[pnfs] spNFS: fsid value changes which causes errors

[Dean Hildebrand]

> And I should have added, it would be nice to add various performance
> matrix testing in our interop work to deal with the classic
> performance incompatibilities that are found in heterogeneous
> implementations.  I have spoken with some about this idea and
> potentially using filebench workloads to drive this work.
>
> Let me know if there is interest...
>   
Maybe this is something we can discuss at Connectathon (types of 
workloads, multi-client/single client, etc)?  We will be showing up only 
with laptops, so I don't think we can do *performance* testing there, 
but it would be nice to get the ball rolling.

Dean

> Spencer
>
>   
>> On Mar 24, 2008, at 9:08 PM, neutron wrote:
>>
>>     
>>> Hi all,
>>>
>>> I'm running IOzone throughput test on top of spNFS.
>>>
>>> After the client performs some IO operations in the spnfs mount  
>>> point, the spnfs falls into an error.
>>> I traced into the kernel, and found the direct cause of this error  
>>> as below.
>>>
>>> (MDS side)
>>> nfsd4_getdevinfo():
>>>            receives a pnfs_fsid:pnfs_devid with values as  
>>> ffff8100784d19e8:1, instead of the correct value of 0:1. Therefore  
>>> rqst_exp_find(rqstp, FSID_NUM, &fsidv) fails to find the export  
>>> point of a file system with that fsid.
>>>
>>> (Client side)
>>> nfs41_xdr_dec_pnfs_getdeviceinfo()
>>>   -- > decode_getdeviceinfo() --> decode_op_hdr() fails, and  
>>> decode_op_hdr() returns value -2  (NFSERR_NOENT)
>>>
>>> It looks like that the "fsid" value at client side changes,  and  
>>> client sends this fsid value to MDS with a OP_GETDEVICEINFO rpc  
>>> call.  MDS fails to find a FS with this fsid. Hence mds returns an  
>>> error code to client to indicate this mistake.
>>>
>>> Client receives the fsid value from MDS when initializing the  
>>> mount point. After that this value should remain constant.  I'm  
>>> not sure why this value should change. Any suggestions? thanks!
>>>
>>>
>>>
>>> _______________________________________________
>>> pNFS mailing list
>>> pNFS at linux-nfs.org
>>> http://linux-nfs.org/cgi-bin/mailman/listinfo/pnfs
>>>       
>
> _______________________________________________
> pNFS mailing list
> pNFS at linux-nfs.org
> http://linux-nfs.org/cgi-bin/mailman/listinfo/pnfs
>