On Thu, May 03, 2001 at 02:14:57AM +0200, Oskar Sandberg wrote:
> On Wed, May 02, 2001 at 12:45:12PM -0400, Tavin Cole wrote:
> > requests means you end up having to switch back and forth b/t REQUESTING
> > and TRANSFERRING -- yet in FCPClient and InternalClient you can only go
> > through the Transfer* states once, which makes it very difficult to
> > logically get to TRANSFERRING a second time if the request is restarted.
> > So what we have here is an FNPClient artifact.
>
> How so? A request can go between Pending and Transfering an infinite
> number of times in the internal node states, why not in Internal client?
Sorry, I was thinking of inserts, in which case the data is only transferred
into the node's store once from the provided Buckets. Hence the sequence
of Transfer* events can only occur once except for FNPClient.
Since the node only communicates the restarting of queries back to the
initiator, how would we even keep InternalClient or FCPClient posted about
the node's activity transferring the insert out of the store to other nodes?
Anyway, I think my point still makes sense. We don't need the TRANSFERRING
state in Request because the Transfer* events and the RestartedEvent carry
all the necessary information. And without making some probably useless
and inelegant extensions to the FeedbackToken system in the node states,
there's no way for FCPClient or InternalClient to logically deal with
the TRANSFERRING state for inserts.
--
# tavin cole
#
# "The process of scientific discovery is, in effect,
# a continual flight from wonder."
# - Albert Einstein
_______________________________________________
Devl mailing list
[EMAIL PROTECTED]
http://lists.freenetproject.org/mailman/listinfo/devl
