Re: Software RAID on SPARC64
On Tuesday 19 July 2005 09:17, Turbo Fredriksson wrote: > > Yes. For MD devices on SPARC, the (first) partition _MUST_ start on '1', > not '0' as is the default. > Thanks for that everyone, that was indeed the problem. Regards, David. -- David Johnson www.david-web.co.uk - My Personal Website www.ethereye.org.uk - EtherEye Network Host Checker www.penguincomputing.co.uk - Need a Web Developer? -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Software RAID on SPARC64
Quoting Martin <[EMAIL PROTECTED]>: > On Mon, 2005-07-18 at 14:06 -0400, Sam Creasey wrote: >> Not sure this is actually it (I'm just guessing here), but are you >> starting the raid partition on cylinder 0? Or cyl 1? I *think* (though >> I'm not sure) that starting on cylinder 0 actually puts the disklabel into >> the beginning of the partition itself. This isn't a problem for UFS, as >> UFS leaves 8k unused at the beginning of the FS, but it might be a problem >> for md. >> >> Again, that's just a shot in the dark, take it for what it's worth. > > This sounds familiar. I certainly remember having to go back and > re-partition starting on 1 not 0 to avoid eating the disklabel and > partition table. Can't remember what the symptoms where but I'm pretty > sure it was on the machine that runs a RAID array... Yes. For MD devices on SPARC, the (first) partition _MUST_ start on '1', not '0' as is the default. - s n i p - [EMAIL PROTECTED]:~# mdadm -D /dev/md/0 | grep /dev/scsi 0 8 340 active sync /dev/scsi/host3/bus0/target2/lun0/part2 1 8 1301 active sync /dev/scsi/host3/bus0/target10/lun0/part2 2 8 2262 active sync /dev/scsi/host4/bus0/target2/lun0/part2 3 65 503 active sync /dev/scsi/host4/bus0/target10/lun0/part2 4 8 242 -1 spare /dev/scsi/host4/bus0/target3/lun0/part2 5 8 50 -1 spare /dev/scsi/host3/bus0/target3/lun0/part2 [EMAIL PROTECTED]:~# fdisk -l /dev/scsi/host4/bus0/target2/lun0/disc Disk /dev/scsi/host4/bus0/target2/lun0/disc (Sun disk label): 64 heads, 32 sectors, 8635 cylinders Units = cylinders of 2048 * 512 bytes Device FlagStart EndBlocks Id System /dev/scsi/host4/bus0/target2/lun0/disc1 1 257262144 83 Linux native /dev/scsi/host4/bus0/target2/lun0/disc2 257 8635 8579072 fd Linux raid autodetect /dev/scsi/host4/bus0/target2/lun0/disc3 0 8635 88422405 Whole disk - s n i p - -- [Hello to all my fans in domestic surveillance] Semtex kibo CIA killed PLO Nazi NORAD explosion Ft. Bragg Clinton 767 SEAL Team 6 subway congress [See http://www.aclu.org/echelonwatch/index.html for more about this] -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Software RAID on SPARC64
On Mon, 18 Jul 2005 17:57:22 +0100, David Johnson wrote: > I've been having problems with software RAID (using mdadm) on Sarge on > an Ultra Enterprise 450. There seems to be a bug somewhere causing > corruption of Sun disk labels. > > I start with 8 SCSI disks with valid Sun disk labels and one partition > filling each disk. The partition types are set to "Linux RAID > autodetect". What do the partition tables look like? You may know this already, but you need to start the first partition on each disk at cylinder 1 instead of cylinder 0 to leave room for the disk label. Apparently, Ext2/3 leaves some free space at the start of partitions for this sort of thing, but RAID doesn't. S. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Software RAID on SPARC64
