Re: [wsrf] ISSUE : setResourceProperties atomicity

[Steve Graham <sggraham@us.ibm.com>]

Are you suggesting 1) and 2) in your
email are alternatives or complementary?

I don't see the purpose of having an
attribute on the operation.  Maybe I don't understand your suggestion.
Would this be a decoration of the wsdl operation?  Why would the policy
assertion of <wsrf-rp-pol:SetResourcePropertiesAtomicity>true</wsrf-rp-pol:SetResourcePropertiesAtomicity>
alone be sufficient?

++++++++
Steve Graham
(919)254-0615 (T/L 444)
STSM, On Demand Architecture
Member, IBM Academy of Technology
<Soli Deo Gloria/>
++++++++

"Glen Daniels" <gdaniels@sonicsoftware.com>

10/28/2004 08:23 PM

To	<wsrf@lists.oasis-open.org>
cc
Subject	[wsrf] ISSUE : setResourceProperties atomicity

I'm sitting here at the WS-Notification meeting, and the following topic
has just come up.

The atomicity of setResourceProperties() in WSRP is currently undefined
(I can't get to the current spec right now (no net) to point at a
particular section, sorry).  An implementation is free to either do
them
all at once, guaranteeing atomicity, or to do them one at a time, so
that some of them might have worked and others did not at the end of the
invocation.  In other words, it's a QoS issue as to whether a given
implementation decides to do it one way or the other.

This seems like a fine thing, except that apparently there is not yet a
plan for either a) a metadata/policy assertion which indicates a given
resource will do setResourceProperties in an atomic (or not) way, or b)
a way to indicate in a given setResourceProperties() invocation that the
requester requires atomicity.  I would like to raise an issue that
both
of these things should be specified in WSRP.

In particular, I'm suggesting that the group:

1) Define a normative way to indicate in metadata that a resource will
do atomic setResourceProperties() invocations (this is basically a
boolean element with a well known QName)

2) Define a boolean "atomic" flag/attribue on the
setResourceProperties() operation, the semantics of which indicate that
a successful result means ALL sets were done, and a failed result means
NO sets were done.

IMHO, without this stuff, the usefulness of a single set operation for
multiple resource properties is severely restricted.

--Glen