#!/usr/bin/env python # -*- coding: utf-8 -*- #------------------------------------------------------------------------------- # This file is part of Code_Saturne, a general-purpose CFD tool. # # Copyright (C) 1998-2018 EDF S.A. # # This program is free software; you can redistribute it and/or modify it under # the terms of the GNU General Public License as published by the Free Software # Foundation; either version 2 of the License, or (at your option) any later # version. # # This program is distributed in the hope that it will be useful, but WITHOUT # ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS # FOR A PARTICULAR PURPOSE. See the GNU General Public License for more # details. # # You should have received a copy of the GNU General Public License along with # this program; if not, write to the Free Software Foundation, Inc., 51 Franklin # Street, Fifth Floor, Boston, MA 02110-1301, USA. #------------------------------------------------------------------------------- try: import ConfigParser # Python2 configparser = ConfigParser except Exception: import configparser # Python3 import datetime import fnmatch import os import subprocess import sys import platform import tempfile python_version = sys.version[:3] import cs_batch #=============================================================================== # Utility functions #=============================================================================== def abs_exec_path(path): """ Find an executable in the system path. """ abs_path = None if os.path.isabs(path): return path else: try: for d in os.getenv('PATH').split(): f = os.path.join(d, path) if os.path.isfile(f): return f except Exception: pass return None #--------------------------------------------------------------------------- def separate_args(s): """ Separate arguments that may contain whitespace, depending on whether whitespace is protected or not by ", ', and \ characters. If quotes are found after the beginning of a string, such as in --option="string 1", do not remove them. """ l = [] if s: a = '' sep = False protected = False in_quotes = '' for i in range(len(s)): if protected: a += s[i] protected = False else: if s[i] == '\\' and s[i+1:i+1].isalnum(): protected = True elif s[i] == '"' or s[i] == "'": if in_quotes == s[i]: a += s[i] in_quotes = '' elif in_quotes != '': a += s[i] else: a += s[i] in_quotes = s[i] elif in_quotes != '': a += s[i] elif (s[i] == ' ' or s[i] == '\t'): if a != '': if (a[0] == a[-1:]) and (a[0] == '"' or a[0] == "'"): l.append(a[1:-1]) else: l.append(a) a = '' else: a += s[i] if a != '': l.append(a) return l #--------------------------------------------------------------------------- def enquote_arg(s): """ Add quotes around argument if it contains whitespace, leave it unchanged otherwise; if the argument already contains unprotected quotes, do not add any more (so for example --option="string 1" is unchanged). """ if s: if (s.find(' ') > -1): protect = False for i in range(len(s)): if s[i] == '\\': protect = not protect if not protect and s[i] == '"': return s return '"' + s + '"' else: return s else: return s #------------------------------------------------------------------------------- def assemble_args(cmd): """ Assemble separate arguments. """ l = '' for s in cmd: if (s.find(' ') > -1): l += ' ' + enquote_arg(s) else: l += ' ' + s return l.strip() #------------------------------------------------------------------------------- # Update command line arguments for a given value #------------------------------------------------------------------------------- def update_command_single_value(args, options, value): """ Adds, updates, or removes parts of a command to pass a given value. The command is provided as a list, and options defining a value may be defined as a tuple (to allow for multiple variants). In all cases, this function assumes the option is followed by a single value argument. Options and values may be defined as separate (successive) arguments, or be separated by a '=' character when the matching option ends with '='. To allow both syntaxes, a given option may be passed both with and without that separator; for example: options = ('--param=', '--param', '-p'). If no option was previously present and a value is added, the first syntax of the options tuple will be used. """ i = -1 # Update first occurence if value: val_s = str(value) for opt in options: if opt[-1:] == '=': l = len(opt) for arg in args: if arg[:l] == opt: i = args.index(arg) break if i > -1: args[i] = opt + val_s else: if args.count(opt) > 0: i = args.index(opt) if i+1 < len(args): args[i+1] = val_s else: args.append(val_s) i = i+1 if i > -1: i = i + 1 break # Append if none found if i == -1: opt = options[0] if opt[-1:] == '=': args.append(opt + val_s) else: args.append(opt) args.append(val_s) i = len(args) # Remove excess occurences for opt in options: j = i if opt[-1:] == '=': l = len(opt) while j < len(args): if args[j][:l] == opt: args.pop(j) else: j = j+1 else: while j < len(args): if args[j] == opt: args.pop(j) # option if j < len(args): args.pop(j) # matching value else: j = j+1 # Special cases at end i = 0 args_tail = [] while i < len(args): if args[i][0] == '$' or args[i] == '&': a = args.pop(i) args_tail.append(a) else: i = i+1 args += args_tail # Return updated list return args #------------------------------------------------------------------------------- # Get a single value from a command line #------------------------------------------------------------------------------- def get_command_single_value(args, options, default=None): """ Obtain a value from a command line if availble, or using a default