# Standard imports.
import os
import logging
from optparse import OptionParser, OptionGroup, Option
from os.path import abspath, basename, splitext
import sys
import time
# PySPH imports.
from pysph.base.config import get_config
from pysph.base import utils
from pysph.base.nnps import BoxSortNNPS, LinkedListNNPS
from pysph.solver.controller import CommandManager
from utils import mkdir, load
# conditional parallel imports
from pysph import Has_MPI, Has_Zoltan
if (Has_MPI and Has_Zoltan):
from pysph.parallel.parallel_manager import ZoltanParallelManagerGeometric
import mpi4py.MPI as mpi
def list_option_callback(option, opt, value, parser):
val = value.split(',')
val.extend( parser.rargs )
setattr( parser.values, option.dest, val )
logger = logging.getLogger(__name__)
##############################################################################
# `Application` class.
##############################################################################
[docs]class Application(object):
""" Class used by any SPH application.
"""
def __init__(self, fname=None, domain=None):
""" Constructor
Parameters
----------
fname : str
file name to use.
domain : pysph.nnps.DomainManager
A domain manager to use. This is used for periodic domains etc.
"""
self.is_periodic = False
self.domain = domain
if domain is not None:
self.is_periodic = domain.is_periodic
self._solver = None
self._parallel_manager = None
if fname == None:
fname = splitext(basename(abspath(sys.argv[0])))[0]
self.fname = fname
self.args = sys.argv[1:]
# MPI related vars.
self.comm = None
self.num_procs = 1
self.rank = 0
if Has_MPI:
self.comm = comm = mpi.COMM_WORLD
self.num_procs = comm.Get_size()
self.rank = comm.Get_rank()
self._log_levels = {'debug': logging.DEBUG,
'info': logging.INFO,
'warning': logging.WARNING,
'error': logging.ERROR,
'critical': logging.CRITICAL,
'none': None}
self._setup_optparse()
self.path = None
self.particles = []
self.inlet_outlet = []
def _setup_optparse(self):
usage = """
%prog [options]
Note that you may run this program via MPI and the run will be
automatically parallelized. To do this run::
$ mpirun -n 4 /path/to/your/python %prog [options]
Replace '4' above with the number of processors you have.
Below are the options you may pass.
"""
parser = OptionParser(usage)
self.opt_parse = parser
# Add some default options.
# -v
valid_vals = "Valid values: %s"%self._log_levels.keys()
parser.add_option("-v", "--loglevel", action="store",
type="string",
dest="loglevel",
default='info',
help="Log-level to use for log messages. " +
valid_vals)
# --logfile
parser.add_option("--logfile", action="store",
type="string",
dest="logfile",
default=None,
help="Log file to use for logging, set to "+
"empty ('') for no file logging.")
# -l
parser.add_option("-l", "--print-log", action="store_true",
dest="print_log", default=False,
help="Print log messages to stderr.")
# --final-time
parser.add_option("--tf", action="store",
type="float",
dest="final_time",
default=None,
help="Total time for the simulation.")
# --timestep
parser.add_option("--timestep", action="store",
type="float",
dest="time_step",
default=None,
help="Timestep to use for the simulation.")
# --adaptive-timestep
parser.add_option("--adaptive-timestep", action="store_true",
dest="adaptive_timestep", default=None,
help="Use adaptive time stepping.")
parser.add_option("--no-adaptive-timestep", action="store_false",
dest="adaptive_timestep", default=None,
help="Do not use adaptive time stepping.")
# --cfl
parser.add_option("--cfl", action="store", dest="cfl", type='float',
default=0.3,
help="CFL number for adaptive time steps")
# -q/--quiet.
parser.add_option("-q", "--quiet", action="store_true",
dest="quiet", default=False,
help="Do not print any progress information.")
# --disable-output
parser.add_option("--disable-output", action="store_true",
dest="disable_output", default=False,
help="Do not dump any output files.")
# -o/ --fname
parser.add_option("-o", "--fname", action="store",
dest="output", default=self.fname,
help="File name to use for output")
# --pfreq.
parser.add_option("--pfreq", action="store",
dest="freq", default=None, type="int",
help="Printing frequency for the output")
# -d/ --detailed-output.
parser.add_option("-d", "--detailed-output", action="store_true",
dest="detailed_output", default=None,
help="Dump detailed output.")
# --output-remote
parser.add_option("--output-dump-remote", action="store_true",
dest="output_dump_remote", default=False,
help="Save Remote particles in parallel")
# --directory
parser.add_option("--directory", action="store",
dest="output_dir", default=self.fname+'_output',
help="Dump output in the specified directory.")
