ASIO: Detecting broken connections
Hi, I was wondering how I can find out with ASIO whether a connection is broken. I have a client connected to the server that continuously streams data. I have to know at the client-side when the server has hung up. I tried to use an async_write and after 5 erroneous tries the running thread for the connection is terminated. Like this: void connection::run_thread() { while (!stopped()) { event_ptr e = fetch_event(); if (!stopped()) { int write_tries = 0; boost::asio::async_write(_socket, boost::asio::buffer(*e, e->length()), boost::bind(&connection::handle_write, this, boost::asio::placeholders::error, e, write_tries)); } } } void connection::handle_write(const boost::system::error_code &err, const event_ptr event, int write_tries) { if (write_tries == max_write_tries) stopped(true); // thread will break out of its main-loop after being notified if (stopped()) return; // no meaning to continue if (err) { // incremenent write_tries and retry async_write boost::asio::async_write(_socket, boost::asio::buffer(*event, event->length()), boost::bind(&connection::handle_write, this, boost::asio::placeholders::error, event, ++write_tries)); } } I think I am misunderstanding something. I thought that after a async_write the registered handler will be invoked. But the handler is NEVER invoked somehow, although data is send from the client to the server. I have the following questions: - Why is the handler never invoked? When is it supposed to be invoked? - What is the best way to find out a connection has been reset by the server? And client? Thanks alot, Andrej ___________________________________________________________ Yahoo! Mail is the world's favourite email. Don't settle for less, sign up for your free account today http://uk.rd.yahoo.com/evt=44106/*http://uk.docs.yahoo.com/mail/winter07.htm...
Andrej van der Zee <mavdzee@yahoo.co.uk> wrote:
I was wondering how I can find out with ASIO whether a connection is broken. I have a client connected to the server that continuously streams data.
The canonical way to consistently detect broken TCP connections is to always have a read on the socket - it will return 0 on broken connection (in usual BSD sockets type reads - I would expect an appropriate error parameter set in ASIO). Since you're doing async operations this is pretty easy. Many times a write will only buffer the data (at the OS / TCP driver level), and an error is not detected until a later write (and if the client socket is in a "half shutdown" mode, it might be quite a while before the TCP keepalive timer pops and the socket is fully destroyed). This is absent special ASIO capabilities - if ASIO is doing a read "on your behalf", then the problem is the "not completely destroyed connection", which is an OS / TCP "feature" and nothing that ASIO can help with. I haven't found specific details relating to this in ASIO, but that may only be because I haven't read enough of the ASIO documentation. Chris K (or others), clarifications, corrections, or comments? Cliff
On 6/29/07, Cliff Green <cliffg@codewrangler.net> wrote:
Andrej van der Zee <mavdzee@yahoo.co.uk> wrote:
I was wondering how I can find out with ASIO whether a connection is broken. I have a client connected to the server that continuously streams data.
The canonical way to consistently detect broken TCP connections is to always have a read on the socket - it will return 0 on broken connection (in usual BSD sockets type reads - I would expect an appropriate error parameter set in ASIO). Since you're doing async operations this is pretty easy.
Many times a write will only buffer the data (at the OS / TCP driver level), and an error is not detected until a later write (and if the client socket is in a "half shutdown" mode, it might be quite a while before the TCP keepalive timer pops and the socket is fully destroyed).
This is absent special ASIO capabilities - if ASIO is doing a read "on your behalf", then the problem is the "not completely destroyed connection", which is an OS / TCP "feature" and nothing that ASIO can help with. I haven't found specific details relating to this in ASIO, but that may only be because I haven't read enough of the ASIO documentation.
Chris K (or others), clarifications, corrections, or comments?
You test for a disconnection with Asio the same way you would with BSD sockets - you need to leave a long-running read operating going. Check out this post from the Asio-users mailing list: http://osdir.com/ml/lib.boost.asio.user/2006-11/msg00035.html Richard Cliff
_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
What do you mean with a long-running read operation?
In my setup, the client only writes data to the
server. Should I start a thread with a synchronous
asio::read on the socket? Or can I do this with an
asio::asyn_read too? Is this possible while writing
over the same socket at the same time?
Thanks,
Andrej
--- Richard Dingwall <rdingwall@gmail.com> wrote:
> On 6/29/07, Cliff Green <cliffg@codewrangler.net>
> wrote:
> >
> > Andrej van der Zee <mavdzee@yahoo.co.uk> wrote:
> > > I was wondering how I can find out with ASIO
> whether a
> > > connection is broken. I have a client connected
> to the
> > > server that continuously streams data.
> >
> > The canonical way to consistently detect broken
> TCP
> > connections is to always have a read on the socket
> - it
> > will return 0 on broken connection (in usual BSD
> sockets
> > type reads - I would expect an appropriate error
> parameter
> > set in ASIO). Since you're doing async operations
> this is
> > pretty easy.
