diff options
| -rw-r--r-- | DOCUMENTATION | 94 |
1 files changed, 50 insertions, 44 deletions
diff --git a/DOCUMENTATION b/DOCUMENTATION index fb435d1..9d346b4 100644 --- a/DOCUMENTATION +++ b/DOCUMENTATION @@ -1,57 +1,63 @@ SquashFu - an alternative backup solution Inspired by http://forums.gentoo.org/viewtopic-p-4732709.html -Requirements: aufs, aufs2-util, squashfs-tools, rsync +Requirements: aufs, aufs2-util, squashfs-tools, rsync, bc Goal: To create a backup solution which provides incremental backups and compression, and which also provides an easy way to roll back. Design: - A directory structure is created as follows: + A directory structure is created as follows (with some terminology included): backup_root/ - |- seed.sfs - |- ro/ - |- rw/ - |- bins/ - |-1/ <-- Monday incremental - |-2/ <-- Tuesday incremental - |-3/ <-- Wednesday incremental - |-4/ <-- Thursday incremental - |-5/ <-- Friday incremental - |-6/ <-- Saturday incremental - |-7/ <-- Sunday incremental - - seed.sfs is created from an initial backup and compressed using SquashFS, which i - is simply a filesystem which focuses on compression, but is read only. It's mounted, + |- seed.sfs <-- squash, or seed + |- .bin.list <-- bin inventory list (or binventory) + |- ro/ <-- squash mount point + |- rw/ <-- union mount point + |- .bins/ <-- incrementals + |-1/ + | ..... + | ..... + | ..... + | ..... + | ..... + |-n/ + + seed.sfs is created from an initial backup and compressed using SquashFS, which is + simply a filesystem which focuses on compression, but is read only. It's mounted, using a loopback device, on ro/. Using aufs2, a union mount is formed by ro/ and each - of the numbered bins, each corresponding to a day. - - Wait, wat? What the heck IS a union mount? In simple terms, a union combines multiple - directories (which themselves could also be mount points) into a single resultant - mount point. Using our directory structure above as an example, each day (bin) is - overlayed (in order, its important) on the base (seed.sfs) to create what appears - to be an up to date backup in rw/. Skip the rest if you've had enough. - - At the time of the backup, the day is determined in order to prepare the union. If - today is Thursday, `date +%u` returns 4. In order for rsync to properly create our - incremental, we need to compare current data with the seed plus incrementals leading - up to Thursday, so we create our Aufs mount with the branches bins/4, bins/3, bins/2, - bins/1, and ro/ (in that order). When rsync now writes to the resulting union, aufs - happily receives the data into the first available (writable) branch of the union, - which is bins/4. The next day, the union mount is reformed adding the next bin... - - If and when you want to roll back, the process is simple. Mount the seed plus - the necessary bins to increment the seed up to the day of interest. - -Warnings: - It is imperative that you do NOT touch the individual branches. You can and will - cause irrepairable damage to the union, rendering your incrementals worthless. - + of the numbered bins, each corresponding to an incremental backup. + + At the time of the backup, the next available bin is created and logged to an inventory + sheet with a timestamp. A union is created with all the available bins, mounted in + reverse chronological order on top of the seed (newest to oldest) on rw/. At this point, + the union represents the state of your files at the end of the last backup. The newest + branch is marked as read/write, and rsync is called. Because this top branch is the + only writable location in the union, the files rsync generates with the -u (update) + flag are placed into this branch. The backup finishes, and the union and seed are + unmounted. + + At this point, Squashfu ensures compliance with the user's settings of MAX_BINS. If + the current number of used bins exceeds this value, a new seed is generated. The + number of old incrementals merged into the new seed is determined by the difference + between MAX_BINS and MIN_BINS in the config file. In this way, you always have + MIN_BINS available to roll back to, but you're not forced to recompress your seed + at every backup -- an operation that may take a long time depending on how big + your backup source is. + + If and when you want to roll back, execute Squashfu with the -R action, and supply + the number of bins you want to roll back. The bins are ordered chronologically, + and the oldest "number_of_bins - bins_to_rollback" are mounted on the union mount + point. + +WARNING: + You should not, under any circumstances, add or remove files contained in the bins, + nor should you alter your binventory's time stamps. Doing so can result in non-recoverable + damage to the integrity of the backups. Further reading: - http://en.wikipedia.org/wiki/Aufs - http://en.wikipedia.org/wiki/UnionFS - http://aufs.sourceforge.net/ - http://en.wikipedia.org/wiki/SquashFS - http://en.wikipedia.org/wiki/Rsync + http://en.wikipedia.org/wiki/Aufs + http://en.wikipedia.org/wiki/UnionFS + http://aufs.sourceforge.net/ + http://en.wikipedia.org/wiki/SquashFS + http://en.wikipedia.org/wiki/Rsync |