Possible to convert to RAID without a complete reinstall?

plexustech

Here's my scenario: I have a working gentoo system with an 80G drive, and the traditional 3 partitions on it (boot/swap/everything else). I also have an identical, second 80G drive, which I'd like to plug in and use in a RAID configuration.

Is there any way I can convert the system over to a RAID system without nuking everything and starting an installation from scratch? I'm not keen on starting over after all the effort it's cost me to get the first version going.

So what I'm asking is, is it possible to create the necessary partitions/files on the blank drive, copy my working system onto that, and then nuke the original drive and redo it so that its partitions match the new one's?

This way, if possible, I'm off the air for the minimum amount of time.
_________________
Idiot Filter: "Ya, we run the C++ operating system on a QNX platform over FDDI twisted pair at 600 MIPS." If they swallow that, hang up.

DarrenM · Posted: Mon Oct 13, 2003 11:32 am Post subject:

I think it depends on which type of raid you want. For mirroring (raid 1) your raid controller should take care of the copying/mirroring for you but I don't think you could do it for a striped raid (raid 0).

plexustech · Posted: Mon Oct 13, 2003 12:38 pm Post subject:

Perhaps I should have been more specific - I'd like to do this with software RAID, not by using any hardware controllers. I'm leaning towards RAID1.
_________________
Idiot Filter: "Ya, we run the C++ operating system on a QNX platform over FDDI twisted pair at 600 MIPS." If they swallow that, hang up.

NickDaFish · Posted: Mon Oct 13, 2003 6:58 pm Post subject:

I would also like to know.... I've been trying to do the same thing myself. I have made *some* progress... it I do work it out I'll post a follow up.

NickDaFish · Posted: Mon Oct 13, 2003 9:29 pm Post subject:

Ok, I think I have this somewhat nailed down. This is a *quick* overview, I am not finished working this all out.

To start off with we have two disks, sda and sdb. sda1 is the boot, sda2 is swap and sda3 is the root. sdb is blank.

Format sdb to have the same partition sizes as sda.

Create single disk mirror arrays on sdb.

DarrenM · Posted: Mon Oct 13, 2003 11:17 pm Post subject:

The Software Raid HOWTO has a section on converting a non-raid installation to software raid 1 here.

NickDaFish · Posted: Tue Oct 14, 2003 1:56 am Post subject:

NickDaFish · Posted: Wed Oct 22, 2003 10:27 pm Post subject:

I have been working on a HowTo for this. It's not really finished but people have been asking.....

The more readable form is here:
http://wiki.nickdafish.com/public/HowToRaid

The raw (un-rendered) text isn't pretty but it is readable.... When I have finished (yeah right!) the How To I will try and make a phpBB version. Any commects/connections/etc please let me know....