> >
> > Many times a write will only buffer the data (at
> the OS /
> > TCP driver level), and an error is not detected
> until a
> > later write (and if the client socket is in a
> "half
> > shutdown" mode, it might be quite a while before
> the TCP
> > keepalive timer pops and the socket is fully
> destroyed).
> >
> > This is absent special ASIO capabilities - if ASIO
> is
> > doing a read "on your behalf", then the problem is
> the
> > "not completely destroyed connection", which is an
> OS /
> > TCP "feature" and nothing that ASIO can help with.
> I
> > haven't found specific details relating to this in
> ASIO,
> > but that may only be because I haven't read enough
> of the
> > ASIO documentation.
> >
> > Chris K (or others), clarifications, corrections,
> or
> > comments?
>
>
> You test for a disconnection with Asio the same way
> you would with BSD
> sockets - you need to leave a long-running read
> operating going.
>
> Check out this post from the Asio-users mailing
> list:
>
http://osdir.com/ml/lib.boost.asio.user/2006-11/msg00035.html
>
> Richard
>
> Cliff
> >
> > _______________________________________________
> > Boost-users mailing list
> > Boost-users@lists.boost.org
> >
>
http://lists.boost.org/mailman/listinfo.cgi/boost-users
> >
> > _______________________________________________
> Boost-users mailing list
> Boost-users@lists.boost.org
>
http://lists.boost.org/mailman/listinfo.cgi/boost-users
___________________________________________________________
Yahoo! Answers - Got a question? Someone out there knows the answer. Try it
now.
http://uk.answers.yahoo.com/
On 6/29/07, Andrej van der Zee <mavdzee@yahoo.co.uk> wrote: > > What do you mean with a long-running read operation? A read operation that waits indefinitely for any data the server might send. In my setup, the client only writes data to the > server. Should I start a thread with a synchronous > asio::read on the socket? You could, but... Or can I do this with an > asio::asyn_read too? async_read() is better because you don't have to spawn another thread :) Is this possible while writing > over the same socket at the same time? > Yes it is. Thanks, > Andrej I recommend you join the asio-users mailing list here: https://lists.sourceforge.net/lists/listinfo/asio-users I'm not sure if Chris checks boost-users that often, and he is much better at answering questions! Richard --- Richard Dingwall <rdingwall@gmail.com> wrote: > > > On 6/29/07, Cliff Green <cliffg@codewrangler.net> > > wrote: > > > > > > Andrej van der Zee <mavdzee@yahoo.co.uk> wrote: > > > > I was wondering how I can find out with ASIO > > whether a > > > > connection is broken. I have a client connected > > to the > > > > server that continuously streams data. > > > > > > The canonical way to consistently detect broken > > TCP > > > connections is to always have a read on the socket > > - it > > > will return 0 on broken connection (in usual BSD > > sockets > > > type reads - I would expect an appropriate error > > parameter > > > set in ASIO). Since you're doing async operations > > this is > > > pretty easy. > > > > > > Many times a write will only buffer the data (at > > the OS / > > > TCP driver level), and an error is not detected > > until a > > > later write (and if the client socket is in a > > "half > > > shutdown" mode, it might be quite a while before > > the TCP > > > keepalive timer pops and the socket is fully > > destroyed). > > > > > > This is absent special ASIO capabilities - if ASIO > > is > > > doing a read "on your behalf", then the problem is > > the > > > "not completely destroyed connection", which is an > > OS / > > > TCP "feature" and nothing that ASIO can help with. > > I > > > haven't found specific details relating to this in > > ASIO, > > > but that may only be because I haven't read enough > > of the > > > ASIO documentation. > > > > > > Chris K (or others), clarifications, corrections, > > or > > > comments? > > > > > > You test for a disconnection with Asio the same way > > you would with BSD > > sockets - you need to leave a long-running read > > operating going. > > > > Check out this post from the Asio-users mailing > > list: > > > http://osdir.com/ml/lib.boost.asio.user/2006-11/msg00035.html > > > > Richard > > > > Cliff > > > > > > _______________________________________________ > > > Boost-users mailing list > > > Boost-users@lists.boost.org > > > > > > http://lists.boost.org/mailman/listinfo.cgi/boost-users > > > > > > _______________________________________________ > > Boost-users mailing list > > Boost-users@lists.boost.org > > > http://lists.boost.org/mailman/listinfo.cgi/boost-users > > > > ___________________________________________________________ > Yahoo! Answers - Got a question? Someone out there knows the answer. Try > it > now. > http://uk.answers.yahoo.com/ > _______________________________________________ > Boost-users mailing list > Boost-users@lists.boost.org > http://lists.boost.org/mailman/listinfo.cgi/boost-users >
Thanks. I did some tests in the meanwhile and
asio::asyn_read() does detect broken connections very
well, without spanning an extra thread.
