4.5.0 hung on futex under load

[Stefan Metzmacher]

Hi Andrew,

> We upgraded from v4.4.4 to v4.5.0 last Saturday.
> 
> On Monday during a period of heavy load, we found Samba unable to
> complete the setup of new connections.  As clients kept retrying, the
> number of smbd processes built up to nearly 10,000.  We had to stop
> Samba (took a while!) and restart.
> 
> The same thing happened on Tuesday, and I was able to gather some
> information before shutting down Samba.  I reverted to v4.4.4, and the
> problem has not returned.
> 
> I have attached my smb.conf file.  We run winbind (just for managing DC
> connections), nmbd, and smbd.  The operating system is fully-patched
> Debian 8 with kernel 3.16.36-1+deb8u1.
> 
> While the problem was occurring, I strace'd the parent smbd process and
> captured the child process straces.  They all end with:
> 
> futex(0x7fa4db6130a8, FUTEX_WAIT, 2147513662, NULL <detached ...>
> 
> (the same uaddr and val on all of them)
> 
> I have attached one of these trace files as an example.  They are all
> identical except for IP addresses and pid numbers.
> 
> I have also attached the lsof of the same process as the attached trace
> file.

A "thread apply all bt full" from gdb would be useful.

> I ran smbstatus while this was happening, but I don't see these hung
> processes listed there.
> 
> So far as I know, existing connections continued to work.
> 
> Is this a known issue with v4.5.0?  I couldn't find anything in bugzilla
> relating to this.

Maybe it related to
https://bugzilla.samba.org/show_bug.cgi?id=12045

metze



-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: OpenPGP digital signature
URL: <http://lists.samba.org/pipermail/samba-technical/attachments/20161015/0434f57e/signature.sig>