[Gluster-devel] Occasional I/O error

[Снежана Бекова]

  Hello,
I'm running glusterfs 1.4.0pre5 (glusterfs--mainline--3.0--patch-359)  
and fuse-2.7.3glfs10 on 2 machines with AFR with client side  
replication. My test setup is: 2 glusterfs servers and 2 glusterfs  
clients,i.e. the two mashines (server1 and server2) are configured as  
server and client. I was getting occasional Input/output error when  
listing glisterfs (afr) directory on server1.

  The glusterfs client log messages are:
2008-10-06 12:55:02 E [afr_self_heal.c:123:afr_lds_setdents_cbk]  
afr-wwwroot: op_ret=-1 op_errno=17
2008-10-06 12:55:02 E [afr_self_heal.c:123:afr_lds_setdents_cbk]  
afr-wwwroot: op_ret=-1 op_errno=17
2008-10-06 12:55:02 E [fuse-bridge.c:398:fuse_entry_cbk]  
glusterfs-fuse: 196: LOOKUP() / => -1 (Input/output error)

  The messages in the glusterfs server log are:
2008-10-06 12:54:01 C [posix.c:2756:ensure_file_type] wwwroot: entry  
/wwwroot//xxx.xxxx.xx is a different type of file than expected

In afr directory there are many symlinks and the file  
/wwwroot//xxx.xxxx.xx is a symlink. So I must stop glusterfs client  
and server processes on server1, remove the symlinks, start them again  
and listing the glusterfs directory to remove the problem.

  My config files on the two client and server mashines are:
cat /etc/glusterfs/glusterfs-server.vol
volume wwwroot
  type storage/posix                  
  option directory /wwwroot           
end-volume

  volume server
  type protocol/server
  option transport-type tcp/server
  subvolumes wwwroot
  option auth.addr.wwwroot.allow 10.0.0.*,127.0.0.1
end-volume

  cat /etc/glusterfs/glusterfs-client.vol
volume client-server1-wwwroot
  type protocol/client
  option transport-type tcp/client
  option remote-host 127.0.0.1   
  option remote-subvolume wwwroot
end-volume

  volume client-server2-wwwroot
  type protocol/client
  option transport-type tcp/client
  option remote-host 10.0.0.100  
  option remote-subvolume wwwroot
end-volume

  volume afr-wwwroot
  type cluster/afr
  subvolumes client-server1-wwwroot client-server2-wwwroot
end-volume

I think the problem does not exist on version 1.3.12.
Maybe it is a bug or can you help me what is wrong?

  Thanks,
Snezhana