DRBD and MySQL - Heartbeat Setup
February 11, 2010
Heartbeat automates all the moving parts and can work as well with the MySQL master-master active/passive solution as well as it can with the MySQL & DRBD solution. It manages the virtual IP address used by the database, directs DRBD to become primary, or relinquish primary duties, mounts the /dev/drbd0 device, and starts/stops MySQL as needed.
DRBD provides an alternative high availability solution for MySQL. By effectively putting a mirrored filesystem beneath the database, the Linux operating system is thus replicating all of your data without the database even knowing about it. In our first two articles on the topic, we discussed some of the strengths and weaknesses of native MySQL statement based replication and then compared and contrasted those with the DRBD based solution. The advantages are in simplification of management, and elimination of some of the anomalous behavior of MySQL replication. The potential tradeoffs though are some performance impact, both in how your filesystem will then respond as it has to wait for acknowledgement of writes on remote destinations as well as a performance hit when you failover as the target MySQL database is starting up fresh, and has to warm its cache before performance will equal that on the former primary.
The last piece of the puzzle is the heartbeat software. Heartbeat automates all the moving parts that we've discussed so far. What's more, heartbeat can also be used with the native MySQL master-master active/passive replication solution, so it's likely to be in the mix, whatever high availability solution you go with.
What is heartbeat?
Heartbeat is a daemon that runs on all of the nodes in your cluster. It communicates with other nodes in the cluster to decide who will be primary, and what that primary will do. If one of the nodes becomes unresponsive, it will automatically remove it as primary, and make one of the other nodes take over. It also exposes a virtual IP address for your primary MySQL database to run on.
How does it work with DRBD and MySQL?
As you'll recall from previous articles, when switching to an alternate primary, one must (a) stop the database (b) unmount /dev/drbd0 and (c) tell the drbd daemon to be secondary, that is relinquish primary duties. On the alternate node, you must (d) tell drbd that it is now primary (e) mount /dev/drbd0 and finally (f) startup MySQL. Heartbeat performs all of these steps, but further it exposes a virtual IP address upon which to run your MySQL database. Each of the nodes in your cluster will have their my.cnf configured for an alternate virtual IP address of your choosing. Heartbeat will notify, manage, and take over that virtual IP address for use of the primary server before starting MySQL for you.
What that means is that when your mysql client, be it the mysql shell, apache webserver, or what have you, connects to the database, it will do so through this virtual IP address. When heartbeat moves that to a new machine, your application will transparently start hitting that alternate database without even knowing the rug was pulled out from under it. Ingenious!
Setup and Configuration
There are three main files that control heartbeat and we'll look at each of them in turn.
-- code section -- /etc/ha.d/ha.cf keepalive 1 # time between heartbeat packets deadtime 10 # How fast heartbeat notices dead node warntime 5 # alert about a late heartbeat initdead 120 # time for heartbeat to report dead node udpport 694 # UDP port for heartbeat ping 192.168.1.4 # virtualbox host ip address (of your laptop) bcast eth0 # interface to use for heartbeat packets # names of our node hostnames node adelie node african respawn hacluster /usr/lib/heartbeat/ipfail use_logd yes # send messages to /var/log/messages #use_logd no # send messages to /var/log/hb.log logfile /var/log/hb.log # various heartbeat system messages debugfile /var/log/heartbeat-debug.log # debug messages --
The above ha.cf file tells heartbeat how to startup and it's own default settings for pinging, noticing nodes are out, switching, switching back and the names of those nodes and ports. It doesn't tell heartbeat anything about the resources it will manage. We'll look at that next.
-- code section -- /etc/ha.d/haresources adelie IPaddr::192.168.1.30/24/eth0 drbddisk::mysql \ Filesystem::/dev/drbd0::/mnt/mysql::ext2::defaults mysqld --
The above lines tell heartbeat (a) what virtual IP address to manage, (b) what device and filesystem to mount and/or unmount and how and then (c) what service to start, which for us is our database.
-- code section -- /etc/ha.d/authkeys auth 2 2 crc --
This snippet basically disables authentication. Obviously if you are using this in production you'll want to configure this, but for our purposes in the virtualbox demo, it won't be necessary.
Once you've edited these three files on both of your nodes, you'll want to stop MySQL and unmount the filesystem, then startup heartbeat as follows:
$ kill %1 # assuming you've run mysqld_safe & $ umount /mnt/mysql $ /etc/init.d/heartbeat start
You will then want to monitor /var/log/messages or hb.log to see what is happening.
$ tail -f /var/log/messages
Heartbeat will take a moment to startup, and you'll see messages to that affect. It will wait for a period of time, and decide who to make the primary, then start services on that node, and start MySQL on it's new virtual IP address.
Lastly verify that you can connect to MySQL on the virtual IP address:
$ mysql --host=192.168.1.30 --port=3306
And that's it, you've done it!
Heartbeat brings together the last piece of the puzzle in setting up a high availability cluster for MySQL using DRBD. What makes it even better is that it works as well with MySQL standard replication as it does with DRBD so you have options there as well. Lastly, by exposing a virtual IP address and managing the starting and stopping of all the components, heartbeat nicely keeps everything seamless, thus providing a single target virtual database to hit, but multiple underlying server nodes to protect you, eliminating a single point of failure.
» See All Articles by Columnist Sean Hull