MySQL Posts

Discovered something neat with the new version of MySQL and thought it warranted a mention. Storing tree structures in a relational database is a common use case across many different areas of tech. The problem comes when you need to construct a query based on a subset of that tree. But MySQL 8 has some nice new features that makes doing this a breeze.

Here’s a little quickie for you. Say you have a small MySQL server floating around your house that you want to have regular backups of. You do want regular backups right? In my case, the biggest motivation was wanting a regular way to grab a recent MySQL dump of an internal tool I use at home to develop against. After poking around the Internet a bit, I was surprised that, other than mysqldump itself, there doesn’t seem to be a simple tool out there that you can slam into a cronjob and let it do it’s thing. So, like any good hacker, I decided to brew my own. After all, when you have 256,428 different solutions, why not make solution 256,429? :)

So I was recently helping a client with an issue in MySQL where a migration failed to transfer the full contents of some fields. This amounted to a little over 1% of the total messages transferred. In doing some research, we discovered that the one thing every message had in common was the presence of multi-byte (high unicode) characters. In many cases, this was due to a user pasting some text from Microsoft Word.

So I came across an interesting quirk in MySQL the other day. Let’s say you have a table schema and some values that look like this: +-------------------+------------------+------+-----+---------+-------+ | Field             | Type             | Null | Key | Default | Extra | +-------------------+------------------+------+-----+---------+-------+ | page_id       | varchar(30)      | YES  |     | NULL    |       | | clicks            | int(10) unsigned | YES  |     | NULL    |       | +-------------------+------------------+------+-----+---------+-------+ +---------+--------+ | page_id | clicks | +---------+--------+ | 1 | NULL | +---------+--------+ And then let’s say you pass the following SQL statement to MySQL: update page_click_count set clicks = clicks + 1 where page_id=1; If you come from a loosely-typed language such as PHP, you would probably expect clicks for page_id 1 to now be 1. But that’s not the case in MySQL. After the query is run, the table will still look like this: +---------+--------+ | page_id | clicks | +---------+--------+ | 1 | NULL | +---------+--------+ Not only does the query fail, but it fails with no warnings given. It appears that mathematical operations on null values silently fail. There are a couple of ways around this. The first and most obvious is to set NOT NULL and a default value on the column. In the example above, this would work. The NULL value in that field becomes a 0 and you can to normal mathematical operations on it. But what happens if, for whatever reason, you can’t do that? We actually have this situation in a few places at dealnews, where NULL represents a distinct value of that field that is different from 0. In this case, you can use COALESCE() to fill in the appropriate value for the field. update page_click_count set clicks = coalesce(clicks, 0) + 1 where page_id=1; Edit: Brian Moon informs me that this is actually part of the SQL specification. So hooray for specifications. Still, it’s kind of arcane; in working with MySQL (and PHP) for a decade now, this is the first time I’ve ever actually encountered this. Hopefully this helps someone who was as confused as I was.

In some ways, after years of doing programming and scripting, I’m now sort of rediscovering the power of the shell. Tonight, I was working on my server and remembered that I needed to start backing up my MySQL databases (which you do also … right?). So instead of writing a script to do that, with a little research, I was able to come up with a way to: Dump each database to a separate SQL file, with a timestamp. bzip the file. Keep 5 days worth of backups for each database, rotating the oldest backup off. Here’s what I came up with: cd /backup/mysql; for i in $(mysql -BNe 'show databases' -u root -p<password>); do mysqldump -u root -p<password> $i | bzip2 > $i-`date +"%Y%m%d"`.sql.bz2; rm -rf $i-`date -d "-5 day" +"%Y%m%d"`.sql.bz2; done > /dev/null 2>&1 Shoved that in my crontab. Works great. Linux rocks.

I’ll be attending MySQL Conference in Santa Clara, California this year. This will actually be my first time attending this conference, so I’m looking forward to it. Also, my coworker Brian Moon will be speaking at the conference, “What is memcached and What Does It Do,” so pop in and see him as well!

In working on a side project with a few friendly developers, we decided to set up a Subversion repository and a Trac bug and issue tracker. Both of these, in normal setups, rely on HTTP authentication. So, being that we already had an authentication database as part of the project, my natural first thought was to find a way to authenticate Trac and Subversion of these against our existing MySQL authentication database rather than to rely on Apache passwd files that would have to be updated separately. Surprisingly, this was more difficult than it sounded. My first thought was to try mod_auth_mysql. However, from the front page, it looks as if this project has not been updated since 2005 and is likely not being actively maintained. Nonetheless, I gave it a shot and, surprisingly, got it mostly working against Apache 2.2.14. Notice I said “mostly.” It would authenticate about 50% of the time, while filling the Apache error logs with fun things like: [Sat Feb 13 11:11:27 2010] [error] [client -.-.-.-] MySQL ERROR: Lost connection to MySQL server at 'reading initial communication packet', system error: 0 [Sat Feb 13 11:11:28 2010] [notice] child pid 19074 exit signal Segmentation fault (11) [Sat Feb 13 11:34:14 2010] [error] [client -.-.-.-] MySQL ERROR: Lost connection to MySQL server during query: [Sat Feb 13 11:34:15 2010] [error] [client -.-.-.-] MySQL ERROR: MySQL server has gone away:` Rather than tear into this and try to figure out why a 5-year-old auth module isn’t working against far newer code, and with very little to actually go on, I just concluded that it wasn’t compatible and looked for a different solution. That’s when I came across mod_authnz_external. If your’e not familiar with this module, what it allows you to do is auth against a program or script running on your system, therefore allowing you to auth against anything you want - a script talking to a database, PAM system logins, LDAP, pretty much anything you have access to. All you have to do is write the glue code. In pipe mode, mod_authnz_external uses pwauth format, where it passes the username and password to stdin, each separated with a newline. It uses exit codes to return back to Apache whether or not the login was valid. Knowing that, it’s pretty easy to write a little script to intercept the username/password, run a query, and return the login. #!/usr/bin/php <?php` include "secure_prepend.php"; include "database.php"; $fp=fopen("php://stdin","r"); $username = stream_get_line($fp,1024,"\n"); $password = stream_get_line($fp,1024,"\n"); $sql = "select user_id from users where username='%s' and password='%s' and disabled=0"; $sql = sprintf($sql, $db->escape_string($username), $db->escape_string($password)); $user = $db->get_row($sql); if(!empty($user)) { exit(0); } exit(1); ?> Then, you just hook this into your Apache config for Trac or Subversion: AddExternalAuth auth /path/to/authenticator/script SetExternalAuthMethod auth pipe <Location /> DAV svn SVNPath /path/to/svn AuthName "SVN" AuthType Basic AuthBasicProvider external AuthExternal auth require valid-user </Location> Restart, and it should be all working. Some may argue that the true “right” way to do this is LDAP. But with just three of us, LDAP is overkill, especially when we already have the rest of the database stuf in place. The big advantage to this, even over mod_auth_mysql, is the amount of processing you can do on login. You basically can run any number of queries in your authenticator script - rather than just one. You can update with last login or last commit date, for instance. Or you can join tables for group checking; say you want someone to have access to Trac, but not Subversion. You can do that with this.