Hi, I''m using Lustre 1.8.2 on SLES 11, with the following rpms installed: lustre-ldiskfs-3.0.9-2.6.27.39_0.3_lustre.1.8.2_default lustre-modules-1.8.2-2.6.27.39_0.3_lustre.1.8.2_default kernel-default-extra-2.6.27.39-0.3_lustre.1.8.2 kernel-default-base-2.6.27.39-0.3_lustre.1.8.2 lustre-1.8.2-2.6.27.39_0.3_lustre.1.8.2_default kernel-default-2.6.27.39-0.3_lustre.1.8.2 e2fsprogs-1.41.6.sun1-0suse For the OSTs, I have SAN disks presented to two hosts. Reading the Manual "Using MMP" it states, that mmp gets enabled automatically when I use the -- failover parameter to specify the failover host. I have to admit, after formatting the OST, I did not checked whether it is enabled or not. I manage Lustre mounting on the OST hosts via pacemaker, using the Lustre resource script, and it happened, that one of the OSTs was mounted on both hosts at the same time. So I reformatted the OST this way: mkfs.lustre --fsname=WBSTOR01 --ost --param="failover.node=192.168.24.22 at tcp" --mgsnode=192.168.8.150 at tcp --mkfsoptions="-m 0 -b 4096 -E stride=16" -- mountfsoptions="stripe=192,errors=remount-ro,extents,mballoc" --reformat /dev/lvm/WBOST01/WBOST01 And afterwards checking, dumpe2fs shows: dumpe2fs -h /dev/lvm/WBOST01/WBOST01 | grep -i features dumpe2fs 1.41.6.sun1 (30-May-2009) Filesystem features: has_journal ext_attr resize_inode dir_index filetype extent sparse_super large_file uninit_bg The manual says, MMP is activated automatically on new filesystems, and I thought it has been a new filesystem, also when I use the --reformat parameter. Now I enabled MMP manually: tune2fs -O mmp /dev/lvm/WBOST01/WBOST01 And I also see the option testing for it: dumpe2fs -h /dev/lvm/WBOST01/WBOST01 | grep -i features dumpe2fs 1.41.6.sun1 (30-May-2009) Filesystem features: has_journal ext_attr resize_inode dir_index filetype needs_recovery extent mmp sparse_super large_file uninit_bg And I see MMP working, mounting it on the first server prevents a second mount on the other at the same time. So I am wondering whether I missed sth.? regards, Sebastian
On Mon, 2010-03-29 at 10:49 +0100, Sebastian Reitenbach wrote:> Hi,Hi,> mmp gets enabled automatically when I use the -- > failover parameter to specify the failover host.Indeed.> So I reformatted the OST this way: > mkfs.lustre --fsname=WBSTOR01 --ost --param="failover.node=192.168.24.22 at tcp" > --mgsnode=192.168.8.150 at tcp --mkfsoptions="-m 0 -b 4096 -E stride=16" -- > mountfsoptions="stripe=192,errors=remount-ro,extents,mballoc" --reformat > /dev/lvm/WBOST01/WBOST01 > > And afterwards checking, dumpe2fs shows: > dumpe2fs -h /dev/lvm/WBOST01/WBOST01 | grep -i features > dumpe2fs 1.41.6.sun1 (30-May-2009) > Filesystem features: has_journal ext_attr resize_inode dir_index filetype > extent sparse_super large_file uninit_bg > > The manual says, MMP is activated automatically on new filesystems...when they are formatted with the --failnode option set. You didn''t use the --failnode option. You set the failover.node parameter. Having not looked at the actual code, I''d guess that only the --failnode option sets MMP and parameters are not considered for the purpose of MMP setting. The moral of the story is to use --failnode, not set the failover.node parameter when formatting filesystem. And file a bug about lack of MMP when setting the parameter instead of using --failnode. :-) b. -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 197 bytes Desc: This is a digitally signed message part Url : http://lists.lustre.org/pipermail/lustre-discuss/attachments/20100329/4d2e526d/attachment.bin
Hi, On Monday 29 March 2010 12:56:49 pm Brian J. Murrell wrote:> On Mon, 2010-03-29 at 10:49 +0100, Sebastian Reitenbach wrote: > > Hi, > > Hi, > > > mmp gets enabled automatically when I use the -- > > failover parameter to specify the failover host. > > Indeed. > > > So I reformatted the OST this way: > > mkfs.lustre --fsname=WBSTOR01 --ost > > --param="failover.node=192.168.24.22 at tcp" --mgsnode=192.168.8.150 at tcp > > --mkfsoptions="-m 0 -b 4096 -E stride=16" -- > > mountfsoptions="stripe=192,errors=remount-ro,extents,mballoc" --reformat > > /dev/lvm/WBOST01/WBOST01 > > > > And afterwards checking, dumpe2fs shows: > > dumpe2fs -h /dev/lvm/WBOST01/WBOST01 | grep -i features > > dumpe2fs 1.41.6.sun1 (30-May-2009) > > Filesystem features: has_journal ext_attr resize_inode dir_index > > filetype extent sparse_super large_file uninit_bg > > > > The manual says, MMP is activated automatically on new filesystems > > ...when they are formatted with the --failnode option set. You didn''t > use the --failnode option. You set the failover.node parameter. > > Having not looked at the actual code, I''d guess that only the --failnode > option sets MMP and parameters are not considered for the purpose of MMP > setting. > > The moral of the story is to use --failnode, not set the failover.node > parameter when formatting filesystem. And file a bug about lack of MMP > when setting the parameter instead of using --failnode. :-)thanks for pointing out. with the --failnode parameter everything works like a charm. I opened bug report #22490 in the lustre bugzilla. regards, Sebastian
On Mon, 2010-03-29 at 15:11 +0100, Sebastian Reitenbach wrote:> thanks for pointing out.NP.> with the --failnode parameter everything works like a > charm.Great!> I opened bug report #22490 in the lustre bugzilla.Thanx much! b. -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 197 bytes Desc: This is a digitally signed message part Url : http://lists.lustre.org/pipermail/lustre-discuss/attachments/20100329/6cdc4763/attachment.bin