Incremental backups

How I organised backups of my machine

Update: since April 2016, I've moved to a new backup system, based on rsync --link-dest parameter. I believe it's better for numerous reasons, so please use that!

Based on a short note from my Russian blog, let's make a good backupping system.

Humble beginning

We can start from a simple one-liner:

tar --create --gzip --ignore-failed-read --one-file-system --preserve-permissions --sparse --exclude=*cache* --exclude=*trash* --exclude=home/*/.ssh/‌* --ignore-case --listed-incremental=/backup/weekly --file=/backup/weekly-$(date +%G-%V).tar.gz /home /etc /usr/local/bin

It creates files with names like weekly-2015-39.tar.gz in the /backup/ folder. They are compressed (note --gzip argument), each of them has only files changed since last archive creation (note --listed-incremental= argument), and they have some other useful features - you can check the man tar page for other arguments.

You should run this command weekly. Stick it to /etc/cron.weekly/, for example.

Recoverability

Let's start with adding some recovery information for the case of broken bits. par2create is the tool to create a file with such recovery information (you're encouraged to check how to recover them yourself - I won't tell you).

That's done by a simple command

par2create -r10 -n1 {filename}

where -r10 means "recover from the loss of up to 10% of original file", and -n1 means "create only one recovery file"

To make it play nicely, the script grows a bit:

from="/home /etc /usr/local/bin"
cd /backup # destination for archives
file="weekly-$(date +%G-%V).tar.gz"
tar --create --gzip --ignore-failed-read --one-file-system --preserve-permissions --sparse --exclude=*cache* --exclude=*trash* --exclude=home/*/.ssh/‌* --ignore-case --listed-incremental=weekly --file=$file $from
par2create -r10 -n1 $file

In addition to files mentinoed above, it will also create files like weekly-2015-39.tar.gz.par2 and weekly-2015-39.tar.gz.vol000+200.par2 with some recovery information.

Space concerns

Ok, so incremental archives work like this: first, tar creates a "base" archive, which contains all files (and the weekly file with some metadata) and then, every week it creates an incremental archive with files changed during this week (and also updates the weekly file).

Eventually used space grows, and you might think that it's time to recreate the base archive and delete all incremental ones.

To do this, let's compare the sizes of "base" and all "incremental" archives, and when sum of incremental gets bigger then the base - it's time to clean up.

So we will do something like this:

let size_base=$(du -c base-*.tar.gz | grep total)
let size_incr=$(du -c weekly-*.tar.gz | grep total)
if [ $size_inc -gt $size_base ]; then
    # Clean up and create new base archive
else
    # Create weekly archive as normal
fi

Since we're creating archives in two different places - it's time to move relevant lines to a separate function.

And the whole script will look like this (note that I also added a command to create the "base" archive when there is no "weekly" file):

from="/home /etc /usr/local/bin"
cd /backup # destination for archives

archive() {
    tar --create --gzip --ignore-failed-read --one-file-system --preserve-permissions --sparse --exclude=*cache* --exclude=*trash* --exclude=home/*/.ssh/‌* --ignore-case --listed-incremental=$1 --file=$2 $from
    par2create -r10 -n1 $2
}

week=$(date +%G-%V)

if [ ! -f weekly ]; then
    # First run - create base archive
    archive weekly base-$week.tar.gz
else
    # Check sizes
    let size_base=$(du -c base-*.tar.gz | grep total)
    let size_incr=$(du -c weekly-*.tar.gz | grep total)
    if [ $size_inc -gt $size_base ]; then
        # Clean up and create new base archive
        delete_list="$(ls weekly-* base-*)"
        rm weekly
        archive weekly base-$week.tar.gz
        rm $delete_list
    else
        # Create weekly archive
        archive weekly weekly-$week.tar.gz
    fi
fi

Daily backups

Let's also add some daily backups. But keeping in mind that some files might change every day, instead of just renaming weekly to daily, let's make them also incremental, but based on latest weekly file.

Like this:

#!/bin/bash

from="/home /etc /usr/local/bin"
cd /backup # destination for archives

archive() {
    tar --create --gzip --ignore-failed-read --one-file-system --preserve-permissions --sparse --exclude=*cache* --exclude=*trash* --exclude=home/*/.ssh/‌* --ignore-case --listed-incremental=$1 --file=$2 $from
    par2create -r10 -n1 $2
}

dow=$(date +%w)
week=$(date +%G-%V)

if [ ! -f weekly ]; then
    # First run - create base archive
    archive weekly base-$week.tar.gz
    cp weekly daily
elif [ $dow == 0 ]; then
    # Sunday - check sizes
    let size_base=$(du -c base-*.tar.gz | grep total)
    let size_incr=$(du -c weekly-*.tar.gz | grep total)
    if [ $size_inc -gt $size_base ]; then
        # Clean up and create new base archive
        delete_list="$(ls weekly-* base-*)"
        rm weekly
        archive weekly base-$week.tar.gz
        rm $delete_list
    else
        # Create weekly archive
        archive weekly weekly-$week.tar.gz
    fi
    rm daily*
    cp weekly daily
else
    # Create daily archive
    archive daily daily-$dow.tar.gz
fi

At the end you can also add some command to rsync your backups to remote server.

In next article I'll cover more advanced topics, like:

encryption
multiple machines backups
separating settings, code, and data
...and maybe something more

Stay tuned!

Thanks to my Linux friend for his valuable input on this subject