Re: [ptp-dev] parallel debugger broken in 5.0.1 (was RE: slides posted)

[Galen Arnold <arnoldg@xxxxxxxxxxxxx>]

Jay,

I rebuilt the sdm with the 5.0.1 version and that fixed it.  Updating had allowed me to run with the 5.0.0 sdm.  Nice catch.  Nothing to see here, move along.
It would be nice if an update could catch this issue.

-Galen

Galen Arnold
system engineer
NCSA

----- Original Message -----
From: "Jay Alameda" <jalameda@xxxxxxxxxxxxxxxxx>
To: "Galen Arnold" <arnoldg@xxxxxxxxxxxxx>
Cc: "Parallel Tools Platform general developers" <ptp-dev@xxxxxxxxxxx>
Sent: Wednesday, July 13, 2011 7:21:41 PM
Subject: RE: parallel debugger broken in 5.0.1 (was RE: slides posted)

Galen,

One quick question: did you update the SDM using the ptp-proxy-code from 
http://download.eclipse.org/tools/ptp/builds/5.0.1/I.I201107131351/index.html 
(which is found by following the links from 
http://wiki.eclipse.org/PTP/builds/5.0.1) - RC4 hit today, the specific 
build link has ptp-proxy components as a bundle... there may be an issue 
with out of sync components, which can cause strange issues...

Jay


-----Original Message-----
From: Galen Arnold [mailto:arnoldg@xxxxxxxxxxxxx]
Sent: Wednesday, July 13, 2011 7:18 PM
To: Jay Alameda
Cc: Parallel Tools Platform general developers
Subject: Re: parallel debugger broken in 5.0.1 (was RE: slides posted)

Jay,

This is with the latest updates from : 
http://download.eclipse.org/tools/ptp/updates/indigo_5.0.1

I tested parallel debug with stock 5.0.0 from the public download site and 
it all worked, then updated to 5.0.1 and it all didn't (parallel debug).

Again, symptom: ranks die at breakpoint (using the stock built-in mpi 
hello_world c project).  Parallel runtime seems fine for 5.0.1 with a normal 
run configuration for openmpi on localhost.

It was broken on my 5.0.1 test machine at work (x86 linux) and I thought I'd 
see if it was an issue with that low powered laptop so I tried on a big 
x86_64 linux box here at home and reproduced the broken-ness upon upgrading 
to 5.0.1.

The parallel runtime is openmpi via system monitor all on localhost 
(simplest possible resource mgr as far as I can tell).

-Galen

Galen Arnold
system engineer
NCSA

----- Original Message -----
From: "Jay Alameda" <jalameda@xxxxxxxxxxxxxxxxx>
To: "Galen Arnold" <arnoldg@xxxxxxxxxxxxx>
Cc: "Parallel Tools Platform general developers" <ptp-dev@xxxxxxxxxxx>
Sent: Wednesday, July 13, 2011 7:10:54 PM
Subject: parallel debugger broken in 5.0.1 (was RE: slides posted)

Thank you, Galen, I take it that this is with 5.0.1 RC4 that is posted.
Greg, this is something you'll want to see -
I *think* tomorrow, I'll have the students use the temporary update site,
and update their installations.  I'm really nervous as we've not had
sufficient time to pound on 5.0.1 (or even 5.0.0 for that matter, in the
manner of the full tutorial).  On the other hand, if they don't know about
the temporary update site, perhaps we can glide through with 5.0.0... and
when we are happy with 5.0.1, and release it, then the update procedure will
then work (right now, all that will happen is that 5.0.0 gets installed
again, so that the pump is primed for the update mechanisms to work, for the
std ptp indigo update site).

My heart did skip a beat when I saw your mail come through...

Jay


-----Original Message-----
From: Galen Arnold [mailto:arnoldg@xxxxxxxxxxxxx]
Sent: Wednesday, July 13, 2011 7:06 PM
To: Jay Alameda
Cc: Parallel Tools Platform general developers
Subject: Re: slides posted

Jay,

5.0.1 breaks parallel debug.  It works in 5.0.0 .
The symptom is that the ranks all die in 5.0.1 when you hit a breakpoint.
This is with openmpi resource mgr via system monitor.  I'm not going to
bother with the parallel debug for tomorrow.
It looks like we'll have plenty of material.  I might try to work something
up again on Fri. afternoon or maybe not--it would mean backing up to 5.0.0 .

-Galen

Galen Arnold
system engineer
NCSA

----- Original Message -----
From: "Jay Alameda" <jalameda@xxxxxxxxxxxxxxxxx>
To: "Galen Arnold" <arnoldg@xxxxxxxxxxxxx>
Cc: "Jay Alameda" <jalameda@xxxxxxxxxxxxxxxxx>
Sent: Wednesday, July 13, 2011 6:58:18 PM
Subject: slides posted

Galen,



I managed to get some version (not happy with the versions posted, not yet
at least!) to http://wiki.eclipse.org/PTP/tutorials/TG11, at the moment,
I'm planning on backing out the additional slides to start remote
synchronized projects (module 3) and going through module 4, there are a
number of things in that module that need updating (ARGH).  5, 6, 7 are
unchanged from SC10, and we aren't presenting them anyway, they are bonus
slides. and, I did modify the wrap up to reflect what is going on today a
bit more accurately.  I think I must add synchronized projects to module 3
for TG11; I plan on doing some testing with trestles in the process of
working up module 4 tonight before going home.  I'm going to stage source
in my home directory.  Hair standing on end.



Jay