Re: [ptp-user] Need help about setting up and debugging!

["Fabio Meneghini" <fab.meneghini@xxxxxxxxx>]

yeah, that's right.....so what?

Fabio

2008/6/27 Beth Tibbitts <tibbitts@xxxxxxxxxx>:

Dumb question: you're running Eclipse on the frontend node, right? you have an ORTE@Local local resource manager (not remote)

...Beth

Beth Tibbitts (859) 243-4981 (TL 545-4981)
High Productivity Tools / Parallel Tools http://eclipse.org/ptp
IBM T.J.Watson Research Center
Mailing Address: IBM Corp., 745 West New Circle Road, Lexington, KY 40511

"Fabio Meneghini" <fab.meneghini@xxxxxxxxx>

"Fabio Meneghini" <fab.meneghini@xxxxxxxxx>

Sent by: ptp-user-bounces@xxxxxxxxxxx

06/27/08 10:31 AM

Please respond to
PTP User list <ptp-user@xxxxxxxxxxx>

To		"PTP User list" <ptp-user@xxxxxxxxxxx>
cc
Subject	Re: [ptp-user] Need help about setting up and debugging!

Greg,
I just did what you told me to. I tried on each of the 8 nodes (included the frontend), to run the sequential gdb debugger (from the console terminal) with the program, and it's perfectly working, from the start to the end. I'm able to watch variables, set breakpoints wherever I want...
What else could I try?
Thanks,
Fabio

2008/6/27 Greg Watson <g.watson@xxxxxxxxxxxx>:

Fabio,

I get virtually identical output to you up to the "don't know how to handle OS specific section" line, so it looks like this is causing the problem.

Can you try the following:

1. Log into one of the nodes and run 'gdb myprog' where myprog is the MPI program you're trying to run.
2. Set a breakpoint in main using the command 'set breakpoint main'.
3. Start the program using 'run'.

Let me know if you see any error messages or if gdb fails.

Thanks,

Greg

On Jun 27, 2008, at 7:31 AM, Fabio Meneghini wrote:

Hi Greg,
first thanks for you helping me,
BTW, the machine is a linux cluster, composed by 8 single-processor computers interconnected, frontend included.
Some moths ago I found that there was no gdb installed on the nodes, but only in the frontend (which is usually part of the computational nodes). So I copied the gdb binary from the frontend to the local folders on the other nodes. Now, when I access (through ssh or rsh) each node, I am able to run the gdb on that node, simply with the command /usr/bin/gdb
So I suppose this might be ok.

Regarding my citation: nevermind. Now, don't ask me why, the situation is somewhat different:
First, as I start Eclipse, the following message appears in the terminal: "ORTE@Local: proxy connected"
ok. That's awesome

Then, it's time to start parallel debugging! then I launch the parallel debugger, with 8 processors, and:
as the usual "Debugger has started...etc etc" window appears, it disappears almost instantly, and an error message raises up:
"Handling debug Event....
reason:
See Details..."
So I go to see the terminal, and the following results the output from the very moment I made the debugger run:

debug: waiting for connect
debug: received connected event
Msg: EventRequestManager - addEventRequest(): Request: Start debugger request in
status [5] for 0-7.
**** Msg: AbstractEventManager - registerEventRequest(): Request: Start debugger
request in status [5] for 0-7.
Msg: SDMEventManager - update(): Event: EVENT_DBG_OK transid=1 8:FF
Msg: SDMEventManager - fireEvent(): added PDIEvent: org.eclipse.ptp.debug.intern
al.core.pdi.event.ConnectedEvent@4365d26c
Msg: NotifyJob - size of runnables: 3
**** Msg: AbstractEventManager - notifyEventRequest(): Request: Start debugger r
equest in status [3] for .
Msg: NotifyJob - size of runnables: 3
Msg: EventRequestManager - addEventRequest(): Request: Set Function Breakpoint R
equest in status [5] for 0-7.
**** Msg: AbstractEventManager - registerEventRequest(): Request: Set Function B
reakpoint Request in status [5] for 0-7.
Msg: EventRequestManager - addEventRequest(): Request: Resume request in status
[5] for 0-7.
Msg: SDMEventManager - update(): Event: EVENT_DBG_OK transid=1 8:FF
**** Msg: AbstractEventManager - notifyEventRequest(): Request: Set Function Bre akpoint Request in status [3] for .
**** Msg: AbstractEventManager - notifyEventRequest(): Request: Resume request i n status [3] for 0-7.
Msg: NotifyJob - size of runnables: 5
================= TIME RESUME: 1214569632785
ORTE@Local: BFD: /lib64/ld-linux-x86-64.so.2: don't know how to handle OS specif ic section `.gnu.hash' [0x6ffffff6]
Msg: SDMEventManager - update(): Event: EVENT_DBG_EXIT transid=1 8:FF 177
Msg: SDMEventManager - fireEvent(): added PDIEvent: org.eclipse.ptp.debug.intern al.core.pdi.event.DestroyedEvent@3bb505fe
Msg: EventRequestManager - addEventRequest(): Request: Stop debugger request in status [5] for .
Msg: SDMEventManager - fireEvent(): added PDIEvent: org.eclipse.ptp.debug.intern al.core.pdi.event.DisconnectedEvent@72282e3b
**** Msg: AbstractEventManager - notifyEventRequest(): Request: Stop debugger re quest in status [3] for .
calling session finish
Msg: NotifyJob - size of runnables: 4
debug: received disconnected event
ORTE@Local: sdm: shutdown completed


