Sunday, 9 August 2009

Mysql Cluster: The definitive HOWTO

This guide was written over a year ago and is an excellent introduction however it has not been updated since! If you have any questions please email me or Buy my book (US).


You MUST have a third server as a managment node but this can be shut down after the cluster starts. Also note that I do not recommend shutting down the managment server (see the extra notes at the bottom of this document for more information). You can not run a mysql cluster with just two servers And have true redundancy.

You should also disable SELinux or use the following rules (many thanks to Robin Bowes for letting me know of these):

allow mysqld_t port_t:tcp_socket name_connect;
allow mysqld_t var_lib_t:file append;
allow mysqld_t var_lib_t:sock_file create;
allow mysqld_t var_lib_t:file read;
allow mysqld_t var_lib_t:sock_file unlink;
allow mysqld_t var_lib_t:file { getattr write };

Although it is possible to set the cluster up on two physical servers you WILL NOT GET the ability to "kill" one server and for the cluster to continue as normal. For this you need a third server running the managment node.

I am going to talk about three servers,

Servers 1 and 2 will be the two that end up "clustered". This would be perfect for two servers behind a loadbalancer or using round robin DNS and is a good replacement for replication. Server 3 needs to have only minor changes made to it and does NOT require a mysql install. It can be a low-end machine and can be carrying out other tasks.

STAGE 1: Install mysql on the first two servers:

Complete the following steps on both mysql1 and mysql2:

groupadd mysql
useradd -g mysql mysql
cd /usr/local/
tar -zxvf mysql-max-5.0.15-linux-i686-glibc23.tar.gz
rm mysql-max-5.0.15-linux-i686-glibc23.tar.gz
ln -s mysql-max-5.0.15-linux-i686-glibc23 mysql
cd mysql
scripts/mysql_install_db --user=mysql
chown -R root .
chown -R mysql data
chgrp -R mysql .
cp support-files/mysql.server /etc/rc.d/init.d/
chmod +x /etc/rc.d/init.d/mysql.server
chkconfig --add mysql.server

Do not start mysql yet.

STAGE 2: Install and configure the managment server

You need the following files from the bin/ of the mysql directory: ndb_mgm and ndb_mgmd. Download the whole mysql-max tarball and extract them from the bin/ directory.

mkdir /usr/src/mysql-mgm
cd /usr/src/mysql-mgm
tar -zxvf mysql-max-5.0.15-pc-linux-gnu-i686.tar.gz
rm mysql-max-5.0.15-pc-linux-gnu-i686.tar.gz
cd mysql-max-5.0.15-pc-linux-gnu-i686
mv bin/ndb_mgm .
mv bin/ndb_mgmd .
chmod +x ndb_mg*
mv ndb_mg* /usr/bin/
rm -rf /usr/src/mysql-mgm

You now need to set up the config file for this managment:

mkdir /var/lib/mysql-cluster
cd /var/lib/mysql-cluster
vi [or emacs or any other editor] config.ini

Now, insert the following (changing the bits as indicated):

# Managment Server
# Storage Engines
DataDir= /var/lib/mysql-cluster
# 2 MySQL Clients
# I personally leave this blank to allow rapid changes of the mysql clients;
# you can enter the hostnames of the above two servers here. I suggest you dont.

Now, start the managment server:


This is the mysql managment server, not maganment console. You should therefore not expect any output (we will start the console later).

STAGE 3: Configure the storage/SQL servers and start mysql

On each of the two storage/SQL servers ( and enter the following (changing the bits as appropriate):

vi /etc/my.cnf

Enter i to go to insert mode again and insert this on both servers (changing the IP address to the IP of the managment server that you set up in stage 2):


Be aware that anything in the [mysql_cluster] section will override the defaults in [mysql], so if you introduce a nodeid and then try to run multiple daemons on the same machines you should be aware of this!

Now, we make the data directory and start the storage engine:

mkdir /var/lib/mysql-cluster
cd /var/lib/mysql-cluster
/usr/local/mysql/bin/ndbd --initial
/etc/rc.d/init.d/mysql.server start

If you have done one server now go back to the start of stage 3 and repeat exactly the same procedure on the second server.
NOTE that you should ONLY use --initial if you are either starting from scratch or have changed the config.ini file on the managment.

STAGE 4: Check its working

You can now return to the managment server (mysql3) and enter the managment console:


Enter the command SHOW to see what is going on. A sample output looks like this:

[root@mysql3 mysql-cluster]# /usr/local/mysql/bin/ndb_mgm
-- NDB Cluster -- Management Client --
ndb_mgm> show
Connected to Management Server at: localhost:1186
Cluster Configuration
[ndbd(NDB)] 2 node(s)
id=2 @ (Version: 5.0.15, Nodegroup: 0, Master)
id=3 @ (Version: 5.0.15, Nodegroup: 0)

[ndb_mgmd(MGM)] 1 node(s)
id=1 @ (Version: 5.0.15)

[mysqld(API)] 2 node(s)
id=4 (Version: 5.0.15)
id=5 (Version: 5.0.15)


If you see

not connected, accepting connect from 192.168.0.[1/2/3]

in the first or last two lines they you have a problem. Please email me with as much detail as you can give and I can try to find out where you have gone wrong and change this HOWTO to fix it.

If you are OK to here it is time to test mysql. On either server mysql1 or mysql2 enter the following commands: Note that we have no root password yet.

use test;
INSERT INTO ctest () VALUES (1);
SELECT * FROM ctest;

You should see 1 row returned (with the value 1).

