[ogsa-wg] ... without wsa:Address profile, AbstractName is meaningless ...
Steve Loughran
steve_loughran at hpl.hp.com
Fri Nov 11 04:20:27 CST 2005
Frank Siebenlist wrote:
> Without the proper uniqueness guarantees for the wsa:Address, the use of
> ws-naming's AbstractName is meaningless.
>
> The EPR with an embedded AbstractName essentially describes a binding
> between the Address and the AbstractName.
> One could argue that this binding is a bit of a flaky assertion without
> any way to specify time-validity or issuer.
> As a result, one can not see from an EPR alone when it was issued or
> whether it is still valid or not.
>
> An EPR-minter can decide to create a new EPR for a resource and reuse an
> Address that was used for an other resource before.
> This would yield the undesirable situation that you would have two EPRs
> with identical Addresses and two different AbstractNames associated with
> different resources, and the client would have no way to see which one
> of the EPRs is valid...
>
> For example, if I have two EPRs with the same Address, where one EPR
> includes an AbstractName that identifies MyBankAccount and the other
> includes an AbstractName identifying YourBankAccount, then one of us
> will not be happy with that situation...
>
> In order to avoid this ambiguity, we need the guarantee from the
> EPR-minter that Address values will not be reused for different
> resources: for all times, the Address should either refer to that one
> and only resource, or it should be invalid (and it is allowed to change
> between those two states).
> Furthermore, the EPR-minters should ensure that globally unique
> Addresses are used for a resource such that different EPR-minters do not
> (accidentally) use the same Address for different resources.
> The described uniqueness properties of the Address constitutes a
> required EPR-minter profile for the use of AbstractNames.
>
> ---
>
> This has been reiterated in a number of emails, but i wanted to call it
> out separately to facilitate the discussion whether the observation
> holds water and if so, whether such a EPR-minter Address-profile should
> become part of the ws-naming spec.
One thing to remember is that WSDM defines every resource as having a
ResourceId, and says that that ResID must be unique.
So in the CDDLM deploy API we have
-a resource ID that is used to uniquely identify a created System.
-an operation on a portal that maps from a resource ID to an address
The latter effectively allows multiple portals to manage the same
cluster by giving you a more robust means to access things inside it.
However, once you have an entity that can be accessed by >1 Address, you
have to consider how WS Resource Lifetime interacts. Does the resource
at address A have the same lifetime as address B: if I go A:<Destroy>
does that destroy B too? And if the lease to B expires, does A go away
too? Or do you have to model the thing at the end of every address not
as the resource with which you are interacting with, but as a transient
view of the resource, a view with its own lifetime.
-steve
> PS. Note that everywhere I wrote "Address" above, it should probably be
> substituted by "Address+ReferenceParameters", but this simplification
> doesn't change the observation.
that depends on how you implement WSRF. Mine only uses the address, and
so should in theory interop better with other WS-A implementations. But
I do use generated GUIDs in that address, as it is the best way to
guarantee uniqueness.
More information about the ogsa-wg
mailing list