This is a Python script that acts as an interface between BackupPC and Amazon S3. It uses BackupPC's archive function to extract a tarball and split it into chunks, like the normal archive function. Then, the chunks are encrypted using gpg and transmitted to S3 using Boto.
I wrote this script some years ago, and can't remember how to get it going. But, here's going to be my best guess :-)
You will need Python, Boto, and a working BackupPC installation.
Note: Python 2.6+ and Boto 2.0+ are required for recent changes, which include multiprocessing support. I may make these optional later on, but until then, tag stable-20110610 is what was running before I decided to mess with things!
Something like this seems like a good idea:
cd /usr/local/src/ git clone git://github.com/rtucker/backuppc-archive-s3.git
Then create a link from
ln -s /usr/local/src/backuppc-archive-s3/BackupPC_archiveHost_s3 /usr/share/backuppc/bin/
Create a file in this directory called
secrets.py, based upon the
secrets.py.origfile. It should have your AWS Access and Shared keys, a passphrase that will be used to encrypt the tarballs.
accesskey = 'ASDIASDVINASDVASsvblahblah' sharedkey = '889rv98rv8fmasmvasdvsdvasdv' gpgsymmetrickey = 'hunter2'
Previously, you could use a
speedfileto change the permitted upstream bandwidth on the fly. This was cantankerous and was ultimately dropped in September 2011. See tag stable-20110610 if you need this functionality (and open an issue to let me know!), or take a look at The Wonder Shaper to limit throughput on a system-wide level.
From the BackupPC configuration interface, go to
Edit Hostsand add a new host,
archiveS3, which looks like the existing
archivehost. Save this, select the
archives3host, and then
Edit Configfor that host.
Change the settings on each tab as follows:
XferMethod: archive ArchiveDest: /var/lib/backuppc/archives3 ArchiveComp: bzip2 ArchiveSplit: 500 ArchiveClientCmd: $Installdir/bin/BackupPC_archiveHost_s3 $tarCreatePath $splitpath $parpath $host $backupnumber $compression $compext $splitsize $archiveloc $parfile *
That should be just about it. Note that
ArchiveDestis where it will stage the tarballs before it uploads them; this must have enough disk space for your archive!
ArchiveSplitis the size of each tar file, in megabytes; you may want to adjust this for your needs. Also, the
ArchiveClientCmdis the default, except with the
Go to the main page for the
archives3host and click
Start Archive. To start with, just tick the box next to the smallest backup you have, then
Archive selected hosts. Go with the defaults (which look suspiciously like what you set on the Xfer tab, do they not? :-) and then
Start the Archive.
Watch syslog and hopefully everything will work. If it does not, there will be decent debugging output in the archive job's log, viewable via the BackupPC console.
There is a companion script,
backup-manager.py, that can be used to see
what's on S3. Run it with no arguments to get a listing of backups and
their ages, or use the
--help argument to see what it can do.
The "crown jewel" of this whole system is the
script command, which
produces a script that can be used to restore a backup. It uses S3's
Query String Request Authentication
mechanism to generate temporary URLs to download each file required to
restore a backup.
Each night, from
cron, I run a script:
#!/bin/sh BACKUPMGR=/path/to/backup-manager.py # Delete all backups older than 30 days. $BACKUPMGR delete --age=30 # Create restore scripts, valid for one week, for all of my computers cd /home/rtucker/Dropbox/RestoreScripts/ $BACKUPMGR --expire=604800 --host=gandalf script > restore_gandalf.sh $BACKUPMGR --expire=604800 --host=witte script > restore_witte.sh # etc, etc # Output a list of what's on the server $BACKUPMGR
The output of this is mailed to me, so I always know what's going on!
BackupPC is written in Perl. Why is this thing written in Python?
I know Python much better than I know Perl, so I wrote it in Python. The good news is that BackupPC doesn't care, but it does mean this probably won't be part of the BackupPC main distribution any time soon.
Is this project dead?
You could say that. A lot of my projects are one-off scripts that solve a very specific need I have, and I don't put too much thought into making them useful for other people. This script works for me and (sorta) meets my needs, so that's where it is.
What changed in September 2011?
I got tired of seeing a square-wave pattern on my throughput graphs, and so I modified the system to use Python's multiprocessing library. It will now run GPG encryption jobs in the background, with as many CPUs as you have available, while transmitting files.
This probably isn't a problem for anyone else, but my BackupPC server is slow (exactly one "Intel(R) Pentium(R) 4 CPU 1.70GHz") and is behind a very asymmetric cable modem connection.