 sec)

mysql> SELECT 1/0;
+------+
| 1/0  |
+------+
| NULL |
+------+
1 row in set, 1 warning (0.00 sec)

"What? WTF man, that’s just a warning, I expected an error!" You say. I hear you. Let’s look at this warning

mysql> show warnings;
+---------+------+---------------+
| Level   | Code | Message       |
+---------+------+---------------+
| Warning | 1365 | Division by 0 |
+---------+------+---------------+
1 row in set (0.00 sec)

Hm, it seems like the SQL mode just produces warnings, but let’s try it with a table

mysql> CREATE TABLE test(value int);
Query OK, 0 rows affected (0.04 sec)

mysql> INSERT INTO test(value) VALUES(1/0);
ERROR 1365 (22012): Division by 0

It seems SQL mode won’t cut it all times. Simple SELECT statements will not be fully covered for these kinds of errors, only your table data (what you INSERT and UPDATE) will.

Let’s confirm this by looking at another example with SELECT s and table data.

mysql> SELECT count(*)/0 FROM test;
+------------+
| count(*)/0 |
+------------+
|       NULL |
+------------+
1 row in set, 1 warning (0.01 sec)

mysql> SELECT value/0 FROM test WHERE value=1;
+---------+
| value/0 |
+---------+
|    NULL |
+---------+
1 row in set, 1 warning (0.00 sec)

That’s it indeed, for SELECT s all we get is a warning, but at least we’re covered from data corruption. It is not perfect but should help a bit, also if your database interface allows, you can tell it to convert the warnings into errors in which case you would get "full protection". In the absence of that feature a technique that I’ve seen some people use is to try to cover for the zero case by using IF s or NULLIF s, more or less like so:

SELECT 1/nullif(some_column, 0); -- returns null
-- OR
SELECT 1/if(some_column = 0, 1, some_column); -- returns 1

I do not like this technique as I have to remember to do that, plus it makes the query ugly. But since practicality speak lauder than my tastes, ultimately, I have to go with the less worse solution, which in this case seems to be the last one.

this is just an example, and setting the SQL mode for the session suffices, but in a real world scenario you should set your SQL mode on the server’s configuration file, so it affects every single connection and the settings can persist after reboots.

'' = 0

mysql> SELECT ''=0;
+------+
| ''=0 |
+------+
|    1 |
+------+
1 row in set (0.01 sec)

It could be said that this is not a problem since many programming languages do this. But the thing is that SQL is not a programming language (each database vendor adds their own procedural extensions to SQL to make it be more like a programming language and cater for the "limitations" of pure SQL) and furthermore my expectations about how things should work in the database are completely different from how they should work in a programming language.

Anyhow, the kicker is that the behavior is inconsistent when compared with most programming languages, particularly when the operands for the equality operator are numbers and strings. See the following sample

mysql> SELECT 'password'=0;
+--------------+
| 'password'=0 |
+--------------+
|            1 |
+--------------+

Two values of different data types, one falsy and another truthy are being compared somehow and are considered equal. I’d get it if we were comparing a string of numbers like '1' with a number like 1, but this…​ this is weird.

Unfortunately, I no longer remember exactly what was the case, but I’ve had a situation in the past where this caused me to waste hours to figure out. All I recall is there was a simple mistake of swapping the values for the fields in the WHERE clause, which caused the query to produce correct results sometimes but fail unpredictably at others.

What is the way around this? Being careful and minding warnings.

Zeros in dates and timestamps '0000-00-00 00:00:00'

This is another weird thing about MySQL, it allows for invalid dates containing zeroes. There are claims for legitimate good cases for having this "feature", but perhaps I haven’t lived long enough to see one just yet. Regardless, the situation is the one bellow:

mysql> CREATE TABLE test (birth_day date, created_at datetime);
Query OK, 0 rows affected (0.02 sec)

mysql> INSERT INTO test VALUES('0000-00-00', '0000-00-00 00:00:00');
Query OK, 1 row affected (0.00 sec)

mysql> INSERT INTO test VALUES('2000-10-00', '0000-00-00 19:30:00');
Query OK, 1 row affected (0.00 sec)

mysql> SELECT * FROM test;
+------------+---------------------+
| birth_day  | created_at          |
+------------+---------------------+
| 0000-00-00 | 0000-00-00 00:00:00 |
| 2000-10-00 | 0000-00-00 19:30:00 |
+------------+---------------------+
2 rows in set (0.00 sec)

Suppose you’re called in to analyze this data. How would you interpret it?

