[undertow-dev] SSE close task

[Nel Taurisson]

Ok, thank you
Nel

2015-11-24 5:46 GMT+01:00 Stuart Douglas <sdouglas at redhat.com>:

>
>
> ----- Original Message -----
> > From: "Nel Taurisson" <nel.taurisson at gmail.com>
> > To: "Stuart Douglas" <sdouglas at redhat.com>
> > Cc: undertow-dev at lists.jboss.org
> > Sent: Tuesday, 24 November, 2015 3:22:02 PM
> > Subject: Re: [undertow-dev] SSE close task
> >
> > Ok, I understand : the only way to figure out that a client connection is
> > closed is by sending a message as sending will eventually fail.
> >
> > Reading what your saying, even if the client is disconnected, the first
> > attempt to send a message will not fail (from the server point of view),
> is
> > this correct ?
>
> Generally no, depending on how the client closed the connection. If the
> client sends an RST immediately on close then the first write will fail.
> Otherwise the RST will not be sent till the client receives a message from
> the server, so from the servers point of view it is the second write that
> fails.
>
> >
> > Does that mean that the server cannot fully determine that the client
> > received a message ?
>
> No, but you are never going to get the sort of guarantee anyway, as the
> client could crash in between an sending an ACK and actually processing the
> data.
>
> Stuart
>
> >
> > Thanks
> > Nel
> >
> >
> >
> > 2015-11-24 1:16 GMT+01:00 Stuart Douglas <sdouglas at redhat.com>:
> >
> > > I have commented on the JIRA, but will include it here for anyone who
> is
> > > following:
> > >
> > > The close listener is called, however due to the nature of SSE and TCP
> > > this behavior is not super intuitive.
> > >
> > > Basically when the close() is called on the client side the client
> sends a
> > > FIN to the server, however as the server is not expecting any more data
> > > from the client this has no effect (in fact the client may have already
> > > sent a FIN if connection:close was sent).
> > >
> > > From the servers point of view the socket is still functioning
> normally,
> > > up until the point where it attempts to send some data. When this
> happens
> > > the client will response with an RST. Unfortunately the server still
> has no
> > > idea that the socket is broken, as the RST arrives after the data has
> been
> > > sent (unless the data is very large). When the server attempts to send
> for
> > > the second time after the client has closed the connection it will
> receive
> > > an IOException from the RST, the connection will be closed and the
> listener
> > > invoked.
> > >
> > > If you want to detect this in a timely manner your only real option is
> > > some kind of heartbeat message, due to the way TCP and SSE work there
> is
> > > not really any other options.
> > >
> > >
> > > Stuart
> > >
> > > ----- Original Message -----
> > > > From: "Nel Taurisson" <nel.taurisson at gmail.com>
> > > > To: "Stuart Douglas" <sdouglas at redhat.com>
> > > > Cc: undertow-dev at lists.jboss.org
> > > > Sent: Tuesday, 24 November, 2015 8:31:35 AM
> > > > Subject: Re: [undertow-dev] SSE close task
> > > >
> > > > Ok, thanks, here's the JIRA with a failing test attached as a patch:
> > > >
> > > > https://issues.jboss.org/browse/UNDERTOW-589
> > > >
> > > > Nel
> > > >
> > > > 2015-11-23 7:46 GMT+01:00 Stuart Douglas <sdouglas at redhat.com>:
> > > >
> > > > > This sounds like a bug, can you file a JIRA?
> > > > >
> > > > > https://issues.jboss.org/projects/UNDERTOW
> > > > >
> > > > > Stuart
> > > > >
> > > > > ----- Original Message -----
> > > > > > From: "Nel Taurisson" <nel.taurisson at gmail.com>
> > > > > > To: undertow-dev at lists.jboss.org
> > > > > > Sent: Monday, 23 November, 2015 5:00:10 PM
> > > > > > Subject: [undertow-dev] SSE close task
> > > > > >
> > > > > > Hi,
> > > > > >
> > > > > > I'm playing with the SSE handler and I can't figure how the close
> > > task is
> > > > > > working. I thought the close task would be invoked when the
> client
> > > closes
> > > > > > the connection but it doesn't seem to be the case.
> > > > > >
> > > > > > I slightly modified the example code to register a close task
> when a
> > > > > > connection is initiated, here's the code :
> > > > > >
> > > > > > ServerSentEventConnectionCallback callback = new
> > > > > > ServerSentEventConnectionCallback() {
> > > > > > @Override
> > > > > > public void connected(ServerSentEventConnection connection,
> String
> > > > > > lastEventId) {
> > > > > > System.out.println("adding close task on connection...");
> > > > > > connection.addCloseTask(channel -> System.out.println("goodbye
> cruel
> > > > > > world..."));
> > > > > > }
> > > > > > };
> > > > > > final ServerSentEventHandler sseHandler =
> > > > > > Handlers.serverSentEvents(callback);
> > > > > >
> > > > > > It appears to me that the close task is only invoked when the
> server
> > > > > stops,
> > > > > > but now when the client closes. Is it the expected behavior ? If
> so,
> > > is
> > > > > > there any other way to be notified of a client disconnection ?
> > > > > >
> > > > > > Thanks a lot.
> > > > > >
> > > > > > Nel
> > > > > >
> > > > > >
> > > > > > _______________________________________________
> > > > > > undertow-dev mailing list
> > > > > > undertow-dev at lists.jboss.org
> > > > > > https://lists.jboss.org/mailman/listinfo/undertow-dev
> > > > >
> > > >
> > >
> >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.jboss.org/pipermail/undertow-dev/attachments/20151124/62a466d0/attachment-0001.html