[Python-Dev] Windows buildbots randomly die with twisted ConnectionLost errors?

["Martin v. Löwis"]

> I'm not sure if I should read into the fact that it's occurring after
> networking-oriented tests like test_httplib and test_ftplib.  Running
> rt.bat on the resulting build manually doesn't indicate any errors in
> these tests.  Have other Windows buildbot owners seen this?

Notice that it also occurs in other steps, e.g.

http://www.python.org/dev/buildbot/all/x86%20W2k8%20trunk/builds/50/step-compile/0

Please understand that the basic principle of buildbot is that the 
slaves connect to the master, and that the master relies on the
slaves keeping the connection up.

Could it be that you are behind some firewall infrastructure which 
suddenly decides that certain TCP connections are idle/expired/closed?
In that case, the firewall might reject further messages from the
master, causing the master to believe that the connection was lost.

In that case, the slave should send repeated keepalive messages
(or the firewall be reconfigured to not close connections to
dinsdale:buildbot). It seem buildbot already has support for
keepalives, although the frequency might be too low.

In any case, you should check the twistd log files around the time
of the connection loss whether it shows any problems from the
client side as well.

Regards,
Martin