# --openmp
parser.add_option("--openmp", action="store_true", dest="with_openmp",
default=None, help="Use OpenMP to run the "\
"simulation using multiple cores.")
parser.add_option("--no-openmp", action="store_false", dest="with_openmp",
default=None, help="Do not use OpenMP to run the "\
"simulation using multiple cores.")
# Restart options
restart = OptionGroup(parser, "Restart options",
"Restart options for PySPH")
restart.add_option("--restart-file", action="store", dest="restart_file",
default=None,
help=("""Restart a PySPH simulation using a specified file """)),
restart.add_option("--rescale-dt", action="store", dest="rescale_dt",
default=1.0, type="float",
help=("Scale dt upon restarting by a numerical constant"))
parser.add_option_group( restart )
# NNPS options
nnps_options = OptionGroup(parser, "NNPS", "Nearest Neighbor searching")
# --nnps
nnps_options.add_option("--nnps", dest="nnps",
choices=['box', 'll'],
default='ll',
help="Use one of box-sort ('box') or "\
"the linked list algorithm ('ll'). "
)
# --fixed-h
nnps_options.add_option("--fixed-h", dest="fixed_h",
action="store_true", default=False,
help="Option for fixed smoothing lengths")
nnps_options.add_option("--cache-nnps", dest="cache_nnps",
action="store_true", default=False,
help="Option to enable the use of neighbor caching.")
nnps_options.add_option(
"--sort-gids", dest="sort_gids", action="store_true",
default=False, help="Sort neighbors by the GIDs to get "\
"consistent results in serial and parallel (slows down a bit)."
)
parser.add_option_group( nnps_options )
# Zoltan Options
zoltan = OptionGroup(parser, "PyZoltan",
"Zoltan load balancing options")
zoltan.add_option("--with-zoltan", action="store_true",
dest="with_zoltan", default=True,
help=("""Use PyZoltan for dynamic load balancing """))
zoltan.add_option("--zoltan-lb-method", action="store",
dest="zoltan_lb_method", default="RCB",
help=("""Choose the Zoltan load balancnig method"""))
# --rcb-lock
zoltan.add_option("--rcb-lock", action="store_true", dest="zoltan_rcb_lock_directions",
default=False,
help=("Lock the directions of the RCB cuts"))
# rcb--reuse
zoltan.add_option("--rcb-reuse", action='store_true', dest="zoltan_rcb_reuse",
default=False,
help=("Reuse previous RCB cuts"))
# rcb-rectilinear
zoltan.add_option("--rcb-rectilinear", action="store_true", dest='zoltan_rcb_rectilinear',
default=False,
help=("Produce nice rectilinear blocks without projections"))
# rcb-set-direction
zoltan.add_option("--rcb-set-direction", action='store', dest="zoltan_rcb_set_direction",
default=0, type="int",
help=("Set the order of the RCB cuts"))
zoltan.add_option("--zoltan-weights", action="store_false",
dest="zoltan_weights", default=True,
help=("""Switch between using weights for input to Zoltan.
defaults to True"""))
zoltan.add_option("--ghost-layers", action='store', dest='ghost_layers',
default=3.0, type='float',
help=('Number of ghost cells to share for remote neighbors'))
zoltan.add_option("--lb-freq", action='store', dest='lb_freq',
default=10, type='int',
help=('The frequency for load balancing'))
zoltan.add_option("--zoltan-debug-level", action="store",
dest="zoltan_debug_level", default="0",
help=("""Zoltan debugging level"""))
parser.add_option_group( zoltan )
# Options to control parallel execution
parallel_options=OptionGroup(parser, "Parallel Options")
# --update-cell-sizes
parallel_options.add_option("--update-cell-sizes", action='store_true',
dest='update_cell_sizes', default=False,
help=("Recompute cell sizes for binning in parallel"))
# --parallel-scale-factor
parallel_options.add_option("--parallel-scale-factor", action="store",
dest="parallel_scale_factor", default=2.0, type='float',
help=("""Kernel scale factor for the parallel update"""))
# --parallel-output-mode
parallel_options.add_option("--parallel-output-mode", action="store",
dest="parallel_output_mode", default=None,
help="""Use 'collected' to dump one output at
root or 'distributed' for every processor. """)
parser.add_option_group( parallel_options )
# solver interfaces
interfaces = OptionGroup(parser, "Interfaces",
"Add interfaces to the solver")
interfaces.add_option("--interactive", action="store_true",
dest="cmd_line", default=False,
help=("Add an interactive commandline interface "
"to the solver"))
interfaces.add_option("--xml-rpc", action="store",
dest="xml_rpc", metavar='[HOST:]PORT',
help=("Add an XML-RPC interface to the solver; "
"HOST=0.0.0.0 by default"))
interfaces.add_option("--multiproc", action="store",
dest="multiproc", metavar='[[AUTHKEY@]HOST:]PORT[+]',
default="pysph@0.0.0.0:8800+",
help=("Add a python multiprocessing interface "
"to the solver; "
"AUTHKEY=pysph, HOST=0.0.0.0, PORT=8800+ by"
" default (8800+ means first available port "
"number 8800 onwards)"))
interfaces.add_option("--no-multiproc", action="store_const",
dest="multiproc", const=None,
help=("Disable multiprocessing interface "
"to the solver"))
parser.add_option_group(interfaces)
# solver job resume support
parser.add_option('--resume', action='store', dest='resume',
metavar='COUNT|count|?',
help=('Resume solver from specified time (as stored '
'in the data in output directory); count chooses '
'a particular file; ? lists all '
'available files')
)
def _process_command_line(self):
""" Parse any command line arguments.
