This article contains my notes and detailed instructions on setting up a MySQL cluster. After reading it, you should have a good understanding of what a MySQL cluster is capable of, how and why it works, and how to set one of these bad boys up. Note that I'm primarily a developer, with an interest in systems administration but I think that every developer should be able to understand and set up a MySQL cluster, at least to make the dev environment more robust.

Notes

• In short, a MySQL cluster allows a user to set up a MySQL database shared between a number of machines. Here are some benefits:
  • High availability. If one or some of the machines go down, the cluster will stay up, as long as there is at least one copy of all data still present. The more redundant copies of data there are, the more machines you can afford to lose.
  • Scalability. Distributed architecture allows for load balancing. If your MySQL database is getting hit with lots of queries, consider setting up a cluster to spread this load in almost linear fashion. A 4 node cluster should be able to handle twice as many queries as a 2 node cluster.
  • Online backups.
  • Full support for transactions.
• Must-have manual: MySQL Clustering by Alex Davies and Harrison Fisk, MySQL Press.
• First and foremost, I would like to get this out of the way (from MySQL Clustering):
  • Response time with MySQL Cluster is quite commonly worse than it is with the traditional setup. Yes, response time is quite commonly worse with clustering than with a normal system. If you consider the architecture of MySQL Cluster, this will begin to make more sense.
  • When you do a query with a cluster, it has to first go to the MySQL server, and then it goes to storage nodes and sends the data back the same way. When you do a query on a normal system, all access is done within the MySQL server itself. It is clearly faster to access local resources than to read the same thing across a network. Response time is very much dependant on network latency because of the extra network traffic. Some queries may be faster than others due to the parallel scanning that is possible, but you cannot expect all queries to have a better response time.
  • So if the response time is worse, why would you use a cluster? First, response time isn't normally very important. For the vast majority of applications, 10ms versus 15ms isn't considered a big difference.
  • Where MySQL Cluster shines is in relation to the other two metrics: throughput and scalability.
• A typical MySQL cluster setup involves 3 components in at least this configuration:
  • 1 management (ndb_mgmd) node.
    • Management nodes contain the cluster configuration.
    • A management node is only needed to connect new storage and query nodes to the cluster and do some arbitration.
    • Existing storage and query nodes continue to operate normally if the management node goes down.
    • Therefore, it's relatively safe to have only 1 management node running on a very low spec machine (configuring 2 management nodes is possible but is slightly more complex and less dynamic).
    • Interfacing with a management node is done via an ndb_mgm utility.
    • Management nodes are configured using config.ini.
    • My setup here involves 1 management node.
  • 2 storage (ndbd) nodes.
    • You do not interface directly with those nodes, instead you go through SQL nodes, described next.
    • It is possible to have more storage nodes than SQL nodes.
    • It is possible to host storage nodes on the same machines as SQL nodes.
    • It is possible, although not recommended, to host storage nodes on the same machines as management nodes.
    • Storage nodes will split up the data between themselves automatically. For example, if you want to store each row on 2 machines for redundancy (NoOfReplicas=2) and you have 6 storage nodes, your data is going to be split up into 3 distinct non-intersecting chunks, called node groups.
    • Given a correctly formulated query, it is possible to make MySQL scan all 3 chunks in parallel, thus returning the result set quicker.
    • Node groups are formed implicitly, meaning you cannot assign a storage node to a specific node group. What you can do, however, is manipulate the IDs of the nodes in such a way that the servers you want will get assigned to the node groups you want. The nodes having consecutive IDs get assigned to the same node group until there are NoOfReplicas nodes in a node group, at which point a node group starts.
    • Storage nodes are configured using /etc/my.cnf. They are also affected by settings in config.ini on the management node.
    • My setup here involves 4 storage nodes.
  • 2 query (SQL) nodes.
    • SQL nodes are regular mysqld processes that access data in the cluster. You guessed it right – the data sits in storage nodes, and SQL nodes just serve as gateways to them.
    • Your application will connect to these SQL node IPs and will have no knowledge of storage nodes.
    • It is possible to have more SQL nodes than storage nodes.
    • It is possible to host SQL nodes on the same machines as storage nodes.
    • It is possible, although not recommended, to host SQL nodes on the same machines as management nodes.
    • SQL nodes are configured using /etc/my.cnf. They are also affected by settings in config.ini on the management node.
    • My setup here involves 4 SQL nodes.
• Normally a cluster doesn't want to start if not all the storage nodes are connected (from MySQL Clustering).
  • Therefore, the cluster waits longer during the restart if the nodes aren't all connected so that the other storage nodes can connect. This period of time is specified in the setting StartPartialTimeout, which defaults to 30 seconds. If at the end of 30 seconds, a cluster is possible (that is, it has one node from each node group) and it can't be in a network partitioned situation (that is, it has all of one node group), the cluster will perform a partial cluster restart, in which it starts up even though storage nodes are missing.
  • If the cluster is in a potential network partitioned setup, where it doesn't have all of a single node group, then it will wait even longer, with a setting called StartPartitionedTimeout, which defaults to 60 seconds.
• Adding databases propagates to all SQL nodes (at least with the latest version of MySQL), so when you create a new database, you only need to do it once on any SQL node. However, users dont propagate, so each SQL node will need to have its own users set up. Warning: do NOT try to change the MySQL internal tables (the ones in database mysql) to type ndbcluster as the cluster will break.
• I will think of something else to put here.

