[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
|
Re: [ptp-user] parallel application run failure
|
Brett
I installed the 08/05/09 PTP build and ran both my test case and yours.
Both ran successfully and I see console output.
One more question. When you set up the resources tab in the run
configuration with the PE options you wanted, do you remember what
specific ones you set? The minimal set, which I generally use are the host
list file and number of tasks on the tasks tab in the resources pane.
I just committed the latest version of the proxy code, which adds some
timing code and an optimization one of the messages that might be goving
you trouble. I'm not sure this will fix your problem, but it might be
worth a try. You can extract it from cvs (the
org.eclipse.ptp.rm.ibm.pe.proxy project, src/ptp_ibmpe_proxy.c source). I
can also email you the source file directly or you can pick up the next
PTP nightly build, which I think would include this.
Greg, if this doesn't solve the problem, I need to see the messages that
are logged by the PTP trace and by the PE proxy. Can you remind me what
file I have to edit and what options to set within that file to turn on
this tracing?
Dave
From:
Brett Bode <bbode@xxxxxxxxxxxxx>
To:
PTP User list <ptp-user@xxxxxxxxxxx>
Date:
08/12/2009 11:25 AM
Subject:
Re: [ptp-user] parallel application run failure
Sent by:
ptp-user-bounces@xxxxxxxxxxx
On Aug 11, 2009, at 5:52 PM, Dave Wootton wrote:
>
> Brett
> I have a small MPI program, two tasks, I ran that outputs a single
> line of text, about 100 bytes long to stdout. If the program runs
> for a few seconds, then I see the output appear in the console view,
> assuming I have checked the 'Display combined output in a console
> view'.
>
> A few questions for you
>
> Is your program writing to stdout using stdio functions (printf,
> etc) or writing to file descriptor 1 using write() system calls,
> assuming C/C++ code. If it's Fortran code are you using write/print
> statements?
I wrote a simple test case this morning that exhibits the same issue:
#include <stdio.h>
int main(void) {
printf("Simple IO test to see if this appears on the
Eclipse console
\n");
sleep(10);
printf("Bye bye!\n");
}
>
> With the PTP runtime jobs view open, normal behavior should be
> seeing is that a blue job icon briefly appears in the left hand pane
> of teh jobs view. After a couple seconds, it should turn into a
> green triangle and green diamond icons appear in the right hand pane
> of the jobs view, 1 per task. When the job completes, the job icon
> and the diamond icons should turn red. Are you seeing this sequence
> or something else?
>
I see the blue icon change to a green triangle, but there are no
diamonds (green or otherwise) in the jobs view.
> Once the job completes, if you double click on the leftmost red
> diamond icon, which represents task 0, a 'Process details' window
> should appear in the editor pane of the perspective, with the job
> output appearing there. You should also be able to do this when the
> icon is green and watch stdout in real time. Do you see anything
> there?
Still no diamonds??
>
> If you open the error log view, do you see any errors (red icons)
> logged related to the run?
I am seeing some rather non-descript errors. They seem to appear one
per process when the run starts and again when it ends. The errors
simply say "ProcessChange: unknown process 8" and an exception trace
is not available.
> Does it appear that the program is otherwise running to completion
> (other data files it writes other than stdout/stderr get created
> correctly)?
My previous test appeared to, but I don't think this one is staying
running 10 seconds so I would say no. The infinite loop test stayed
running until I told it to stop.
>
> I'm seeing the output almost immediately after I start the program.
> If you change your test program so that it runs for something like
> 10 seconds instead of running quickly, or an infinite loop, does
> that affect the output?
> Dave
>
>
>
> Re: [ptp-user] parallel application run failure
>
> Brett Bode
> to:
> PTP User list
> 08/11/2009 03:09 PM
>
> Sent by:
> ptp-user-bounces@xxxxxxxxxxx
>
> Please respond to PTP User list
>
>
>
>
>
>
> It can be a short running job, but I have modified it to go into an
> infinite loop as well. When it's setup as a short running task the
> second console window does not open at all. The traceback shows up for
> both cases. In any event I would hope that I would be able to see the
> output of even something as trivial as hostname.
>
> Brett
> On Aug 11, 2009, at 2:05 PM, Greg Watson wrote:
>
> > Brett,
> >
> > Is this a short running job (e.g. hello world)? If so, it's possible
> > the program is exiting before any output can be displayed. You could
> > test this by adding a sleep(1) at the end of the program.
> >
> > Greg
> >
> > On Aug 11, 2009, at 2:32 PM, Brett Bode wrote:
> >
> >> Hello all,
> >> I am having a problem where I am attempting to run a simple MPI
> >> application remotely on an AIX 6.1 system utilizing the PE RM proxy
> >> (this is with the 8/5 PTP 3.0 build and corresponding RDT server,
> >> etc). It appears that the application launches and runs
> >> successfully, but I do not receive any output back to the console
> >> in eclipse. I have tried several things including specifying a file
> >> for the output and error (that does appear to work), but have had
> >> no luck getting output back to the eclipse console.
> >>
> >> If I set the app up to go into a loop I see a console list the
> >> build results (that works fine) and then a new console opens up but
> >> remains empty. If I set the app up to run (quickly) to termination
> >> the 2nd console doesn't open at all. The only clue I see is a
> >> traceback on the server side in the window running the RDT server
> >> process:
> >>
> >> java.io.IOException: Pipe closed
> >> at java.io.PipedInputStream.read(PipedInputStream.java:263)
> >> at
> >> com.ibm.jvm.io.ProcessInputStream.read(ProcessInputStream.java:92)
> >> at java.io.FilterInputStream.read(FilterInputStream.java:89)
> >> at
> >> org
> >> .eclipse
> >> .rse
> >> .internal
> >> .dstore
> >> .universal
> >> .miners.command.OutputHandler.readLines(OutputHandler.java:168)
> >> at
> >> org
> >> .eclipse
> >> .rse
> >> .internal
> >> .dstore
> >> .universal.miners.command.OutputHandler.handle(OutputHandler.java:
> 91)
> >> at org.eclipse.dstore.core.model.Handler.finish(Handler.java:
> >> 102)
> >> at
> >> org
> >> .eclipse
> >> .rse
> >> .internal
> >> .dstore
> >> .universal.miners.command.OutputHandler.handle(OutputHandler.java:
> >> 110)
> >> at org.eclipse.dstore.core.model.Handler.run(Handler.java:135)
> >>
> >> I do have the "Display combined output in console view" box
> >> checked, though I have also tried it without it checked.
> >>
> >> Brett
> >>
> >>
> >> _______________________________________________
> >> ptp-user mailing list
> >> ptp-user@xxxxxxxxxxx
> >> https://dev.eclipse.org/mailman/listinfo/ptp-user
> >
> > _______________________________________________
> > ptp-user mailing list
> > ptp-user@xxxxxxxxxxx
> > https://dev.eclipse.org/mailman/listinfo/ptp-user
> >
>
> _______________________________________________
> ptp-user mailing list
> ptp-user@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/ptp-user
>
> _______________________________________________
> ptp-user mailing list
> ptp-user@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/ptp-user
_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-user
