partition table disappears

[Joe_P]

Hi all,

We have the following problem.

Setup:

RedHat 4.4 ia64 Itanium (rx2620).
EVA4000
Oracle 10G + OCFS2

We created and presented a Vdisk to 2 hosts.
Partitioned the Vdisk on host 1 (using fdisk option 'n')
Formatted the partition for OCFS2 (using ocfs2console).
Mounted the filesystem on host 1.
Mounted the filesystem on host 2.
Everything worked fine.

We rebooted host 1.
After that host 1 cannot mount the filesystem anymore, giving error:
"mount.ocfs2: Device name specified was not found while opening device /dev/sdi1"

It turns out that the partition table seems to be gone:

fdisk -l /dev/sdi
Disk /dev/sdi: 10.7 GB, 10737418240 bytes
64 heads, 32 sectors/track, 10240 cylinders
Units = cylinders of 2048 * 512 = 1048576 bytes

Device Boot Start End Blocks Id System



We didn't reboot host 2 and the disk is still mounted and accessible from host 2, even though fdisk -l doesn't show a partition table on host 2 either.

I don't know what went wrong here.
Has anyone ever experienced a similar problem?

Thanks,
Martijn.

[Steven E. Protter]

Shalom,

I ran into a very similar problem with ocfs2 hosted on a SAN using emulex cards.

I was forced to build a new initrd file with the emulex hba cards built into modeprobe.conf

It is best to build the fdisk with the second node shut down. After build, run partprobe, and boot to see if it remains in the fdisk -l output.

SEP

[Joe_P]

Thanks for your fast reply.
We're using Qlogic hba's.

Were you able to access the partition table again after rebuilding the initrd?

Regards,
Martijn.

[Rick Beldin]

I was involved in an issue once where a similar situation had occurred. The customer was running RHEL3 w/ Oracle 9iRac and had configured one of the drives as an OCFS 'raw' partiton. In this configuration, there was no 'partition table' as OCFS was given direct access to the device. After a reboot for periodic maintenance the OCFS drive became unusuable.

Oracle's position on this issue was that raw device access to a disk with OCFS was problematic and that OCFS was more reliable with partition table. In the OCFS config, the customer needed to specify a device such as /dev/sda1 as opposed to the block device /dev/sda.

Once the customer configured OCFS in this manner, they had no further data corruption...