My Setup

This is my sample configuration with sample IPs:

• mysql-5.1.22-rc-linux-i686-icc-glibc23
• 1x management node (OpenSUSE): 10.0.0.1
• 4x storage (ndbd) nodes (OpenSUSE): 10.0.0.2, 10.0.0.3, 10.0.0.4, 10.0.0.5.
• 4x query (SQL) nodes (OpenSUSE): 10.0.0.2, 10.0.0.3, 10.0.0.4, 10.0.0.5.
• NoOfReplicas = 2, meaning there will be 2 copies of all data and therefore 4/2=2 node groups.
• Cluster data will sit in /var/lib/mysql-cluster.

Sample Screenshot

Here is a sample screenshot of another one of my configurations showing a similar setup. This is the output of show on the management node:

Setup Instructions

On the management node (as root):

1
2
3
4
5
6
7

groupadd mysql
useradd -g mysql mysql
mkdir -p /root/src/
cd /root/src/
wget http://dev.mysql.com/get/Downloads/MySQL-5.1/mysql-5.1.22-rc-linux-i686-icc-glibc23.tar.gz/from/http://mysql.he.net/
tar xvzf mysql-*.tar.gz
rm mysql-*.tar.gz

• ndb_mgmd is the management server
• ndb_mgm is the management client

1
2
3
4
5

cp mysql-*/bin/ndb_mg* /usr/bin/
chmod +x /usr/bin/ndb_mg*
mkdir /var/lib/mysql-cluster
chown mysql:mysql /var/lib/mysql-cluster
vi /var/lib/mysql-cluster/config.ini

Download /var/lib/mysql-cluster/config.ini

1
2
3

ndb_mgmd -f /var/lib/mysql-cluster/config.ini
ndb_mgm
show

On each storage and SQL node (as root):

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

groupadd mysql
useradd -g mysql mysql
cd /usr/local
wget http://dev.mysql.com/get/Downloads/MySQL-5.1/mysql-5.1.22-rc-linux-i686-icc-glibc23.tar.gz/from/http://mysql.he.net/
tar xvzf mysql-*.tar.gz
rm mysql-*.tar.gz
ln -s `echo mysql-*` mysql
cd mysql
chown -R root .
chown -R mysql data
chgrp -R mysql .
scripts/mysql_install_db --user=mysql
cp support-files/mysql.server /etc/init.d/
chmod +x /etc/init.d/mysql.server
vi /etc/my.cnf

