[GFS-WG] RNS v1.1 update
Hideo Matsuda
matsuda at ist.osaka-u.ac.jp
Wed Mar 24 21:35:17 CDT 2010
Dear Tatebe-san,
Thanks a lot for your updating.
Regarding the case in the rename operation, I would like to
support the 2nd choice (fix the order of operations) since
it gives the same semantics as the sequential operations.
Best regards,
Hideo Matsuda
> Hi all,
>
> I updated the RNS 1.1 spec and OGSA WSRF BP rendering docs based
> on the discussion during OGF28 and comments in the tracker.
>
> http://forge.ogf.org/sf/go/artf6426
> http://forge.ogf.org/sf/go/artf6427
>
> The Change includes
>
> - Section 1.1, "a pseudo-schema" is changed to "a pseudo-UML"
> - all "RNSResponseEntry" are changed to "RNSEntryResponse"
> - all "rns:response-entry" are changed to "rns:entry-response"
> - Section 2.3.2, add a missing "</s11:Envelope>" in an example
> of an add request message
> - all "an lookup" are changed to "a lookup"
> - all "an rename" are changed to "a rename"
> - remove Section 3
>
> Regarding "failure", there is a comment as follows
>
>> all the content in the spec up till page 18, use “fault”,
>> including example, but from section 4 onward, all the faults
>> become “failure”, for consistency’s sake, and a common
>> sense in SOAP, “fault” is better here.
>
> I leave it as it is, since this document tries to describe a
> more general way. Please send comments if you think it should
> be changed.
>
> Finally, regarding "uniqueness" of the entry-name, the discussion
> of OGF28 suggests to mention this not only in the add opration
> but also in the rename operation.
>
> When editing about this, I found we need to clarify the case such
> that the target entry exists. In the POSIX, it will be atomically
> replaced. If we follow the POSIX semantics, there is an
> undeterministic condition when renaming multiple entries in a bulk
> operation. For example, when there are two rename requests to the
> same target name, both requests will succeed but the result is
> different depending on which one is applied first.
>
> I guess there are two possible solutions:
>
> 1. Do not follow the POSIX semantics. If the target entry exists,
> rename fails with RNS Entry Exists Failure.
>
> 2. Fix the order of operations in bulk operations. If we assume
> operations are processed in the sequence order provided by the
> input argument.
>
> How do you think about it?
>
> Thanks,
> Osamu
>
More information about the gfs-wg
mailing list