Not sure this is actually it (I'm just guessing here), but are you starting the raid partition on cylinder 0? Or cyl 1? I *think* (though I'm not sure) that starting on cylinder 0 actually puts the disklabel into the beginning of the partition itself. This isn't a problem for UFS, as UFS leaves 8k unused at the beginning of the FS, but it might be a problem for md. Again, that's just a shot in the dark, take it for what it's worth. -- Sam On Mon, 18 Jul 2005, David Johnson wrote: > Hi all, > > I've been having problems with software RAID (using mdadm) on Sarge on an > Ultra Enterprise 450. There seems to be a bug somewhere causing corruption of > Sun disk labels. > > I start with 8 SCSI disks with valid Sun disk labels and one partition filling > each disk. The partition types are set to "Linux RAID autodetect". > > I can create a RAID 5 array without problems. This is the command I'm using: > mdadm -C -l5 -n7 > -x1 /dev/md0 /dev/sdd1 /dev/sde1 /dev/sdf1 /dev/sdg1 /dev/sdh1 /dev/sdi1 > /dev/sdj1 /dev/sdk1 > > If I now reboot, everything is OK and the array is detected and configured. > > The problem occurs when I format the array. E.g.: > mkfs.ext2 /dev/md0 > > This works OK and no errors are given. However when the system is rebooted, > around half the disks now have corrupt disk labels. If I re-create the disk > labels and restore the partition tables, everything is fine when I reboot. > The array is started correctly and the disk labels are maintained on > subsequent reboots. Only formatting the array causes the problem. > > I have tried with other filing systems and the same happens. Not all the disks > are affected, only around half of them (the last time I did it, sd[efghi] > were affected). > > I'd be interested to know if anyone else seen this issue. > > Regards, > David. > > -- > David Johnson > www.david-web.co.uk - My Personal Website > www.ethereye.org.uk - EtherEye Network Host Checker > www.penguincomputing.co.uk - Need a Web Developer? > > > -- > To UNSUBSCRIBE, email to [EMAIL PROTECTED] > with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED] > -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Re: Software RAID on SPARC64
On Mon, 2005-07-18 at 14:06 -0400, Sam Creasey wrote: > Not sure this is actually it (I'm just guessing here), but are you > starting the raid partition on cylinder 0? Or cyl 1? I *think* (though > I'm not sure) that starting on cylinder 0 actually puts the disklabel into > the beginning of the partition itself. This isn't a problem for UFS, as > UFS leaves 8k unused at the beginning of the FS, but it might be a problem > for md. > > Again, that's just a shot in the dark, take it for what it's worth. This sounds familiar. I certainly remember having to go back and re-partition starting on 1 not 0 to avoid eating the disklabel and partition table. Can't remember what the symptoms where but I'm pretty sure it was on the machine that runs a RAID array... Cheers, - Martin -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Software RAID on SPARC64
Hi all, I've been having problems with software RAID (using mdadm) on Sarge on an Ultra Enterprise 450. There seems to be a bug somewhere causing corruption of Sun disk labels. I start with 8 SCSI disks with valid Sun disk labels and one partition filling each disk. The partition types are set to "Linux RAID autodetect". I can create a RAID 5 array without problems. This is the command I'm using: mdadm -C -l5 -n7 -x1 /dev/md0 /dev/sdd1 /dev/sde1 /dev/sdf1 /dev/sdg1 /dev/sdh1 /dev/sdi1 /dev/sdj1 /dev/sdk1 If I now reboot, everything is OK and the array is detected and configured. The problem occurs when I format the array. E.g.: mkfs.ext2 /dev/md0 This works OK and no errors are given. However when the system is rebooted, around half the disks now have corrupt disk labels. If I re-create the disk labels and restore the partition tables, everything is fine when I reboot. The array is started correctly and the disk labels are maintained on subsequent reboots. Only formatting the array causes the problem. I have tried with other filing systems and the same happens. Not all the disks are affected, only around half of them (the last time I did it, sd[efghi] were affected). I'd be interested to know if anyone else seen this issue. Regards, David. -- David Johnson www.david-web.co.uk - My Personal Website www.ethereye.org.uk - EtherEye Network Host Checker www.penguincomputing.co.uk - Need a Web Developer? -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]