[WiP] wipe existing RAID and disks before creating a new one. #5
Conversation
| if os.path.exists(raid_device): | ||
| util.runCmd2(['mdadm', '--stop', '--force', raid_device]) | ||
| for dev in members: | ||
| util.runCmd2(['dd', 'if=/dev/zero', 'of=%s' % dev, 'bs=512', 'count=1']) |
There was a problem hiding this comment.
From what I understand, we zero out the 512 first bytes of each partition? What about mdadm superblocks that would be at the end of the partitions? Someone suggested to write zeroes there too, should we?
There was a problem hiding this comment.
it's the first 512 bits of each disk, that destroys their partition too.
There was a problem hiding this comment.
Ah, I thought we were going to iterate over the partitions in case they were part of a raid array (instead of the whole disk itself), to avoid having the mdadm superblock "resurrect" by chance if the new partition scheme is similar to the old one. But maybe destroying the partition table is enough...
There was a problem hiding this comment.
good point, let me think.
basically, I was trying not to ever parse the existing RAID configuration, and only decide things based on new information. Maybe it's wrong.
There was a problem hiding this comment.
What I meant is not to parse the existing RAID configuration but rather to read the partition table, and for each partition found, destroy the first and last bytes of the partition just in case there would be a superblock there.
Then maybe also destroy the partition table itself like you did and also probably the end of the disk in case there would be a superblock for a disk-wide RAID.
There was a problem hiding this comment.
Another thought (from the discussion in the issue), it is possible that mdadm would automatically detect a RAID array and start it, which means that we'd need to stop it before wiping things, to avoid errors, like they did in xcp-ng/xcp#107 (comment)
There was a problem hiding this comment.
yes, that's the parsing issue, we have to look at all the RAID in the system, parse their configuration, and only stop those whose array include one of the disks we are about to destroy.
Ought to be a solution to xcp-ng/xcp#107
Still need testing.