On Wed Nov 1 17:07:11 2006, Michal 'vorner' Vaner wrote:
On Wed, Nov 01, 2006 at 06:07:39PM +0100, Tobias Markmann wrote:
> Isn't that a TCP problem since that can happen to any protocol
which is
> based to TCP?
Well, it is partly implementation problem, many OSes (as I heard)
are
able to tell you how much was already delivered and if you remember
what
part of data was what stanza, you can resend it after reconnection.
But that is bit more work, of course, and alot more data.
No OS can tell you what's been delivered, but some might be able to
tell you what hasn't been sent, and what hasn't been acknowledged. I
looked for how to do this on Linux, which usually provides the
richest API to the network layer, but I couldn't find anything to
tell me either.
But this isn't quite the same thing anyway - you want to know what
stanzas have been accepted - what happens if the ACKs get lost, or
the server dies?
Consider ESMTP, which has got data level acknowledgement. There's a
long-known problem whereby after DATA (and these days, BDAT and
BURL), there's a chance that you'll lose the connection before you
get the 2xx acknowledgement from the server. This is on the increase
again, partly due to the preference for protocol-level rejections
instead of DSNs, partly due to the marked increase in usage of ESMTP
over things like GPRS.
It's important to note that this specifically is about hop-by-hop,
and not end-to-end, which are different problems entirely. Finding
out if the guy you're talking to is still connected is quite easy,
just send an IQ (in principle *any* IQ), and you'll see.
Hop-by-hop tests are quite easy, too, but there's a gotcha - when
they fail, you want to know which stanzas you need to resend. And
XMPP does not provide any mechanism for that, and nor do pings.
My last suggestion - adding a sequence attribute to stanzas - didn't
seem to impress most people, partly because it requires servers to
rewrite stanzas between hops.
If instead the sender appends a distinct stanza (which could be an
iq, or could be something else) to every TCP segment sent, which
itself contains a sequence, then that can be used as the restart
token with almost precisely the same effect, and requires no
rewriting of stanzas.
So, the sender appends, for instance, <iq type='set'
id='ping123'><ping xmlns='urn:xmpp:ping' sent='1' recv='47'/></iq> to
each send() call's payload, and the receiver can then note this
simply, and respond with an iq reply when it suits it, which also
contains sent and recv sequence counts.
Loosely, you'd add that to the end of each TCP packet, in practise
about every 1.5k or at the end of each send() should be quite safe.
Dave.
--
Dave Cridland - mailto:[EMAIL PROTECTED] - xmpp:[EMAIL PROTECTED]
- acap://acap.dave.cridland.net/byowner/user/dwd/bookmarks/
- http://dave.cridland.net/
Infotrope Polymer - ACAP, IMAP, ESMTP, and Lemonade