= RAIDing up a system HowTo =
This How To covers changing your single disk system with no fault tollerance to a pair of mirrored disks, one of them being the orriginal disk.
Yes I'm mirroring the swap.
This How To assumes that you have two disks. One that the system was installed on (hde) and one new blank disk (hdg) of equal or greater size (Ideally it would be an identical disk).
There are three basic steps in this How To....
1. Make 'broken' RAID1 arrays on your new disk.
1. Populate the 'broken' arrays from the old disk
1. Reboot the system onto the 'broken' arrays
1. Fix the 'broken' arrays by adding the old disk to the arrays
I reffer to the arrays as 'broken' because they are RAID1 (mirror) arrays made with only one disk untill the last step where you add the old disk to the array. RAID1 requires 2 disks, the idea being if either disk dies the other can be used in the mean time.
----
[[TableOfContents]]
----
== Preperation ==
First thing's first. We have to partition up the new disk just like the old one. Then we have to setup the arrays.
=== Setup machine ===
* Compile a kernel with md and RAID1
* emerge mdadm
=== Setup disks ===
* Inital disk - hde, new raid disk - hdg
* fdisk hde and take close note of the number of cylinders for each partition.
* fdisk hdg and set it up
* make each of the partitions
* make them all type 0xfd
Type 0xfd is 'Linux raid autodetect' - Important
* mark the boot partition as bootable
* write the partition table
* Reboot if required (Eg the partition table cannot be re-read)
=== Setup RAID arrays ===
* Create a RAID1 array from each partition (With a missing disk)
* mdadm --create /dev/md1 --level=1 --raid-disks=2 missing /dev/hdg1
* Should say 'mdadm: array /dev/md1 started.'
* mdadm --create /dev/md2 --level=1 --raid-disks=2 missing /dev/hdg2
* mdadm --create /dev/md3 --level=1 --raid-disks=2 missing /dev/hdg3
* Format each RAID1 array with the right FS
* mke2fs /dev/md1
* mkswap /dev/md2
* mke2fs -j /dev/md3
----
== Populating the RAID arrays and cut over ==
Note that you may want to have a liveCD around just in case you have problems in this bnxt section.... Fro example I have in the past forgot to setup grub propperly. If you do not follow all the steps here you could be left with a non-working system. However your orriginal disk should still have all of your data on it. It's not untill the next phase that the orriginal data is lost, when the old disk is added to the new RAID arrays as the 'other' (previously missing) disk. This step should reboot you onto the RAID1 arrays.
=== Setup Boot ===
* Update the kernel's root param
* Mount /boot
* Edit /boot/grub/grub.conf and change root= to /dev/md3[[BR]]
You may want to skip this step if our feeling jumpy. If you skip this step you will not reboot onto the RAID array. If you skip this step and reboot, the sync from hde3 to /dev/md3 that you will preform in a few steps will be out of sync. If you skip this you will need to recopy hde3 to /dev/md3
* Copy files
* Mount /dev/md1 in /mnt/temp (Create if it's not already there)
* Copy the files with '{{{cp -ax /boot/* /mnt/temp/}}}'
* Grub the boot partition (not array) /dev/hdg1
Ok.. this is a little strange but it has to be done. Grub installs a cunk of data on the boot partition that cannot be copyed in the normall way. They easyist way to get it on to the new disk is using grub. Basicly follow the instructions that you followed in the gentoo install docs. If you use LILO then your on your own.
{{{
# grub
Probing devices to guess BIOS drives. This may take a long time.
root (hd1,0)
setup (hd1,0)
quit
}}}
=== Alter fstab ===
* update /etc/fstab[[BR]]
You may want to skip this step if our feeling jumpy. If you skip this step you will not reboot onto the RAID array. If you skip this step and reboot, the sync from hde3 to /dev/md3 that you will preform in a few steps will be out of sync. If you skip this you will need to recopy hde3 to /dev/md3
* Change /boot from hde1 to /dev/md1
* Change swap from hde2 to /dev/md2
* Change / from hde3 to /dev/md3
=== Copy / ===
* Shutdown your system into single user mode[[BR]]
You can do this without shutting down but things might be missed in the copy.
* '{{{shutdown now}}}'
* Mount /dev/md3 in /mnt/temp
* Copy
* '{{{cp -ax / /mnt/temp}}}'[[BR]]
This will not copy /proc or /dev becuase of the 'x'. You don't want to copy them anyhow.
=== Reboot ===
When the machine reboots it will be running entirly on the new RAID1 arrays, your old disk should be untouched. If all else fails all you have to do to revert to your old system is edit /boot/grub/grub.conf and /etc/fstab and reboot.
Note: although you will be using the new RAID1 arrays for almost everything you may well have booted from your 'old' hde1.
----
== Add the old disk to the RAID1 arrays ==
Only continue on in this section if everythign is WORKING. This section will kill off the data on your old disks!
=== Check it out! ===
* Have a look at dmesg
Check for problems! Below is some normall output from dmesg for the first boot. You will get warnings because there is only one disk in each of the arrays and there should be two.
{{{
md: considering ide/host2/bus1/target0/lun0/part1 ...
md: adding ide/host2/bus1/target0/lun0/part1 ...
md: created md1
md: bind<ide/host2/bus1/target0/lun0/part1,1>
md: running: <ide/host2/bus1/target0/lun0/part1>
md: ide/host2/bus1/target0/lun0/part1's event counter: 00000002
md: RAID level 1 does not need chunksize! Continuing anyway.
md1: max total readahead window set to 124k
md1: 1 data-disks, max readahead per data-disk: 124k
raid1: device ide/host2/bus1/target0/lun0/part1 operational as mirror 1
raid1: md1, not all disks are operational -- trying to recover array
raid1: raid set md1 active with 1 out of 2 mirrors
md: updating md1 RAID superblock on device
md: ide/host2/bus1/target0/lun0/part1 [events: 00000003]<6>(write) ide/host2/bus1/target0/lun0/part1's sb offset: 40064
md: recovery thread got woken up ...
md1: no spare disk to reconstruct array! -- continuing in degraded mode
md2: no spare disk to reconstruct array! -- continuing in degraded mode
md3: no spare disk to reconstruct array! -- continuing in degraded mode
md: recovery thread finished ...
}}}
* Get details and examine with mdadm
* '{{{mdadm -D /dev/md1}}}'
* '{{{mdadm -E /dev/hdg1}}}'
=== Change your old disk's setup ===
* fdisk /dev/hde
* Change all the partition types to 0xfd
* Write the partition table
* You may have to reboot if the partition table could not be re-read
=== Add the old disk to the array ===
* Use mdadm to add the old partitions to the RAID1 arrays
* mdadm /dev/md1 --add /dev/hde1
* mdadm /dev/md2 --add /dev/hde2
* mdadm /dev/md3 --add /dev/hde3
* Watch the sync in the logs.....
* tail -f /var/log/everything[[BR]]
For each disk you should see something like this:
{{{
[kernel] md3: no spare disk to reconstruct array! -- continuing in degraded mode
[kernel] md: trying to hot-add ide/host2/bus0/target0/lun0/part3 to md3 ...
[kernel] md: recovery thread got woken up ...
[kernel] disk 22, s:0, o:0, n:0 rd:0 us:0 dev:[dev 00:00]
[kernel] md: md3: sync done.
}}}
== Partay ==
P-A-R-T.....Why? Becuase your machine is now mirrored.

