def stop( self ):
console_logger.info( "Stopping DRM4G .... " )
console_logger.debug( "Stopping gwd .... " )
if self.is_alive():
with open(self.gwd_pid) as pid_file:
pid = next(pid_file)
processes_to_kill = [pid]
while processes_to_kill:
for process in processes_to_kill:
try:
cmd = "ps ho pid --ppid %s" % (process)
out , _ = exec_cmd( cmd )
processes_to_kill = [line.lstrip() for line in out.splitlines()] + processes_to_kill
processes_to_kill.remove(process)
os.kill( int(process), signal.SIGTERM )
except OSError as exc:
if "No such process" in str(exc):
continue
raise
while self.is_alive() :
time.sleep(1)
try:
os.remove(self.gwd_pid)
except OSError:
console_logger.error(" '%s' does not exist", self.gwd_pid)
console_logger.info( " OK" )
else:
console_logger.info(" WARNING: daemon is already stopped")
def run(arg):
try:
daemon = Daemon()
if not daemon.is_alive():
raise Exception('DRM4G is stopped. ')
if arg['submit']:
dependencies = '-d "%s"' % ' '.join(
arg['--dep']) if arg['--dep'] else ''
number_of_tasks = '-n %s' % arg['--ntasks'] if arg[
'--ntasks'] else ''
cmd = 'gwsubmit %s -v %s %s' % (dependencies, arg['<template>'],
number_of_tasks)
elif arg['list']:
cmd = 'gwps -o Jsetxjh '
if arg['<job_id>']:
cmd = cmd + arg['<job_id>'][0]
elif arg['history']:
cmd = 'gwhistory %s' % (arg['<job_id>'][0])
elif arg['log']:
directory = join(
DRM4G_DIR_VAR,
'%d00-%d99' % (int(int(float(arg['<job_id>'][0])) / 100),
int(int(float(arg['<job_id>'][0])) / 100)),
arg['<job_id>'][0], 'job.log')
if not exists(directory):
raise Exception('There is not a log available for this job.')
cmd = 'cat %s' % (directory)
else:
cmd = 'gwkill -9 %s' % (' '.join(arg['<job_id>']))
out, err = exec_cmd(cmd)
console_logger.info(out)
if err:
console_logger.info(err)
except Exception as err:
console_logger.error(str(err))
def run(arg):
try:
daemon = Daemon()
daemon.stop()
sleep(2.0)
daemon.clear()
except Exception as err:
console_logger.error(str(err))
def run(arg):
try:
Daemon().stop()
agent = Agent()
if agent.is_alive():
agent.stop()
except Exception as err:
console_logger.error(str(err))
def update_hosts(self):
"""
Forces the host list to be updated
"""
try:
GwImMad().do_DISCOVER("discover - - -", False)
#GwImMad().do_MONITOR("monitor 2 %s -" % resname)#, False)
except Exception as err:
console_logger.error( "Could not update hosts:\n%s", str(err))
def run(arg):
try:
daemon = Daemon()
if not daemon.is_alive():
raise Exception('DRM4G is stopped.')
config = Configuration()
config.load()
if config.check():
raise Exception("Review the configuration of '%s'." %
(arg['<resource_name>']))
if arg['<resource_name>'] not in config.resources:
raise Exception("'%s' is not a configured resource." %
(arg['<resource_name>']))
lrms = config.resources.get(arg['<resource_name>'])['lrms']
communicator = config.resources.get(
arg['<resource_name>'])['communicator']
if lrms != 'cream' and lrms != 'rocci' and communicator == 'local':
raise Exception("'%s' does not have an identity to configure." %
(arg['<resource_name>']))
if lrms == 'cream' or lrms == 'rocci':
comm = config.make_communicators()[arg['<resource_name>']]
if communicator == 'op_ssh':
#paramiko will always be used to renew the grid certificate
config.resources.get(
arg['<resource_name>'])['communicator'] = 'pk_ssh'
comm = config.make_communicators()[arg['<resource_name>']]
proxy = Proxy(config.resources[arg['<resource_name>']], comm)
config.resources.get(
arg['<resource_name>'])['communicator'] = communicator
config.make_communicators()
if communicator != 'local':
agent = Agent(config.resources[arg['<resource_name>']])
if arg['init']:
if communicator != 'local':
agent.start()
agent.add_key(arg['--lifetime'])
agent.copy_key()
if lrms == 'cream' or lrms == 'rocci':
proxy.configure()
proxy.create(arg['--lifetime'])
elif arg['delete']:
if lrms == 'cream' or lrms == 'rocci':
proxy.destroy()
if communicator != 'local':
agent.delete_key()
else:
if communicator != 'local':
agent.list_key()
if lrms == 'cream' or lrms == 'rocci':
proxy.check()
except Exception as err:
console_logger.error(str(err))
def run(arg):
try:
daemon = Daemon()
if not daemon.is_alive():
raise Exception('DRM4G is stopped.')
cmd = 'gwhost '
if arg['<hid>']:
cmd = cmd + arg['<hid>']
out, err = exec_cmd(cmd)
console_logger.info(out)
if err:
console_logger.info(err)
except Exception as err:
console_logger.error(str(err))
def _start():
# 's' option generates Bourne shell commands on stdout
out , err = exec_cmd( 'ssh-agent -s ' )
console_logger.debug( out )
match = re.search( 'SSH_AUTH_SOCK=(?P<SSH_AUTH_SOCK>[^;]+);.*' \
+ 'SSH_AGENT_PID=(?P<SSH_AGENT_PID>\d+);', out, re.DOTALL)
if match :
console_logger.debug( " OK" )
self.agent_env = match.groupdict()
console_logger.debug( ' Agent pid: %s', self.agent_env['SSH_AGENT_PID'])
else:
console_logger.error( err )
raise Exception( ' Cannot determine agent data from output: %s' % out )
with open( self.agent_file , 'w') as f:
f.write( self.agent_env['SSH_AGENT_PID'] + '\n' + self.agent_env['SSH_AUTH_SOCK'] )
def check_frontends( self ) :
"""
Check if the frontend of a given resource is reachable.
"""
self.check( )
communicators = self.config.make_communicators()
for resname, resdict in sorted( self.config.resources.items() ) :
if resdict[ 'enable' ] == 'true' :
communicator = communicators.get( resname )
try :
# if resdict[ 'communicator' ] == 'op_ssh' :
# #it will use im's socket instead of creating a new one
# communicator.parent_module = 'im'
# communicator.configfile = join(DRM4G_DIR, 'etc', 'openssh_im.conf')
communicator.connect()
console_logger.info( "Resource '%s' :", resname )
console_logger.info( "--> The front-end '%s' is accessible\n", communicator.frontend )
except Exception :
console_logger.error( "Resource '%s' :", resname )
console_logger.error( "--> The front-end '%s' is not accessible\n", communicator.frontend )
def run(arg):
try:
config = Configuration()
resource = Resource(config)
if arg['edit']:
resource.edit()
else:
daemon = Daemon()
if not daemon.is_alive():
raise Exception('DRM4G is stopped.')
elif arg['check']:
resource.check_frontends()
elif arg['create']:
resource.create_vms()
elif arg['destroy']:
resource.destroy_vms()
elif arg['--all']:
resource.list_resources()
else:
resource.list()
except Exception as err:
console_logger.error(str(err))
def run(arg):
try:
Daemon().start()
except Exception as err:
console_logger.error(str(err))