Add any new options before this is called. This also sets up
the logging automatically.
"""
try:
# If this is being run inside an IPython console or notebook
# then this is defined and we should not parse the command line
# arguments.
__IPYTHON__
except NameError:
(options, args) = self.opt_parse.parse_args(self.args)
else:
(options, args) = self.opt_parse.parse_args([])
self.options = options
# Setup logging based on command line options.
level = self._log_levels[options.loglevel]
#save the path where we want to dump output
self.path = abspath(options.output_dir)
mkdir(self.path)
if level is not None:
self._setup_logging(options.logfile, level,
options.print_log)
def _setup_logging(self, filename=None, loglevel=logging.WARNING,
stream=True):
""" Setup logging for the application.
Parameters
----------
filename : The filename to log messages to. If this is None
a filename is automatically chosen and if it is an
empty string, no file is used
loglevel : The logging level
stream : Boolean indicating if logging is also printed on
stderr
"""
# logging setup
logger.setLevel(loglevel)
# Setup the log file.
if filename is None:
filename = splitext(basename(sys.argv[0]))[0] + '.log'
if len(filename) > 0:
lfn = os.path.join(self.path,filename)
format = '%(levelname)s|%(asctime)s|%(name)s|%(message)s'
logging.basicConfig(level=loglevel, format=format,
filename=lfn, filemode='a')
if stream:
logger.addHandler(logging.StreamHandler())
def _create_inlet_outlet(self, inlet_outlet_factory):
"""Create the inlets and outlets if needed.
This method requires that the particles be already created.
The `inlet_outlet_factory` is passed a dictionary of the particle
arrays. The factory should return a list of inlets and outlets.
"""
if inlet_outlet_factory is not None:
solver = self._solver
particle_arrays = dict([(p.name, p) for p in self.particles])
self.inlet_outlet = inlet_outlet_factory(particle_arrays)
# Hook up the inlet/outlet's update method to be called after
# each stage.
for obj in self.inlet_outlet:
solver.add_post_step_callback(obj.update)
def _create_particles(self, particle_factory, *args, **kw):
""" Create particles given a callable `particle_factory` and any
arguments to it.
"""
solver = self._solver
options = self.options
rank = self.rank
# particle array info that is used to create dummy particles
# on non-root processors
particles_info = {}
# Only master actually calls the particle factory, the rest create
# dummy particle arrays.
if rank == 0:
if options.restart_file is not None:
data = load(options.restart_file)
arrays = data['arrays']
solver_data = data['solver_data']
# arrays and particles
particles = []
for array_name in arrays:
particles.append( arrays[array_name] )
# save the particles list
self.particles = particles
# time, timestep and solver iteration count at restart
t, dt, count = solver_data['t'], solver_data['dt'], solver_data['count']
# rescale dt at restart
dt *= options.rescale_dt
solver.t, solver.dt, solver.count = t, dt, count
else:
self.particles = particle_factory(*args, **kw)
# get the array info which will be b'casted to other procs
particles_info = utils.get_particles_info(self.particles)
# Broadcast the particles_info to other processors for parallel runs
if self.num_procs > 1:
particles_info = self.comm.bcast(particles_info, root=0)
# now all processors other than root create dummy particle arrays
if rank != 0:
self.particles = utils.create_dummy_particles(particles_info)
def _do_initial_load_balancing(self):
""" This will automatically distribute the particles among processors
if this is a parallel run.