Cheers,
Andrej
--- Richard Dingwall <rdingwall@gmail.com> wrote:
> On 6/29/07, Andrej van der Zee <mavdzee@yahoo.co.uk>
> wrote:
> >
> > What do you mean with a long-running read
> operation?
>
>
> A read operation that waits indefinitely for any
> data the server might send.
>
> In my setup, the client only writes data to the
> > server. Should I start a thread with a synchronous
> > asio::read on the socket?
>
>
> You could, but...
>
> Or can I do this with an
> > asio::asyn_read too?
>
>
> async_read() is better because you don't have to
> spawn another thread :)
>
> Is this possible while writing
> > over the same socket at the same time?
> >
>
> Yes it is.
>
> Thanks,
> > Andrej
>
>
>
> I recommend you join the asio-users mailing list
> here:
>
https://lists.sourceforge.net/lists/listinfo/asio-users
>
> I'm not sure if Chris checks boost-users that often,
> and he is much better
> at answering questions!
>
> Richard
>
> --- Richard Dingwall <rdingwall@gmail.com> wrote:
> >
> > > On 6/29/07, Cliff Green
> <cliffg@codewrangler.net>
> > > wrote:
> > > >
> > > > Andrej van der Zee <mavdzee@yahoo.co.uk>
> wrote:
> > > > > I was wondering how I can find out with ASIO
> > > whether a
> > > > > connection is broken. I have a client
> connected
> > > to the
> > > > > server that continuously streams data.
> > > >
> > > > The canonical way to consistently detect
> broken
> > > TCP
> > > > connections is to always have a read on the
> socket
> > > - it
> > > > will return 0 on broken connection (in usual
> BSD
> > > sockets
> > > > type reads - I would expect an appropriate
> error
> > > parameter
> > > > set in ASIO). Since you're doing async
> operations
> > > this is
> > > > pretty easy.
> > > >
> > > > Many times a write will only buffer the data
> (at
> > > the OS /
> > > > TCP driver level), and an error is not
> detected
> > > until a
> > > > later write (and if the client socket is in a
> > > "half
> > > > shutdown" mode, it might be quite a while
> before
> > > the TCP
> > > > keepalive timer pops and the socket is fully
> > > destroyed).
> > > >
> > > > This is absent special ASIO capabilities - if
> ASIO
> > > is
> > > > doing a read "on your behalf", then the
> problem is
> > > the
> > > > "not completely destroyed connection", which
> is an
> > > OS /
> > > > TCP "feature" and nothing that ASIO can help
> with.
> > > I
> > > > haven't found specific details relating to
> this in
> > > ASIO,
> > > > but that may only be because I haven't read
> enough
> > > of the
> > > > ASIO documentation.
> > > >
> > > > Chris K (or others), clarifications,
> corrections,
> > > or
> > > > comments?
> > >
> > >
> > > You test for a disconnection with Asio the same
> way
> > > you would with BSD
> > > sockets - you need to leave a long-running read
> > > operating going.
> > >
> > > Check out this post from the Asio-users mailing
> > > list:
> > >
> >
>
http://osdir.com/ml/lib.boost.asio.user/2006-11/msg00035.html
> > >
> > > Richard
> > >
> > > Cliff
> > > >
> > > >
> _______________________________________________
> > > > Boost-users mailing list
> > > > Boost-users@lists.boost.org
> > > >
> > >
> >
>
http://lists.boost.org/mailman/listinfo.cgi/boost-users
> > > >
> > > >
> _______________________________________________
> > > Boost-users mailing list
> > > Boost-users@lists.boost.org
> > >
> >
>
http://lists.boost.org/mailman/listinfo.cgi/boost-users
> >
> >
> >
> >
>
___________________________________________________________
> > Yahoo! Answers - Got a question? Someone out there
> knows the answer. Try
> > it
> > now.
> > http://uk.answers.yahoo.com/
> > _______________________________________________
> > Boost-users mailing list
> > Boost-users@lists.boost.org
> >
>
http://lists.boost.org/mailman/listinfo.cgi/boost-users
> >
> > _______________________________________________
> Boost-users mailing list
> Boost-users@lists.boost.org
>
http://lists.boost.org/mailman/listinfo.cgi/boost-users
___________________________________________________________
Yahoo! Mail is the world's favourite email. Don't settle for less, sign up for
your free account today http://uk.rd.yahoo.com/evt=44106/*http://uk.docs.yahoo.com/mail/winter07.html
participants (3)
-
Andrej van der Zee -
Cliff Green -
Richard Dingwall