[Gluster-users] glusterfs-3.4.2.qa4: problem accessing files when adding a replica brick

[Johan Huysmans]

Hi All,

I'm testing the 3.4.2.qa4 to see if the "Structure Needs cleaning" fix 
is working.
During my deployement I got a new problem which wasn't there in 3.4.1

I have a KVM server running multiple vhosts with there disk stored as a 
file on a fuse-glusterfs mountpoint.
These machines are running without any problems when 1 brick is configured.

I add an new brick as replica 2 with following command:
# gluster volume add-brick testvolume replica 2 SRV-2:/gluster/brick2

 From that moment my virtual machine (BSC-1) looses there disk (ata 
device error).
When I stop and restart my virtual machines the problem is solved.

I've remount the mountpoint with DEBUG resulting in big logfile:
http://tempsend.com/53ABEE479C

This are some main errors:
[2013-12-19 11:49:21.440618] W [fuse-resolve.c:546:fuse_resolve_fd] 
0-fuse-resolve: migration of basefd (ptr:0x15ed8ec inode-gfid:e31eb690-0
b47-4204-8e6c-7a10116458c8) did not complete, failing fop with EBADF 
(old-subvolume:testvolume-0 new-subvolume:testvolume-1)
[2013-12-19 11:49:21.440731] W [fuse-bridge.c:2167:fuse_writev_cbk] 
0-glusterfs-fuse: 47363: WRITE => -1 (Invalid argument)
[2013-12-19 11:49:21.444920] W [fuse-resolve.c:546:fuse_resolve_fd] 
0-fuse-resolve: migration of basefd (ptr:0x15ed8ec inode-gfid:e31eb690-0
b47-4204-8e6c-7a10116458c8) did not complete, failing fop with EBADF 
(old-subvolume:testvolume-0 new-subvolume:testvolume-1)
[2013-12-19 11:49:21.445042] W [fuse-bridge.c:2167:fuse_writev_cbk] 
0-glusterfs-fuse: 47365: WRITE => -1 (Bad file descriptor)
[2013-12-19 11:49:21.449096] W [fuse-resolve.c:546:fuse_resolve_fd] 
0-fuse-resolve: migration of basefd (ptr:0x15ed8ec inode-gfid:e31eb690-0
b47-4204-8e6c-7a10116458c8) did not complete, failing fop with EBADF 
(old-subvolume:testvolume-0 new-subvolume:testvolume-1)



When reading and writing small files (every 0.5 sec) during that 
add-brick command I don't see any problems.
When have a file open with vim during the add-brick , it gives an error 
when closing the file:
E72: Close error on swap file

Is this normal behaviour or a bug (as this was not a problem in a 
previous version I suspect a bug)
Can this be fixed in 3.4.2 or is there some tweaking that can prevent 
this from happening ?

Thanks for your time!
Greetings,
Johan Huysmans