Unfriendly HEAD is unfriendly

[Simon Marlow]

Roman Leshchinskiy wrote:

> Does anyone actually pay attention to the bots? Yesterday:
> 
>> x86-64 Linux head:            lost
>> x86 Windows head:             lost
>> x86 Windows head fast:        fail (failed stage1) fail (failed 
>> stage1) lost fail (failed stage1) lost lost
>> fast486 head:                 fail (failed stage2)
>> gabor head:                   fail (failed darcs)
>> kgardas head:                 fail (failed stage1)
>> mnemosyne x86-64 Gentoo head: fail (failed stage2)
>> tnaur PPC OSX head 2:         fail (failed stage2)
>> tnaur x86 Linux head:         pass
>> x86-64 Linux head unreg:      lost
> 
> I've went through my ghc-cvs archives but couldn't find when all bots 
> last ran successfully.

Sadly, there are two problems affecting the main buildbots here at MSRC:

1. the firewall randomly closes connections (I have a long-standing support 
ticket open - they suspect a bug in the firewall software but haven't been 
able to find it yet)
2. BuildBot can't cope with dropped connections in the middle of a build. 
Again I have a ticket open against BuildBot but fixing it apparently 
requires large amounts of infrastructure refactoring, so it's currently 
scheduled for the next-but-one major release of BuildBot, which could be 
months or years away.

which together mean that most of our buildbot builds fail with "lost". 
Sigh.  For this reason I've left our old nightly builds in place, on 
x86/Linux, and I also don't tend to look much at buildbot results much, 
which is bad.

If anyone can think of a way to work around this, I'm open to suggestions. 
  I've heard that using SSH in tunneling mode might help, but will it 
really re-connect automatically and invisibly to the client?

Cheers,
	Simon