Puggles · n00b Joined: 30 Dec 2002 Posts: 24 Location: Florida

I just want to say NickDaFish's guide worked perfectly. I went from 2.4.20 with EVMS doing RAID1 to 2.6.3 with the kernel doing RAID1 without losing any data! (But I did do a complete /home backup onto DVD+R beforehand..

)

NickDaFish · Posted: Wed Mar 17, 2004 11:48 pm Post subject:

Please note that the most up to date version is here:
http://wiki.nickdafish.com/public/HowToRaid
(Just realised I posted the wrong link!)

I'm glad people are finding it usefull :wink:

krypt · n00b Joined: 18 Feb 2004 Posts: 3

First thanks NickDaFish for the howto...

Now the Problem :-):

Everything works fine up to the Point where to create the FileSytems on the new Raid 1 Partitions.

esoteriskdk · Posted: Mon Mar 29, 2004 4:42 pm Post subject:

I have the exact same problem, I tried to simulate a diskcrash on a RAID5, it worked, but I managed to b0rk everything anyway. So now I want to start from a clean slate.

krypt · n00b Joined: 18 Feb 2004 Posts: 3

@esoteriskdk

Had the Same Problem, upgrading to xfsprogs 2.6.3 (ACCEPT_KEYWORDS="~x86"' emerge xfsprogs for the x86 Plattform) solved it.This is a known Bug on the XFS developer Mailinglist.

It happens when there had been an other Partition with another Filestystem on the Disk.
Don't forget to use the -f Switch to force the Creation even with the 2.6.3 version of xfsprogs.

But im not pleased with the stability of the XFS Filesystem on Softwareraid. Had a lot of Failures and data leakage. Switcht to Reiserfs without being unhappy about the Decision up to now.[/b]
_________________
JabberID: alex@alraune.org

esoteriskdk · Posted: Mon Mar 29, 2004 5:58 pm Post subject:

starachna · Posted: Sun Apr 04, 2004 11:43 am Post subject: thanks NickDaFish

thank you dude, your howto helped me a great deal! awsome work!
_________________
http://www.3am.co.za - za psy trance

Sunthief · Posted: Mon Jan 09, 2006 7:31 am Post subject:

Hi I am trying to do the mirroring raid on my sun e250 system and had a few concerns. I have already created the subpartitions on the soon to be added drive and copied over my boot (which since its a sun I am using /) to the newly made /dev/md0. My concern is after I did my copy of all the files I noticed some differences in the sizes of the mounted file systems. Heres some output that might help: