Re: [nfsv4] [FedFS] Meeting Minutes, 9/30/2010

[Chuck Lever <[e-mail]>]

On Sep 30, 2010, at 3:51 PM, James Lentini wrote:

> 
> FedFS Meeting Minutes, 9/30/2010
> --------------------------------
> 
> Attendees
> ---------
> 
> Andy Adamson (NetApp)
> Craig Everhart (NetApp)
> Sorin Faibish (EMC)
> James Lentini (NetApp)
> Chuck Lever (Oracle)
> Paul LeMahieu (EMC)
> Peter Staubach (EMC)
> Robert Thurlow (Oracle)
> Mark Uddin (EMC)
> 
> Minutes
> -------
> 
> + IETF Note Well Agreement
> 
>  This is a reminder that our discussions are governed by the 
>  IETF Note Well Agreement. See:
> 
>    http://www.ietf.org/NOTEWELL.html
> 
>  We will start each week's meeting with this announcement.
> 
> + October Bake-a-thon Planning
> 
>  Mark Uddin has been configuring equipment for the Bake-a-thon. He has 
>  installed and configured a DNS server on a Windows 2008 Server for 
>  use testing the NFS DNS SRV record. This machine will also be used 
>  for other purposes: VMware vCenter, pNFS testing.
> 
>  Mark asked if this DNS server will be useful for the NFS DNS SRV testing.
>  Rob expected that it would be fine if it could return a standard DNS SRV 
>  record.
> 
>  Mark reported that the NFS DNS SRV record is not setup yet on the server. 
>  This DNS server will also be the DHCP server for the event.
> 
>  We haven't tested this configuration yet, but we would expect it to work
> 
>  Chuck asked if want to do IPv6 testing and if so does this DNS server
>  support IPv6?
> 
>  Mark was unsure of the IPv6 support. We will check on this.
> 
>  Chuck requested that we have a procedure for installing an SRV record on
>  this system. 
> 
>  Mark asked for a pointer to the NFS DNS SRV specification.
> 
>  [Editor's Note: the draft is here:
>   http://tools.ietf.org/html/draft-ietf-nfsv4-federated-fs-dns-srv-namespace-05]
> 
>  Craig Everhart will send Sorin an email with the important points 
>  of the NFS DNS SRV record.
> 
>  Sorin reported that the Bake-a-thon talk agenda is available on 
>  the website:
> 
>  http://www.citi.umich.edu/projects/nfsv4/citi_bakeathon/15th_nfsv4_bakeathon.html
> 
>  There will be conference rooms for break-out meeting with polycom 
>  phones if someone needs to use them.
> 
>  The wireless network will allow for outbound VPN traffic.
> 
>  The wired network configuration has already been defined. Sorin will 
>  send specifics on the IP addresses. 
> 
>  There is time on Friday to discuss ad hoc issues and summarize.
> 
>  Sorin will send his cell phone in case any issues arise during the weekend.
> 
> + FEDFS_*_REPLICATION
> 
>  Craig noticed two items on a second reading of the draft.
> 
>  First, the description of the FEDFS_*_REPLICATION procedures is relatively 
>  clear that they set the FSN for a present fileset, but don't manage a replication
>  relationship between fileservers. However, he could imagine some confusion 
>  in this area and may propose additional text to make this point more obvious.
> 
>  Secondly, Craig pointed out that the FEDFS_CREATE_REPLICATION procedure's 
>  behavior when an "replication" FSN is already set on a fileset is undefined.
>  The FEDFS_CREATE_JUNCTION procedure will fail with FEDFS_ERR_EXIST. We agreed 
>  that the behavior of FEDFS_CREATE_JUNCTION and FEDFS_CREATE_REPLICATION 
>  do not need to be the same. For FEDFS_CREATE_REPLICATION, Craig proposed that
>  this procedure act a if it were setting an attribute and so it should 
>  succeed if called on a fileset with an existing "replication" FSN.
> 
>  Rob agreed with Craig's proposed behavior.
> 
>  We discussed if implementing the FEDFS_*_REPLICATION procedures 
>  are required (MUSTs) or optional (SHOULDs) for internal nodes in 
>  the namespace.
> 
>  Rob observed that if NFS4ERR_MOVED is not returned for an object, the 
>  NFS server is not required to support fs_locations. Therefore, the 
>  FEDFS_*_REPLICATION procedures should be optional (SHOULDs or even MAYs).
> 
>  We agreed to list these as SHOULDs in the next revision.
> 
>  Chuck asked if this attribute follows a copy of a filset? Chuck observed 
>  that one way to implement this in Linux is to use a filesystem's UUID to 
>  location its "replication" FSN value. With that implementation, it would 
>  be somewhat cumbersome for a copy of the filesystem (with different UUID)
>  to locate its "replication" FSN value and return the same fs_locations 
>  value.
> 
>  Rob noted that NFSv4's replication support is designed to work across 
>  heterogeneous servers. In many instances, a "cp -r" command will be 
>  sufficient to create a replica. For that reasons, it seemed prudent 
>  to pursue Linux designs other than the one described above.
> 
>  Chuck asked if there recommendations on how to generate a UUID. 
>  James suggested that folks review recommendations in Section 4.2.1.1 
>  of draft-ietf-nfsv4-federated-fs-protocol-09 and reply with 
>  feedback.
> 
>  We noted that the FEDFS_*_REPLICATION procedures have not be 
>  implemented in either Linux or Solaris yet.
> 
> + Admin Draft Update
> 
>  Draft -06 was posted on 9/29 with the following changes:
> 
> - Each procedure now lists the errors it is allowed to return
> 
> - In 5.2.1, the description of FEDFS_DELETE_JUNCTION
>   indicated that if the targeted object was not a 
>   junction, the error would be FEDFS_ERR_INVAL. In 
>   similar situations in other parts of the specification, 
>   the error is FEDFS_ERR_NOTJUNCT. The FEDFS_DELETE_JUNCTION
>   description was corrected to use the FEDFS_ERR_NOTJUNCT 
>   error.
> 
> - The NOTEMPTY and NOTDIR error codes were removed since 
>   they were no longer in use (text that referred to them 
>   was removed from FEDFS_CREATE_JUNCTION in the previous revision).  
> 
> - Error codes (FEDFS_ERR_NSDB_REFERRAL, ...) and text (5.3.2) was 
>   added for LDAP referrals.
> 
> - Type names used by both the JUNCTION and REPLICATION functions were 
>   renamed for consistency (see CreateJunctionArgs, FedFsLookupArgs, 
>   and FedFsLookupRes).
> 
> - Added a FEDFS_ERR_DELAY and FEDFS_ERR_NOTSUPP error values
> 
> - Added cache lookup error codes (FEDFS_ERR_NO_CACHE, 
>   FEDFS_ERR_UNKOWN_CACHE, and FEDFS_ERR_NO_CACHE_UPDATE)
> 
> - Added a synopsis for each procedure
> 
> - Added a description of the FEDFS_NULL procedure (5.1) 
> 
> Chuck had a list of questions for the group.
> 
> - Should there be an error code for when an NSDB is missing an NCE?
>   This seemed like a reasonable addition. We'll add an FEDFS_ERR_NONCE 
>   error to every procedure where NOFSN or NOFSL are allowed.
> 
> - What should be the error code if the NSDB returns a 
>   permission error? We agreed that FEDFS_ERR_NSDB_LDAP_VAL can be used 
>   for this case.
> 
> - What should be the error if an FSL object is missing information? There 
>   is an existing error code, FEDFS_ERR_NSDB_RESPONSE, for malformed 
>   responses. We'll expand on the description of this error to note some of 
>   situations (like this one) in which it should be used.

...and clarify the differences between FEDFS_ERR_NSDB_RESPONSE, FEDFS_ERR_NSDB_FAULT, and FEDFS_ERR_NSDB_LDAP{_VAL}.

In addition, the NSDB_RESPONSE error code can be returned by the ADMIN protocol's FEDFS_LOOKUP_JUNCTION procedure.  The specific question I had is how an NFS fileserver should reply to GETATTR(fs_locations) if the NSDB returns obviously broken FSL information.

Some convention for the NFS server's reply in this case should be agreed upon outside of the ongoing FedFS discussion.

> + Multi-domain Access Draft update
> 
>  Andy has updated the multi-domain access draft. He has sent an email 
>  to the WG reflector summarizing the changes and asking for feedback on 
>  certain points. We will schedule time to discuss this at a future 
>  meeting.
> 
> + Meeting Schedule
> 
>  Our next meeting will be on October 14.
> _______________________________________________
> nfsv4 mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/nfsv4

-- 
chuck[dot]lever[at]oracle[dot]com