"""
# Instantiate the Parallel Manager here and do an initial LB
num_procs = self.num_procs
options = self.options
solver = self._solver
comm = self.comm
self.pm = None
if num_procs > 1:
options = self.options
if options.with_zoltan:
if not (Has_Zoltan and Has_MPI):
raise RuntimeError("Cannot run in parallel!")
else:
raise ValueError("""Sorry. You're stuck with Zoltan for now
use the option '--with_zoltan' for parallel runs
""")
# create the parallel manager
obj_weight_dim = "0"
if options.zoltan_weights:
obj_weight_dim = "1"
zoltan_lb_method = options.zoltan_lb_method
zoltan_debug_level = options.zoltan_debug_level
zoltan_obj_wgt_dim = obj_weight_dim
# ghost layers
ghost_layers = options.ghost_layers
# radius scale for the parallel update
radius_scale = options.parallel_scale_factor*solver.kernel.radius_scale
self.pm = pm = ZoltanParallelManagerGeometric(
dim=solver.dim, particles=self.particles, comm=comm,
lb_method=zoltan_lb_method,
obj_weight_dim=obj_weight_dim,
ghost_layers=ghost_layers,
update_cell_sizes=options.update_cell_sizes,
radius_scale=radius_scale,
)
### ADDITIONAL LOAD BALANCING FUNCTIONS FOR ZOLTAN ###
# RCB lock directions
if options.zoltan_rcb_lock_directions:
pm.set_zoltan_rcb_lock_directions()
if options.zoltan_rcb_reuse:
pm.set_zoltan_rcb_reuse()
if options.zoltan_rcb_rectilinear:
pm.set_zoltan_rcb_rectilinear_blocks()
if options.zoltan_rcb_set_direction > 0:
pm.set_zoltan_rcb_directions( str(options.zoltan_rcb_set_direction) )
# set zoltan options
pm.pz.Zoltan_Set_Param("DEBUG_LEVEL", options.zoltan_debug_level)
pm.pz.Zoltan_Set_Param("DEBUG_MEMORY", "0")
# do an initial load balance
pm.update()
pm.initial_update = False
# set subsequent load balancing frequency
lb_freq = options.lb_freq
if lb_freq < 1 : raise ValueError("Invalid lb_freq %d"%lb_freq)
pm.set_lb_freq( lb_freq )
# wait till the initial partition is done
comm.barrier()
# set the solver's parallel manager
solver.set_parallel_manager(self.pm)
######################################################################
# Public interface.
######################################################################
def set_args(self, args):
self.args = args
[docs] def add_option(self, opt):
""" Add an Option/OptionGroup or their list to OptionParser """
if isinstance(opt, OptionGroup):
self.opt_parse.add_option_group(opt)
elif isinstance(opt, Option):
self.opt_parse.add_option(opt)
else:
# assume a list of Option/OptionGroup
for o in opt:
self.add_option(o)
[docs] def setup(self, solver, equations, nnps=None, inlet_outlet_factory=None,
particle_factory=None, *args, **kwargs):
"""Setup the application's solver.
This will parse the command line arguments (if this is not called from
within an IPython notebook or shell) and then using those parameters
and any additional parameters and call the solver's setup method.
Parameters
----------
solver: pysph.solver.solver.Solver
The solver instance.
equations: list
A list of Groups/Equations.
nnps: pysph.base.nnps.NNPS
Optional NNPS instance. If None is given a default NNPS is created.
inlet_outlet_factory: callable or None
The `inlet_outlet_factory` is passed a dictionary of the particle
arrays. The factory should return a list of inlets and outlets.
particle_factory : callable or None
If supplied, particles will be created for the solver using the
particle arrays returned by the callable. Else particles for the
solver need to be set before calling this method
args:
extra positional arguments passed on to the `particle_factory`.
kwargs:
extra keyword arguments passed to the `particle_factory`.
Examples
--------
>>> def create_particles():
... ...
...
>>> solver = Solver(...)
>>> equations = [...]
>>> app = Application()
>>> app.setup(solver=solver, equations=equations,
... particle_factory=create_particles)
>>> app.run()
"""
start_time = time.time()
self._solver = solver
solver_opts = solver.get_options(self.opt_parse)
if solver_opts is not None:
self.add_option(solver_opts)
self._process_command_line()
options = self.options
# Setup configuration options.
if options.with_openmp is not None:
get_config().use_openmp = options.with_openmp
# Create particles either from scratch or restart
self._create_particles(particle_factory, *args, **kwargs)
