DSML
Teleconference Minutes
1) List of attendees
James Tauber, Chaired
Christine Thomlinson
Nigel Jacobs
Mark Wahl
Keither Attenborough
Jeff Bohren
Mark Brown
Andy Harjanto
Shon Vella
Vinnie Ryan
John McGarvey
Gavenraj Sodhi
Rob Weltman
Alexis Bor
2) Minutes
First Issues:
Batching is part of DSML core?
- Key requirements of DSML (core level) – Should be transport independent.
- Must have batching. Not every client will have it but should be defined at the bottom.
o Should include request ID then define mechanism that would do batching in other schemas.
- If we go down this path, it may be more complicated. Give one reason for discussion.
- Biggest difference may happen if ordering is different
o Define a batching mechanism for any of the likely transports.
- Principal should be tried to establish minimum set of exchanges that need to occur. Independent of how the exchanges may encapsulated or orchestrated. Maybe it should be associated at the transport level.
- What is in the core?
o Suggested as comprehensive so any more DSML is not added in the transport.
o Example: Microsoft proposal and iPlanet proposal
§ MS – cannot send DSML request without putting in DSML envelope
· Not clear why you need envelope
§ As shown as agnostic in iPlanet’s proposal
o What is exported from DSML XML Schema?
§ Request and responses
§ Groups of those
§ Exporting both
o Trying to define a general purpose encoding.
§ Want to eliminate the envelope?
· Some semantics may not be appropriate.
§ In principle we need to come to a commonality.
o Core problem is what’s being exchanged, XSD files.
§ Microsoft’s proposal is superset of Sun’s proposal.
§ At an impasse at this point.
§ None of the semantics were covered in new version.
§ We need to arrive at some consensus how to edit the document.
§ Do we want to have a core proposal for async/sync?
· Sun’s position: Batch schema and applied semantics with it:
o Separated in two schemas and develop operation grouping settings.
o What is core and what is optional core? – Comment
o Need to talk about SOAP binding?
§ How does a client discover what services are supported by talking to a server?
o SOAP with batching would have better performance and soap without batching would not have as good of performance? - Comment
o In IETF – transactional, distributed operations across multiple servers
o What is an example of batching where current batch envelope does not support? – Which you can augment from IETF batching.
§ Possibility that you can respond to search results or responses in the batch, you would have high latency. Please do all these operations is not adequate enough. With even addition of asynchronous, this would not resolve this.
- Voting Question: Would schema addition be mandatory or optional? Should every implementation include every operation as specified? Is packaging mechanism mandatory?
o For soap binding? - Need a specific packaging
§ Define header – for versioning and features
§ Define body – reference to DSML request/response…
· Microsoft reference is to an envelope in addition to or exclusively.
· Suppose we have request with a series of responses.
o Must include request for each issued or have a soap that doesn’t include all that.
· SOAP has been less open for asynchronous.
o SOAP is one way but you could use underlying transport.
§ Does not have request/response semantics, you use envelope
§ Unless you use something like SOAP with http.
o What happens with SOAP 1.2 with MIME?
· May end up with packaging for search responses.
§ Does everyone agree to having a binding for SOAP over http
§ LDIF with pointy brackets.
· Are they going to transport same payloads?
o Too early to tell because we haven’t through that exercise – Comment
o No concrete bindings as of yet. – Comment
· What is progress of SOAP binding document from iPlanet?
o Need to discuss semantics.
§ Draft will be out in the next few days.
o Pending SOAP binding document, we will come back to this issue.
- Highlighting thoughts from document iPlanet sent out yesterday.
o Item 12: Deleted object class instructions as discussed in teleconference.
§ Question of how to refer schema information within an attached schema should be revisited; we should revisit the requirements as to the reference mechanism. (OID and URI)
§ Proposing: Don’t need OC value definition at the schema level.
· Suggesting that this will be expressed in XML schema?
· Question is whether we will revisit this.
§ Choice of referencing mechanism or what’s on the other end, URI, is orthogonal.
§ Eliminating OID and use URI??
· No. OID is still permitted as in the common case of DSML v2 implementations.
§ What has changed?
· Removed object class definitions
o Trying to get error response to local errors
§ Moved non-LDAP error codes to error response
- Any other thoughts??
o No.