Greetings, We are running a lustre setup on top of ZFS.
Mounting a lustre snapshot on a client and thereafter running:
'lfs data_version -n SNAPSHOT_FILEPATH'
Makes the MDS server fail with a kernel taunt (dump_stack call) thrown from
'lustre/osd-zfs/osd_handler.c : osd_trans_create':
------------------------------------------------------------------------------------
if (dt->dd_rdonly) {
CERROR("%s: someone try to start transaction under "
"readonly mode, should be disabled.\n",
osd_name(osd_dt_dev(dt)));
dump_stack();
RETURN(ERR_PTR(-EROFS));
}
------------------------------------------------------------------------------------
In the above (lfs dataversion) example we can prevent the kernel taunt by
mounting the snapshot (client side) with 'noatime', but that doesn't help us if
we try to run:
`lfs getstripe -d SNAPSHOT_FILEPATH'
We have traced the problem and found that it is the function calls 'mdt_open.c
: mo_attr_set' and 'mdt_open.c : mo_xattr_set' that triggers the kernel taunt.
Our MDS crashed hard after running the above tests on millions of (snapshot)
files simultaneously and we had to restore it from generic ZFS snapshots.
Is this a know issue ?
Best Regards,
Martin
signature.asc
Description: Message signed with OpenPGP
_______________________________________________ lustre-discuss mailing list lustre-discuss@lists.lustre.org http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org
