[lustre-discuss] MDT backup using dump
Dauchy, Nathan (ARC-TNC)[Computer Sciences Corporation]
nathan.dauchy at nasa.gov
Thu Apr 21 19:36:39 PDT 2016
As an alternative to using tar and file level backups (see my posting earlier today), we are investigating options that might allow backups to be completed in a shorter amount of time, yet not consume as much space as a full dd of the device. In particular, we are considering the "dump" utility. It has the added benefit of supporting incremental backups that might come in handy. Package is here:
The use of "dump" for Lustre was discussed 7 years ago here:
And support for ext4 was added to dump/restore in 2009:
Has anyone successfully used the "dump" approach in the last few years with recent Lustre versions?
Our current platform is based on CentOS-6.7 and lustre-2.5.3. The stock OS version of dump doesn't seem to understand ext4/ldiskfs... and I ran into problems building the latest dump from source as follows:
# lvcreate -l95%FREE -s -p r -n mdt_snap /dev/mapper/nbptest--vg-mdttest
# blockdev --setro /dev/mapper/nbptest--vg-mdt_snap
# mount -t ldiskfs -o ro,noload /dev/mapper/nbptest--vg-mdt_snap /mnt/snap
# dump -0uf /tmp/mdt_backup/mdt_backup.dump /mnt/snap
DUMP: Date of this level 0 dump: Tue Apr 19 16:20:07 2016
DUMP: Dumping /dev/mapper/nbptest--vg-mdt_snap (/mnt/snap) to /tmp/mdt_backup/mdt_backup.dump
DUMP: Unsupported feature(s) 0x1000 in filesystem
/dev/mapper/nbptest--vg-mdt_snap: Filesystem has unsupported feature(s) while opening filesystem
DUMP: The ENTIRE dump is aborted.
# cd /tmp/dump-0.4b44
traverse.c: In function 'convert_dir':
traverse.c:1061: error: invalid type argument of '->' (have 'int')
traverse.c:1061: warning: passing argument 1 of 'ext2_get_dirent_size' makes pointer from integer without a cast
/usr/include/ext2fs/ext2_fs.h:885: note: expected 'struct ext2_dir_entry *' but argument is of type 'int'
make: *** [traverse.o] Error 1
make: Leaving directory `/tmp/dump-0.4b44/dump'
make: *** [all] Error 1
# rpm -qf /usr/include/ext2fs/ext2_fs.h
(most current e2fsprogs available from https://downloads.hpdd.intel.com/public/e2fsprogs/)
Has ldiskfs diverged such that this a bad path to go down? (last release of dump was back in 2011)
Is there a simple fix in there somewhere that I'm missing?
All suggestions are much appreciated!
More information about the lustre-discuss