tinc - Jan 2003 - Problem with 'resource temporarily unavailable'

We've got problems running tinc over DSL lines using low cost 
routers/modems (at leat I think it's a hardware problem.)

We get error messages such as the following:

Jan 15 11:36:05 stv100 tinc.stv100[2784]: Sending meta data to vpn 
(162.37.22.162 port 2007) failed: Resource temporarily unavailable

This doesn't seem to stop tinc -- it keeps going, but the interruption 
is enough that rsync thinks the line failed so it waits a minute to retry.  

We did use rsync v2.5.5 in one instance to throttle the bandwidth down 
to 20KB which worked, but in other locations the problem wouldn't go away.

I've added a delay to meta.c so if it errors out, retry in 1 second 
(using usleep(1000)).  I also made an it log every time it was going to 
retry and if it fails, log the failure.  the code looks like this:

 if(write(c->socket, bufp, length) < 0) {
                syslog(LOG_ERR, _("Sending meta data to %s (%s) failed 
will retry is 1000ms"), c->name,
                                 c->hostname);
                usleep(1000);
                if(write(c->socket, bufp, length) < 0) {
                        syslog(LOG_ERR, _("Sending meta data to %s (%s) 
failed: %s"), c->name,
                                   c->hostname, strerror(errno));
                        return -1;
                }
        }

we also changed the mtu on both sides to 1000 -- hoping to possibly 
remove problems there.

We now get lots of 'will retry in 1000ms' messages, and very few 
'failed' messages.  

There are more messages on the sending side of 'Metadata socket error 
for vpn' Connection reset by peer' and the other side showing 'Bogus
data received'

I'll be checking into this further to see if I can find a way to 
overcome the problem.

Thanks!

Ed
edk@cendatsys.com

Tinc:         Discussion list about the tinc VPN daemon
Archive:      http://mail.nl.linux.org/lists/
Tinc site:    http://tinc.nl.linux.org/

On Wed, Jan 15, 2003 at 01:10:24PM -0600, Edward King wrote:
> We've got problems running tinc over DSL lines using low cost 
> routers/modems (at leat I think it's a hardware problem.)
> 
> We get error messages such as the following:
> 
> Jan 15 11:36:05 stv100 tinc.stv100[2784]: Sending meta data to vpn 
> (162.37.22.162 port 2007) failed: Resource temporarily unavailable
That usually indicates that no more network buffers are available,
either because the machine is really low on memory or if more data is
being sent out by tinc than the network can handle.

tinc normally shouldn't send very much meta data (only a ping once a
minute and some updates if a connection to another tinc daemon is made
or broken) unless you're using the TCPonly option. Is it really
necessary for you to use it? What kind of traffic are you sending over
the VPN?
> I've added a delay to meta.c so if it errors out, retry in 1 second 
> (using usleep(1000)).  I also made an it log every time it was going to 
> retry and if it fails, log the failure.  the code looks like this:
usleep(1000) sleeps for 1 millisecond.
> we also changed the mtu on both sides to 1000 -- hoping to possibly 
> remove problems there.
That probably doesn't matter much.
> We now get lots of 'will retry in 1000ms' messages, and very few 
> 'failed' messages.  
That's interesting.
> There are more messages on the sending side of 'Metadata socket error 
> for vpn' Connection reset by peer' and the other side showing
'Bogus
> data received'
You really shouldn't see that unless errors were introduced in the TCP
stream... Do those errors also occur less frequently with the usleep()
than without?

-- 
Met vriendelijke groet / with kind regards,
    Guus Sliepen <guus@sliepen.eu.org>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url :
http://brouwer.uvt.nl/pipermail/tinc/attachments/20030117/c7185010/attachment.pgp