# This must be done before the initial load balancing
# as the inlets will create new particles.
self._create_inlet_outlet(inlet_outlet_factory)
self._do_initial_load_balancing()
# setup the solver using any options
self._solver.setup_solver(options.__dict__)
# fixed smoothing lengths
fixed_h = solver.fixed_h or options.fixed_h
if nnps is None:
kernel = self._solver.kernel
cache = options.cache_nnps
# create the NNPS object
if options.nnps == 'box':
nnps = BoxSortNNPS(
dim=solver.dim, particles=self.particles,
radius_scale=kernel.radius_scale, domain=self.domain,
cache=cache, sort_gids=options.sort_gids
)
elif options.nnps == 'll':
nnps = LinkedListNNPS(
dim=solver.dim, particles=self.particles,
radius_scale=kernel.radius_scale, domain=self.domain,
fixed_h=fixed_h, cache=cache,
sort_gids=options.sort_gids
)
# once the NNPS has been set-up, we set the default Solver
# post-stage callback to the DomainManager.setup_domain
# method. This method is responsible to computing the new cell
# size and doing any periodicity checks if needed.
solver.add_post_stage_callback( nnps.update_domain )
# inform NNPS if it's working in parallel
if self.num_procs > 1:
nnps.set_in_parallel(True)
# save the NNPS with the application
self.nnps = nnps
dt = options.time_step
if dt is not None:
solver.set_time_step(dt)
tf = options.final_time
if tf is not None:
solver.set_final_time(tf)
# Setup the solver output file name
fname = options.output
if Has_MPI:
rank = self.rank
if self.num_procs > 1:
fname += '_' + str(rank)
# set the rank for the solver
solver.rank = self.rank
solver.pid = self.rank
solver.comm = self.comm
# set the in parallel flag for the solver
if self.num_procs > 1:
solver.in_parallel = True
# output file name
solver.set_output_fname(fname)
# disable_output
solver.set_disable_output(options.disable_output)
# output print frequency
if options.freq is not None:
solver.set_print_freq(options.freq)
# output printing level (default is not detailed)
if options.detailed_output is not None:
solver.set_output_printing_level(options.detailed_output)
# solver output behaviour in parallel
if options.output_dump_remote:
solver.set_output_only_real( False )
# output directory
solver.set_output_directory(abspath(options.output_dir))
# set parallel output mode
if options.parallel_output_mode is not None:
solver.set_parallel_output_mode(options.parallel_output_mode)
# Set the adaptive timestep
if options.adaptive_timestep is not None:
solver.set_adaptive_timestep(options.adaptive_timestep)
# set solver cfl number
solver.set_cfl(options.cfl)
# setup the solver. This is where the code is compiled
solver.setup(particles=self.particles, equations=equations, nnps=nnps, fixed_h=fixed_h)
# add solver interfaces
self.command_manager = CommandManager(solver, self.comm)
solver.set_command_handler(self.command_manager.execute_commands)
if self.rank == 0:
# commandline interface
if options.cmd_line:
from pysph.solver.solver_interfaces import CommandlineInterface
self.command_manager.add_interface(CommandlineInterface().start)
# XML-RPC interface
if options.xml_rpc:
from pysph.solver.solver_interfaces import XMLRPCInterface
addr = options.xml_rpc
idx = addr.find(':')
host = "0.0.0.0" if idx == -1 else addr[:idx]
port = int(addr[idx+1:])
self.command_manager.add_interface(XMLRPCInterface((host,port)).start)
# python MultiProcessing interface
if options.multiproc:
from pysph.solver.solver_interfaces import MultiprocessingInterface
addr = options.multiproc
idx = addr.find('@')
authkey = "pysph" if idx == -1 else addr[:idx]
addr = addr[idx+1:]
idx = addr.find(':')
host = "0.0.0.0" if idx == -1 else addr[:idx]
port = addr[idx+1:]
if port[-1] == '+':
try_next_port = True
port = port[:-1]
else:
try_next_port = False
port = int(port)
interface = MultiprocessingInterface((host,port), authkey,
try_next_port)
self.command_manager.add_interface(interface.start)
logger.info('started multiprocessing interface on %s'%(
interface.address,))
end_time = time.time()
self._message("Setup took: %.5f secs"%(end_time - start_time))
[docs] def run(self):
"""Run the application.
"""
start_time = time.time()
self._solver.solve(not self.options.quiet)
end_time = time.time()
self._message("Run took: %.5f secs"%(end_time - start_time))
[docs] def dump_code(self, file):
"""Dump the generated code to given file.
"""
file.write(self._solver.sph_eval.ext_mod.code)
def _message(self, msg):
if self.options.quiet:
return
if self.num_procs == 1:
logger.info(msg)
print msg
elif (self.num_procs > 1 and self.rank in (0,1)):
s = "Rank %d: %s"%(self.rank, msg)
logger.info(s)
print s