otherwise The command is provided as a list, and options defining a value may be defined as a tuple (to allow for multiple variants). In all cases, this function assumes the option is followed by a single value argument. Options and values may be defined as separate (successive) arguments, or be separated by a '=' character when the matching option ands with '='. To allow both syntaxes, a given option may be pased both with and without that separator; for example: options = ('--param=', '--param', '-p'). """ for opt in options: if opt[-1:] == '=': l = len(opt) for arg in args: if arg[:l] == opt: return arg[l:] else: if args.count(opt) > 0: i = args.index(opt) if i+1 < len(args): return args[i+1] return default #--------------------------------------------------------------------------- def get_shell_type(): """ Get name of current shell if available. (Bourne shell variants are handled, C-shell variants are not). """ if sys.platform.startswith('win'): return None user_shell = os.getenv('SHELL') if not user_shell: user_shell = '/bin/sh' elif user_shell[-3:] == 'csh': user_shell = '/bin/sh' return user_shell #------------------------------------------------------------------------------- def append_shell_shebang(l): """ Append lines for shell shebang or '@echo off' for a Windows COMMAND. """ if sys.platform.startswith('win'): l.append('@echo off') else: user_shell = get_shell_type() l.append('#!' + user_shell + '\n\n') l.append('') #------------------------------------------------------------------------------- def write_shell_shebang(fd): """ Write the shell shebang or '@echo off' for a Windows COMMAND. """ if sys.platform.startswith('win'): fd.write('@echo off\n\n') else: user_shell = get_shell_type() fd.write('#!' + user_shell + '\n\n') #------------------------------------------------------------------------------- def append_script_comment(l, comment): """ Add a comment in a script buffer list in the correct form. (starting with a '#' for Linux shells and 'rem' for Windows COMMAND). """ if sys.platform.startswith('win'): l.append('rem ' + comment) else: l.append('# ' + comment) #------------------------------------------------------------------------------- def write_script_comment(fd, comment): """ Write a comment in a script in the correct form. (starting with a '#' for Linux shells and 'rem' for Windows COMMAND). """ if sys.platform.startswith('win'): fd.write('rem ') else: fd.write('# ') fd.write(comment) #------------------------------------------------------------------------------- def write_export_env(fd, var, value): """ Write the correct command so as to export environment variables. """ if not value: value = '' if sys.platform.startswith('win'): export_cmd = 'set ' + var + '=' + value else: if get_shell_type()[-3:] == 'csh': # handle C-type shells export_cmd = 'setenv ' + var + ' ' + value else: # handle Bourne-type shells if value: export_cmd = 'export ' + var + '=' + value else: export_cmd = 'unset ' + var export_cmd = export_cmd + '\n' fd.write(export_cmd) #------------------------------------------------------------------------------- def prepend_path_command(var, user_path): """ Determine the correct command so as to export PATH-type variables. """ # Caution: Windows PATH must NOT must double-quoted to account for blanks # ======= in paths, contrary to UNIX systems if sys.platform.startswith('win'): export_cmd = 'set ' + var + '=' + user_path + ';%' + var + '%' else: if get_shell_type()[-3:] == 'csh': # handle C-type shells export_cmd = 'setenv ' + var + ' "' + user_path + '":$' + var else: # handle Bourne-type shells export_cmd = 'export ' + var + '="' + user_path + '":$' + var return export_cmd #------------------------------------------------------------------------------- def write_prepend_path(fd, var, user_path): """ Write the correct command so as to export PATH-type variables. """ export_cmd = prepend_path_command(var, user_path) + '\n' fd.write(export_cmd) #------------------------------------------------------------------------------- def clean_path(path): """ Remove duplicates from path. """ # Case for a string if type(path) == str: if not path: return '' if sys.platform.startswith('win'): s = ';' else: s = ':' vals = path.split(s) clean_path(vals) value = '' for v in vals: if value: value += s + v else: value = v return value # Case for a list elif type(path) == list: i = len(path) - 1 while i > -1: if path.count(i) > 1 or not path[i]: path.pop(i) i -= 1 return path #------------------------------------------------------------------------------- def get_script_positional_args(): """ Write the positional arguments with a newline character. """ if sys.platform.startswith('win'): args = '%*' else: args = '$@' return args #------------------------------------------------------------------------------- def get_script_return_code(): """ Write the return code with a newline character. """ if sys.platform.startswith('win'): ret_code = '%ERROR_LEVEL%' else: ret_code = '$?' return ret_code #------------------------------------------------------------------------------- def run_command(args, pkg = None, echo = False, stdout = sys.stdout, stderr = sys.stderr, env = None): """ Run a command. """ if echo == True: if type(args) == str: stdout.write(str(args) + '\n') else: l = '' for s in args: if (s.find(' ') > -1): l += ' ' + enquote_arg(s) else: l += ' ' + s stdout.write(l.strip() + '\n') # Modify the PATH for relocatable installation: add Code_Saturne "bindir" if pkg != None: if pkg.config.features['relocatable'] == "yes": if sys.platform.startswith("win"): sep = ";" else: sep = ":" saved_path = os.environ['PATH'] os.environ['PATH'] = pkg.get_dir('bindir') + sep + saved_path # As a workaround for a bug in which the standard output an error # are "lost" (observed in an apparently random manner, with Python 2.4), # we only add the stdout and stderr keywords if they are non-default. kwargs = {} if (stdout != sys.stdout): kwargs['stdout'] = stdout if (stderr != sys.stderr): kwargs['stderr'] = stderr returncode = 1 try: if type(args) == str: p = subprocess.Popen(args, shell=True, executable=get_shell_type(), universal_newlines=True, env = env, **kwargs) else: p = subprocess.Popen(args, universal_newlines=True, env = env, **kwargs) p.communicate() returncode = p.returncode except Exception as e: import traceback exc_type, exc_value, exc_traceback = sys.exc_info() traceback.print_exception(exc_type, exc_value, exc_traceback, limit=2, file=sys.stderr) print("") print("Failed calling subprocess with:") print(" args = " + str(args)) print(" env = " + str(env)) print(" kwargs = " + str(kwargs)) sys.exit(1) # Reset the PATH to its previous value if pkg != None: if pkg.config.features['relocatable'] == "yes": os.environ['PATH'] = saved_path return returncode #------------------------------------------------------------------------------- def get_command_output(cmd): """ Run a command and return it's standard output. """ p = subprocess.Popen(cmd, shell=True, executable=get_shell_type(), stdout=subprocess.PIPE, stderr=subprocess.PIPE, universal_newlines=True) output = p.communicate() if p.returncode != 0: sys.stderr.write(output[1] + '\n') return '' else: return output[0] #------------------------------------------------------------------------------- def get_command_outputs(cmd): """ Run a command and return it's standard and error outputs. """ p = subprocess.Popen(cmd, shell=True, executable=get_shell_type(), stdout=subprocess.PIPE, stderr=subprocess.STDOUT, universal_newlines=True) return p.communicate()[0] #------------------------------------------------------------------------------- def set_modules(pkg): """ Set environment modules if present. """ if pkg.config.env_modules == "no": return cmd_prefix = pkg.config.env_modulecmd cmds = ['purge'] for m in pkg.config.env_modules.strip().split(): cmds.append('load ' + m) for cmd in cmds: (output, error) = subprocess.Popen([cmd_prefix, 'python'] + cmd.split(), universal_newlines=True, stdout=subprocess.PIPE).communicate() exec(output) #------------------------------------------------------------------------------- def source_shell_script(path): """ Source shell script. """ if not os.path.isfile(path): sys.stderr.write('Warning:\n' + ' file ' + path + '\n' + 'not present, so cannot be sourced.\n\n') if sys.platform.startswith('win'): return user_shell = os.getenv('SHELL') if not user_shell: user_shell = '/bin/sh' cmd = ['source ' + path + ' && env'] p = subprocess.Popen(cmd, shell=True, executable=user_shell, stdout=subprocess.PIPE, stderr=subprocess.PIPE, universal_newlines=True) output = p.communicate()[0] for line in output.splitlines(): (key, _, value) = line.partition("=") # For paths, cleanup (remove multiple values) first if key[-4:] == 'PATH': value = clean_path(value) # Add key, value os.environ[key] = value # additional handling for Python path if key == 'PYTHONPATH': vals = value.split(':') vals.reverse() for v in vals: if v: sys.path.insert(0, v) sys.path = clean_path(sys.path) #------------------------------------------------------------------------------- def get_rcfile(pkg): """ Get path to file rcfile in preferences file if present. """ config = configparser.ConfigParser() config.read(pkg.get_configfiles()) if config.has_option('install', 'rcfile'): rcfile = config.get('install', 'rcfile') if not os.path.isabs(rcfile): rcfile = '~/.' + rcfile return rcfile return '' #------------------------------------------------------------------------------- def source_rcfile(pkg): """ Source user environement if defined by rcfile in preferences file. """ rcfile = get_rcfile(pkg) if rcfile: source_shell_script(rcfile) #------------------------------------------------------------------------------- def source_syrthes_env(pkg): """ Source SYRTHES environment """ # Determine SYRTHES home syrthes_home = None # Home based on configuration info cfg_syrthes_home = None config = configparser.ConfigParser() config.read(pkg.get_configfiles()) if config.has_option('install', 'syrthes'): cfg_syrthes_home = os.path.join(config.get('install', 'syrthes')) if not os.path.isdir(cfg_syrthes_home): sys.stderr.write("\nIncorrect install/syrthes entry " + "specified in one of:\n") for f in pkg.get_configfiles(): sys.stderr.write(" " + f + "\n") sys.stderr.write("Directory: '" + cfg_syrthes_home + "' does not exist.\n") sys.exit(1) cfg_syrthes_home = None else: syrthes_home = cfg_syrthes_home # Home based on current environment # Try to base Syrthes version on path used to import the syrthes module, # for consistency with case creation parameters; it this is not enough, # use existing environment or load one based on current configuration. try: for p in sys.path: if p[-14:] == '/share/syrthes' or p[-14:] == '\share\syrthes': syr_profile = os.path.join(p[:,-14], 'bin', 'syrthes.profile') if os.path.isfile(syr_profile): source_shell_script(syr_profile) except Exception: pass env_syrthes_home = os.getenv('SYRTHES4_HOME') if not syrthes_home: syrthes_home = env_syrthes_home if not syrthes_home: sys.stderr.write("\nCannot locate SYRTHES installation.\n" + " Either the install/syrthes entry must be " + "specified in one of:\n") for f in pkg.get_configfiles(): sys.stderr.write(" " + f + "\n") sys.stderr.write(" or a syrthes.profile file must be sourced.