Bug 791159

+++ This bug was initially created as a clone of Bug #769323 +++

Description of problem:

When a temporarily missing RAID 1 component becomes available, an incremental assembly should consult the write intent map and resynchronize the device with those blocks recorded as changed.  Instead, "mdadm -I --run /dev/sdc1" declares the temporarily missing component "invalid".

Version-Release number of selected component (if applicable):

mdadm-3.2.2-125.fc16.x86_64
kernel-3.1.5-6.fc16.x86_64

How reproducible:

Always.

Steps to Reproduce:
1.  Create a RAID 1 device as so:

mdadm --create /dev/md0 --level=1 --raid-devices=2 /dev/sdb1 --write-mostly /dev/sdc1 --bitmap=/var/preserve/vg_extern0 --assume-clean

3. Record the result of "mdadm -D --brief /dev/md0" in /etc/mdadm.conf:

ARRAY /dev/md0 metadata=1.2 bitmap=/var/preserve/extern0 name=myhost:md0 UUID=c84d260e:f10d4241:af5e0c24:1f18d143

2. Use the RAID device (i.e., create a logical volume group containing it, put a file system on one of the logical volumes).  This works as expected.

3. Shut down the system.

4. Physically disconnect the /dev/sdc device.  Leave /dev/sdb connected.

5. Boot the system.  /dev/md0 will be assembled and run with the one available component.

6. Use the RAID 1 device.  Activate the logical volume, mount the filesystem from it, and write a file in that file system.

7. Shut down the system.

8. Physically reattach the /dev/sdc device.

9. Boot the system.

10. /dev/md0 will be created with only the /dev/sdb component participating.

11.  sudo mdadm -I --run /dev/sdc1 
  
Actual results:

mdadm: failed to add /dev/sdc1 to /dev/md0: Invalid argument.

Expected results:

/dev/sdc1 to participate in the RAID 1 array, and shown to by synchronizing in results of "cat /proc/mdstat".  If the writes made to the one RAID component were small, the synchronization will be finished before you can observe the synchronization state, and the RAID device will contain both components up to date "[UU]".

Additional info:

This worked as expected in Fedora 14.

--- Additional comment from Jes.Sorensen on 2012-01-03 05:33:13 EST ---

Stephen,

What is the output from mdadm when you try to add the second device?
It has to be more than just 'Invalid argument'

Second, you shouldn't need to add --run when specifying -I, I don't know
if that makes any difference.

Jes

--- Additional comment from Jes.Sorensen on 2012-01-03 08:15:21 EST ---

Stephen,

Ok, I tried reproducing the bug here, and it looks like I am seeing the
exact same as you.

I will try and investigate further.

Jes

--- Additional comment from Jes.Sorensen on 2012-01-03 10:33:45 EST ---

Ok, this is strange.

If I try to add the drive this way:

mdadm -I --run /dev/sdf3

I get the same error as you. On the other hand if I do the following, it
works just fine:

mdadm -a /dev/md42 /dev/sdf3

Looks like something isn't detected correctly by either the kernel or mdadm
when doing an Incremental add. This happens with the latest mdadm from
Neil's git tree and Linus' top of tree kernel as well btw.

Jes

--- Additional comment from sschaefer on 2012-01-04 16:35:50 EST ---

Thanks for pursuing this and sorry for not responding; I'd been under the weather.  I'll see if the "-a" version does what I need - if the write intent map gets ignored, I end up with a 16 hour sync that needs to finish in a 10 hour window.

--- Additional comment from Jes.Sorensen on 2012-01-05 03:43:09 EST ---

Stephen,

No worries, I've been battling being under the weather myself. I think
you have a genuine bug here, but -a seems to work as a workaround for me.
I think it respects the bitmap, but I would certainly appreciate it if you
can confirm that it does.

I will continue investigating why -I doesn't do the right thing.

Cheers,
Jes

--- Additional comment from sschaefer on 2012-01-05 11:49:28 EST ---

I'm pleased to report that, with the -a workaround, the write intent map works as hoped.  Thanks,

    - Stephen

--- Additional comment from Jes.Sorensen on 2012-01-06 01:32:24 EST ---

Stephen,

Thanks for confirming this. This means at least the core infrastructure
seems to be working as expected. I'll look into why it doesn't do the
right thing with -I

Cheers,
Jes