Am 25.06.2012 22:23, schrieb Josef Bacik:

On Mon, Jun 25, 2012 at 02:20:31PM -0600, Stefan Priebe wrote:

quoted

Am 25.06.2012 22:11, schrieb Josef Bacik:

quoted

On Mon, Jun 25, 2012 at 01:33:09PM -0600, Stefan Priebe wrote:

quoted

With v3.4 the same. Can't go back more as this really results in very
fast corruption. Any ideas how to debug?

What workload are you running?  I have a ssd here with discard support I can try
and reproduce on.  Thanks,

i'm using fio with 50 jobs and randwrite of 4k blocks in ceph but i
don't know which load ceph then exactly generates. ;-(

Thats fine, I have this handy "create a local ceph cluster" script from an
earlier problem, just send me your fio job and I'll run it locally.  Thanks,

OK my fio job is running in a KVM using a RBD:

fio --filename=$DISK --direct=1 --rw=randwrite --bs=4k --size=200G 
--numjobs=50 --runtime=300 --group_reporting --name=file1

Backed by 3 server with 4 OSDs (all intel SSDs) each running btrfs on
top.

THANKS!

Greets
Stefan