This seems like a bad usage of the typing system. If we’re going to represent missing data why not simply use NULL, since that is precisely what it is for?!

In the same token, its contradictory to mandate that a field be NOT NULL, but then go and keep invalid values on it. We would be respecting the constraint but at the expense of littering data with insignificant and hard (if at all) interpretable values.

I remember working on an HR system where zeroes where allowed in the dates. Whenever a date was missing, 0000-00-00 was used instead, and as a result queries for computing the candidates experience would bring inconsistent results.

How to avoid this? Set SQL mode to include STRICT_ALL_TABLES, NO_ZERO_DATE and NO_ZERO_IN_DATE, so that it complains appropriately upon the presence of incorrect date values. Let’s try it:

mysql> SET SESSION sql_mode = 'NO_ZERO_DATE,NO_ZERO_IN_DATE,STRICT_ALL_TABLES';
Query OK, 0 rows affected, 1 warning (0.00 sec)

mysql> INSERT INTO test VALUES('2000-10-00', '0000-00-00 19:30:00');
ERROR 1292 (22007): Incorrect date value: '2000-10-00' for column 'birth_day' at row 1
You must combine all these 3 sql modes. Without strict mode MySQL will still behave incorrectly and raising warning but ultimately no protection is provided. See bellow:
mysql> SET SESSION sql_mode = 'NO_ZERO_DATE,NO_ZERO_IN_DATE';
Query OK, 0 rows affected, 1 warning (0.00 sec)

mysql> INSERT INTO test VALUES('2000-10-00', '0000-00-00 19:30:00');
Query OK, 1 row affected, 2 warnings (0.00 sec)

mysql> SELECT * FROM test;
+------------+---------------------+
| birth_day  | created_at          |
+------------+---------------------+
| 0000-00-00 | 0000-00-00 00:00:00 |
+------------+---------------------+
1 row in set (0.00 sec)

Note how not only it "simply" raised warnings, but it also replaced our values with zeros, which is much worse than what we had to begin with.

TIMESTAMP vs. DATETIME

Some people unknowingly use these data types as if they were synonymous, but in reality they’re different and appropriate for different usage scenarios. The sample bellow should clarify what I mean:

SET SESSION time_zone='+2:00';

CREATE TABLE dates (
    date_timestamp timestamp,
    date_datetime datetime
)

-- Inserting the exact same value to both columns
INSERT INTO dates (date_timestamp, date_datetime) VALUES ('2017-07-09 20:11:00', '2017-07-09 20:11:00');

mysql> SELECT * FROM dates;
+---------------------+---------------------+
| date_timestamp      | date_datetime       |
+---------------------+---------------------+
| 2017-07-09 20:11:00 | 2017-07-09 20:11:00 |
+---------------------+---------------------+
1 row in set (0.00 sec)

SET SESSION time_zone='+4:00';

mysql> SELECT * FROM dates;
+---------------------+---------------------+
| date_timestamp      | date_datetime       |
+---------------------+---------------------+
| 2017-07-09 22:11:00 | 2017-07-09 20:11:00 |
+---------------------+---------------------+
1 row in set (0.00 sec)

MySQL converts TIMESTAMP values from the current time zone to UTC for storage, and back from UTC to the current time zone for retrieval. (This does not occur for other types such as DATETIME.) By default, the current time zone for each connection is the server’s time. The time zone can be set on a per-connection basis. As long as the time zone setting remains constant, you get back the same value you store. If you store a TIMESTAMP value, and then change the time zone and retrieve the value, the retrieved value is different from the value you stored. This occurs because the same time zone was not used for conversion in both directions. The current time zone is available as the value of the time_zone system variable.