Download /etc/my.cnf

1
2
3
4
5
6
7
8
9

mkdir /var/lib/mysql-cluster
chown mysql:mysql /var/lib/mysql-cluster
cd /var/lib/mysql-cluster
su mysql
/usr/local/mysql/bin/ndbd --initial # start the storage node and force it to (re)read the config
exit
echo "/usr/local/mysql/bin/ndbd" > /etc/init.d/ndbd
chmod +x /etc/init.d/ndbd
/etc/init.d/mysql.server restart # start the query node

SUSE:

1
2
3
4

chkconfig --add mysql.server # this is SUSE's way of starting applications on system boot
chkconfig --add ndbd
chkconfig --list mysql.server
chkconfig --list ndbd

Ubuntu:

1
2
3
4
5

sudo apt-get install sysv-rc-conf # this is chkconfig's equivalent in Ubuntu
sysv-rc-conf mysql.server on
sysv-rc-conf ndbd on
sysv-rc-conf --list mysql.server
sysv-rc-conf --list ndbd

That's it! At this point you should go back to the management console that you logged into earlier (ndb_mgm) and issue the 'show' command again. If everything is fine, you should see your data and SQL nodes connected. Now you can login to any SQL node, make some users, and create new ndb tables. If you're experiencing problems, do leave a message in the comments.

In the next mysql cluster article, I will explore various cluster error messages I have encountered as well as config file tweaking. Now go and spend some time outside in the sun – life is too short to waste it at a dark office.

• MySQL Conference Liveblogging: Portable Scale-out Benchmarks For MySQL (Wednesday 10:50AM)
• MySQL Conference Liveblogging: Performance Guide For MySQL Cluster (Tuesday 10:50AM)
• MySQL Conference Liveblogging: The Future Of MySQL (Tuesday 11:55AM)
• MySQL Falcon Storage Engine Enters Beta Stage.
• MySQL Slave Lag (Delay) Explained And 7 Ways To Battle It

If you are seeing similar problems, here's the fix. After digging around a bit, I have discovered that the problem was incorrect data encoding. My character set was set to KOI8-R while ncurses expected UTF-8. Here is how to change the corresponding setting in putty:

… and SecureCRT:

Now restart YaST and voila:

P.S. You may be wondering why my screen (the GNU one) has a handy footer bar and how to configure it. Not to worry, I will be posting more info on it soon.

• How To Check If The Local SVN Revision Is Up-To-Date
• Crazy Optical Illusion That Makes You Hallucinate – For Real
• Supercharge Your GNU Screen With A Power "Taskbar" And Never Feel Lost Again
• Mastering The Linux Shell – Bash Shortcuts Explained (Now With Cheat Sheets)
• How To Fix Palm Pre's "Error Sending Mail" Problem

cpan

Install a module:

cpan install MODULE

Upgrade a module:

cpan upgrade MODULE

Reinstall a module or force install in case of failed tests:

force install MODULE

See a list of upgradable modules:

r

See cpan configuration (that's the letter 'o'):

o conf

Update an option in cpan configuration:

o conf OPTION_NAME OPTION_VALUE

It is always nice to:

upgrade CPAN
install Bundle::CPAN

If there's an error making a Perl module, it can be caused by a missing make path in cpan configuration. In cpan, type in:

o conf

which will show all cpan options, then:

o conf make /usr/bin/make
o conf commit

A good option is:

o conf prerequisites_policy follow

By default, cpan will prompt to install every dependecy while installing a module. This option will automatically install dependencies without prompting.

