CoreOS mdadm RAID1 for ROOT

README.md

CoreOS mdadm RAID1 for ROOT

• Setup SoftwareRAID for the CoreOS root partition without data-loss.
• Only for ext4!
• The Trick is to use the right volume label and fs-types :)
• We assume CoreOS is on /dev/sdaX and the second RAID Device is /dev/sdb
• Drawback: only / gets mirrored, Node goes down when sda fails

1. Boot into any recovery system like Grml

2. If not already done: install CoreOS

3. Backup /dev/sda9

   mkdir -p /root/sda9
   mount /dev/sda9 /mnt || exit 1
   rsync -a /mnt/* /root/sda9
   umount /dev/sda9

4. Prepare the partition layout

    # delete sda9
    sgdisk /dev/sda --delete=9
   
    # create new /dev/sda9 partition (max size)
    START=`sgdisk /dev/sda -f`
    END=`sgdisk /dev/sda -E`
    sgdisk /dev/sda --new=9:$START:$END --type=9:fd00
    sleep 0.5
    partprobe /dev/sda
   
    # Remove partition table from sdb
    sgdisk --clear -g /dev/sdb || exit 1
   
    # create a partition sdb9 with the size of sda9
    PARTDATA=( $(sgdisk -i 9 /dev/sda | grep 'Partition size' || exit 1) )
    SECTORS=$(expr ${PARTDATA[2]} + 2048)
    sgdisk /dev/sdb -a 2048 --new=9:2048:$SECTORS --type=9:fd00 || exit 1
    sleep 0.5
    partprobe /dev/sdb

5. Create the RAID

    mdadm --create /dev/md0 --level=1 --raid-devices=2 /dev/sda9 /dev/sdb9
    mkfs.ext4 -I 128 -L ROOT /dev/md0

6. Copy the Data

    mount /dev/md0 /mnt
    "rsync -a /root/sda9/* /mnt" || exit 1
    umount /mnt

7. Wait until resync is finished

   watch -n 1 cat /proc/mdstat

8. reboot!

I'm attempting this using Core itself as the in-RAM system. Two things are happening currently:

1. rsync doesn't support ACLs and throws an error on the copy. Removing the A flag lets it succeed, but I don't know what repercussions that might have. Is there another tool that might work for this? I would think maybe dd would be appropriate here, but I'm not certain.
2. On reboot the machine hangs. Oddly the last output is looking for a PS/2 Mouse device, so I presume it is some issue before that which is really crashing it. I'll try to get into a rescue system and parse logs soon.

Thanks for your work on this. I'm hoping to refine it into something we can build into our provisioning.

you just backup the sda9 partition.
sda1 or other partition also can backup use this solution?
thanks.

coreos doesn't seem to init mdadm on boot, as a result the boot fails and you're dropped into the dracut shell (after a large delay as it tries to look for ROOT by-label which isn't there. It should just be returning as soon as it isn't found, though...).

Running a mdadm --assemble --scan on that shell and exiting out allows the normal boot to continue.

Any way to automate this?

Edit (Answering my own question):

Pass rd.auto to the kernel with https://coreos.com/os/docs/latest/other-settings.html#adding-custom-kernel-boot-options

Thanks to kayrus@freenode for assisting me with this.