Which one to use? It depends on your situation and needs. Let that guide your choices and you should be fine.

UTF8 is not UTF8 aka Can you INSERT 💩?

This is serious, can you INSERT 💩 in your table? …​No?

Why not? It’s UTF-8 right? I saw you doing the CHARSET thing when you created your table…​

To be fair, encodings, unicode, character sets and collations make my head hurt and I’m not a smart guy so I’ll just give you the bottom line and refer to a place where you can know more.

Bottom line is: if you created a table with CHARSET uft8 then it won’t work with 💩, that is, you’re not supporting all characters in unicode, and so people cannot leave emojis on comments, or write asian kanjis or characters, on your site/app. This is because UTF8 (the real one) is utf8mb4, not utf8 as is said in many places on the internet.

Let’s do the test.

-- Lets try uft8

CREATE DATABASE test;

USE test;

CREATE TABLE poo_utf8 (
    contents varchar(191)
) CHARSET=utf8 COLLATE=utf8_unicode_ci;

INSERT INTO poo_utf8(contents) VALUES ('big ol pile of 💩');
Query OK, 1 row affected, 1 warning (0.01 sec)

Oh, lovely…​ let’s query it then

mysql> SELECT * FROM poo_utf8;
+------------------+
| contents         |
+------------------+
| big ol pile of ? |
+------------------+
1 row in set (0.01 sec)

What the 💩? Where is my * 💩? Where did it go?

Told you

-- Now let's try utf8mb4
CREATE TABLE poo_utf8mb4 (
    contents varchar(191)
) CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci;

INSERT INTO poo_utf8mb4(contents) VALUES ('big ol pile of 💩');

SELECT * FROM poo_utf8mb4;
mysql> SELECT * FROM poo_utf8mb4;
+---------------------+
| contents            |
+---------------------+
| big ol pile of 💩     |
+---------------------+
1 row in set (0.00 sec)

(sigh) There’s my lovely 💩.

Don’t let anyone take your 💩. Use utf8mb4 and utf8mb4_unicode_ci.

Of course 💩 was just an example. If you want to support any character you need to use "proper" UTF8.

BTW on MySQL 8 this is going to be the default, but we all know everyone must do ceremonies and rituals prior to migrating, so…​

To know the exact details of why this is so check out https://mathiasbynens.be/notes/mysql-utf8mb4

The CHECK constraint is only parsed but ignored in the end

This has been in MySQL since forever and not even MySQL 8 will fix it. MySQL parses the CHECK constraints when defining tables but it doesn’t enforce them. They’re just there but do nothing.

The example bellow depicts the behavior. On it we imagine defining a people table. Its just supposed to keep the id, name and gender of for each person. In order to save a bit of space we want to constrain the value that can go in gender to m or f, standing for male and female. We use the CHECK constraint for it.

CREATE TABLE people (
  id INT NOT NULL PRIMARY KEY auto_increment,
  name varchar(100),
  gender char(1) CHECK (gender IN ('m', 'f'))
);

mysql> INSERT INTO people (NAME, gender) VALUE ('Paulo', 'h');
Query OK, 1 row affected (0.00 sec)

mysql> SELECT * FROM people;
+----+-------+--------+
| id | name  | gender |
+----+-------+--------+
|  1 | Paulo | h      |
+----+-------+--------+
1 row in set (0.00 sec)

As you saw all went well, except MySQL didn’t cry when I said my gender was h --for human.

How to go about this? Know that CHECK constraints in MySQL are just for show. You’ll need to find another way instead, perhaps triggers or some code in your application.

Aggregations without GROUP BY

The problem with allowing this kind of stuff is with the results provided. Let’s suppose we have some table to keep records of candidates, those candidates for something…​ say vacancies. The candidates can go change through various stages from registered to hopefully (selected).

