[Buildbot-devel] dangling processes

[Daniel]

Greetings, all.

I have been in the process of moving buildbot from one XP-64bit system  
to another, in order to spread the load between our systems.  I  
finally have buildbot running on the new system, but it is highly  
unreliable.  What I find is that processes are left-over from an  
earlier build, so later builds get "permission denied" errors trying  
to remove a directory.  The build then terminates in "exception" status.

Our buildbot system builds on Linux (SLES10, AMD, 64-bit), OS X  
(Leopard, Intel, 64-bit), and Windows (XP, 64-bit).  The original  
Windows build slave does not exhibit these hangs, and neither Linux  
nor OS X exhibits these hangs.

I am using procexp to see which process(es) is/are holding on to the  
files in question, and to kill those processes.  This usually lets a  
single build complete, but rarely does it allow more than one build to  
complete.  So far I have not been able to determine a pattern to when  
the files are locked (that is, when the dangling processes are not  
killed off) and when they are not (that is, when the dangling  
processes are all properly killed off).

I've done Google searches and have only discovered one other person  
with this issue, but there was no discussion on how the issue was  
resolved.

Anybody have any ideas of what I should look into here?  Any  
assistance would be appreciated.

Oh, we are on buildbot 0.7.7.

Thanks.

Daniel