My understanding is that the current procedure to remove a write journal device is to first fail it then remove it. When I tried that mdadm hanged when I tried to remove the journal device. I'm running kernel 4.12.5-300.fc26.x86_64 and mdadm - v4.0 - 2017-01-09.
My original objective was to test array reshaping with a journal device in use. My attempt to add a device failed so I decided to try removing the journal, doing a reshape, then re-adding the journal.
What is the correct procedure for removing a journal device?
Here's what I did...
[root@build ~]# mdadm --readonly /dev/md1
[root@build ~]# cat /proc/mdstat
Personalities : [raid6] [raid5] [raid4]
md1 : active (read-only) raid5 sdf1[5](S) sdb1[0] sdd1[3] sde1[4](J) sdc1[1]
4190208 blocks super 1.2 level 5, 512k chunk, algorithm 2 [3/3] [UUU]
unused devices: <none>
[root@build ~]# mdadm /dev/md1 --fail /dev/sde1
mdadm: set /dev/sde1 faulty in /dev/md1
[root@build ~]# mdadm /dev/md1 --remove /dev/sde1
...hangs here...
The journal device is now marked failed but it isn't removed.
[root@build ~]# cat /proc/mdstat
Personalities : [raid6] [raid5] [raid4]
md1 : active (read-only) raid5 sdf1[5](S) sdb1[0] sdd1[3] sde1[4](J)(F) sdc1[1]
4190208 blocks super 1.2 level 5, 512k chunk, algorithm 2 [3/3] [UUU]
unused devices: <none>
The following shows up in the logs:
[ 393.707053] md/raid:md1: Disk failure on sde1, disabling device.
md/raid:md1: Operation continuing on 3 devices.
[ 393.714055] md/raid:md1: Disabling writeback cache for degraded array.
--Larkin