[Buildbot-devel] Intermittent Connection.Lost error with Windows slaves

[SZENTE Balint]

Hello!


I can confirm this "Connection to the other side  
was lost in a non-clean fashion." behavior.

Setup:
* Master and slave on the same Gentoo x86_64 box (Core i7-2600, 8GB RAM)
* Windows XP under VirtualBox (running always)
* Ubuntu x86 11.10 under VirtualBox (running always)
* CentOS x86_64 6.3 under VirtualBox (running always)
* Buildbot 0.8.5, Buildslave Buildbot-Slave 0.8.5

No SSH tunneling, no latent buildslave.

Some facts:
 * The "connection lost" issue happens *only* with Windows slave. On
   the Linux slave never happened (for more than a year).
 * The time in which the slave get lost varies between 3-30 minutes,
   and it happens quite randomly. There are builds that take 6-15
   minutes, and does not happen, and sometimes it happens in 3 minutes.
   I did not find any pattern.
 * When this lost error appears, the Windows slave usualy works alone
   (it remains behind usually from the Linux machines), so we can
   eliminate the fact that the other 2 slaves have high CPU usage.
   Mostly they are idle, when the Windows lost appear.

My first impression is that maybe Twisted has some issues with Windows.
It would be very nice to sort out this problem.

Regards,
Balint Szente

On Thu, 06 Sep 2012 13:02:15 -0600
Tom Prince <tom.prince at ualberta.net> wrote:

> Dmitry Mikhin <dmitry.mikhin at gmail.com> writes:
> > - Windows XP under VirtualBox (LibVirtSlave)
> 
> Can you reproduce this behavior without using a latent slave? 
> 
>   Tom