\n") sys.exit(1) # Now source environment if not done already or different if syrthes_home != env_syrthes_home: syr_profile = os.path.join(config.get('install', 'syrthes'), 'bin', 'syrthes.profile') source_shell_script(syr_profile) # Finally, ensure module can be imported syr_datapath = os.path.join(syrthes_home, os.path.join('share', 'syrthes')) if sys.path.count(syr_datapath) > 0: sys.path.remove(syr_datapath) sys.path.insert(0, syr_datapath) #------------------------------------------------------------------------------- def get_parent_process_path(): """ Retrieve parent script path, when possible """ path = None if sys.platform.startswith('win'): try: f = open("/proc/" + str(os.getppid()) + "/cmdline") l = f.readlines() l = f.read() path = l.split('\x00')[1] except Exception: pass return path #------------------------------------------------------------------------------- class batch_info: #--------------------------------------------------------------------------- def __init__(self): """ Get batch system information. """ self.batch_type = None self.submit_dir = None self.job_file = None self.job_name = None self.job_id = None self.queue = None # Check for specific batch environments s = os.getenv('LSB_JOBID') # LSF if s != None: self.batch_type = 'LSF' self.submit_dir = os.getenv('LS_SUBCWDIR') self.job_file = os.getenv('LSB_JOBFILENAME') self.job_name = os.getenv('LSB_JOBNAME') self.job_id = os.getenv('LSB_BATCH_JID') self.queue = os.getenv('LSB_QUEUE') if self.batch_type == None: s = os.getenv('PBS_JOBID') # PBS if s != None: self.batch_type = 'PBS' self.submit_dir = os.getenv('PBS_O_WORKDIR') self.job_name = os.getenv('PBS_JOBNAME') self.job_id = os.getenv('PBS_JOBID') self.queue = os.getenv('PBS_QUEUE') if self.batch_type == None: s = os.getenv('OAR_JOBID') # OAR if s != None: self.batch_type = 'OAR' self.submit_dir = os.getenv('OAR_WORKING_DIRECTORY') self.job_name = os.getenv('OAR_JOBNAME') self.job_id = os.getenv('OAR_JOBID') if self.batch_type == None: s = os.getenv('LOADL_JOB_NAME') # LoadLeveler if s != None: self.batch_type = 'LOADL' self.job_file = os.getenv('LOADL_STEP_COMMAND') self.submit_dir = os.getenv('LOADL_STEP_INITDIR') self.job_name = os.getenv('LOADL_JOB_NAME') self.job_id = os.getenv('LOADL_STEP_ID') self.queue = os.getenv('LOADL_STEP_CLASS') if self.batch_type == None: s = os.getenv('SGE_TASK_ID') # Sun Grid Engine if s != None: self.batch_type = 'SGE' self.submit_dir = os.getenv('SGE_O_WORKDIR') self.job_name = os.getenv('JOB_NAME') self.job_id = os.getenv('JOB_ID') self.queue = os.getenv('QUEUE') if self.batch_type == None: s = os.getenv('SLURM_JOBID') # SLURM if s != None: self.batch_type = 'SLURM' self.submit_dir = os.getenv('SLURM_SUBMIT_DIR') self.job_name = os.getenv('SLURM_JOB_NAME') self.job_id = os.getenv('SLURM_JOBID') self.queue = os.getenv('SLURM_PARTITION') #--------------------------------------------------------------------------- def get_remaining_time(self): """ Get remaining time if available from batch system. """ rtime = None if self.batch_type == 'PBS': cmd = "qstat -r $PBS_JOBID | grep $PBS_JOBID" \ + " | sed -e's/ \{1,\}/ /g' | cut -d ' ' -f 9" rtime = get_command_output(cmd) elif self.batch_type == 'SLURM': cmd = "squeue -h -j $SLURM_JOBID -o %L" # In case of job array, check on first line rs = get_command_output(cmd).splitlines()[0] rtime = cs_batch.parse_wall_time_slurm(rs) return rtime #------------------------------------------------------------------------------- class resource_info(batch_info): #--------------------------------------------------------------------------- def __init__(self, n_procs = None, n_procs_default = None, n_threads = None): """ Get execution resources information. """ batch_info.__init__(self) self.manager = None self.n_procs = None self.n_nodes = None self.n_threads = None self.n_procs_ri = None self.n_threads_ri = None # Pre-set the number of threads if OMP_NUM_THREADS given if n_threads == None: s = os.getenv('OMP_NUM_THREADS') if s != None: n_threads = int(s) # If obtained from an environment variable, express # the hosts file using a shell variable rather than # an absolute name (for use in generated scripts). self.hosts_file = None self.hosts_list = None # Check for resource manager # Test for SLURM (Simple Linux Utility for Resource Management). s = os.getenv('SLURM_NPROCS') if s != None: self.manager = 'SLURM' self.n_procs = int(s) s = os.getenv('SLURM_NNODES') if s != None: self.n_nodes = int(s) else: s = os.getenv('SLURM_NNODES') if s != None: self.manager = 'SLURM' self.n_nodes = int(s) s = os.getenv('SLURM_TASKS_PER_NODE') if s != None: # Syntax may be similar to SLURM_TASKS_PER_NODE=2(x3),1" # indicating three nodes will each execute 2 tasks and # the fourth node will execute 1 task. self.n_procs = 0 for s0 in s.split(','): i = s0.find('(') if i > -1: self.n_procs += int(s0[0:i])*int(s0[i+2:-1]) else: self.n_procs += int(s0) else: self.n_procs = self.n_nodes if self.manager == 'SLURM': s = os.getenv('SLURM_CPUS_PER_TASK') if s != None: self.nthreads = int(s) # Test for Platform LSF or OpenLava. if self.manager == None and self.batch_type == 'LSF': self.manager = 'LSF' self.n_procs = 0 self.n_nodes = 0 s = os.getenv('LSB_MCPU_HOSTS') if s != None: mcpu_list = s.split(' ') self.n_nodes = len(mcpu_list)/2 for i in range(self.n_nodes): self.n_procs += int(mcpu_list[i*2 + 1]) else: s = os.getenv('LSB_HOSTS') if s != None: hl = s.split(' ') self.n_procs_from_hosts_list(hl, True) # Test for IBM LoadLeveler. if self.manager == None and self.batch_type == 'LOADL': s = os.getenv('LOADL_TOTAL_TASKS') if s != None: self.manager = 'LOADL' self.n_procs = int(s) s = os.getenv('LOADL_BG_SIZE') if