[mosh-devel] Mosh connections didn't come back after ca. 18000 sec over 2x NAT

[Axel Beckert]

Hi Keith,

On Sun, Dec 30, 2012 at 01:59:53PM -0500, Keith Winstein wrote:
> Something is awry here but I am a little confused, especially about
> why packets are being sent so often. I wonder if somehow two
> sessions got slotted in to the same spot, somehow, on the
> intermediate NATted link. But really I don't quite understand what
> is going on.

I just checked, there's definitely just one mosh-server process
running. There's no NAT on that box, just bridging.

> Is the display (visible on the client) updating the screen often or
> at all?

Once a second. There's a screen session running which has a status bar
with a clock with seconds in it. I often use that to check if the a
connection is alive.

And now that you mention it: That clock still works, i.e. it's updated
on the client. It updates once a second, but the blue bar on top is
still there and says "mosh: Last reply 45804 seconds ago. [To quit:
Ctrl-^ .]". But nothing I type is shown on the terminal while the
clock still continues to update the time every second.

I also still have more or less idle sessions to one host open, but
they are actually three sessions to the same host, so I thought it may
be harder to debug and chose the one host, where I only have one mosh
session open and not even a second, new one.

I could also kill two out of three of these mosh session on that host,
so that only one remains. One without ticking clock, i.e. an idle one,
if that would help.

> We may need to figure out a way for you to break in to the server
> with gdb and turn on verbose debugging. Is the server running the
> wheezy 1.2.3 package on x86, amd64, or another architecture?

It's a Squeeze amd64 server running mosh 1.2.2-1~bpo60+1. But I also
have a second session to a Debian Unstable box (also amd64) having
1.2.3-1 running on the server side. So I don't think that the
server-side version makes a difference here, at least not between
1.2.2 and 1.2.3.

Actually I still have at least 5 sessions open with that symptom (no
need to reset them, so I forgot in my initial mail, that there's more
than one session :-) to three different hosts. All amd64, 2x Squeeze
with the backports package and one Sid with the 1.2.3-1 package.

What I also noticed: All sessions I had open, showed these symptoms,
but for all sessions either one of the following was true, too: There
was more than one mosh session from the same client host open at the
same time. Or more than one mosh session went through one bridging
computer (the one from whose session the tcpdumps are) through which
mosh sessions from the same client to computers behind it's bridge
were open, too. Not sure if that's important, but since you mentioned
mixing up of sessions, I thought, it might be relevant.

		Kind regards, Axel
-- 
/~\  Plain Text Ribbon Campaign                   | Axel Beckert
\ /  Say No to HTML in E-Mail and News            | abe at deuxchevaux.org  (Mail)
 X   See http://www.asciiribbon.org/              | abe at noone.org (Mail+Jabber)
/ \  I love long mails: http://email.is-not-s.ms/ | http://noone.org/abe/ (Web)