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

Dmitry Mikhin dmitry.mikhin at gmail.com
Fri Sep 7 13:24:57 UTC 2012


---------- Forwarded message ----------
From: Dmitry Mikhin <dmitry.mikhin at gmail.com>
Date: Fri, Sep 7, 2012 at 8:06 PM
Subject: Re: [Buildbot-devel] Intermittent Connection.Lost error with
Windows slaves
To: SZENTE Balint <balint at szentedwg.ro>


OK, a bit more testing here: I started the slave manually and switched
buildbot to use it as a plain BuildSlave, no latent slaves. Connection
was again over PuTTY and MyEnTunnel. Initially behaved better, I was
able to complete one 3-hour plus build end-to-end, then, on the
manually triggered re-build, again fell with ConnectionLost error. On
Windows side MyEnTunnel reported "plink.exe: FATAL ERROR: server
unexpectedly closed network connection". (The server had almost no
activity apart from this build at this moment.)

So, yes, behavior is reproducible without latent slaves, although the
"always up" slave was a bit more stable. Still, I cannot see any
pattern in the moment of failure, it sometime happens early, sometimes
not. Thus, "more stable" can be a coincidence. Also, I never
experienced this "ConnectionLost" problem with other virtual slaves
(although they use KVM-QEMU, not VirtualBox and comparison is not
fair).

Now I'm running the same manually started slave with direct
master-slave connection, no tunnel. Will see how it goes.

Still at loss how to approach the debugging, it's a pipeline with too
many components.

Dmitry




More information about the devel mailing list