S3QL comes with a few contributed programs that are not part of the core distribution (and are therefore not installed automatically by default), but which may nevertheless be useful. These programs are in the contrib directory of the source distribution or in /usr/share/doc/s3ql/contrib if you installed S3QL from a package.
This program measures S3QL write performance, uplink bandwidth and compression speed to determine the limiting factor. It also gives recommendation for compression algorithm and number of upload threads to achieve maximum performance.
This program physically duplicates Amazon S3 bucket. It can be used to migrate buckets to a different storage region or storage class (standard or reduced redundancy).
pcp.py is a wrapper program that starts several rsync processes to copy directory trees in parallel. This is important because transferring files in parallel significantly enhances performance when copying data from an S3QL file system (see Improving copy performance for details).
To recursively copy the directory /mnt/home-backup into /home/joe using 8 parallel processes and preserving permissions, you would execute
pcp.py -a --processes=8 /mnt/home-backup/ /home/joe
This is an example script that demonstrates how to set up a simple but powerful backup solution using S3QL and rsync.
The s3_backup.sh script automates the following steps:
The backups are stored in directories of the form YYYY-MM-DD_HH:mm:SS and the expire_backups.py command is used to delete old backups.
expire_backups.py is a program to intelligently remove old backups that are no longer needed.
To define what backups you want to keep for how long, you define a number of age ranges. expire_backups ensures that you will have at least one backup in each age range at all times. It will keep exactly as many backups as are required for that and delete any backups that become redundant.
Age ranges are specified by giving a list of range boundaries in terms of backup cycles. Every time you create a new backup, the existing backups age by one cycle.
Example: when expire_backups is called with the age range definition 1 3 7 14 31, it will guarantee that you always have the following backups available:
If you do backups in fixed intervals, then one cycle will be equivalent to the backup interval. The advantage of specifying the age ranges in terms of backup cycles rather than days or weeks is that it allows you to gracefully handle irregular backup intervals. Imagine that for some reason you do not turn on your computer for one month. Now all your backups are at least a month old, and if you had specified the above backup strategy in terms of absolute ages, they would all be deleted! Specifying age ranges in terms of backup cycles avoids these sort of problems.
expire_backups usage is simple. It requires backups to have names of the forms year-month-day_hour:minute:seconds (YYYY-MM-DD_HH:mm:ss) and works on all backups in the current directory. So for the above backup strategy, the correct invocation would be:
expire_backups.py 1 3 7 14 31
When storing your backups on an S3QL file system, you probably want to specify the --use-s3qlrm option as well. This tells expire_backups to use the s3qlrm command to delete directories.
expire_backups uses a “state file” to keep track which backups are how many cycles old (since this cannot be inferred from the dates contained in the directory names). The standard name for this state file is .expire_backups.dat. If this file gets damaged or deleted, expire_backups no longer knows the ages of the backups and refuses to work. In this case you can use the --reconstruct-state option to try to reconstruct the state from the backup dates. However, the accuracy of this reconstruction depends strongly on how rigorous you have been with making backups (it is only completely correct if the time between subsequent backups has always been exactly the same), so it’s generally a good idea not to tamper with the state file.
For a full list of available options, run expire_backups.py –help.
s3ql_upstart.conf is an example upstart job definition file. It defines a job that automatically mounts an S3QL file system on system start, and properly unmounts it when the system is shut down.