[pnfs] pNFS testing at connectathon
Lisa Week
Lisa.Week at Sun.COM
Mon Jan 15 19:07:50 EST 2007
Hi Rahul,
Iyer, Rahul wrote On 01/10/07 14:42,:
>Hi Lisa,
>The last time we tried interoperability testing at the Bakeathon, we
>couldn't do I/O. Just to recap, our client couldn't do any I/O because
>your server returned NFS4ERR_TRY_LATER on a LAYOUTGET. The reason for
>this was that that our client could issue more than one LAYOUTGET per
>file. If your server received a LAYOUTGET for a file for which a layout
>was already handed to the concerned client, then it returns
>NFS4ERR_TRY_LATER.
>
Yes, our server behavior was probably wrong. Multiple LAYOUTGETs issued
from the same client for the same file should succeed.
>We will work on trying to make sure we don't do this,
>however, I do not believe it is illegal for a client to issue more than
>one LAYOUTGET for the same file and byte range; at least the spec
>doesn't outlaw this.
>
>
If a client issues a LAYOUTGET for a file and byte range that it already
has a layout for, the files based server should just reply to the
LAYOUTGET request with the layout that has already been handed out.
Correct?
>Could you please look into this?
>
>
As far as what was happening at Bakeathon (just from memory)...
I think the root cause of why our server was returning
NFS4ERR_LAYOUTTRYLATER to you was that upon receipt of the LAYOUTGET
from your client, we would record that a layout was requested for the
the file and byte range specified and then we would contact all of the
data servers included in the layout. When doing this, we hung trying to
contact one of the data servers and couldn't respond to the LAYOUTGET.
Then, I believe the LAYOUTGET request was retransmitted by your client
and that is when we responded with NFS4ERR_LAYOUTTRYLATER.
We (well actually Robert Gordon) are changing our server side LAYOUTGET
processing so I don't think we'll see this error at Connectathon.
Thanks,
Lisa
>>-----Original Message-----
>>From: Iyer, Rahul
>>Sent: Wednesday, December 20, 2006 2:37 PM
>>To: Lisa.Week at Sun.COM
>>Cc: pnfs at linux-nfs.org
>>Subject: Re: [pnfs] pNFS testing at connectathon
>>
>>OK, then -08 is what we'll work with.
>>Regards
>>Rahul
>>
>>
>>
>>
>>>-----Original Message-----
>>>From: Lisa Week [mailto:Lisa.Week at Sun.COM]
>>>Sent: Wednesday, December 20, 2006 2:35 PM
>>>To: Iyer, Rahul
>>>Cc: William A. (Andy) Adamson; pnfs at linux-nfs.org
>>>Subject: Re: [pnfs] pNFS testing at connectathon
>>>
>>>Hi Rahul,
>>>
>>>
>>>
>>>>I was working of the minorversion1-07 dot-x file. Will it be
>>>>
>>>>
>>>too much
>>>
>>>
>>>>of a problem for us to stick to this one? I haven't read the
>>>>
>>>>
>>>-08 dot-x
>>>
>>>
>>>>file yet. I'm yet to integrate the sessions code with pNFS and I'd
>>>>appreciate if I didn't need to go implement new ops, not to
>>>>
>>>>
>>>mention any
>>>
>>>
>>>>XDR changes, given the remaining time. So, unless it will be
>>>>
>>>>
>>>a lot of
>>>
>>>
>>>>trouble, I'd like to stick with the -07 dot-x.
>>>>
>>>>
>>>>
>>>>
>>>We have already moved our implementation forward to the
>>>minorversion1-08 dot-x file. Sorry, but it is doubtful
>>>
>>>
>>that we can go
>>
>>
>>>back to the
>>>minorversion1-07 dot-x.
>>>
>>>Given that we have already moved forward to -08, it would take some
>>>work to revert our implementation back to -07.
>>>Unfortunately, the work to revert to -07 would just put us behind
>>>where we are right now and it is hard to justify that.
>>> Therefore, we would like to take -08 based implementations to
>>>Connectathon.
>>>
>>>So, what do you think?
>>>
>>>Thanks,
>>>Lisa
>>>
>>>
>>>
>>>>
>>>>
>>>>
>>>>
>>>>>-----Original Message-----
>>>>>From: Lisa Week [mailto:Lisa.Week at Sun.COM]
>>>>>Sent: Tuesday, December 19, 2006 10:18 PM
>>>>>To: Iyer, Rahul
>>>>>Cc: William A. (Andy) Adamson; pnfs at linux-nfs.org
>>>>>Subject: Re: [pnfs] pNFS testing at connectathon
>>>>>
>>>>>Hi Rahul,
>>>>>
>>>>>Thanks for the info.
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>>>The only thing I wonder is - could we get more details
>>>>>>>
>>>>>>>
>>about what
>>
>>
>>>>>>>Sessions ops you are planning to have implemented at
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>Connectathon.
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>>>More specifically, what ops do you mean when you say "the
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>sessions op
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>>>to replace the clientid we've added to several pNFS operations"?
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>I'm doing the sessions bit for the client right now. The
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>ops I support
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>>are create_clientid, create_session, sequence and (not yet, but
>>>>>>hopefully) destroy session. On the callback path
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>CB_SEQUENCE will be
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>>all I support.
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>Would you be able to tell me which version of the dot-x
>>>>>
>>>>>
>>>file that you
>>>
>>>
>>>>>are using for your Sessions implementation (i.e.
>>>>>which minor version draft does it correspond to)?
>>>>>
>>>>>I apologize if I have misunderstood something, but...
>>>>>By the time Connectathon comes around do you intend to be on the
>>>>>draft-ietf-nfsv4-minorversion1-08 dot-x file? In other
>>>>>
>>>>>
>>>words, can you
>>>
>>>
>>>>>commit to supporting EXCHANGE_ID in lieu of CREATE_CLIENTID?
>>>>>
>>>>>Just for clarification, here is the dot-x file that we are working
>>>>>toward: http://nfsv4-editor.org/draft-08/nfsv41.x
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>>>Also, are you planning to have both client and server
>>>>>>>
>>>>>>>
>>>with pNFS and
>>>
>>>
>>>>>>>Sessions functionality at Connectathon?
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>Yes. I will be integrating sessions and pNFS on the
>>>>>>
>>>>>>
>>Linux client.I
>>
>>
>>>>>>will be doing the same on the Netapp server too. However,
>>>>>>
>>>>>>
>>>I have put
>>>
>>>
>>>>>>in minimal stubs for the Linux server. One of Andy, Marc or
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>Benny mostly
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>>will integrate this with pNFS.
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>Okay.
>>>>>
>>>>>Thanks,
>>>>>Lisa
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>_______________________________________________
>>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
>
>
More information about the pNFS
mailing list