Bug #2596

mds: spinning on restart

Added by Sage Weil over 11 years ago. Updated about 7 years ago.

Can't reproduce
Target version:
% Done:


Community (dev)
Affected Versions:
Labels (FS):
Pull request ID:
Crash signature (v1):
Crash signature (v2):


from ML:

On Fri, 15 Jun 2012, Amon Ott wrote:
> Hello all,
> I have seen this for a long time, but never investigated further. After stabl$
> test runs for several days, this is our last known show stopper before using  
> Ceph in production. We are running 0.47.2 on 32 Bit.
> If we restart MDS (or all ceph daemons) on all nodes, one after another or al$
> together, they first recover and then the active one starts to spin with full 
> cpu and does not answer any more. After a while, the next takes over, starts  
> to spin, etc., until the whole cluster is unusable. This is completely
> reproducable and happens even without any active client.
> As ecpected, ceph -w shows lots of
> "2012-06-15 11:35:28.588775   mds e959: 1/1/1 up {0=3=up:active(laggy or
> crashed)}" 
> It does not help to stop all services on all nodes for minutes or longer and
> to restart them - MDS will restart spinning. But: If we reboot the whole
> cluster, everything goes back to work.
> Today's MDS log is available at
> Is this a known problem? It has been with us for a looong time now, but since
> rebooting used to help, we never tracked it down.


#1 Updated by Amon Ott over 11 years ago

gdb is not helpful here, process seems to be spinning in syscall:
(gdb) thread apply all bt

Thread 1 (process 14820):
#0 0x55ba0422 in __kernel_vsyscall ()
#1 0x5574df4b in ?? ()
Backtrace stopped: previous frame inner to this frame (corrupt stack?)

strace shows:
futex(0x16b44d1c, FUTEX_WAIT_PRIVATE, 1, NULL

So it seems that a futex deadlock occurred. It might be that it only happens if a client had been connected before.

Will try to get a full strace with spin tomorrow.

#2 Updated by Sage Weil about 11 years ago

  • Status changed from New to Can't reproduce

#3 Updated by John Spray about 7 years ago

  • Project changed from Ceph to CephFS
  • Category deleted (1)

Bulk updating project=ceph category=mds bugs so that I can remove the MDS category from the Ceph project to avoid confusion.

Also available in: Atom PDF