Hudson jobs lose working directory sporadically [message #1503193] |
Mon, 08 December 2014 03:38  |
Eclipse User |
|
|
|
Hi,
I'm currently having problems with Hudson jobs losing their working directory from time to time - it seems that something is removing the working directory of the job while it is running. Since I cannot entirely be sure that it's a Hudson problem, I wanted to ask beforehand if anyone has seen a problem like that before.
The jobs in question contain "Execute shell" steps that run a Python script which in turn forks other processes (something wrapped in shell scripts). There are no special settings regarding working directories in the "Advanced Job Options" section.
Occasionally - very rarely and in different jobs - the script fails when forking other commands with an error message indicating that the working directory of the shell just forked is gone, i.e.:
Hudson -> Execute shell -> Python -> shell wrapper -> fails because $PWD is gone.
sh: 0: getcwd() failed: No such file or directory
shell-init: error retrieving current directory: getcwd: cannot access parent directories: No such file or directory
This error message seems to indicate that a shell was started from a directory that had already been deleted at the time. Since none of the steps changes its working directory, I assume that there is some sort of race condition going on regarding the working directory of the Hudson job. No other build of that job was running at the same time. The problem is not reproducible: on re-executing the failed jobs, they worked ok.
We're running Hudson 3.1.2 on several nodes with build slaves being attached through ssh and slave.jar. The build slaves use local directories (i.e., no NFS or anything) as their file system base.
Does this behavior ring a bell with anyone?
Thanks and best regards,
Christoph
|
|
|
|
|
Powered by
FUDForum. Page generated in 0.07713 seconds