[lustre-discuss] Can't get zfs OST to mount on a freshly built server

[Peter Kjellström]

On Wed, 13 May 2015 10:16:07 -0400
Bob Ball <ball at umich.edu> wrote:

> OK, so, I am seeing EXACTLY the issue reported at the end of LU-6452
> 8 minutes after it was closed by Andreas Dilger.
> https://jira.hpdd.intel.com/browse/LU-6452
> 
> There is no response.  Is there a solution?

As I read it that's what you get when your osd_zfs module does not
match your spl/zfs.

We build a set of kernel+spl+zfs+lustre then leave it alone until we
build another set. (currently using this combo in production for lustre
on zfs: 2.6.32-504.16.2+0.6.4.1-1+2.5.3-5chaos)

Also note that IPoIB is very broken on the -504.8.1 kernel.

/Peter K

> This is Lustre 2.7.0 with (now) zfs 0.6.4.1-1, which was current when 
> the server was built.  I see a number of recent Emails about updates
> to Lustre sources against this zfs version, but is there a solution
> for the standard set of 2.7.0 Lustre rpms?  Or any solution that will
> get me un-stuck?
> 
> [root at umdist01 ~]# cat /etc/redhat-release
> Scientific Linux release 6.6 (Carbon)
> [root at umdist01 ~]# uname -r
> 2.6.32-504.8.1.el6_lustre.x86_64
> 
> I note that my first OSS server, initially built in March against zfs 
> 0.6.3-1, seems to be happy and definitely does not exhibit this
> issue. That server has now also updated to 0.6.4.1, and I am loathe
> to reboot it now (up for 40 days) for fear of losing access to all of
> its stored files.  Perhaps I should downgrade to the earlier zfs and
> cut off automatic zfs updates from that repo?
> 
> Thanks,
> bob
> 
> _______________________________________________
> lustre-discuss mailing list
> lustre-discuss at lists.lustre.org
> http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org