[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
|
Re: [ptp-dev] Invoking SDM debugger to debug PE applictaion fails with rc -1
|
Dave,
The routing file is (currently) generated by the PTP debugger when the
job state changes to RUNNING. If the job's processes have not been
added to the model at this point, then when the debugger queries the
model it will not find any processes, and hence write out an empty
routing file.
Model elements should behave like this:
1. When a job submitted, create a job model element with state PENDING
2. When the job has been scheduled for execution, change the job model
element state to STARTED
3. Create a process model element for each process, with state STARTED
4. When the processes start executing, change the process model
element state to RUNNING
5. When all processes have started, change the job model element to
RUNNING
6. When a processes terminates, change the process model element state
to EXITED
7. When all processes have terminated, change the job model element
state to TERMINATED
Can you check that the PE resource manager follows this?
Regards,
Greg
On Nov 24, 2008, at 1:42 PM, Dave Wootton wrote:
Greg
I'm looking at Ping's problem debugging PE applications using the SDM
debugger where the master process startup fails with rc -1.
I verified that the code in my PE proxy which attempted to write the
routing file is ifdef-ed out of the executable, so my proxy is not
writing
anything to the routing_file. However, the routing_file is corrupt,
consisting only of a '0' followed by newline.
I wrote a wrapper script to invoke my proxy under control of strace
so I
could see the I/O and process-related calls made by it. I have a trace
that shows all opens, reads and writes, along with exec calls. The
only
open of the routing_file is for read-only, and that is in the SDM
process
itself. So my proxy is not corrupting the routing_file by writing
over it.
Can you see any cases where the front end code would generate a
routing_file where it thinks that there are zero processes in the
application? I might not be passing across an event in the correct
format
to let the debugger know what the processes are, resulting in a
corrupt
file. I don't know which messages the front end debugger code might be
looking for to build the routing file, so any guidance you have
would be
helpful.
The other possibility is that if some of the data in the messages is
in
binary format rather than ASCII text, then there might be a byte
ordering
(endian) problem, since this is x86 Linux, while I was able to debug
successfully all last week with ppc Linux.
I haven't tested my proxy's invocation of the debugger on x86 Linux at
all, since I didn't have access to a x86 system with PE installed. I
will
try to get access to a system here and see if I can get the failure
locally.
Dave
_______________________________________________
ptp-dev mailing list
ptp-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-dev
