Re: vkernel crash?

[Mike Wolman <mike@xxxxxxxxx>]

Hi,

A quick follow up:

I was unable to see what the actual console was displaying as the
screen was in suspend and i could not awaken it.

I have tried to repeat the problem for the last few hours and so
far have not been able reproduce the issue.

Mike.

On Sun, 18 Mar 2007, Matthew Dillon wrote:

> :Hi,
> :
> :I was having an little play with vkernels - pretty much followed the man page -
> :except i was using vn1 and not vn0 for the file system.
> :
> :After doing some scping and other network related testing i closed my xterm
> :controlling with the vkernel console in it by mistake - i was also logged into
> :the physical machine via another ssh and pretty much as soon as i closed the
> :xterm all connections to the main dfly box were lost as well and i am now
> :unable to ping or see it on the network (the machine had been up for the last 4
> :or 5 days after and rebuild to 1.9-dev via cvsup on wednesday from a fresh 1.8
> :install).
> :
> :Unfortunately i am not in the same location as the machine at the moment
> :so cannot tell if it has produced a proper crash or has just thrown the machine
> :off the network.  However i guess this should be pretty easy for anyone to
> :reproduce - simply start a vkernel from a ssh login and terminate the ssh
> :connection.
>
>    I haven't been able to reproduce it yet.  I have a test box and tried
>    running a basic vkernel in an xterm ssh'd to the test box, then closing
>    the xterm.  I also tried terminating the ssh process, but that didn't
>    generate a crash either.  The vkernel running on the test box terminated
>    but the test box was otherwise ok.
>
>    Given your description, there is clearly a vkernel related bug here
>    and hopefully you or I or someone else will be able to reproduce it and
>    get a definitive crash dump or other information.
>
> :Should vkernels be able to run in a 'headless' mode and any messages be
> :directed to a file?
>
>    Yes, definitely.  It would be easy for the vkernel to detect whether
>    its 'console' is a tty or something else (like a regular file or logging
>    socket), and then operate the console as a logging-only output device.
>
>    Maybe 5 minutes of coding to do that... I want to encourage other
>    developer to sink their teeth into the vkernel but if nobody submits
>    or commits code to do that in the next few days, give me a reminder and
>    I'll do it.
>
> 					-Matt
> 					Matthew Dillon
> 					<dillon@backplane.com>