• How To Install The Latest SOAP::Lite Using Perl CPAN
• Do NOT Use This Perl Module: Passwd::Unix
• How To Make Your Site Lightning Fast* By Compressing (deflate/gzip) Your HTML, Javascript, CSS, XML, etc In Apache
• [Perl] How To Get The Path Of An Included Library (.pm), Regardless Of Current Directory
• How To Download And Install Windows Media Player 11 Bypassing WGA

Full description is available here: http://sysbench.sourceforge.net/docs/

install mysql, mysql-devel
wget http://downloads.sourceforge.net/project/sysbench/sysbench/0.4.12/sysbench-0.4.12.tar.gz
tar xvzf sysbench*gz
cd sysbench*
./configure && make install

mysql tests

This will run 10 separate consecutive mysql tests using an InnoDB table type, each with 100 mysql threads, doing a total of 1000 various SQL operations per test. Then it will print the total time they took to finish:

sysbench --test=oltp --mysql-user=USER --mysql-password=PASS --mysql-db=test \
  --mysql-host='HOST' --mysql-table-engine=innodb prepare
 
time perl -e "foreach(1..10){print \`sysbench --max-requests=1000 --test=oltp \
  --mysql-user=USER --mysql-password=PASS --mysql-db=test --mysql-host='HOST' \
  --mysql-table-engine=innodb --num-threads=100 run\`}"
 
sysbench --test=oltp --mysql-user=USER --mysql-password=PASS --mysql-db=test \
  --mysql-host='HOST' cleanup

• MySQL Falcon Storage Engine Enters Beta Stage.
• How *Not* To Implement A Web Application That Handles External Authentication, Using BeTwittered.com As An Example
• MySQL Conference Liveblogging: Introduction To The BLOB Streaming Project (Wednesday 3:00PM)
• MySQL Conference Liveblogging: Benchmarking Tools (Wednesday 4:25PM)
• MySQL Conference Liveblogging: Portable Scale-out Benchmarks For MySQL (Wednesday 10:50AM)

• introduce the new Linux section of the site (specifically dedicated to openSUSE).
• provide a very short description of Linux and compare it to other operating systems.
• answer some installation questions.

I'm a Windows user, I admit it. Not because I enjoy frequent reboots, freezes, and other unexplainable quirks. It's mostly because I have so many programs I'm used to, it would be impossible to switch to anything else, and I know ins and outs that allow me to be very comfortable with the Windows. I'm talking 20-30 programs I'm not willing to give up any time soon.

However, I also have deep respect for *nix based systems. I started using them back in college and continued later on when I worked at Intel (Solaris) and other jobs. Stability and abundance of open-source software that just works attracted me to Linux. Finally, being a user just wasn't going to cut it: I had way too many questions about how things work, how the OS boots, what the conventions are, etc, etc, etc. So, finally, a couple of days ago I picked openSUSE 10.2 as my distro (distribution) of choice to play around with. The Linux section here on the site has been empty for a while, so I'm going to fill it up by documenting all my experiences and questions throughout the whole learning process as seen with the eyes of an amateur, which, I think, is what most tutorials are missing. So, let's begin.

OK, so… what are all these distributions? Ubuntu, who hasn't heard that by now? SUSE? Fedora? Why can't Linux be just Linux, just like Windows is just Windows? In short, in my opinion, because nobody but Microsoft has access to the Windows code (technically, after that leak a couple of years ago, everyone had a chance to see kernel source, but nobody can actually modify the OS or release another one based on it). Linux is open source, anybody can contribute to its development. That's the beauty of Linux: collaboration of hundreds thousands of people often working on their own time with passion no employee on payroll can match. Some motivated individuals go even further: they compile a set of Linux components that they think would work best together and release it as a distribution. openSUSE is a fine example of that. A bunch of smart people over at Novell got together and produced openSUSE a while ago, with its main goals being:

• Make openSUSE the easiest Linux distribution for anyone to obtain and the most widely used open source platform.
• Provide an environment for open source collaboration that makes openSUSE the world's best Linux distribution for new and experienced Linux users.
• Dramatically simplify and open the development and packaging processes to make openSUSE the platform of choice for Linux hackers and application developers.

