Re: [Evolution] IMAP scanning gets stuck

["Ujwal S. Sathyam" <ujwal netbrowser com>]

On Fri, 2001-11-23 at 17:04, Jeffrey Stedfast wrote:
> On Fri, 2001-11-23 at 17:31, Dan Winship wrote:
> > On Fri, 2001-11-23 at 19:03, Jeffrey Stedfast wrote:
> > > I wish there was a way to manually set the tcp timeout, but there isn't.
> >
> > We could do a non-blocking connect followed by a select with a timeout.
> >
> > However, the tcp timeout isn't 30 minutes anyway, so something else is
> > going on here.
>
> Yea, 30 minutes does seem to be a bit high... but I just assumed he was
> exaggerating.

Yes, I was a tad exaggerating. However, it is still pretty high, long
enough to be annoying.

> from the manpage:
>
>        ETIMEDOUT
>               Timeout while attempting connection. The server may
>               be  too  busy  to accept new connections. Note that
>               for IP sockets the timeout may be  very  long  when
>               syncookies are enabled on the server.
>
> so maybe the server has syncookies enabled? they don't really say what
> "very long" is :-)
>
> Anyways, we already set the socket to O_NONBLOCK before we connect() and
> do a select() on the socket after afaik, so this doesn't seem to work on
> linux systems.

The timeout on a blocking connect() in Linux can vary depending on the
destination IP address. I have seen timouts of 14 minutes for
destination outside the LAN. In my code, I usually set the socket to
O_NONBLOCK with fcntl() before doing the connect. I then use poll with a
10 second timeout (I have a pretty busy server), and that has always
worked well for me. 

> Also, I seem to recall reading in some man page or another that the
> linux connect() always blocks anyway? I could have sworn I read this a
> while back after noticing that winsock was able to do non-blocking
> connect()'s.

If you use O_NONBLOCK correctly, connect() can be made non-blocking. As
I said, that is what I do in my code.

> ah, I was right...we *do* use O_NONBLOCK on the socket while
> connect()ing... we also set the timeout to 4 minutes (which is the
> standard tcp timeout for blocking connects) while also allowing the user
> to cancel the connection with the stop button.
>
> I guess we could set the timeout to something smaller, but I don't see
> what that really gains us here (since the user can force a disconnect at
> any time).

How do I do that? I can't even exit Evolution since the shell is waiting
for the mailer to quit, and the mailer is stuck. So I use killev. Pretty
ugly...Also, depending on the user to force a manual disconnect is not
clean IMHO.


> Jeff
>
> > -- Dan


Ujwal