def workflow(self, comb):
""" """
comb_ok = False
logger.info(slugify(comb) + \
' starts to compile')
try:
export = "source /opt/intel/bin/compilervars.sh intel64; "
src_directory = "/home/arrouan/workspace/aevol/git/world/aevol/"
bin_directory = "/home/arrouan/workspace/aevol/compiled_binary/"
configure_option = "--with-tracing --without-x"
if comb['parallel'] == 'tbb':
configure_option += " --with-tbb"
if comb['blas'] == 'openblas':
configure_option += " --with-blas"
elif comb['blas'] == 'mkl':
configure_option += " --with-mkl"
elif comb['blas'] == 'atlas':
configure_option += " --with-atlas"
if comb['experiment'] == 'raevol':
configure_option += " --with-raevol"
if comb['compilator'] == 'intel':
configure_option += " CXX=icc"
full_bin_directory = bin_directory + comb['experiment']+'_'+comb['compilator']+'_'+comb['parallel']+'_'+comb['blas']
try:
os.mkdir(full_bin_directory)
except:
for f in os.listdir(full_bin_directory):
os.remove(full_bin_directory + "/" + f)
p = Process(export+'cd '+src_directory+'; autoreconf; ./configure '+configure_option+'; make clean; make; cp src/aevol_run '+full_bin_directory+'/; cp src/aevol_create '+full_bin_directory+'/')
p.shell = True
#
p.run()
print p.stdout
comb_ok = True
finally:
if comb_ok:
self.sweeper.done(comb)
logger.info(slugify(comb) + \
' has been done')
else:
self.sweeper.cancel(comb)
logger.warning(slugify(comb) + \
' has been canceled')
logger.info(style.step('%s Remaining'),
len(self.sweeper.get_remaining()))
def get_server_ip(host):
"""Get the server IP"""
if isinstance(host, Host):
host = host.address
logger.debug('Retrieving IP from %s', style.host(host))
get_ip = Process('host ' + host + ' |cut -d \' \' -f 4')
get_ip.shell = True
get_ip.run()
ip = get_ip.stdout.strip()
return ip
def get_server_ip(host):
"""Get the server IP"""
if isinstance(host, Host):
host = host.address
logger.debug('Retrieving IP from %s', style.host(host))
get_ip = Process('host ' + host + ' |cut -d \' \' -f 4')
get_ip.shell = True
get_ip.run()
ip = get_ip.stdout.strip()
return ip
def create_diet_architecture_files(self):
logger.info("Create a DIET architecture")
# Architecture without LA
process = Process("./dietg/set_archi_diet_4.sh gridnodes "+str(self.nodes_service[0]))
process.run()
MA_file = "./dietg/cfgs/MA1.cfg"
SeD_file = ['./dietg/cfgs/server.cfg']
logger.info("Create MA file")
set_scheduler(MA_file, self.scheduler)
logger.info("Create Sed files")
for file2 in SeD_file:
# print file2
set_parallel_jobs(file2, self.concLimit)
def get_logs_from_server(self):
distant_file = "/root/MA.stat"
local_folder = "./results_"+self.oargrid_job_id+"_"+self.scheduler+"/"
nb_files = 0
# Get(self.servers, [distant_file])
for host in self.servers:
local_file = local_folder+host+"_"+self.scheduler+"_SeD.stat"
process = Process("scp root@"+host+":"+distant_file+" "+local_file)
process.run()
try: #si le fichier existe
with open(local_file) as fichier:
nb_files += 1
fichier.close
except IOError: #si le fichier n'existe pas
pass
return nb_files
def dnsmasq_server(server, clients=None, vms=None, dhcp=True):
"""Configure a DHCP server with dnsmasq
:param server: host where the server will be installed
:param clients: list of hosts that will be declared in dnsmasq
:param vms: list of virtual machines
"""
logger.debug('Installing and configuring a DNS/DHCP server on %s', server)
test_running = Process('nmap ' + server + ' -p 53 | grep domain')
test_running.shell = True
test_running.run()
if 'open' in test_running.stdout:
logger.info('DNS server already running, updating configuration')
else:
cmd = 'killall dnsmasq; export DEBIAN_MASTER=noninteractive ; ' + \
'apt-get update ; apt-get -y purge dnsmasq-base ; ' + \
'apt-get install -t wheezy -o Dpkg::Options::="--force-confdef" ' + \
'-o Dpkg::Options::="--force-confnew" ' + \
'-y dnsmasq; echo 1 > /proc/sys/net/ipv4/ip_forward '
SshProcess(cmd, server).run()
sites = list(
set([
get_host_site(client)
for client in clients if get_host_site(client)
] + [get_host_site(server)]))
add_vms(vms, server)
if clients:
kill_dnsmasq = TaktukRemote('killall dnsmasq', clients)
for p in kill_dnsmasq.processes:
p.ignore_exit_code = p.nolog_exit_code = True
kill_dnsmasq.run()
resolv_conf(server, clients, sites)
if dhcp:
sysctl_conf(server, vms)
dhcp_conf(server, vms, sites)
logger.debug('Restarting service ...')