If this works, now go to the other server and run the same SELECT and see what you get. Insert from that host and go back to host 1 and see if it works. If it works then congratulations.

The final test is to kill one server to see what happens. If you have physical access to the machine simply unplug its network cable and see if the other server keeps on going fine (try the SELECT query). If you dont have physical access do the following:

ps aux | grep ndbd

You get an output like this:

root      5578  0.0  0.3  6220 1964 ?        S    03:14   0:00 ndbd
root 5579 0.0 20.4 492072 102828 ? R 03:14 0:04 ndbd
root 23532 0.0 0.1 3680 684 pts/1 S 07:59 0:00 grep ndbd

In this case ignore the command "grep ndbd" (the last line) but kill the first two processes by issuing the command kill -9 pid pid:

kill -9 5578 5579

Then try the select on the other server. While you are at it run a SHOW command on the managment node to see that the server has died. To restart it, just issue




Further notes about setup

I strongly recommend that you read all of this (and bookmark this page). It will almost certainly save you a lot of searching.

The Managment Server

I strongly recommend that you do not stop the managment server once it has started. This is for several resons:

  • The server takes hardly any server resources
  • If a cluster falls over, you want to be able to just ssh in and type ndbd to stat it. You dont want to have to start messing around with another server
  • If you want to take backups then you need the managment server up
  • The cluster log is sent to the management server so to check what is going on in the cluster or has happened since last this is an important tool
  • All commands from the ndb_mgm client is sent to the management server and thus no management commands without management server.
  • The managment server is required in case of cluster reconfiguration (crashed server or network split). In the case that it is not running, "split-brain" scenario will occure. The management server arbitration role is required for this type of setup to provide better fault tollerance.

However you are welcome to stop the server if you prefer.

Starting and stopping ndbd automatically on boot

To achieve this, do the following on both mysql1 and mysql2:

echo "ndbd" > /etc/rc.d/init.d/ndbd
chmod +x /etc/rc.d/init.d/ndbd
chkconfig --add ndbd

Note that this is a really quick script. You ought really to write one that at least checks if ndbd is already started on the machine.

Use of hostnames

You will note that I have used IP addresses exclusively throught this setup. This is because using hostnames simply increases the number of things that can go wrong. Mikael Ronström of MySQL AB kindly explains: "Hostnames certainly work with MySQL Cluster. But using hostnames introduces quite a few error sources since a proper DNS lookup system must be set-up, sometimes /etc/hosts must be edited and their might be security blocks ensuring that communication between certain machines is not possible other than on certain ports". I strongly suggest that while testing you use IP addresses if you can, then once it is all working change to hostnames.


Use the following formula to work out the amount of RAM that you need on each storage node:

(Size of database * NumberofReplicas * 1.1) / Number of storage nodes

NumberofReplicas is set to two by default. You can change it in config.ini if you want. So for example to run a 4GB database with NoOfReplicas set to 2 you need just under 9GB of RAM in total (4 * 2 * 1.1), so if you had two storage nodes you would need 4.5GB ram per storage node. For the SQL nodes and managment nodes you dont need much RAM at all.Bear in mind that if you have variable-width fields in MySQL Cluster 4.0 or 5.0 you will find that you will need a LOT more RAM than this formula predicts.

Note: A lot of people have emailed me querying the maths above! Remember that the cluster is fault tollerant, and each piece of data is stored on at least 2 nodes. (2 by default, as set by NumberOfReplicas). So you need TWICE the space you would need just for one copy, multiplied by 1.1 for overhead.

Adding storage nodes

If you decide to add storage nodes, bear in mind that 3 is not an optimal numbers. If you are going to move from two (above) then move to 4.

Adding SQL nodes

If you want to add another SQL node (i.e. you have another server that you want to add to the cluster but you dont need it to act as a storage node), then just add the following to /etc/my.cnf on the server (it must be a mysql-max server):


Then you need to make sure that there is another [MYSQLD] line at the end of config.ini on the managment server. Restart the cluster (see below for an important note) and restart mysql on the new API. It should be connected.

Important note on changing config.ini

If you ever change config.ini you must stop the whole cluster and restart it to re-read the config file. Stop the cluster with a SHUTDOWN command to the ndb_mgm package on the managment server and then restart all the storage nodes.

Some useful configuration options that you will need if you have large tables:

DataMemory: defines the space available to store the actual records in the database. The entire DataMemory will be allocated in memory so it is important that the machine contains enough memory to handle the DataMemory size. Note that DataMemory is also used to store ordered indexes. Ordered indexes uses about 10 bytes per record. Default: 80MB
IndexMemory The IndexMemory is the parameter that controls the amount of storage used for hash indexes in MySQL Cluster. Hash indexes are always used for primary key indexes, unique indexes, and unique constraints. Default: 18MB
MaxNoOfAttributes This parameter defines the number of attributes that can be defined in the cluster. Default: 1000
MaxNoOfTables Obvious (bear in mind that each BLOB field creates another table for various reasons so take this into account). Default: 128

View this page at for further information about the things you can put in the [NDBD] section of config.ini

A note about security

MySQL cluster is not secure. By default anyone can connect to your managment server and shut the whole thing down. I suggest the following precautions:

  • Install APF and block all ports except those you use (do NOT include any MySQL cluster ports). Add the IPs of your cluster machines to the /etc/apf/allow_hosts file.
  • Run MySQL cluster over a second network card on a second, isolated, network.

Other resources

I found the following resources very useful:

No comments: