[lustre-discuss] 2.7.63 compilation problem with OFED-3.18-1-20151109-0727 and ZFS 0.6.5.2; o2ib failure

Nathan R.M. Crawford nrcrawfo at uci.edu
Fri Nov 20 11:36:11 PST 2015


Hi All,

  I'm attempting to install Lustre on ZFS on EL6.7 using Truescale HCAs.
Not surprisingly, this involves several pre-release versions of various
packages. Also not surprisingly, there are some snags.

Version info:
Scientific Linux (CERN) 6.7 using kernel 2.6.32-573.7.1.el6.x86_64
ZFS 0.6.5.2
OFED 3.18-1-20151109-0727 (ib_qib)
Lustre 2.7.63

Lustre configure options:
./configure --disable-ldiskfs --disable-gss --with-o2ib \
--with-zfs=/usr/src/zfs-0.6.5.2 --with-spl=/usr/src/spl-0.6.5.2 \
--with-linux=/usr/src/kernels/2.6.32-573.7.1.el6.x86_64 \
--with-linux-obj=/usr/src/kernels/2.6.32-573.7.1.el6.x86_64

  The compilation issue occurs in the osd-zfs part, but is due to some
configure tests failing on redefined structures/functions in kthread.h.
Elsewhere in configre, HAVE_COMPAT_RDMA is used to avoid the conflicting
OFED and kernel versions of kthread.h, but I think this wasn't carried over
into the zfs parts.

  To work around this, I forced configure to add -DCONFIG_COMPAT_IS_KTHREAD
to EXTRA_OFED_INCLUDE. The redefinition problem goes away and the code
compiles. I can mount a lustre filesystem using tcp on ethernet or IPoIB,
but o2ib fails when trying to mount an OST. The nodes can ping each other
on o2ib, but throw up many errors during the lst brw tests.

  Is there something trivial I'm missing? Otherwise, what is the most
useful way to package up a bug report?

Thanks,
Nate Crawford

-- 

Dr. Nathan Crawford              nathan.crawford at uci.edu
Modeling Facility Director
Department of Chemistry
1102 Natural Sciences II         Office: 2101 Natural Sciences II
University of California, Irvine  Phone: 949-824-4508
Irvine, CA 92697-2025, USA
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20151120/a0220969/attachment.htm>


More information about the lustre-discuss mailing list