[Linux-cluster] NFS and gfs problems

[Bas van der Vlies]

Pavel Stano wrote:
> i have same problem, kernel 2.6.21.6-grsec x86_64 with cluster 1.0.4
> from source = cannot write file over nfs wich is bigger than cca 2kb
> write() return EINVAL or something like that
> i dont try 1.0.3
> 
Pavel,

  Thanks for the info.  Our 1.0.3 version with NFS is very stable. Only the 
last time our filesystems are becoming full and we encounter sky high load 
and hanging NFS-servers. We are now back at this level and done a clean up 
and now everything works as expected again.

I do not know if there is stilol development for the gfs1 STABLE release!

> but i am droping gfs at all from our servers and i am going use xfs on
> one machine + nfs to all other, because gfs and all that cluster stuff
> is not very stable
> 
It is some work, but it is more scalable then one server with serveral 
clients. On our site we have 5 servers with 700 clients.

> Bas van der Vlies wrote:
>> Are we the only site that tried to run cluster version 1.0.4 in combo
>> with  NFS?
>>
>>
>>
>> Bas van der Vlies wrote:
>>> We just downgraded to versions 1.0.3 with some cvs updates kernel
>>> 2.6.17.4
>>> and nfs-kernel-server 1.0.6.
>>>
>>> Now we can run vim/tar/bonnie++ and other unix utilities without any
>>> problem.
>>>
>>> With this version we experienced problems when the gfs filesystems
>>> becomimg
>>> full the nfsd daemons will fail and to the load of the machine becomes
>>> equal to number of nfsd daemoms and the system is unresponsive.
>>>
>>> Regards
>>>
>>> Bas van der Vlies wrote:
>>>> gfs version 1.0.4
>>>> kernel 2.6.20.16
>>>> nfs-common/etch uptodate 1:1.0.10-6+etch.1
>>>> nfs-kernel-server/etch uptodate 1:1.0.10-6+etch.1
>>>>
>>>> We have a five node fileserver cluster. The GFS file system works
>>>> perfectly, but on our clients we have major problems. See examples below
>>>>
>>>> Do other people see this problems and is there a solution for this
>>>> problem?
>>>>
>>>>
>>>> bas at gb2-r39n16:~$ vim j
>>>> E72: Close error on swap filebas at gb2-r39n16:~$
>>>>
>>>> bas at gb2-r39n16:~$ tar cvf bin.tar bin
>>>> bin/
>>>> bin/mpicc-wrapper-data.txt
>>>> bin/mpiexec
>>>> bin/setmvapich
>>>> bin/modenv
>>>> bin/jobnodes
>>>> bin/SMclient
>>>> bin/ssh
>>>> tar: bin.tar: Warning: Cannot close: Invalid argument
>>>> bas at gb2-r39n16:~$
>>>>
>>>>
>>>>
>>>> bas at gb2-r39n16:~$ /usr/sbin/bonnie++
>>>> Writing with putc()...done
>>>> Writing intelligently...done
>>>> Rewriting...Can't read a full block, only got 3944 bytes.
>>>> Can't read a full block, only got 8040 bytes.
>>>> Can't read a full block, only got 3944 bytes.
>>>> Can't read a full block, only got 8040 bytes.
>>>> Can't read a full block, only got 3944 bytes.
>>>> Can't read a full block, only got 3944 bytes.
>>>> Can't read a full block, only got 3944 bytes.
>>>> Can't read a full block, only got 8040 bytes.
>>>> Bad seek offset
>>>> Error in seek(0)
>>>>
> 
> 
> --
> Pavel Stano
> 
> --
> Linux-cluster mailing list
> Linux-cluster at redhat.com
> https://www.redhat.com/mailman/listinfo/linux-cluster


-- 
********************************************************************
*                                                                  *
*  Bas van der Vlies                     e-mail: basv at sara.nl      *
*  SARA - Academic Computing Services    phone:  +31 20 592 8012   *
*  Kruislaan 415                         fax:    +31 20 6683167    *
*  1098 SJ Amsterdam                                               *
*                                                                  *
********************************************************************