Missing disk emulation not working

[rangerco1]

So, I was testing some scenarios to see how this handles failed disks. What I did was to run a disk wipe on one of the data disks to simulate some user error and see how the nonraid handles it.

So when I went to start the array, this is what I got:

TLDR:
I had to unassign the failed disk to start the array, mounted the disks and the disk is blank and files are not being emulated from the parity and remaining data disk. Am I missing something here with the commands to make this work?

DETAILS:
root@vector:/mnt# nmdctl start
Scanning array configuration...
Found disk for slot 0: sda1 (ID: WDC_WD40EFRX-68WT0N0_WD-WCC4EF...)
Skipping slot 1: Disk has been unassigned (status: DISK_NP)
Found disk for slot 2: sdc1 (ID: WDC_WD40EFRX-68WT0N0_WD-WCC4E1...)

Importing 2 disks...
Importing disk to slot 0: sda1
Importing disk to slot 2: sdc1
Successfully imported 2 disk(s)

=== NonRAID Array Status ===

Array State : STOPPED
Array Label : array1
Superblock : /nonraid.dat
Disks Present : 2
Array Health : PARTIAL (2/3 disks imported)
Array Checked : 19 minutes, 56 seconds ago
Array Size : 7453 GB (2 data disk(s))
Parity : Single Parity (3727 GB)

=== Disk Status ===

Slot Status Device Size(GB)

---

P OK sda1 3727
1 DISK_NP none 3727
2 OK sdc1 3727

Error: Cannot start array with missing disks
The following disk slots are defined but not imported: 1
You must either:

1. Import the missing disk(s) with nmdctl import
2. Unassign the missing disk(s) with nmdctl unassign SLOT

So, next I ran the unassign command

root@vector:/mnt# nmdctl unassign 1
Unassigning disk from slot 1...
Successfully unassigned disk from slot 1

=== NonRAID Array Status ===

Array State : STOPPED
Array Label : array1
Superblock : /nonraid.dat
Disks Present : 2
Array Health : PARTIAL (2/3 disks imported)
Array Checked : 21 minutes, 23 seconds ago
Array Size : 7453 GB (2 data disk(s))
Parity : Single Parity (3727 GB)

=== Disk Status ===

Slot Status Device Size(GB)

---

P OK sda1 3727
1 DISABLED none 3727
2 OK sdc1 3727

Followed by another start command

root@vector:/mnt# nmdctl start
Scanning array configuration...
Skipping slot 0: Already has device sda1 imported
Skipping slot 1: Disk has been unassigned (status: DISK_NP_DSBL)
Skipping slot 2: Already has device sdc1 imported
All disks already imported (2 disk(s) found)

=== NonRAID Array Status ===

Array State : STOPPED
Array Label : array1
Superblock : /nonraid.dat
Disks Present : 2
Array Health : PARTIAL (2/3 disks imported)
Array Checked : 21 minutes, 47 seconds ago
Array Size : 7453 GB (2 data disk(s))
Parity : Single Parity (3727 GB)

=== Disk Status ===

Slot Status Device Size(GB)

---

P OK sda1 3727
1 DISABLED none 3727
2 OK sdc1 3727

Array is in normal STOPPED state
Array started successfully

=== NonRAID Array Status ===

Array State : STARTED
Array Label : array1
Superblock : /nonraid.dat
Disks Present : 2
Array Health : DEGRADED (Invalid: 1, Disabled: 1)
Array Checked : 21 minutes, 47 seconds ago
Array Size : 7453 GB (2 data disk(s))
Parity : Single Parity (3727 GB)

=== Disk Status ===

Slot Status Device Size(GB) Disk Name FS Mountpoint Usage

---

P OK sda1 3727 none P - -
1 DISABLED none 3727 nmd1p1 xfs unmounted -
2 OK sdc1 3727 nmd2p1 xfs unmounted -

And next, I did the mount command

root@vector:/mnt# nmdctl mount
=== Mounting NonRAID Array Disks ===

Using mount prefix: /mnt/disk

Slot 1 (nmd1p1): Mounting XFS to /mnt/disk1... SUCCESS
Slot 2 (nmd2p1): Mounting XFS to /mnt/disk2... SUCCESS

Mount summary:
2 disks mounted successfully
0 disks skipped (already mounted or no filesystem)

Except that, it mounts the disks, but does not emulate any of the missing files from disk1. It just has an empty directory.
Am I missing something here? Is there something else that I need to do to have the missing file emulated from the data and parity?

[qvr]

What I did was to run a disk wipe on one of the data disks to simulate some user error and see how the nonraid handles it.

How did you wipe the disk, and especially which device name did you wipe? If you wiped /dev/nmd1p1 that would have written the "wipe" into parity, so the end result of empty emulated disk would be correct.

[rangerco1]

OK, that might explain it then. I had originally tried to wipe the underlying disk /dev/sdb, but it wasn't doing anything, so I also wiped the /dev/nmd1p1 device. Which also didn't do anything. So I rebooted the server, and the array was showing the failed device.

I used this to wipe the disk, and then again on the nmd1p1 device.
dd if=/dev/zero of=/dev/sdb count=1k bs=1k

I will try my test again, but leave the nmd1p1 device alone. Thanks for the quick reply.

[qvr]

Yeah, sounds like the second wipe wiped the emulated disk. :)

Also, if you just stop the array and wipe the underlying disk without unassigning the disk first, the array would still probably start as it would not be aware that the disk has been wiped. Parity check at this point would start showing a lot of errors, and any new writes (like creating the filesystem again) would corrupt existing parity.

Correct order would be to stop the array, unassign the disk, wipe the physical disk, start the array, and the array would start in disk disabled mode, emulating the unassigned disk contents.

(That's basically what I do in the DKMS integration tests )

[qvr]

Btw, nmdctl 1.17.0 improves the "UX" around unassigning disks and how they are displayed, so you can give that a try too.