s != None: self.manager = 'LOADL' self.n_nodes = int(s) if not self.n_procs: s = os.getenv('LOADL_PROCESSOR_LIST') if s != None: self.manager = 'LOADL' hl = s.strip().split(' ') self.n_procs_from_hosts_list(hl, True) if self.n_nodes and not self.n_procs: if n_procs: self.n_procs = n_procs elif os.getenv('LOADL_BG_SIZE'): # Blue Gene/Q self.n_procs = self.n_nodes*16 if n_threads: if n_threads > 4: self.n_procs = self.n_nodes*16*4/n_threads s = os.getenv('LOADL_HOSTFILE') if s != None: self.manager = 'LOADL' self.hosts_file = '$LOADL_HOSTFILE' # Test for TORQUE or PBS Pro. if self.manager == None and self.batch_type == 'PBS': s = os.getenv('PBS_NODEFILE') if s != None: self.manager = 'PBS' self.hosts_file = '$PBS_NODEFILE' s = os.getenv('PBS_NUM_NODES') if s != None: self.n_nodes = int(s) s = os.getenv('PBS_NUM_PPN') if s != None: self.n_procs = int(s)*self.n_nodes # Test for OAR if self.manager == None and self.batch_type == 'OAR': s = os.getenv('OAR_NODEFILE') if s != None: self.manager = 'OAR' self.hosts_file = '$OAR_NODEFILE' # Test for Sun (/Oracle/Whowever keeps this zombie going) Grid Engine if self.manager == None and self.batch_type == 'SGE': s = os.getenv('NSLOTS') if s != None: self.n_procs = int(s) s = os.getenv('NHOSTS') if s != None: self.n_nodes = int(s) s = os.getenv('TMPDIR') if s != None: s += '/machines' if os.path.isfile(s): self.manager = 'SGE' self.hosts_file = '$TMPDIR/machines' else: s = os.getenv('PE_HOSTFILE') if s != None: self.manager = 'SGE' self.hosts_file = '$PE_HOSTFILE' # Check hosts file presence if self.hosts_file == '$TMPDIR/machines': if not os.path.isfile(os.getenv('TMPDIR') + '/machines'): self.hosts_file = None elif self.hosts_file != None: if self.hosts_file[0] == '$': if not os.path.isfile(os.getenv(self.hosts_file[1:])): self.hosts_file = None elif not os.path.isfile(self.hosts_file): self.hosts_file = None # Determine number of processors from hosts file or list if self.n_procs == None: if self.hosts_file != None: self.n_procs_from_hosts_file(self.hosts_file) elif self.hosts_list != None: self.n_procs_from_hosts_list(self.hosts_list) # keep track of automatically determined values self.n_procs_ri = self.n_procs self.n_threads_ri = self.n_threads # Check and possibly set number of processes if n_procs != None: if self.n_procs != None: if self.n_procs != n_procs: sys.stderr.write('Warning:\n' +' Will try to use ' + str(n_procs) + ' processes while resource manager (' + self.manager + ')\n allows for ' + str(self.n_procs) + '.\n\n') self.n_procs = n_procs if self.n_procs == None: self.n_procs = n_procs_default # Check and possibly set number of threads if n_threads != None: if self.n_threads != None and n_threads != self.n_threads: sys.stderr.write('Warning:\n' +' Will try to use ' + str(self.n_threads) + ' threads per task while resource manager (' + self.manager + ')\n allows for ' + str(n_threads) + '.\n\n') self.n_threads = n_threads #--------------------------------------------------------------------------- def n_procs_per_node(self): """ Determine number of processors per node. """ ppn = None if self.n_procs != None and self.n_nodes != None: ppn = self.n_procs / self.n_nodes return ppn #--------------------------------------------------------------------------- def n_procs_from_hosts_file(self, hosts_file): """ Compute number of hosts from a hostsfile. """ self.n_procs = 0 if hosts_file == '$TMPDIR/machines': path = os.getenv('TMPDIR') + '/machines' elif hosts_file[0] == '$': path = os.getenv(hosts_file[1:]) else: path = hosts_file f = open(path, 'r') for line in f: self.n_procs += 1 f.close() #--------------------------------------------------------------------------- def n_procs_from_hosts_list(self, hosts_list, is_copy=False): """ Determine number of processors and nodes from hosts list. """ self.n_procs = len(hosts_list) self.n_nodes = 1 # If the hosts list is not already a copy, build one so # that sorting will not alter the original list. if is_copy == False: hl = [] for s in hosts_list: hl.append(s) else: hl = hosts_list hl.sort() for i in range(self.n_procs - 1): if hl[i] != hl[i+1]: self.n_nodes += 1 #--------------------------------------------------------------------------- def get_hosts_list(self): """ Get execution resources information. """ hosts_list = None # Hosts list may already have been defined by constructor if self.hosts_list != None: hosts_list = self.hosts_list # Check for resource manager and eventual hostsfile elif self.manager == 'SLURM': hosts_list = get_command_output('srun hostname -s').split() hosts_list.sort() elif self.manager == 'LSF': s = os.getenv('LSB_MCPU_HOSTS') if s != None: mcpu_list = s.split(' ') hosts_list = [] for i in range(len(mcpu_list)/2): host = mcpu_list[i*2] count = int(mcpu_list[i*2 + 1]) for j in range(count): hosts_list.append(host) else: s = os.getenv('LSB_HOSTS') if s != None: hosts_list = s.split(' ') elif self.manager == 'LOADL': hosts_list = [] s = os.getenv('LOADL_PROCESSOR_LIST') if s != None: hosts_list = s.split(' ') return hosts_list #--------------------------------------------------------------------------- def get_hosts_file(self, wdir = None): """ Returns the name of the hostsfile associated with the resource manager. A hostsfile is built from a hosts list if necessary. """ hosts_file = self.hosts_file if self.hosts_file == None: hosts_list = self.get_hosts_list() if hosts_list != None: if wdir != None: hosts_file = os.path.join(wdir, 'hostsfile') else: hosts_file = 'hostsfile' f = open(hosts_file, 'w') # If number of procs not specified, determine it by list if self.n_procs == None or self.n_procs < 1: n_procs = 0 for host in hosts_list: f.write(host + '\n') n_procs += 1 # If the number of procs is known, use only beginning of list # if it contains more entries than the required number, or # loop through list as many time as necessary to reach # prescribed proc count else: proc_count = 0 while proc_count < self.n_procs: for host in hosts_list: if proc_count < self.n_procs: f.write(host + '\n') proc_count += 1 f.close() self.hosts_file = hosts_file return hosts_file #------------------------------------------------------------------------------- # MPI environments and associated commands #------------------------------------------------------------------------------- MPI_MPMD_none = 0 MPI_MPMD_mpiexec = (1<<0) # mpiexec colon-separated syntax MPI_MPMD_configfile = (1<<1) # mpiexec -configfile syntax MPI_MPMD_script = (1<<2) class mpi_environment: def __init__(self, pkg, resource_info=None, wdir = None): """ Returns MPI environment info. """ # Note that self.mpiexec_n will usually be ' -n ' if present; # blanks are used to separate from the surrounding arguments, # but in the case of srun, which uses a -n instead # of -n syntax, setting it to ' -n' will be enough. self.type = pkg.config.libs['mpi'].variant self.bindir = pkg.config.libs['mpi'].bindir self.gen_hostsfile = None self.del_hostsfile = None self.mpiboot = None self.mpihalt = None self.mpiexec = None self.mpiexec_opts = None self.mpiexec_n = None self.mpiexec_n_per_node = None self.mpiexec_separator = None self.mpiexec_exe = None self.mpiexec_args = None self.mpmd = MPI_MPMD_none self.info_cmds = None # Initialize options based on system-wide or user configuration config = configparser.ConfigParser() config.read(pkg.get_configfiles()) if config.has_section('mpi'): for option in config.items('mpi'): k = option[0] v = option[1] if not v: v = None elif v[0] in ['"', "'"]: v = v[1:-1] if k == 'mpmd': self.mpmd = eval('MPI_MPMD_' + v) else: self.__dict__[k] = v # We may have a specific syntax for tasks per node if # a launcher from a resource manager is used: if resource_info and self.mpiexec: if os.path.basename(self.mpiexec)[:4] == 'srun': if resource_info.n_procs_ri != None \ and resource_info.n_procs_ri != resource_info.n_procs: self.mpiexec_n = ' --ntasks=' elif not self.mpiexec_n_per_node: self.mpiexec_n_per_node = ' --ntasks-per-node=' # Initialize based on known MPI types, or default. init_method = self.__init_other__ if len(self.type) > 0: mpi_env_by_type = {'MPICH':self.__init_mpich2_3__, 'MPICH2':self.__init_mpich2_3__, 'Intel_MPI':self.__init_mpich2_3__, 'MSMPI':self.__init_msmpi__, 'OpenMPI':self.__init_openmpi__, 'BullxMPI':self.__init_openmpi__, 'BGQ_MPI':self.__init_bgq__, 'Platform_MPI':self.__init_platform_mpi__} if self.type in mpi_env_by_type: init_method = mpi_env_by_type[self.type] p = os.getenv('PATH').split(':') if len(self.bindir) > 0: p = [self.bindir] init_method(p, resource_info, wdir) # Overwrite options based on system-wide or user configuration if config.has_section('mpi'): for option in config.items('mpi'): k = option[0] v = option[1] if not v: v = None elif v[0] in ['"', "'"]: v = v[1:-1] if k == 'mpmd': self.mpmd = eval('MPI_MPMD_' + v) else: self.__dict__[k] = v # Now adjust mpiexec_n_per_node base on available info: # leave value alone if digit (ppn value) already present, # complete or remove string otherwise. if self.mpiexec_n_per_node: if not self.mpiexec_n_per_node[-1:].isdigit(): ppn = resource_info.n_procs_per_node() if ppn: self.mpiexec_n_per_node += str(ppn) else: self.mpiexec_n_per_node = None #--------------------------------------------------------------------------- def __get_mpiexec_absname__(self, p): """ Build absolute pathname matching mpiexec command. """ absname = '' if self.mpiexec != None: if os.path.isabs(self.mpiexec): absname = self.mpiexec else: for d in p: absname = os.path.join(d, self.mpiexec) if os.path.isfile(absname): break else: absname = '' return absname #--------------------------------------------------------------------------- def __get_mpich2_3_default_pm__(self, mpiexec_path): """ Try to determine the program manager for MPICH2 or MPICH-3. """ # Only smpd is supported on Windows (or was in MPICH2). if sys.platform.startswith('win'): return 'smpd' # Check if we have a suffix which is self-explanatory i = mpiexec_path.rfind('.') if i > -1: suffix = mpiexec_path[i+1:] if suffix in ['hydra', 'gforker', 'remshell', 'smpd', 'mpd']: return suffix # If command is an external wrapper, use its base name cmd = os.path.basename(mpiexec_path) if cmd not in ['mpiexec', 'mpirun']: return cmd # Use mpichversion/mpich2version preferentially infoname = os.path.join(os.path.split(mpiexec_path)[0], 'mpichversion') if not os.path.isfile(infoname): infoname = os.path.join(os.path.split(mpiexec_path)[0], 'mpich2version') if os.path.isfile(infoname): # If program managers were specified at configure time, # the first is the default info = get_command_outputs(infoname + ' -configure') i = info.find('-with-pm=') if i > -1: s = info[i + len('-with-pm='):] for pm in ['hydra', 'gforker', 'remshell']: if s[0:len(pm)] == pm: return pm # Otherwise, we know the default changed with MPICH2 1.3, # from MPD to Hydra. info = get_command_outputs(infoname + ' -version') v = info.rstrip().split('\t')[1].split('.') if int(v[0]) == 1 and int(v[1]) < 3: return 'mpd' else: return 'hydra' # If MPICH2 / MPICH-3 info is not available, try # to determine this in another way if os.path.islink(mpiexec_path): if os.path.basename(os.path.realpath(mpiexec_path)) == 'mpiexec.py': return 'mpd' info = get_command_outputs(mpiexec_path + ' -help') if info.find('Hydra') > -1: return 'hydra' elif info.find(' smpd ') > -1: return 'smpd' elif info.find('-usize') > -1: return 'gforker' # might also be remshell sys.stderr.write('Warning:\n' + ' Unable to determine MPICH program manager:' + ' assume "Hydra".