cmd = 'service dnsmasq stop ; rm /var/lib/misc/dnsmasq.leases ; ' + \
'service dnsmasq start',
SshProcess(cmd, server).run()
def get_nb_tasks_server(self):
distant_file = "/root/dietg/log/total.jobs"
local_file = "./task_counter"
# Get(self.servers, [distant_file])
for host in self.servers:
process = Process("scp root@"+host+":"+distant_file+" "+local_file)
process.run()
try: #si le fichier existe
with open(local_file) as fichier:
self.nb_tasks[host] = get_nb_tasks(local_file)
fichier.close
except IOError: #si le fichier n'existe pas
self.nb_tasks[host] = 0
try:
os.remove(local_file);
except OSError:
pass
def dnsmasq_server(server, clients=None, vms=None, dhcp=True):
"""Configure a DHCP server with dnsmasq
:param server: host where the server will be installed
:param clients: list of hosts that will be declared in dnsmasq
:param vms: list of virtual machines
"""
logger.debug('Installing and configuring a DNS/DHCP server on %s', server)
test_running = Process('nmap ' + server + ' -p 53 | grep domain')
test_running.shell = True
test_running.run()
if 'open' in test_running.stdout:
logger.info('DNS server already running, updating configuration')
else:
cmd = 'killall dnsmasq; export DEBIAN_MASTER=noninteractive ; ' + \
'apt-get update ; apt-get -y purge dnsmasq-base ; ' + \
'apt-get install -t wheezy -o Dpkg::Options::="--force-confdef" ' + \
'-o Dpkg::Options::="--force-confnew" ' + \
'-y dnsmasq; echo 1 > /proc/sys/net/ipv4/ip_forward '
SshProcess(cmd, server).run()
sites = list(set([get_host_site(client) for client in clients
if get_host_site(client)] + [get_host_site(server)]))
add_vms(vms, server)
if clients:
kill_dnsmasq = TaktukRemote('killall dnsmasq', clients)
for p in kill_dnsmasq.processes:
p.ignore_exit_code = p.nolog_exit_code = True
kill_dnsmasq.run()
resolv_conf(server, clients, sites)
if dhcp:
sysctl_conf(server, vms)
dhcp_conf(server, vms, sites)
logger.debug('Restarting service ...')
cmd = 'service dnsmasq stop ; rm /var/lib/misc/dnsmasq.leases ; ' + \
'service dnsmasq start',
SshProcess(cmd, server).run()
def clean_archi(self):
""" Delete all files related to an existing DIET archi """
logger.info("Clean DIET architecture")
process = Process("./dietg/clean_archi_diet.sh")
process.run()
process = Process("./dietg/clean.sh")
process.run()
process = Process("if [ -e ./tmp ]; then rm ./tmp; fi")
process.run()
def workflow(self, comb):
"""
Compute one application launch
using a given parameter group
"""
comb_ok = False
try:
# Generate configuration file needed by MPI processes
logger.info("Generating assembly file...")
py = comb['cores'] / comb['px']
prepare = Process('cd %s && python %s %d %d %d %d %d %s app.lad' %
(self.workingPath, self.genLadScript, comb['datasize'], comb['datasize'],
comb['datasize'], comb['px'], py, comb['transposition']))
prepare.shell = True
prepare.run()
# Generate the MPI host file
mfile = self.generate_machine_file()
# Start L2C
lad = "./app.lad"
logger.info("Computing...")
res = Process("export OAR_JOB_KEY_FILE=~/.oar_key ; cd %s && l2c_loader -M,-machinefile,%s --mpi -c %d %s" % (self.workingPath, mfile, comb['cores'], lad))
res.shell = True
res.stdout_handlers.append(os.path.join(self.result_dir, slugify(comb) + '.out'))
res.stdout_handlers.append(sys.stdout)
res.stderr_handlers.append(os.path.join(self.result_dir, slugify(comb) + '.err'))
res.stderr_handlers.append(sys.stderr)
res.run()
if not res.ok:
logger.error('Bad L2C termination')
raise Exception('Bad L2C termination')
if len(res.stderr) > 0: # WARNING: when L2C cannot find the LAD file or something strange like this
logger.warning('Not empty error output')
# Clean configuration files
logger.info("Removing assembly files...")