Any hints?
thanks, again.
Fabio


2008/6/27 Greg Watson <g.watson@xxxxxxxxxxxx>:
Fabio,

What sort of machine are you using? Is it a cluster or an 8 cpu SMP?

Can you explain what you mean when you say "I can rapidly see the eight blocks representing the processors (in the machine view) becoming green, then all suddenly stops, after one second."? Do you mean the process icons change color at this point?

The first trace you sent looks ok up to the point:

ORTE@Local: BFD: /lib64/ld-linux-x86-64.so.2: don't know how to handle OS specific section `.gnu.hash' [0x6ffffff6]

This looks like a message from GDB. It's possible GDB is exiting at this point which may be causing the problem. Is there any delay between this message and the next:

Msg: SDMEventManager - update(): Event: EVENT_DBG_EXIT transid=1 8:FF 177

Thanks,

Greg


On Jun 27, 2008, at 5:25 AM, Fabio Meneghini wrote:

P.S. I was forgetting.....
when I hit the cancel button, in the "Debugger has started, waiting for connection...." window, the following appears in the console terminal:
calling session finish
debug: received message event
Msg: EventRequestManager - addEventRequest(): Request: Suspend request in status [5] for 0-4.
Error completing debug job launch: Cannot connect to debugger
Msg: EventRequestManager - addEventRequest(): Request: Terminate request in status [5] for 0-4.
Msg: EventRequestManager - addEventRequest(): Request: Stop debugger request in status [5] for .
**** Msg: AbstractEventManager - registerEventRequest(): Request: Suspend request in status [5] for 0-4.
**** Msg: AbstractEventManager - notifyEventRequest(): Request: Suspend request in status [1] for 0-4.

...


2008/6/27 Fabio Meneghini <fab.meneghini@xxxxxxxxx>:
hi again, now I noticed another weird thing: if I set the number of processors other than 8 (8 is the maximum number of processors of my machine), e.g. 5 or 3, or 2.... when I give the debug command, it stucks in the "Debugger has started, waiting for connection...." window....and it remains there forever, while in the terminal (from which I launched Eclipse) the following appears:

debug: waiting for connect
ORTE@Local: Debugger error: Proxy server error: could not find host
Again, please take note that ORTE@Local is my Resource Manager, that is supposed to be perfectly working, since I use it in the Run Sessions (instead of Debug Sessions) and it works great.

Please, help...I'm going mad...
Best
Fabio

2008/6/27 Fabio Meneghini <fab.meneghini@xxxxxxxxx>:

Hi Clement,
ok, I usually start eclipse from command line, just in order to see events messages...here's the output messages when I try to parallel debug my application:

