What are you protecting?

It is important to refine some terms at the start.
A backup is a redundant, non-functional copy of something important, made recently, that you might need in a hurry.
A code repository is not a backup.

For web applications we have to consider two things: the Database(s) and the Codebase.
These two things run together to provide the application.

At its best, a backup is a point-in-time copy of a functioning system... a snapshot.
This can be a difficult thing to get, especially from a high traffic site. As a system runs, the database is constantly changing or is running slightly behind its save-graph, the codebase changes less often (or not at all) but to some extent it is temporally dependent on the state of the database.

As this is a redundant copy that might be needed it just has to be useful not necessarily perfect. Let's go with hourly backups of the running DB and hourly copies of the application code.
To get a copy of the database, the backup system needs to do a dump. For MySQL there is a tool called mysqldump we can call from a crontab script set to run each hour.

#crontab
5 * * * * /usr/bin/mysqldump --quick -ureadonlyuser -p$PASS innodbDatabase > /data/db_backups/running-backup.sql

Now that we have a new copy each hour we need to put it somewhere safe.
In the same crontab the running-backup.sql file is copied to an Amazon S3(Simple Storage Solution) bucket. This step is really important. If we lose access to the server for any reason we cannot get to the hourly backup of the database. if the backup is in an S3 bucket we can quickly copy it to a new server and install it in a different MySQL server and carry on.

Now the Codebase.
We need to mirror the files to another directory, create a compressed archive of the copied filesystem, then copy the archive to an S3 bucket.
To get a point in time copy of the running code we can and should rely on a tool called rsync. Rsync makes a copy of a filesystem and can follow rules about what to mirror. Rsync is also extremely fast at refreshing the copy. It uses algorithms to calculate hashes for files and parts of files and only copies the parts of the file that have changed between the source and destination directories. This speed is critical as we need to tread lightly on the source code of the running application.

#crontab
5 * * * * /usr/bin/rsync -av --delete /var/www/html /home/ec2-user/rsync-backup

Now that we have our mirror created at the same time as the database dump we need to make a compressed archive using tar and copy it to the S3 bucket.

25 * * * * /bin/tar -pczf /home/ec2-user/site.filesystem.tar.gz /home/ec2-user/rsync-backup/
45 * * * * /usr/bin/s3put running-backups/codebase/microinstance-filesystem-backup.tar.gz /home/ec2-user/site.filesystem.tar.gz

For most purposes, this backup regime is probably good enough. Disaster recovery would involve launching a new LAMP server, installing the running-backup.sql, unpacking the microinstance-filesystem-backup.tar.gz in the apache document root.
There are a couple of failure modes that need to be dealt with. If the S3 bucket gets a bad or empty version of your backups just before the server fails it is possible you don't actually have a useful backup when you need it. To guard against this it is wise to write each new backup file to S3 with some sort of dynamic file name. I could use:

Monday.database.sql
Monday.codebase.tar.gz
Tuesday.database.sql
Tuesday.codebase.tar.gz
etc...

So 24 times each day I have a good backup and I have a good backup for 6 yesterdays until the cycle repeats. However, in the above failure mode I can only count on the backup from yesterday!

I can change this to:
Monday-13-00.database.sql
Monday-14-00.database.sql
etc...

Now I have the ability to step back up to 168 hours into the past to find a good backup and recover with that.
If your application must stay as up-to-date as possible this higher-redundancy system is useful but will result in higher storage fees.

This backup discussion does use AWS specific parts but it doesn't need to. As long as you have a copy of your application automatically copied to a different computer somewhere in the world everything will be ok.

chris.macdonald@@doctrinaire.com.au