res = Process('cd %s && rm -f app.lad*' % self.workingPath)
res.shell = True
res.run()
comb_ok = True
except Exception:
pass
finally:
if comb_ok:
self.sweeper.done(comb)
logger.info(style.host(slugify(comb)) + ' has been done')
else:
self.sweeper.cancel(comb)
logger.warning(style.host(slugify(comb)) + ' has been canceled')
logger.info(style.step('%s Remaining'),
len(self.sweeper.get_remaining()))
def workflow(self, comb):
"""
Compute one application launch
using a given parameter group
"""
comb_ok = False
try:
# Generate configuration file needed by MPI processes
logger.info("Generating assembly file...")
py = comb['cores'] / comb['px']
prepare = Process(
'cd %s && python %s %d %d %d %d %d %s app.lad' %
(self.workingPath, self.genLadScript, comb['datasize'],
comb['datasize'], comb['datasize'], comb['px'], py,
comb['transposition']))
prepare.shell = True
prepare.run()
# Generate the MPI host file
mfile = self.generate_machine_file()
# Start L2C
lad = "./app.lad"
logger.info("Computing...")
res = Process(
"export OAR_JOB_KEY_FILE=~/.oar_key ; cd %s && l2c_loader -M,-machinefile,%s --mpi -c %d %s"
% (self.workingPath, mfile, comb['cores'], lad))
res.shell = True
res.stdout_handlers.append(
os.path.join(self.result_dir,
slugify(comb) + '.out'))
res.stdout_handlers.append(sys.stdout)
res.stderr_handlers.append(
os.path.join(self.result_dir,
slugify(comb) + '.err'))
res.stderr_handlers.append(sys.stderr)
res.run()
if not res.ok:
logger.error('Bad L2C termination')
raise Exception('Bad L2C termination')
if len(
res.stderr
) > 0: # WARNING: when L2C cannot find the LAD file or something strange like this
logger.warning('Not empty error output')
# Clean configuration files
logger.info("Removing assembly files...")
res = Process('cd %s && rm -f app.lad*' % self.workingPath)
res.shell = True
res.run()
comb_ok = True
except Exception:
pass
finally:
if comb_ok:
self.sweeper.done(comb)
logger.info(style.host(slugify(comb)) + ' has been done')
else:
self.sweeper.cancel(comb)
logger.warning(
style.host(slugify(comb)) + ' has been canceled')
logger.info(style.step('%s Remaining'),
len(self.sweeper.get_remaining()))
sed_time = Process('sed -i "s/simulator.duration.*/simulator.duration = ' +
str(1000) + '/g"' +
' load_events_generator/config/simulator.properties').run()
sed_nodes = Process(
'sed -i "s/nodes.number.*/nodes.number = ' + str(n_host) + '/g"' +
' load_events_generator/config/simulator.properties').run()
sed_vms = Process('sed -i "s/vm.number.*/vm.number = ' + str(len(vms)) +
'/g"' +
' load_events_generator/config/simulator.properties').run()
logger.info('Generating events list')
gen_events = Process(
'cd load_events_generator ; ' +
'java -jar load_events_generator.jar vms.list > ../events_load.xml')
gen_events.shell = True
gen_events.run()
tree = ET.parse('events_load.xml')
root = tree.getroot()
events = {}
for event in root.findall('./event'):
events[int(round(float(event.get('time'))))] = {
'vm': event.get('target'),
'load': event.get('value')
}
def set_cpu_load(load, vm_ip, pid):
"""Use cpulimit to change process intensity on vm"""
logger.info('kill cpu_limit on %s and set it to %s', vm_ip, load)
kill_cpu_limit = SshProcess(
f = open('hosts.list')
for line in f:
n_host += 1
f.close()
sed_time = Process('sed -i "s/simulator.duration.*/simulator.duration = ' + str(1000) + '/g"' +
' load_events_generator/config/simulator.properties').run()
sed_nodes = Process('sed -i "s/nodes.number.*/nodes.number = ' + str(n_host) + '/g"' +
' load_events_generator/config/simulator.properties').run()
sed_vms = Process('sed -i "s/vm.number.*/vm.number = ' + str(len(vms)) + '/g"' +
' load_events_generator/config/simulator.properties').run()
logger.info('Generating events list')
gen_events = Process('cd load_events_generator ; ' +
'java -jar load_events_generator.jar vms.list > ../events_load.xml')
gen_events.shell = True
gen_events.run()
tree = ET.parse('events_load.xml')
root = tree.getroot()
events = {}
for event in root.findall('./event'):
events[int(round(float(event.get('time'))))] = {'vm': event.get('target'),
'load': event.get('value')}
def set_cpu_load(load, vm_ip, pid):
"""Use cpulimit to change process intensity on vm"""
logger.info('kill cpu_limit on %s and set it to %s', vm_ip, load)
kill_cpu_limit = SshProcess('ps aux| grep "cpulimit" | grep -v "grep" | awk \'{print $2}\' | xargs -r kill -9',
vm_ip).run()
start_cpu_limit = SshProcess('cpulimit -p ' + str(pid) + ' -l ' + str(load), vm_ip)