Opened 4 years ago

Closed 4 years ago

#1524 closed help (answered)

advanced debug of job

Reported by: swr04ojb Owned by: um_support
Component: UM Model Keywords:
Cc: simon.tett Platform: ARCHER
UM Version: 8.5

Description

Hi NCAS,

we have a bug that is proving tricky to chase. What would be ideal would be if we could step through the routine line-by-line, querying variable values as we did so. The UM vm8.5 atmos-only job is on ARCHER, is such a form of debug possible? I've been referred to this page by you before:

http://www.archer.ac.uk/documentation/best-practice-guide/debug.php

, which I've used for stack-traces, so thank you for that, but I don't know how to implement an interactive debug such as described above. Has anyone attempted this before with the UM on ARCHER? Are there guidance notes somewhere?

cheers,

Oliver

Change History (5)

comment:1 Changed 4 years ago by swr04ojb

UPDATE:

Simon's found that compiling/running the UM on an interactive ARCHER node significantly reduces the development cycle. From a typical turnaround of ~20minutes using the short-queue we're now down to 5 minutes:

a few minutes for the submit from PUMA
+ 30secs for compile
+ ~30-60sec to get say 100 timesteps in & then stop the model to look for errors.

That is nice, but still doesn't get us *inside* the routine as most modern debuggers would allow.

We can switch on debug-flags on individual files, via hand-edit:

http://cms.ncas.ac.uk/ticket/1449

Presumably we'd now just need a way to attaching a debugger to the process-id?

comment:2 Changed 4 years ago by grenville

Oliver

We don't yet have experience with DDT - we previously used totalview.

Grenville

comment:3 Changed 4 years ago by swr04ojb

Grenville,

thanks. I've installed a DDT client, and got it talking to ARCHER okay. I'm at 7.5.5 on:

http://www.archer.ac.uk/documentation/best-practice-guide/debug.php#sec-7.5

which is the step where I would configure it to call the executable, but I'm unsure on what to specify there - should it be the stage_1_submit file, the umuisubmit_run file, the UMScr_Toplevel file, or something else?

cheers,

Oliver

comment:4 Changed 4 years ago by grenville

Oliver

It wants the executable - it's not obvious how to specify that. TotalView? worked differently, in that you'd put a wrapper around aprun (that was easy) - it may be that ddt allows the same method of working, but I don't know yet.

Grenville

comment:5 Changed 4 years ago by grenville

  • Resolution set to answered
  • Status changed from new to closed
Note: See TracTickets for help on using tickets.