After moving from a hard drive to an SSD on my work laptop, I decided to keep the hard drive spinning and use it as a backup for the SSD.
With the following setup, I can pull the SSD out of my laptop and it should still boot up normally with all of my data on the hard drive.
Manually setting up an encrypted root partition
Before setting up the synchronization between the two drives, I had to replicate the partition setup.
I used fdisk, cfdisk and gparted to create the following partitions:
Device Boot Start End Blocks Id System
/dev/sdb1 * 2048 499711 248832 83 Linux
/dev/sdb2 501760 500117503 249807872 5 Extended
/dev/sdb5 503808 500117503 249806848 83 Linux
and then loosely followed
these instructions
to create an encrypted root partition on /dev/sdb5:
$ cryptsetup luksFormat /dev/sdb5
$ cryptsetup luksOpen /dev/sdb5 sdb5_crypt
$ pvcreate /dev/mapper/sdb5_crypt
$ vgcreate akranes2 /dev/mapper/sdb5_crypt
$ vgchange -a y akranes2
$ lvcreate -L247329718272B -nroot akranes2
$ lvcreate -L8468299776B -nswap_1 akranes2
$ mkfs.ext4 -m1 /dev/akranes2/root
Finally, I added the new encrypted partition to the list of drives to bring up at boot time by looking up its UUID:
$ cryptsetup luksUUID /dev/sdb5
and creating a new entry for it in /etc/crypttab.
Copying the boot partition
Setting up the boot partition was much easier because it's not encrypted. All that was needed was to format it and then copy the files over:
$ mkfs.ext2 /dev/sdb1
$ mount /dev/sdb1 /mnt/boot
$ cp -a /boot/* /mnt/boot/
The only other thing to remember is to install grub on the boot loader of
that drive. On modern Debian systems, that's usually just a matter of
running dpkg-reconfigure grub-pc and adding the second drive (/dev/sdb
in my case) to the list of drives to install grub on.
Sync scripts
To keep the contents of the SSD and the hard drive in sync, I set up a
regular rsync of the root and boot partitions using the following mount
points (as defined in /etc/fstab):
/dev/mapper/akranes-root / ext4 noatime,discard,errors=remount-ro 0 1
/dev/mapper/akranes2-root /mnt/root ext4 noatime,errors=remount-ro,nofail 0 2
UUID=0b9109d0-... /boot ext2 defaults 0 2
UUID=6e6f05fb-... /mnt/boot ext2 defaults,nofail 0 2
I use this script (/usr/local/sbin/ssd_boot_backup) for syncing the boot
partition:
#!/bin/sh
if [ ! -e /mnt/boot/hdd.mounted ] ; then
echo "The rotating hard drive is not mounted in /mnt/boot."
exit 1
fi
if [ ! -e /boot/ssd.mounted ] ; then
echo "The ssd is not the boot partition"
exit 1
fi
nice ionice -c3 rsync -aHx --inplace --delete --exclude=/boot/ssd.mounted --exclude=/boot/hdd.mounted --exclude=/boot/lost+found/* /boot /mnt/
and a similar one (/usr/local/sbin/ssd_root_backup) for the root
partition:
#!/bin/sh
if [ ! -e /mnt/root/hdd.mounted ] ; then
echo "The rotating hard drive is not mounted in /mnt/root."
exit 1
fi
if [ ! -e /ssd.mounted ] ; then
echo "The ssd is not the root partition"
exit 1
fi
nice ionice -c3 rsync -aHx --delete --exclude=/dev/* --exclude=/proc/* --exclude=/sys/* --exclude=/tmp/* --exclude=/boot/* --exclude=/mnt/* --exclude=/lost+found/* --exclude=/media/* --exclude=/run/* --exclude=/scratch/* --exclude=/var/tmp/* --exclude=/ssd.mounted --exclude=/var/lock/* /* /mnt/root/
To ensure that each drive is properly mounted before the scripts run, I
created empty ssd.mounted files in the root directory of each of the
partitions on the SSD, and empty hdd.mounted files in the root directory
of the hard drive partitions.
Note that the boot partition rsync command uses the --inplace option.
That's to prevent No space left on device errors on due to the temporary
file copying (the rsync default mode of operation) on small partitions.
Cron jobs
The sync scripts are run every couple of hours through this crontab:
10 */4 * * * root /usr/local/sbin/ssd_boot_backup
20 0,4,8,12,16,20 * * * root /usr/local/sbin/ssd_root_backup
20 2,6,10,14,18,22 * * * root /usr/bin/on_ac_power && /usr/local/sbin/ssd_root_backup
which includes a reduced frequency while running on battery to avoid spinning the hard drive up too much.
Hmm, why do you need all of those "--exclude" options? Shouldn't "-x" be enough to ensure that e.g. /sys is not copied?
Also, is there some particular reason why you are not using "--numeric-ids"?
Regarding the excludes, I think you're right, I might be able to remove most of them given that they would cross filesystem boundaries anyways.
As for the numeric-ids option, I've looked at the manpage, but I honestly don't understand how that would make a difference in this case. Do you think it would help with anything?
I didn't try a real RAID1 setup, but I suspect that I might lose some of the performance advantages of having an SSD if I/O is blocked on the rotating drive.
Also, from a power usage point of view, it would mean that the hard drive needs to be spinning all the time, as opposed to every couple of hours.
I totally agree that a RAID1 would kill the speed of SSD. But then, your post title says RAID1...
Now, I wanted to mention that if you know the name of your device, you can check whether it's a rotational (HDD) or not (SSD). Here is an example of how you get the flag:
alexis~$ cat /sys/block/sda/queue/rotational
The file is either a 0 (SSD) or a 1 (HDD). What you need is the "sda" part of the path. In a shell script, it can be difficult to grab that info. Although a function such as "df ." give you the info, it's not that practical.
Anyway, your solution is pretty much what I had in mind. Thank you.