On 30.03.20 01:35, Andre Tomt wrote:
I'm having a issue here where I am unable to get a journal device
removed from a raid6 array. From what I could gather from mailing list
posts and documentation, one should set the journal mode to
write-through, fail the journal and remove it, then restart the array
(perhaps with force).
But this isnt working. The journal just gets re-added on array
startup. I've done this successfully before, in the same way, I think.
I also tried a bigger hammer, wipefs the journal device too, to "make
sure", the array will come up but refuse any writes.
For now, the journal device has been restored and the array is back to
read-write, but I really need to get it removed at some point
(preferably without re-building metadata with --assume-clean)
Any ideas? Is this way just out of date?
mdadm: 4.1-5ubuntu1
kernel: 5.5.13
# cat /proc/mdstat
md2 : active (auto-read-only) raid6 sdm1[0] sde1[11] sdk1[10] sdt1[9]
sdl1[8] sdc1[7] sdj1[6] sdb1[5] sdu1[4] sds1[3] sdr1[2] sdd1[1]
nvme0n1p1[12](J)
58603894400 blocks super 1.2 level 6, 64k chunk, algorithm 2
[12/12] [UUUUUUUUUUUU]
# echo write-through > /sys/block/md2/md/journal_mode
# mdadm --fail /dev/md2 /dev/nvme0n1p1
mdadm: set /dev/nvme0n1p1 faulty in /dev/md2
# mdadm --remove /dev/md2 /dev/nvme0n1p1
mdadm: hot removed /dev/nvme0n1p1 from /dev/md2
# cat /proc/mdstat
md2 : active (auto-read-only) raid6 sdm1[0] sde1[11] sdk1[10] sdt1[9]
sdl1[8] sdc1[7] sdj1[6] sdb1[5] sdu1[4] sds1[3] sdr1[2] sdd1[1]
58603894400 blocks super 1.2 level 6, 64k chunk, algorithm 2
[12/12] [UUUUUUUUUUUU]
Not know journal well, but I guess it is better to change the
consistency_policy before stop array
by "echo resync > /sys/block/md2/md/consistency_policy" since journal
device is not available.
Thanks,
Guoqing
# mdadm --stop /dev/md2
mdadm: stopped /dev/md2
# mdadm --assemble /dev/md2 --force
mdadm: /dev/md2 has been started with 12 drives and 1 journal.
<-- !?
# cat /proc/mdstat
md2 : active (auto-read-only) raid6 sdm1[0] nvme0n1p1[12](J) sde1[11]
sdk1[10] sdt1[9] sdl1[8] sdc1[7] sdj1[6] sdb1[5] sdu1[4] sds1[3]
sdr1[2] sdd1[1]
^^^
58603894400 blocks super 1.2 level 6, 64k chunk, algorithm 2
[12/12] [UUUUUUUUUUUU]
Okay then. Hammer time. Do all that again, wipefs the journal device,
force start the array:
# mdadm --assemble /dev/md2 --force
mdadm: Journal is missing or stale, starting array read only.
mdadm: /dev/md2 has been started with 12 drives.
# cat /proc/mdstat
md2 : active (read-only) raid6 sdm1[0] sde1[11] sdk1[10] sdt1[9]
sdl1[8] sdc1[7] sdj1[6] sdb1[5] sdu1[4] sds1[3] sdr1[2] sdd1[1]
58603894400 blocks super 1.2 level 6, 64k chunk, algorithm 2
[12/12] [UUUUUUUUUUUU]
Then it is just stuck in read-only.