Imagine I want to get the total of candidates per each category. I could (mistakenly) go with a query like SELECT flow_status, count(*) FROM candidate, missing the GROUP BY.

mysql> SELECT flow_status, count(*) FROM candidate;
+-------------+----------+
| flow_status | count(*) |
+-------------+----------+
| registered  |    10761 |
+-------------+----------+
1 row in set (0.01 sec)

This result is very likely to be wrong. Given that I’m using aggregate functions without grouping I’m going to get just one record back, with the count for all the candidates but only one of flow_status (the first in this case).

It could be said that this is a fault on whoever wrote the query, but I disagree this query should not have been allowed to run in the first place. The parser should’ve rejected it.

There are legitimate cases for using aggregations without grouping, but only aggregations should be allowed then. A basic example would be to know the average height of the candidates and the count fo them. There’s nothing wrong with that. But as soon you an aggregations and non-aggregations without grouping then it is very likely you have problem.

The cure for this problem is the same as for the next point. So, just keep going.

Non-GROUPed-BY nor aggregated columns in SELECT

This is stopped by default starting from MySQL 5.7 as seen these reference articles:

However, for those using versions or that do not their settings right, bellow follows an example of what I mean.

CREATE TABLE invoice_line_items (
    id INT NOT NULL PRIMARY KEY auto_increment,
    invoice_id INT NOT NULL,
    description varchar(100)
);

INSERT INTO invoice_line_items VALUES
    (NULL, 1, 'New socks'),
    (NULL, 1, 'A hat'),
    (NULL, 2, 'Shoes'),
    (NULL, 2, 'T shirt'),
    (NULL, 3, 'Tie');

mysql> SELECT id, invoice_id, description FROM invoice_line_items GROUP BY invoice_id;
+----+------------+-------------+
| id | invoice_id | description |
+----+------------+-------------+
|  1 |          1 | New socks   |
|  3 |          2 | Shoes       |
|  5 |          3 | Tie         |
+----+------------+-------------+

Because MySQL doesn’t enforce the usage the correct behavior of GROUP BY we can easily return incorrect data by accident, such as above. Luckily this behavior has been corrected by default since version 5.7 with the mode ONLY_FULL_GROUP_BY. Setting your SQL mode to include it sort things out for you you.

Data Truncations

MySQL tends to do data truncations whenever a value doesn’t fit a column. Like in the other cases the sin is the silent warning. Let look at an example.

mysql> CREATE TABLE foo (bar VARCHAR(4));
Query OK, 0 rows affected (0.00 sec)

mysql> INSERT INTO foo (bar) VALUES ("12345");
Query OK, 1 row affected, 1 warning (0.00 sec)

mysql> SHOW WARNINGS;
+---------+------+------------------------------------------+
| Level   | Code | Message                                  |
+---------+------+------------------------------------------+
| Warning | 1265 | Data truncated for column 'bar' at row 1 |
+---------+------+------------------------------------------+

Yap, your data was truncated, just like that! And this also happens on

ALTER TABLE foo MODIFY COLUMN bar VARCHAR(2);

You can make MySQL do the right thing by setting the SQL Mode option to include STRICT_TRANS_TABLES or STRICT_ALL_TABLES. The difference is that the former will only enable it for transactional data storage engines. As much as I’m loathed to say it, I don’t recommend using STRICT_ALL_TABLES, as an error during updating a non-transactional table will result in a partial update, which is probably worse than a truncated field. Setting the mode to TRADITIONAL includes both these and a couple of related ones (NO_ZERO_IN_DATE, NO_ZERO_DATE, ERROR_FOR_DIVISION_BY_ZERO)

MySQL uses separate encoding for different parts

In MySQL these components have different encodings:

  • Server

  • Client

  • Connection

  • Database

  • Table

  • Field

mysql> \s
--------------
mysql  Ver 14.14 Distrib 5.7.15, for osx10.11 (x86_64) using  EditLine wrapper