CDT version: 4.0.3.200802251018
debug: waiting for connect
debug: received connected event
Msg: EventRequestManager - addEventRequest(): Request: Start debugger request in status [5] for 0-7.
**** Msg: AbstractEventManager - registerEventRequest(): Request: Start debugger request in status [5] for 0-7.
Msg: SDMEventManager - update(): Event: EVENT_DBG_OK transid=1 8:FF
Msg: SDMEventManager - fireEvent(): added PDIEvent: org.eclipse.ptp.debug.internal.core.pdi.event.ConnectedEvent@615c6e8d
Msg: NotifyJob - size of runnables: 3
**** Msg: AbstractEventManager - notifyEventRequest(): Request: Start debugger request in status [3] for .
Msg: NotifyJob - size of runnables: 3
Msg: EventRequestManager - addEventRequest(): Request: Set Function Breakpoint Request in status [5] for 0-7.
**** Msg: AbstractEventManager - registerEventRequest(): Request: Set Function Breakpoint Request in status [5] for 0-7.
Msg: EventRequestManager - addEventRequest(): Request: Resume request in status [5] for 0-7.
Msg: SDMEventManager - update(): Event: EVENT_DBG_OK transid=1 8:FF
**** Msg: AbstractEventManager - notifyEventRequest(): Request: Set Function Breakpoint Request in status [3] for .
**** Msg: AbstractEventManager - notifyEventRequest(): Request: Resume request in status [3] for 0-7.
Msg: NotifyJob - size of runnables: 5
================= TIME RESUME: 1214557631787
ORTE@Local: BFD: /lib64/ld-linux-x86-64.so.2: don't know how to handle OS specific section `.gnu.hash' [0x6ffffff6]
Msg: SDMEventManager - update(): Event: EVENT_DBG_EXIT transid=1 8:FF 177
Msg: SDMEventManager - fireEvent(): added PDIEvent: org.eclipse.ptp.debug.internal.core.pdi.event.DestroyedEvent@15bde556
Msg: EventRequestManager - addEventRequest(): Request: Stop debugger request in status [5] for .
Msg: SDMEventManager - fireEvent(): added PDIEvent: org.eclipse.ptp.debug.internal.core.pdi.event.DisconnectedEvent@44f4ccc0
**** Msg: AbstractEventManager - notifyEventRequest(): Request: Stop debugger request in status [3] for .
calling session finish
Msg: NotifyJob - size of runnables: 4
ORTE@Local: sdm: shutdown completed
Please note that ORTE@local is the Resource Manager I created. Just One thing to note: When I created this resource manager, there were no possibility to select a Multiplexing option (just "none" selected), in the PTP -> Resource Manager-> Open RTE preferences, the "port forwarding" multiplexing option is selected instead.

What should I do?

Thanks again,
best
Fabio



2008/6/27 Clement Kam Man Chu <clement.chu@xxxxxxxxxxxxxxxxxxxxxx>:

Hi Fabio,

How do you start up eclipse? You can try to start up eclipse on command line, then you can see ptp debug message on command line window. If possible, please send me a copy. Thanks.

Regards,
Clement

Fabio Meneghini wrote:
ok, thank you very much: I'm done with the Resource Manager. Successfully cvreated, following the instructions in the help!!!
Now, I managed to set-up the running configuration, and tried it within the PTP Runtime Perspective, and it works, great!
But, still I can't debug: seems weird, but as I start debugging (with 8 processors), I can rapidly see the eight blocks representing the processors (in the machine view) becoming green, then all suddenly stops, after one second. Actually I'm pretty good in parallel programming, but not THIS good! :-)
I'm sure there is no mistake in the code, and the program itself, since when I run it, either from command line (mpirun, etc etc) , or from the Parallel Run command in eclipse, it does what it's supposed to.
In the console windows no messages appears, where can I look for some error message occurred?
Thanks again,
best
Fabio

2008/6/26 Beth Tibbitts <tibbitts@xxxxxxxxxx <mailto:tibbitts@xxxxxxxxxx>>:


There is another walk-thru of the setup steps in a recent tutorial
we gave.
See the PTP tutorials page: http://wiki.eclipse.org/PTP/tutorials
and get the slides from the TACC tutorial on May 16.




