|
Kurt,
>>> "Kurt D. Zeilenga" <Kurt@OpenLDAP.org> 08/16/00 11:00PM
>>>
[Kurt] Is the (modify/delete) operation still performed as an atomic operation? - The modify/delete operation will be atomic only on a single entry,
and the operation as a whole will have the same semantics as a search
and multiple modifies/deletes from the client. Imposing a total atomicity in my
opinion would be difficult to implement, inefficient, and also unnecessary in
most cases. Directory transaction is probably a better mechanism
to address atomicity of multi-entry operations.
[Kurt] I find the use of controls to return continuations awkward at best. Such controls which extend operations which the "scope" of existing operations should use extended partial responses to return "continuations". In fact, I don't see why we need separate controls/responses for each of these operation. I, personally, would rather see: A) extendedPartialResponses generalized such that they may
returned in response to any operation (after appropriate client solicitation). - Currently a modify or delete operation doesn't return any partial
responses, and hence the desicion to let the response for modify/delete remain
as it is. As the request control is the one that imposes search semantics, the
control response was designed to carry its specific responses, rather than
change the base operation's response. [Kurt] B) scopingControl be defined to allow specification of "scope" information (allowing use of "one" and "subtree" scopes). Per entry/continuation responses may be returned. C) filterControl be defined to allow specification of a filter which must match before performing operation. Per entry/continuation responses may be returned. - This looks like a good idea to me. But will there be any
reallife scenario's where we want to do a filtering, without the
scope specified?
[Kurt] I would note that such extensions must be very explicit
in regards to "atomic" properties of the "operation". I would recommend that only entry level atomic properties be maintained, not operation level atomic properties. - I agree. It would be better to limit atomicity requirements to operation
on a single entry, rather than the whole operation. This is the assumption made
in the document.
- Thanks and Regards,
- Haripriya
|