[gpfsug-discuss] Extended attributes and ACLs with AFM-based "NFS migration"

Loic Tortay tortay at cc.in2p3.fr
Fri Apr 15 17:06:41 BST 2016


Hello,
I have a testbed cluster where I have setup AFM for an incremental NFS
migration between 2 GPFS filesystems in the same cluster.  This is with
Spectrum Scale 4.1.1-5 on Linux (CentOS 7).

The documentation states: "On a GPFS data source, AFM moves all user
extended attributes and ACLs, and file sparseness is maintained."
(SpectrumScale 4.1.1 Advanced Administration Guide, page 226)

If I'm not mistaken, I have a GPFS data source (since I'm doing a
migration from GPFS to GPFS).

While file sparseness is mostly maintained, user extended attributes and
ACLs in the source/home filesystem do not appear to be migrated to the
target/cache filesystem (same goes for basic tests with ACLs):
 % getfattr /fs1/zone1/s04/1900/3e479a3eb2eb92d419f812ba1287e8c6269
 getfattr: Removing leading '/' from absolute path names
 # file: fs1/zone1/s04/1900/3e479a3eb2eb92d419f812ba1287e8c6269
 user.mfiles:sha2-256
 %
While on the target filesystem:
 % getfattr /fs2/zone1/s04/1900/3e479a3eb2eb92d419f812ba1287e8c6269
 %

Am I missing something ?  Is there another meaning to "user extended
attributes" than OS level extended attributes (i.e. non-GPFS xattr) ?


Loïc.
-- 
|       Loïc Tortay <tortay at cc.in2p3.fr> - IN2P3 Computing Centre      |



More information about the gpfsug-discuss mailing list