...Beth

Beth Tibbitts (859) 243-4981 (TL 545-4981)
High Productivity Tools / Parallel Tools http://eclipse.org/ptp
IBM T.J.Watson Research Center
Mailing Address: IBM Corp., 745 West New Circle Road, Lexington,
KY 40511
Inactive hide details for "Fabio Meneghini"

<fab.meneghini@xxxxxxxxx>"Fabio Meneghini"
<fab.meneghini@xxxxxxxxx <mailto:fab.meneghini@xxxxxxxxx>>



*"Fabio Meneghini"
<fab.meneghini@xxxxxxxxx
<mailto:fab.meneghini@xxxxxxxxx>>*

Sent by: ptp-user-bounces@xxxxxxxxxxx
<mailto:ptp-user-bounces@xxxxxxxxxxx>


06/26/08 01:19 PM

Please respond to
PTP User list <ptp-user@xxxxxxxxxxx
<mailto:ptp-user@xxxxxxxxxxx>>





To

"PTP User list" <ptp-user@xxxxxxxxxxx <mailto:ptp-user@xxxxxxxxxxx>>


cc


Subject

Re: [ptp-user] Need help about setting up and debugging!






I just removed the eclipse ganymede, and installed the europa.
Then the same procedure to install the PTP 2.0, with the same
BUILD command. Again, the building seems to go ok: the binaries
are created, BUT, again, whenever I try to set up a new parallel
debug configuration, Eclipse complain about no Resource Manager
selected, showing me an empty list of choices....
I can't really figure out what's wrong....

Fabio

2008/6/26 Fabio Meneghini <_fab.meneghini@gmail.com_
<mailto:fab.meneghini@xxxxxxxxx>>:


yes, I can. mpirun and stuff work great. Now I'll try the
versions of eclipse and cdt you suggested me.
I'll get you informed.
thanks,
best
fabio

2008/6/26 Beth Tibbitts <_tibbitts@xxxxxx.com_
<mailto:tibbitts@xxxxxxxxxx>>:

>Are you asking me if the mpi tests from the command line
run successfully?
Yes, for example can you do mpicc and mpirun from the
command line?
The release notes at
_http://wiki.eclipse.org/PTP/release_notes/2.0#Install_Open_MPI_
cover this.
Let us know how you fare.



...Beth

Beth Tibbitts (859) 243-4981 (TL 545-4981)
High Productivity Tools / Parallel Tools
_http://eclipse.org/ptp_
IBM T.J.Watson Research Center
Mailing Address: IBM Corp., 745 West New Circle Road,
Lexington, KY 40511
Inactive hide details for "Fabio Meneghini"

<fab.meneghini@xxxxxxxxx>"Fabio Meneghini"
<_fab.meneghini@gmail.com_ <mailto:fab.meneghini@xxxxxxxxx>>

*"Fabio
Meneghini"
<**_fab.meneghini@gmail.com_*
<mailto:fab.meneghini@xxxxxxxxx>*>*

Sent by:
_ptp-user-bounces@eclipse.org_
<mailto:ptp-user-bounces@xxxxxxxxxxx>



06/26/08
10:53 AM

Please respond to
PTP User list <_ptp-user@eclipse.org_
<mailto:ptp-user@xxxxxxxxxxx>>



To

"PTP User list" <_ptp-user@eclipse.org_
<mailto:ptp-user@xxxxxxxxxxx>>


cc

Subject

Re: [ptp-user] Need help about setting up and debugging!




Hi Beth
Thank you very much for your quick, kind and clear reply!
So it turns out it's a "versioning" problem, uh?
ok, I'll follow your advices....just one question: what do
you exactly mean with
" Did you confirm that your MPI installation (e.g. OpenMPI)
works from the command line first? "
Are you asking me if the mpi tests from the command line run
successfully?

Thanks again,
best
Fabio Meneghini

2008/6/26 Beth Tibbitts <_tibbitts@xxxxxx.com_
<mailto:tibbitts@xxxxxxxxxx>>:

Hi Fabio,
Sorry you have having problems.
PTP 2.0 was built to run with Eclipse 3.3 - Hmmm
we need to make the pre-reqs more explicit - we
check that your release is 'new enough" but not
'too new" methinks. :)
PTP 2.1, targeted at around August, will work
with Eclipse 3.4 (Ganymede)

Try it with Eclipse 3.3 and CDT _4.0.3._
<http://4.0.0.3/>

Did you walk thru the release notes at
_http://wiki.eclipse.org/PTP/release_notes/2.0?_
Did you confirm that your MPI installation (e.g.
OpenMPI) works from the command line first?

Also, read the help in Eclipse Help > Help
Contents and click on PTP on the left navigation
area, it will walk you through the setup once
you launch eclipse.

