I'm way beyond that problem. I'm still dealing with issues, but a
basic cluster works. I'm dealing with specific server issues.
I really have no bandwidth to make my bproc code accessible right now,
outside the company. It's still unreleased code and in a state of
daily flux.
-- John
On Mon, Mar 26, 2012 at 6:50 PM, Adam Young <ayoung(a)redhat.com> wrote:
On 02/13/2012 11:42 AM, John Hawkes wrote:
>
> Printf works for a process that executes on the master node (perhaps
> doing a printf), does a bproc_move() to a compute node, then does a
> printf on the compute node.
>
> Printf does not work for a process for which bpsh does an execmove to
> a compute node and does a printf from there. In part this failure is
> due to a failing sys_dup2() done on the compute node to connect the
> I/O sockets to the stdin, stdout, and stderr file descriptor numbers,
> because one or more of those stdin/stdout/stderr file descriptors are
> open (even after the kernel/slave.c code closes those file descriptors
> in the newly forked stub). I'm fixing that problem now.
>
John,
I am assuming that you have either had success or were diverted onto a
different project.
If you were successful, can you make the current BProc code accessible? I
will start the process of applying it to the Fedora Kernel, with an eye to
getting it submitted upstream.
If you are still stuck, let me know, and I will help you with the NFS
approach I had working before, that makes it simpler to debug the bproc
problems.