Re: FIN_WAIT_2 and a socket connect that takes an age to close.

Tech-Archive recommends: Fix windows errors by optimizing your registry

From: Andy Coates (nothanks_at_nowhere.com)
Date: 06/16/04 

Date: Wed, 16 Jun 2004 14:54:29 +0100

Cheers guys - the problem is that the client side is synchronous and doesn't
attempt to close the socket unless a write fails. Going to add some
asynchronous handling on the client side.

Thanks again,

Andy.

"Alan J. McFarlane" <alanjmcf@yahoo.com.INVALID> wrote in message
news:40ce1506_3@mk-nntp-1.news.uk.worldonline.com...
> "Andy Coates" <nothanks@nowhere.com> wrote in message
> news:eB2oJ37TEHA.3420@TK2MSFTNGP12.phx.gbl...
> [...]
> > This normally works fine, but I have one client application (for which I
> > can't easily get source) that causes problems on shutdown. This
> > particular
> > socket takes ~minutes before the queued read operations complete. When
I
> > run netstat I see that the socket is in the FIN_WAIT_2 state, which I
> > understand to mean that it has received the ACK from the client.
> >
> But not a FIN from it...
>
> In FIN_WAIT_2, a connection is "half-closed". Remember that TCP is
> full-duplex and allows the connection to be closed in a single direction
> whilst the other direction continues to pass data. So in this case _your_
> application has done SHUTDOWN(WRITE) (at least) and the peer has
> ack'd the resultant FIN.
>
> But the _peer_ has not done similar: the application there has not done
> SHUTDOWN(WRITE) or anything similar in effect. The socket on the peer
> should be in CLOSE_WAIT--that is, "waiting for a connection termination
> request from the local user", as RFC 793 puts it.
>
> The pertinent bit of the sequence diagram is (for the server side):
> ESTABLISHED
> |
> | CLOSE() / send FIN
> \/
> FIN_WAIT_1
> |
> | rcv ACK of FIN / (nothing)
> \/
> FIN_WAIT_2
> |
> | rcv FIN / send ACK
> \/
> TIME_WAIT
> |
> | timeout=2MSL / delete TCB (TCP Control Block)
> \/
> TIME_WAIT
>
>
> > Why is this socket taking so long to shutdown?
> >
> Well, as noted the connection is open and the connection /could feasibly/
> remain in that state for a long time, being used for half-duplex comms
until
> the peer application is finished and calls SHUTDOWN(). However many
> platforms have a timer that kills-off half-closed connections, and
> particularly so when the connection is idle (no data is flowing on the
open
> half). I'm not sure what Windows does in this regard.
>
> Other possibilities are that the peer application has a two minute
timeout.
> I think you'd have to watch the packets to see what causes the local TCP
to
> move from FIN_WAIT_2 to TIME_WAIT---if there's no incoming packet then the
> local stack has just timed it out.
>
> Note, don't be confuse that putative timeout with the timeout in
TIME_WAIT,
> it is quite separate. People often confuse timeouts in half-closed, with
> the normal TIME_WAIT timeout. It is 4 minutes in Windows (2000) btw.
> --
> Alan J. McFarlane
> http://homepage.ntlworld.com/alanjmcf/
> Please follow-up in the newsgroup for the benefit of all.
>
>

Relevant Pages

  • Re: How to know that the server side has close the connection when using java socket?
    ... OP asked how to detect if socket is closed. ... done with timeout. ... Server may choose to not disconnect idle clients. ... server will know if client is idle but ...
    (comp.lang.java.programmer)
  • Re: My first socket question
    ... The benefit of using a timeout to can_read to control program execution ... After the client has been serviced, the program can decide to perform ... socket) into a select, ... if the server has some low priority clean-up work it could be doing ...
    (comp.lang.perl.misc)
  • Re: How to know that the server side has close the connection when using java socket?
    ... Timeout may occure if nothing is written for specified time. ... For example chat client may by connected to the server and not send and messages. ... But if pings are implemented, server will know if client is idle but connected, or if socket was closed. ...
    (comp.lang.java.programmer)
  • Re: this is the error seen on the web service side
    ... Connection Timeout settings? ... > by peer: socket write error ... >> exception on the buffer. ...
    (microsoft.public.dotnet.framework.webservices.enhancements)
  • Re: epoll and timeouts
    ... To serve a client the server has to access some information over the ... recompute the timeout, and continue. ... First, if I add a socket in the middle of epoll_wait, its timeout must ...
    (comp.unix.programmer)