Note that you have to configure at least one
"Resource Manager"
which can be as simple as OpenMPI/ORTE, doesn't
have to be a more complex scheduler.
A "local" resource manager means it (e.g.
OpenMPI) is on the same machine where your
Eclipse workbench (and PTP) are running.
A "remote" resource manager means it's on a
different machine.

ptp_orte_proxy is not meant to be run
stand-alone, it will be launched from PTP.



...Beth

Beth Tibbitts (859) 243-4981 (TL 545-4981)
High Productivity Tools / Parallel Tools
_http://eclipse.org/ptp_
IBM T.J.Watson Research Center
Mailing Address: IBM Corp., 745 West New Circle
Road, Lexington, KY 40511
Inactive hide details for "Fabio Meneghini"

<fab.meneghini@xxxxxxxxx>"Fabio Meneghini"
<_fab.meneghini@gmail.com_
<mailto:fab.meneghini@xxxxxxxxx>>
*"Fabio
Meneghini"
<**_fab.meneghini@gmail.com_*
<mailto:fab.meneghini@xxxxxxxxx>*>*

Sent
by:
_ptp-user-bounces@eclipse.org_
<mailto:ptp-user-bounces@xxxxxxxxxxx>



06/26/08
09:27
AM



Please respond to
PTP User list <_ptp-user@eclipse.org_
<mailto:ptp-user@xxxxxxxxxxx>>


To
_
__ptp-user@eclipse.org_ <mailto:ptp-user@xxxxxxxxxxx>

cc

Subject

[ptp-user] Need help about setting up and debugging!




Hi all,
I've just downloaded the latest version of
eclipse for c++ developer (3.4.0), and the PTP
2.0 correctly installed
Since my installation of openmpi isn't resident
in canonical folder (i.e. /usr/local ), I built
up all the ptp binaries stuff with the command:

CFLAGS=-I/opt/openmpi/include/
LDFLAGS="-L/opt/openmpi/lib
-Wl,-rpath,/opt/openmpi/lib" sh BUILD

compilation, linking and installation seem to be
accomplished. No errors or problems.
So the binaries are actually in the right
folder. Then I tried to parallel debug a program
of mine. No way, because the debug configuration
stucks me with the remark: "No Resource Manager
has been selected". Infact, there is no resource
manager available in the list.
I do not use any external, thirdy party,
Resource Manager, like IBM's or something. And I
actually don't even understand if I need one,
since I didn't realize what you mean with the
_expression_:
" If you don't intend to run parallel
applications locally, there is no need to
install any resource managers ".
I'm supposed to run and debug parallel
applications locally, since I use to locally
access the cluster frontend, so am I supposed to
use a "Resource Manager", or whatever? and if
so, may I use the ptp_orte_proxy as a "Resource
Manager"? .....or what?
BTW, I suspect there is something wrong with the
ptp_orte_proxy, because when I try to execute
from a terminal, it results in " proxy
connection failed"
I know I could sound something sick, but I'm
really frustrated by this problem...all I am
trying to do, from weeks ago, is simply
debugging my parallels programs...but I can't
managed yet.
Can anybody help me, please?
_______________________________________________
ptp-user mailing list_
__ptp-user@eclipse.org_
<mailto:ptp-user@xxxxxxxxxxx>_

__https://dev.eclipse.org/mailman/listinfo/ptp-user_


_______________________________________________
ptp-user mailing list_
__ptp-user@eclipse.org_
<mailto:ptp-user@xxxxxxxxxxx>_

__https://dev.eclipse.org/mailman/listinfo/ptp-user_
_______________________________________________
ptp-user mailing list_
__ptp-user@eclipse.org_ <mailto:ptp-user@xxxxxxxxxxx>_

__https://dev.eclipse.org/mailman/listinfo/ptp-user_


_______________________________________________
ptp-user mailing list_
__ptp-user@eclipse.org_ <mailto:ptp-user@xxxxxxxxxxx>_

__https://dev.eclipse.org/mailman/listinfo/ptp-user_

_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx <mailto:ptp-user@xxxxxxxxxxx>

https://dev.eclipse.org/mailman/listinfo/ptp-user


_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx <mailto: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



--
Clement Kam Man Chu
Research Fellow
Faculty of Information Technology
Monash University, Caulfield Campus
Ph: 61 3 9903 2355


_______________________________________________
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

_______________________________________________
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