"McDonald, Ira" wrote:
> Hi Michael,
> But when you ask for events by (only) "notify-recipient-uri",
> you are certain to get duplicates (because your polling interval
> MUST be shorter than the event persistent, in order for polling
> to work at all).
> The explicit list of pairs of subscription-id and sequence-number
> allows duplicate avoidance. Now you see why the optimization?
Yes, however you'll still get duplicates with the current proposal
for any new subscriptions that you haven't added to the subscription
ID set, and quite frankly that just adds yet another layer of
complexity to the server's processing of the request:
Copy subscription ID set to local list
Copy sequence number set to local list
for each subscription for recipient URI
if subscription ID in local list then
add subscription ID to local list
add sequence number 0 to local list
I'm not sure if this level of complexity is needed or not?
In the case of the wait-mode Get-Notifications, providing the
recipient URI will give you all events from all matching
subscriptions, without duplicates (unless you have to re-poll)
If your client can sort out N subscriptions simultaneously,
then it can probably track the duplicate events itself and
not re-process them. Not the most efficient bandwidth-wise,
but it does reduce the load on the server.
This brings up another issue (maybe it has been addressed, I
don't remember off hand) - if you do a wait-mode Get-Notifications
with a notify-recipient-uri, and initially there are subscriptions
for this URI, do we also return a successful-ok-no-more-events
once all active subscriptions expire (including their events),
or does the Get-Notifications continue indefinitely (or until
the server decides to close the connection)???
-- ______________________________________________________________________ Michael Sweet, Easy Software Products firstname.lastname@example.org Printing Software for UNIX http://www.easysw.com
This archive was generated by hypermail 2b29 : Tue Jul 31 2001 - 11:49:19 EDT