Friday, June 20, 2014

Adding newlines in a SQL mysqldump to split extended inserts

The official mysqldump supports more or less two output styles: separate INSERTs (one insert statement per row) or extended INSERTs (one insert per table). Extended INSERTs are much faster, but MySQL write them all in one line, the result being a SQL very hard to read. Can we get the best of both worlds ?

Separate INSERTs

INSERT INTO mytable (id) VALUES (1);
INSERT INTO mytable (id) VALUES (2);

Extended INSERTs

INSERT INTO mytable (id) VALUES (1),(2);

New-And-Improved INSERTs


Current solutions

Using sed

mysqldump --extended-insert | sed 's/),(/),\n(/g'

Only problem is, lines will be split, even in the middle of strings, altering your data.

Using net_buffer_length

mysqldump --extended-insert --net_buffer_length=5000

mysqldump will make sure lines are not longer than 5000 (or whatever), starting a new INSERT when needed. The problem is that the behaviour is kinda random, diffs are hard to analyze and it may break your data if you are storing columns longer than this.

Writing a parser

This question has been often asked without a proper reply, so I decided to write a simple parser. Precisely, we need to check for quotes, parenthesis, and escape characters.

I first wrote it in PHP:

But then I realized it was too slow, so I rewrote it in C, using strcspn to find string occurence:

The only flaw that I can think of is that the parser will fail if the 10001st character of a line is an escaped quote, it will see it as an unescaped quote.

Happy dumping !

Monday, June 16, 2014

Using PHP’s mb_detect_encoding to cleanup your data

Ever heard of iso-8859-1 ? Yeah… that nightmare… With it, my name ends up more often than not… Sébastien. The computers gurus came up one day with UTF-8 and all our problems should have been solved; one encoding to rule them all.

Sweet, let’s all switch to UTF-8 ! Oh wait… legacy projects… PHP internal encoding is still not UTF-8 and functions like strlen() are still not able to properly process multi-bytes strings. It is being said that UTF-8 should land in PHP 6, but in the mean time, we still have to do something.

I am currently working on a big project with a lot of spaghetti-legacy code with a lot of entry points to the database. Almost all the data is stored in one big table, but not encoded uniformly. When I started working on it, tables had fields with a combinaison of  ascii_bin, utf8_general_ci, latin_general_ci and latin_swedish_ci … and we are in Canada ! In all those fields, data was stored with absolutely no guaranty of its encoding. Data was retrieved and passed through a series of UTF-8 encode/decode and stuff like this:

if (strpos($string, 'é') !== false) {
   $string = utf8_decode($string);

I eventually managed to change every field, change all database connections and remove and traces of encode/decode. However, I still had the problem of having data not encoded properly in some rows/columns. You may or may not be familiar with mb_detect_encoding, here is a very simple trick:

Once you have detected the encoding, use iconv to convert it. This crunched through my 1GB database in no time and I was then sure that everything was in UTF-8.

Of course, this is an example with a database, it can work with any data. This script could also be faster if all updates where done at the same time for each row.

Please, save yourself some trouble, make sure all user content is in UTF-8.