\n\n') return 'hydra' #--------------------------------------------------------------------------- def __init_mpich2_3__(self, p, resource_info=None, wdir = None): """ Initialize for MPICH-3 environment. MPICH2, MPICH-3, and their derivatives allow for different process managers, all or some of which may be built depending on installation options: - HYDRA is the default on Unix-type systems. It natively uses existing daemons on the system such as ssh, SLURM, PBS, etc. - gforker is a simple manager that creates all processes on a single machine (the equivalent seem possible with HYDRA using "mpiexec -bootstrap fork") - remshell is a very simple version of mpiexec which makes use of the ssh command to start processes on a collection of machines. It ignores the command line options which control the environment variables given to MPI programs. A hostsfile by name of machines should contain the list of machines on which to run, one machine name per line (machines may be listed multiple times if necessary). For completeness, we may mention that MPICH2 also included MPD, which was deprecated in MPICH2-1.3 (in October 2010), and SMPD, which could be used both on Windows and Linux. They are not supported anymore, though they can still be handled through post-install settings. """ # Determine base executable paths # Executables suffixes 'mpich' or 'mpich2' may occur in case # of Linux distribution packaging, while 'hydra', 'smpd', # 'gforker', and 'remshell' are defined by the standard MPICH # install and determine the associated launcher. pm = '' absname = '' if self.mpiexec != None: absname = self.__get_mpiexec_absname__(p) pm = self.__get_mpich2_3_default_pm__(absname) else: launcher_names = ['mpiexec.mpich', 'mpiexec.mpich2', 'mpiexec', 'mpiexec.hydra', 'mpiexec.gforker', 'mpiexec.remshell', 'mpirun.mpich2', 'mpirun.mpich', 'aprun', 'mpirun'] for d in p: for name in launcher_names: absname = os.path.join(d, name) if os.path.isfile(absname): pm = self.__get_mpich2_3_default_pm__(absname) # MPD and SMPD are deprecated; avoid them if pm == 'mpd' or pm == 'smpd': continue # Set launcher name if d == self.bindir: self.mpiexec = absname else: self.mpiexec = name break else: absname = '' if self.mpiexec != None: break if self.mpiexec != None: break if (self.mpiexec == None): self.mpiexec = 'mpiexec' basename = os.path.basename(self.mpiexec) # Determine processor count and MPMD handling launcher_base = os.path.basename(self.mpiexec) if not self.mpiexec_n: if launcher_base[:7] == 'mpiexec': self.mpiexec_n = ' -n ' elif launcher_base[:6] == 'mpirun': self.mpiexec_n = ' -np ' if not self.mpmd: if launcher_base[:7] == 'mpiexec': self.mpmd = MPI_MPMD_mpiexec | MPI_MPMD_configfile | MPI_MPMD_script elif launcher_base[:6] == 'mpirun': self.mpmd = MPI_MPMD_script # Other options to add # Resource manager info rm = None if resource_info != None: rm = resource_info.manager if pm == 'hydra': # Nothing to do for resource managers directly handled by Hydra if rm not in ['PBS', 'LOADL', 'LSF', 'SGE', 'SLURM']: hostsfile = resource_info.get_hosts_file(wdir) if hostsfile != None: self.mpiexec += ' -f ' + hostsfile if resource_info != None and not self.mpiexec_n_per_node: ppn = resource_info.n_procs_per_node() if ppn: self.mpiexec_n_per_node = ' -ppn ' + str(ppn) elif pm == 'remshell': hostsfile = resource_info.get_hosts_file(wdir) if hostsfile != None: self.mpiboot += ' --file=' + hostsfile elif pm == 'gforker': hosts = False hostslist = resource_info.get_hosts_list() if hostslist != None: hosts = True else: hostsfile = resource_info.get_hosts_file(wdir) if hostsfile != None: hosts = True if hosts == True: sys.stderr.write('Warning:\n' + ' Hosts list will be ignored by' + ' MPICH gforker program manager.\n\n') # Info commands if self.type == 'MPICH': self.info_cmds = ['mpichversion'] if self.type == 'MPICH2': self.info_cmds = ['mpich2version'] #--------------------------------------------------------------------------- def __init_openmpi__(self, p, resource_info=None, wdir = None): """ Initialize for OpenMPI environment. """ # Determine base executable paths if self.mpiexec != None: absname = self.__get_mpiexec_absname__(p) else: launcher_names = ['mpiexec.openmpi', 'mpirun.openmpi', 'mpiexec', 'mpirun'] for d in p: for name in launcher_names: absname = os.path.join(d, name) if os.path.isfile(absname): if d == self.bindir: self.mpiexec = absname else: self.mpiexec = name break else: absname = '' if self.mpiexec != None: break if self.mpiexec != None: break if (self.mpiexec == None): self.mpiexec = 'mpiexec' if absname: info_name = os.path.join(os.path.dirname(absname), 'ompi_info') else: info_name = 'ompi_info' # Determine processor count and MPMD handling launcher_base = os.path.basename(self.mpiexec) if not self.mpiexec_n: self.mpiexec_n = ' -n ' if resource_info != None and not self.mpiexec_n_per_node: ppn = resource_info.n_procs_per_node() if ppn: self.mpiexec_n_per_node = ' --npernode ' + str(ppn) if not self.mpmd: if launcher_base[:7] == 'mpiexec': self.mpmd = MPI_MPMD_mpiexec | MPI_MPMD_script elif launcher_base[:6] == 'mpirun': self.mpmd = MPI_MPMD_script # Other options to add # Detect if resource manager is known by this Open MPI build if resource_info != None: known_manager = False rc_mca_by_type = {'SLURM':' slurm ', 'LSF':' lsf ', 'LOADL':' loadleveler ', 'PBS':' tm ', 'SGE':' gridengine '} if resource_info.manager in rc_mca_by_type: info = get_command_output(info_name) if