Connection id:      6
Current database:   soma
Current user:       root@localhost
SSL:            Not in use
Current pager:      less
Using outfile:      ''
Using delimiter:    ;
Server version:     5.7.15 MySQL Community Server (GPL)
Protocol version:   10
Connection:     127.0.0.1 via TCP/IP
Server characterset:    utf8mb4
Db     characterset:    utf8
Client characterset:    utf8mb4
Conn.  characterset:    utf8mb4
TCP port:       3306
Uptime:         1 hour 32 min 20 sec

Threads: 5  Questions: 289  Slow queries: 0  Opens: 166  Flush tables: 1  Open tables: 159  Queries per second avg: 0.052
--------------
 CREATE TABLE `demo_encoding` (
  `username` varchar(20) CHARACTER SET latin1 DEFAULT NULL
) ENGINE=InnoDB DEFAULT CHARSET=greek

There’s nothing wrong with the above. In fact it is a feature and I’ve had good legitimate cases for having two different encodings in use for different parts. It’s more a good to know thing, as to improve your decisions on configuration, and helping in managing your expectations upon the server’s behavior regarding this matter.

Booleans are synonymous with Tiny Integers

In MySQL Booleans are synonymous with tiny integers, actually to be precise BOOL s are aliases for TINYINT s. Let’s have a look at the sample code

mysql> CREATE TABLE things (is_fit BOOL);
Query OK, 0 rows affected (0.11 sec)

mysql> DESC things;
+--------+------------+------+-----+---------+-------+
| Field  | Type       | Null | Key | Default | Extra |
+--------+------------+------+-----+---------+-------+
| is_fit | tinyint(1) | YES  |     | NULL    |       |
+--------+------------+------+-----+---------+-------+
1 row in set (0.00 sec)

mysql> SHOW CREATE TABLE things;
+--------+---------------------------------------------------------------------------------------------------+
| Table  | Create Table                                                                                      |
+--------+---------------------------------------------------------------------------------------------------+
| things | CREATE TABLE `things` (
  `is_fit` tinyint(1) DEFAULT NULL
) ENGINE=InnoDB DEFAULT CHARSET=latin1 |
+--------+---------------------------------------------------------------------------------------------------+
1 row in set (0.01 sec)

This is not terrible awful, but if the idea is to save some space why not just use BIT(1)? This would allow us to save space and is more strict, it only allows 1 for true and 0 for false.

Using TINYINT(1) can allow from -128 and 127, and a funny guy can set the field value to something else other than the expect 1 or 0.

Bonus: VARCHAR(255) Obsession

I’ve seen this one over and over and over …​ heck, done it myself …​ anyway

Stop already with VARCHAR(255) Obsession. The thing can go up to 65535. Yap VARCHAR(65535) FTW. Way better than the TEXT you use for most cases.

Only CHAR has a limit of 255.

The limit for VARCHAR was lifted from 255 on MySQL 5.0.3. And it is 21844 when using UTF-8 (the real UTF8 not UTF8 :)).

Why keep MySQL then?

Given all these subtleties why keep MySQL instead of going with something else, seems to be a legitimate question. After all some people may be used to the level of strictness of some other vendor’s products and may not tolerate these "things". But just as in every other technological decision there are many other factors and forces involved — dark and light. And in this particular case I think there are more light forces than dark. Though these are my personal reasons for keeping it I think they will resonate with you:

  • Ecosystem (WordPress, Drupal, <insert-your-favorite-cms-here> )

  • It’s open source, free, has a great community, and lots of resources to learn from on the web.

  • It’s not that bad, you just have to educate yourself, discipline it by setting the SQL mode to be reasonable default server wise, and start taking the warnings it produces seriously. Whenever you MySQL says there was a warning, SHOW it as to better decide how to proceed from there.

I promised not to make comparisons with other products, but I’d be remiss if I didn’t mention that if these issues are really bothering you then perhaps you can try a different distribution of MySQL other than the default one, which usually have more saner defaults out of the box, like: Percona Server, MariaDB, Drizzle or WebScaleSQL. They’re are MySQL after all.


I hope this has been informative for you. Drop a comment bellow if you found something fishy, agree with my views, or have something to add.



Please, use syntax highlighting in your comments, to make them more readable.