[Buildbot-devel] 0.8.3p1 queries

John Earl jearl at airsource.co.uk
Tue Mar 1 19:34:07 UTC 2011

On 28/02/2011 18:01, Dustin J. Mitchell wrote:
> On Mon, Feb 28, 2011 at 11:49 AM, John Earl<jearl at airsource.co.uk>  wrote:
>> An update on this. I disabled the fork(), by just calling launch() in
>> the slave's startup script, and this appears to "fix" the problem for me.
>> Does this trigger any ideas for anyone?
> I'm not sure what you mean here - the startup script is gone by the
> time the slave is started, right?  And you said that this occurs when
> you reconfigure the master, so the slave isn't even disconnected from
> the master.

Inaccurately typed. Occurs when you *restart* the master, so the slave 
*is* disconnected.

The startup script is gone, yes, but somehow the presence/absence of the 
fork() is affecting behaviour when disconnected. The simplest way to 
induce the hang, since discovered, was to start the master, then do 
buildslave start followed immediately by buildslave stop. Cue tight loop 
and hang of the buildslave process after it disconnects.

Could it be something to do with how Twisted 10.2.0 behaves when it's a 
daemon? I've added --nodaemon to the faked twistd arguments in the 
buildslave startup script, and this appears to resolve the hang problem 
on our slaves (all Macs).


More information about the devel mailing list