Don Hayward wrote:
drives (both controllers). fdisk revealed a label of 0x0000 for all 4 drive tables. No logs are available (they were on /var, of course). Is there any way to get the partition tables relabeled as raid-automatic without writing over the rest of the tables -- in hopes there would be something left that would permit mdadm to build the raid? Any ideas or rumors about what might have happened?
If the label you are referring to is the system id, that you can just set it back to 0xfd with fdisk. It contains no more information than "this is a raid-auto partition". The actual raid meta-data is stored somewhere at the end of the devices.
It is kind of curious, that the system id should be wiped, and the rest of the partition table survive. Are you sure, those labels/ids were set to raid-auto before? I believe mdadm doesn't care anyway.
If you remember the chunksize, the order of the drive and the algorithm used (left/right-(as)symmetric) you can just create a new raid and cross your fingers. If nothing but the meta-information was lost -- that's about all it contains.
I don't think there is any chance to recover the apt information unless you recover the whole raid.
Did you run the usual hardware checks? smartctl/badblocks on the drives, memtest on you memory? gcc doesn't segfault on you? To me this sounds more like a hardware failure, than a software bug.
Elmar