weird software RAID issue

[Kakumba]

Ok, I am trying to create a software RAID 5 array during an install of Red Hat Enterprise Linux 4 (Enterprise Server).

I did this previously, and all was well, however I realized that I had screwed up enough things that it was time to reinstall. At the same time, I added another hard drive, for a total of 5 500Gb drives. Now, when I go to reinstall, I get one of 2 errors.

Mainly, I get an error that /dev/md0 already exists. This doesn't really make sense to me, as it is a fresh install, and I am instructing anaconda to re-create the RAID array. I guess when they were formatted, they were set to fd (RAID auto detect), and so when mdadm is run to autodetect RAID arrays (just before you get to the installer), it detects the array, so I cant do much with it.

I managed to convince the installer that /dev/md0 does not exist as a RAID array (once), however at that point, creating the array failed, however the installer would not give me any detail. simply that there was a major error and the install could no continue.

I also get the same error if I try the same partitioning under the Fedora 8 install. I am right now trying to format all drives as ext3, so that hopefully setting them to code 83 (instead of the fd that I suspect they were).

My hardware is this:

DFI SLI-DR Expert motherboard
5x500Gb Seagate Barracuda 7200.11 (4 in the SIL-3114, and one on the NF4 controller)
1x 300Gb Seagate Barracuda 7200.9 (plugged into NF4)
Opty 165 (running at stock)
2x 512Mb DDR400 RAM
NVidia 7300GS

So. my questions are: can anyone help me to resolve why I cannot create the array with 5 disks, even though I have done it a couple of times with 4 disks? I am suspecting that it does not like having drives on more than one controller? I will try to create a 4 drive RAID 5 array using 3 on the SIL-3114 and one on the NF4, and see what happens.

Also, I tried using mdadm to add the last drive to md0, then mdadm -G /dev/md0, however all that did was add the drive as a spare, it did not increase the size of the array as I had expected. Again, anyone know why?

At this point, any advice at all would be useful, I have run through the installer a good 15 times in the last few days.

 

[Nothinman]

I also get the same error if I try the same partitioning under the Fedora 8 install. I am right now trying to format all drives as ext3, so that hopefully setting them to code 83 (instead of the fd that I suspect they were).

Putting a filesystem on them won't affect the partition type, you need to use a partitioning tool like fdisk to do that.

So. my questions are: can anyone help me to resolve why I cannot create the array with 5 disks, even though I have done it a couple of times with 4 disks? I am suspecting that it does not like having drives on more than one controller? I will try to create a 4 drive RAID 5 array using 3 on the SIL-3114 and one on the NF4, and see what happens.

You can add drives on any controller that you like, AFAIK mdadm doesn't even look at that. You're likely just running into a bug with Anaconda with regards to creating RAID arrays on a system that previously housed one.

mdadm has an option to zero the superblock, you might want to boot a LiveCD and try that.

Also, I tried using mdadm to add the last drive to md0, then mdadm -G /dev/md0, however all that did was add the drive as a spare, it did not increase the size of the array as I had expected. Again, anyone know why?

Without seeing the exact options that you used I can't say, but I have grown a RAID5 array in a VM before without issue.

 

[Kakumba]

Cheers for the help. For the first point, what I meant to say was that I was re-partitioning all drives, deleted the software raid partition, and create new partitions with ext3 file systems. So, unless anaconda decides that all I really mean is to create a new file system, and does not properly create new partitions, then it shoudl be ok.

I believe the exact command I used was mdadm --add /dev/mdo /dev/sdf1. This "hot-adds" the partition /dev/sdf1 to the device /dev/md0.

I will try a new install tonight, see what happens.

 

[Kakumba]

Ok, it looks like it is an anaconda issue, as I used the disk druid process to create software raid partitions, however not the raid device. I created this manually, and now have a full 1.8Tb device at /dev/md0.

 

[Nothinman]

I figured as much, I've seen anaconda do some pretty stupid things so I've never been a fan of it.