info.find(rc_mca_by_type[resource_info.manager]) > -1: known_manager = True elif resource_info.manager == 'OAR': self.mpiexec += ' -machinefile $OAR_FILE_NODES' self.mpiexec += ' -mca pls_rsh_agent "oarsh"' known_manager = True if known_manager == False: hostsfile = resource_info.get_hosts_file(wdir) if hostsfile != None: self.mpiexec += ' --machinefile ' + hostsfile # Info commands self.info_cmds = ['ompi_info -a'] #--------------------------------------------------------------------------- def __init_bgq__(self, p, resource_info=None, wdir = None): """ Initialize for Blue Gene/Q environment. """ # Set base executable path self.mpiexec = 'runjob' # Determine processor count and MPMD handling self.mpiexec_n = ' --np ' self.mpmd = MPI_MPMD_configfile rm = None ppn = 1 if resource_info != None: rm = resource_info.manager ppn = resource_info.n_procs_per_node() if rm == 'SLURM': self.mpiexec = 'srun' self.mpiexec_n = ' --ntasks=' if ppn: self.mpiexec_n_per_node = ' --ntasks-per-node=' + str(ppn) else: if ppn != 1: self.mpiexec_n_per_node = ' --ranks-per-node ' + str(ppn) self.mpiexec_separator = ':' # Other options to add # self.mpiexec_exe = '--exe' # self.mpiexec_args = '--args' # self.mpiexec_envs = '--envs OMP_NUM_THREADS=' + str(omp_num_threads) # Info commands #--------------------------------------------------------------------------- def __init_platform_mpi__(self, p, resource_info=None, wdir = None): """ Initialize for Platform MPI environment. The last version of HP MPI is version 2.3, released early 2009. HP MPI was then acquired by Platform MPI (formerly Scali MPI), which merged Scali MPI and HP MPI in Platform MPI 8.0. """ # Determine base executable paths self.mpiexec = 'mpirun' # mpiexec also possible, but fewer options for d in p: if not os.path.isabs(self.mpiexec): absname = os.path.join(d, self.mpiexec) if os.path.isfile(absname): if d == self.bindir: self.mpiexec = absname break # Determine processor count and MPMD handling # Appfile also possible, but uses -np instead of -n self.mpmd = MPI_MPMD_script self.mpiexec_n = '-np' # Determine options to add # If resource manager is used, add options if resource_info != None: if resource_info.manager == 'SLURM': self.mpiexec += ' -srun' self.mpiexec_n = None elif resource_info.manager == 'LSF': self.mpiexec += ' -lsb_hosts' # Info commands #--------------------------------------------------------------------------- def __init_msmpi__(self, p, resource_info=None, wdir = None): """ Initialize for MS-MPI environment. Microsoft MPI is based on standard MPICH2 distribution. It allows only for the smpd process manager that consists of independent daemons, so if a hostsfile is used, it must be passed to mpiexec. """ # On Windows, mpiexec.exe will be found through the PATH variable # so, unset the MPI 'bindir' variable self.bindir = '' # Determine MS-MPI configuration self.mpiexec = 'mpiexec.exe' self.mpmd = MPI_MPMD_mpiexec | MPI_MPMD_configfile self.mpiexec_n = ' -n ' # Other options to add # Resource manager info # Info commands #--------------------------------------------------------------------------- def __init_other__(self, p, resource_info=None, wdir = None): """ Initialize MPI environment info for environments not handled in one the the previous cases. """ # If possible, select launcher based on resource manager or # specific systems (would be better done in derived classes, # but these are cases on systems we have used in the past # but do not currently have access to). if platform.uname()[0] == 'AIX': if abs_exec_path('poe') != None: self.mpiexec = 'poe' self.mpiexec_n = None self.mpmd = MPI_MPMD_script #--------------------------------------------------------------------------- def info(self): """ Outputs MPI environment information in known cases. """ output = '' if self.info_cmds != None: for cmd in self.info_cmds: output += get_command_output(cmd) + '\n' return output #--------------------------------------------------------------------------- def unset_mpmd_mode(self, mpi_mpmd_mode): """ Unset mask allowing a given mpmd mode. """ if self.mpmd & mpi_mpmd_mode: self.mpmd = self.mpmd ^ mpi_mpmd_mode #------------------------------------------------------------------------------- # Execution environment (including MPI, OpenMP, ...) #------------------------------------------------------------------------------- class exec_environment: #--------------------------------------------------------------------------- def __init__(self, pkg, wdir=None, n_procs=None, n_procs_default=None, n_threads=None): """ Returns Execution environment. """ if sys.platform.startswith('win'): self.user = os.getenv('USERNAME') else: self.user = os.getenv('LOGNAME') if not self.user: self.user = os.getenv('USER') if not self.user: try: from pwd import getpwuid getpwuid(os.getuid())[0] except Exception: pass if not self.user: self.user = 'unknown' self.wdir = wdir if self.wdir == None: self.wdir = os.getcwd() self.resources = resource_info(n_procs, n_procs_default, n_threads) self.mpi_env = None # Associate default launcher and associated options based on # known MPI types, use default otherwise self.mpi_env = mpi_environment(pkg, self.resources, wdir) #------------------------------------------------------------------------------- if __name__ == '__main__': import cs_package pkg = cs_package.package() e = exec_environment(pkg) import pprint pprint.pprint(e.__dict__) pprint.pprint(e.resources.__dict__) pprint.pprint(e.mpi_env.__dict__) print(e.mpi_env.info()) #------------------------------------------------------------------------------- # End #-------------------------------------------------------------------------------