"Solver did not exit normally" - Jupyter / Python3 / Ubuntu

Question

I am trying to get my first Pyomo model running on my Ubuntu VM (Azure). I have Python3 and the COIN-OR solvers installed on this machine. No matter what solver I try, I get the same result.

Edit: changing the solver to couenne (it's a nonlinear problem) the Jupyter output looks like this. When I open the log files in the tmp directory, there is nothing in the couenne.log file and the pyomo files are the problem formulation. So I assume that Pyomo isn't communicating with the Couenne solver at all?

Solver log file: '/tmp/tmpezw0sov2_couenne.log'
Solver solution file: '/tmp/tmpq6afa7e8.pyomo.sol'
Solver problem files: ('/tmp/tmpq6afa7e8.pyomo.nl',)
ERROR: Solver (asl) returned non-zero return code (-1)
ERROR: See the solver log above for diagnostic information.
---------------------------------------------------------------------------
ApplicationError                          Traceback (most recent call last)
<ipython-input-6-486e3a9173f4> in <module>()
     20 #instance = model.create_instance()
     21 opt = SolverFactory('couenne', executable = solverpath_exe)
---> 22 opt.solve(model,tee=True,keepfiles=True)
     23 #solver=SolverFactory(solvername,executable=solverpath_exe)

/home/ralphasher/.local/lib/python3.6/site-packages/pyomo/opt/base/solvers.py in solve(self, *args, **kwds)
    598                     logger.error("Solver log:\n" + str(_status.log))
    599                 raise pyutilib.common.ApplicationError(
--> 600                     "Solver (%s) did not exit normally" % self.name)
    601             solve_completion_time = time.time()
    602             if self._report_timing:

ApplicationError: Solver (asl) did not exit normally

Answer 1

The "catch-all" exception is raised because the solver runs as a separate, non-Python process, so Python really can't tell what exactly went wrong with it, it just sees that the process has quit abnormally.

As such, solver log is the thing to go to as this is where the solver itself writes its status updates, so the specific error, whatever it is, should be reflected there.

If solver log is empty, this most probably means that the solver has failed to start at all (if solver process is run with stream redirection, the log file is opened -- thus created -- before the solver command is exec'd, so this is a common symptom when there's a problem with a program's startup). Since pyomo is the thing that starts the solver, getting the details of what exactly happens at the time of solver's start is where the answer lies.

---

According to pyomo solve command — Pyomo 5.6.6 documentation, you can use --info or --verbose command line options to increase the verbosity of the pyomo log.

If that still doesn't produce anything revealing, time to bring out the big guns:

• run pyomo under pdb _{(pyomo is just a script so you can pass it to python -m pdb like any other; make sure to use the same python executable as in the script's shebang)} and step through the code in pyomo machinery to see what exactly it does with the solver process (what info it passes, how it invokes it)
  • you'll be able to see defects in this process if there are (e.g. no info is actually passed) or repeat the same operations by hand to see the result firsthand; and/or
• run the command under strace -f (to also monitor the solver's child process) and see if there are any obvious errors like an error to exec the solver command or any errors opening files and such.