[DRBD-user] Primary freezes on Secondary Shutdown

[Andreas Abele]

Hi,

i am testing drbd at the moment and during the tests i found a serious 
problem:

Setup:

2 Hosts with ubuntu-server-i386-7.10
Tested drbd-version:
drbd8 (but 0.7 shows the same behaviour)
tested volume-types:
drbd-partition on top ov lvm2 (but native /dev/sda give the same result)
tested filesystems:
xfs, jfs, ext3

The problem:
on a running drbd (Primary/Secondary) everything is in sync do a:
1.) cd /mnt/drbd
2.) dd if=/dev/zero of=test6g.dat bs=1024k count=6000
2a.) Everything works fine until a do a shutdown or reboot on the 
secondary during the running dd
When i shut down the secondary i get a so called "stuck on D state" for 
the dd process on the primary.
this means, that i cannot reboot the primary, cause the process hangs in 
an internal kernel loop and the primary freezes (hangs) after some 
minutes. The next step is to boot the secondary again.
What happens then is a kind of split brain:
On primary: (Primary/Unknown)
On Secondary: (Secondary/Primary)
but: The Primary is not accessible

So, because i cann not reboot the primary, i have to do a "Hard Reset". 
After that, the Sync ist starting an everthing seems to be fine again.

But the complete behaviour tells me, that a failure on the secondary 
causes the complete system to stop. I do not think that this is okay. So 
my question is:

Is this a Configuration-Error by me?
Is it a drbd-problem?
Or is this a kernel-Problem (see "Stuck on D state")? And shouldn't drbd 
have mechanisms to avoid these kind of processes during I/O?

BTW: It does not depend on the dd-command. Copying big Files or doing 
rsync of large Directories show the same behaviour.

kind regards
Andreas Abele

-- 
  Die schlimmste aller Weltanschauungen ist die derer,
  die die Welt nie angeschaut haben.
  - Alexander von Humboldt

Andreas Abele
Newton-Abbot-Straße 18
74354 Besigheim
Email: andreas.abele at vprofis.de
Phone: +49 7143 60 204
Mobile: +49 177 634 88 26