Solaris software RAID
Add raid to your Solaris system.
This describes how to install Software raid using Solstice Disksuite. First, you need to grab the DiskSuite packages from Sun. It’s a free download, but you do need to create a sunsolve account to get it.
You can find a download link for the software at http://wwws.sun.com/software/solaris/get.html. It’s about half-way down that page.
Install the packages – There’s an installer included. Don’t bother. Just pkgadd the individual packages. There’s only 5 or so.
Once you’re done that, you’ll need to determine how you want to lay out your disks. The following assumes that:
1 – You have 2 disks – c0t0d0 (disk0) and c0t1d0 (disk1).
2 – The system installed only on disk0, and disk1 is unused.
3 – Each disk has the following slices:
0 – /
1 – swap
2 – whole-disk
3 – unassigned 64-MB
4 – unassigned 64-MB
Adjust the above to match whatever your preferred layout is. This is only for a simple example. Slices 3 and 4 are for Meta-Database logging. If you don’t have 128MB of free space to spare, then try and make some space (ie., sacrafice some swap if you have to).
Disclaimer: Before you continue, I can’t stress enough. Make backups. I always do this procedure during system installation, and I do it routinely. Although your data is supposed to survive and actually migrate without incident, I have seen DiskSuite eat a system once or twice – usually due to operator error.
You need to duplicate your layout from disk0 to disk1. It’s fairly important that the disk geometry matches. Metadevices work at the block-level of the disk, and if one disk has fewer blocks than the other you’ll wind up making a mess. Once you’re sure you’re ready to proceed, dump the layout from disk0 to disk1 thusly:
prtvtoc /dev/rdsk/c0t0d0s2 | fmthard -s - /dev/rdsk/c0t1d0s2
Second, you need to create your meta-databases. This is for logging, and all but eliminates the need for fsck to run after a dirty shutdown. Do the following:
metadb -af -c 2 /dev/dsk/c0t0d0s3 /dev/dsk/c0t0d0s4
metadb -af -c 2 /dev/dsk/c0t1d0s3 /dev/dsk/c0t1d0s4
This adds (-a) 2 (-c for count) meta-databases in each of the slices. If you have more disks, you can span the databases across multiple disks for better performance and fault-tolerance.
The next step is to create your raid-devices. In a two-disk system, you’re stuck with Raid0 and Raid1. Since Raid0 is almost pointless (you’re doing this for redundancy, remember?!), we’ll go with Raid1 – mirrored disks.
We’ll deal with the following raid devices and members:
d0 – / mirror
d10 – /dev/dsk/c0t0d0s0
d20 – /dev/dsk/c0t1d0s0
d1 – swap
d11 – /dev/dsk/c0t0d0s1
d21 – /dev/dsk/c0t1d0s1
The device names are somewhat arbitrary. In a simple setup like this, I use d0 to match up with a mirrored slice0, and d10 to indicate member 1 of d0 (member 1 d0 = d10, member 2 d0 = d20).
So create the raid devices and members:
metainit -f d10 1 1 /dev/dsk/c0t0d0s0
metainit -f d20 1 1 /dev/dsk/c0t1d0s0
metainit -f d0 d10
metainit -f d11 1 1 /dev/dsk/c0t0d0s1
metainit -f d21 1 1 /dev/dsk/c0t1d0s1
metainit -f d1 d11
This initializes the devices. The command “metastat” will show you that the devices exist, but the mirror-halves aren’t attached. So let’s attach them:
metattach -f d0 d10
metattach -f d1 d11
You’ve just attached the first half of the mirror. Yes, this is the disk that you’re currently running on. Your data is still there.
Next, you need to ensure the system will use the metadevices. The root-filesystem is easy:
Next, you need to edit /etc/vfstab to change the swap device to use /dev/md/dsk/d1 as swap. While you’re in there, turn on logging under the mount options for the root filesystem (d0). Double-check that you haven’t screwed up. Save and exit if it all looks good.
Once you’re done, issue the following:
Watch your system come up. There will be some new messages, most notably the kernel complaining about not being able to forceload three raid modules:
forceload of misc/md_trans failed
forceload of misc/md_raid failed
forceload of misc/md_hotspares failed
You can ignore these messages. They’re harmless. Basically, you haven’t created any raid-devices that require those modules so they’re refusing to load.
Now that your system is up (You didn’t mess up vfstab, did you?!), you need to finish off the process. Log in and do this:
metattach -f d0 d20
metattach -f d1 d21
You’ll notice that your system is now a little slower, both commands took a moment to return, and your disks are going nuts. Look at the output of “metastat” and you’ll see why – your disks are syncing.
You’ll need to install the bootsector to your second disk so that you can boot from it. This is fairly easy to do:
installboot /usr/platform/`uname -i`/lib/fs/ufs/bootblk /dev/rdsk/c0t1d0s0
You might also want to set the OBP to boot from disk1 if it can’t boot from disk0. If you bring the machine to the OBP (ok) prompt via init 0, you can enter the following:
setenv boot-device disk disk1
This will set up a failover boot to disk1. The very last command there will also boot from disk1, proving to you that this works. Do be sure to substitute the correct disk for “disk1″.
You now have your root filesystem and swap space sitting on raid1 volumes. This means that losing a disk no longer means that you have to rebuild your system. Now you just need to replace a disk.
I strongly suggest that you read through DiskSuite’s docbook at Sun’s online documentation site (docs.sun.com). There’s a lot more that you can do with it that’s not covered here. Oh, and you’ll probably want to read up on how to actually replace a failed disk.
(If there’s enough interest, I might be coerced into posting a howto on that)
Follow the reactions below and share your own thoughts.