See this openSUSE.org about page for more info.

Installing Linux is easier than ever. Back in the day, Linux puzzled average users. Nowadays, for everyday activities Linux is as easy to use as Windows (or Mac if you're into big shiny objects). A short installation guide by Novel is available here and a very nice long guide by HowToForge is available here. If you breeze through the installation steps, you should have a shiny desktop ready to rock and roll in about 30 minutes. While the installation itself is very easy (just burn a CD/DVD from http://download.opensuse.org and reboot the computer), I had a couple of questions about things I wasn't quite sure about. Here they are:

• You're prompted to select your Desktop. The choices are KDE, GNOME, and Other (the Other menu has serveral options, one being no graphical desktop, which you would use if you didn't want a GUI at all, for example for server installs or if you imagined myself as an uber leet hax0r). A graphical desktop is a bit like a skin for the OS. Each also includes different sets of tools, for example KDE's default editor is KEdit while GNOME's is gedit. There are other differences, but for now we have to pick one. The good news is that you can install both later via the software update tool called YaST (actually it's yast2 in SUSE 10.2, yast is just a symlink to /sbin/yast2) and switch between them with ease. For my particular installation, I picked KDE as the desktop manager, which I ended up enjoying quite a lot.

• Correct partitioning of my 110GB hard drive raised the most questions out of all the installation prompts. After consulting with a friendly sysadmin from work, I ended up with
  • 1.0GB swap partition with swap filesystem. I have 2GB RAM, so 1GB swap in case I run out of RAM should suffice.
  • 100MB /boot partition with ext3 filesystem. /boot is where a boot loader like GRUB will reside. We don't want to piss off GRUB, so we'll give it a separate partition.
  • ###MB /tmp partition with ext3 filesystem. /tmp stores temp files for all users. It is better to assign a separate partition to it so that malicious users cannot potentially flood it with useless data and cause the server to run out of space. Don't make this space too small though, so that legitimate users don't run out of temp space.
  • ###MB /var partition with ext3 filesystem. /var can also run the server out of space quickly if a runaway log file fills it up (normally, in /var/log). Assign a few gigs to this dir depending on total space.
  • The rest of the drive I assigned to / with ext3 filesystem. Some people also recommend assigning a separate partition to /home, which is probably a good idea if you're planning on having multiple users on the box. I picked ext3 after reading up online about various filesystem types. ReiserFS was not really an option anymore, since it wasn't too stable yet, and the creator of Reiser was recently arrested for murdering his wife. 2 days after the news broke out, SUSE creators officially announced ext3 as the recommended SUSE filesystem. When ext4 comes out, ext3 partitions will be easily upgradeable to ext4 without the need for repartitioning. The only drawback that I could find for ext3 is the ~32k directory and file limit in a single directory. Oh well, as long as I know about it, I'll be able to craft my applications with that in mind.
  • Update 3/20/2008: I found a great Filesystem Hierarchy Standard document that explains all the dirs in detail here.

• Since I was planning on letting my router (Linksys WRT54G with HyperWRT firmware) handle the firewall duties, I disabled the firewall in the Network part of the installation. It's probably a good idea to leave it on and open ports like SSH (22), FTP (20-21), or HTTP (80) manually. I'm probably going to turn the firewall back on after a couple of weeks once I get the hang of everything.

Well, that should do it for this introduction. In the following articles, I will share my experiences dealing with something something specific, writing as I learn myself. They may be dedicated to software installation and usage, for example, proftpd ftp server. In other articles I may visit Linux concepts, like how to make programs start up at boot time.

• How To Resize/Grow VMware Linux Disks and Partitions
• Ubuntu Laptop Installation
• The Repositories For apt-get In SUSE 10.2
• Beer Planet is Fast Again
• The Most Awesome VPN Tip: How To Make Windows Automatically Use Your Local WiFi/LAN Connection Directly For Requests That Don't Need To Go Through VPN