[Buildbot-devel] Non-fatal disconnects?

[Benoit Sigoure]

Quoting Minesh Patel <mpatel at 2wire.com>:

> Could it be possible to use a utility like screen 
> http://www.howtoforge.com/linux_screen to be able to re-attach to a 
> lost process. This would be helpful for those people that have 
> firewall/drop connections issues. Any thoughts...
>
> Thanks,
> Minesh
>

Hello
(please could you remove unecessary quotes and avoid top posting? thanks)

The URL you posted explains how to use screen so that applications continue to
run despite the connection dropping once in a while. This is has nothing to do
with buildbot because what they are describing there is that, when you work
remotely using SSH, if your connection drops for long enough, and the ssh
channel gets closed because of this, then the remote SSH daemon will terminate
the session which will kill all processes in that session. Using screen helps
in this regarde because the processes you run in a screen don't belong to your
session but to the screen's own session (which won't get killed by sshd).

With BuildBot it's not working this way though, because BuildBot slaves 
are not
part of your session (when you lunch them, they continue to run in background,
like daemons do, which is also what screen does in some respect), so that when
you log out, they can continue to run.

So putting the BuildBot slaves in a screen would not help, but if you have a
fragile connection, and if BuildBot is quite easily disturbed by connections
going up and down very shortly once in a while, then encapsulating the
connection from a slave to its master can help, because SSH is quite good at
keeping connections open. When an SSH connection gets closed, there are very
good chances that the connection was closed for a real reason, not just 
because
of a fragile network.

Cheers,

-- 
SIGOURE Benoit aka Tsuna
   _____
  /EPITA\ Promo 2008, LRDE