[Linux-cluster] GFS on iSCSI+LVM

[Michael Weitzel]

Hi,

after some time of successfull stress-testing I got an I/O-error after
some time and the following message in the system log:

GFS: fsid=Isabella:301bis304.3: fatal: invalid metadata block
GFS: fsid=Isabella:301bis304.3:   bh = 39254347 (type: exp=4, found=0)
GFS: fsid=Isabella:301bis304.3:   function = gfs_get_meta_buffer
GFS: fsid=Isabella:301bis304.3:   file =
/usr/src/packages/BUILD/cluster-cvs.STABLE/gfs-kernel/src/gfs/dio.c,
line = 1223
GFS: fsid=Isabella:301bis304.3:   time = 1139576449
GFS: fsid=Isabella:301bis304.3: about to withdraw from the cluster
GFS: fsid=Isabella:301bis304.3: waiting for outstanding I/O
GFS: fsid=Isabella:301bis304.3: telling LM to withdraw
lock_dlm: withdraw abandoned memory
GFS: fsid=Isabella:301bis304.3: withdrawn

gfs_fsck was able to correct this error. My "playground"-config is
probably rather unusual *g*
(I'm trying to use spare disks in cluster nodes to emulate a SAN):

 - 4 nodes of a cluster export spare disks via iSCSI (ietd)
 - 8 nodes (the 4 ietd-nodes included) import the iSCSI-targets
   with open-iscsi (v. 1.0-485) (nodes do not import their own exported
   disks)
 - the four local/imported disks form a LVM volume-group with a single
   large logical volume with GFS/lock_dlm/clvmd

Is clock-syncronization important? What caused this error?
Are there any experiences with iSCSI+Linux Software RAID (Level 1)+GFS?
-- 
Michael Weitzel