def persistent_connection(resource_config, logger):
"""Keep running a persistent master connection"""
sleep_secs = 300
hostname = resource_config['HOSTURL']
# Mark this session as a multiplexing master to avoid races:
# see further details in shared/ssh.py
resource_config['SSHMULTIPLEXMASTER'] = True
while True:
try:
logger.debug('connecting to %s' % hostname)
(exit_code,
executed) = execute_on_resource('sleep %d' % sleep_secs, False,
resource_config, logger)
if 0 != exit_code:
msg = 'ssh multiplex %s: %s returned %i' % \
(hostname, executed, exit_code)
print msg
# make sure control_socket was cleaned up
host = resource_config['HOSTURL']
identifier = resource_config['HOSTIDENTIFIER']
unique_id = '%s.%s' % (host, identifier)
control_socket = \
os.path.join(configuration.resource_home,
unique_id, 'ssh-multiplexing')
try:
os.remove(control_socket)
except:
pass
sleep(sleep_secs)
except StandardError, err:
msg = '%s thread caught exception (%s) - retry later' % \
(hostname, err)
print msg
logger.error(msg)
sleep(sleep_secs)
def persistent_connection(resource_config, logger):
"""Keep running a persistent master connection"""
sleep_secs = 300
hostname = resource_config['HOSTURL']
# Mark this session as a multiplexing master to avoid races:
# see further details in shared/ssh.py
resource_config['SSHMULTIPLEXMASTER'] = True
while True:
try:
logger.debug('connecting to %s' % hostname)
(exit_code, executed) = execute_on_resource('sleep %d'
% sleep_secs, False, resource_config, logger)
if 0 != exit_code:
msg = 'ssh multiplex %s: %s returned %i' % (hostname,
executed, exit_code)
print msg
# make sure control_socket was cleaned up
host = resource_config['HOSTURL']
identifier = resource_config['HOSTIDENTIFIER']
unique_id = '%s.%s' % (host, identifier)
control_socket = \
os.path.join(configuration.resource_home,
unique_id, 'ssh-multiplexing')
try:
os.remove(control_socket)
except:
pass
sleep(sleep_secs)
except StandardError, err:
msg = '%s thread caught exception (%s) - retry later'\
% (hostname, err)
print msg
logger.error(msg)
sleep(sleep_secs)
ssh_command = \
'ssh -X %s@%s \\"cd %s; mv -f %s job-dir_%s; cd job-dir_%s; chmod +x %s; bash -c \'./%s\'\\"'\
% (
exe['execution_user'],
exe['execution_node'],
exe['execution_dir'],
dest_filename,
localjobname,
localjobname,
dest_filename,
dest_filename,
)
logger.info('execute_on_resource: %s' % ssh_command)
exit_code = -1
try:
(exit_code, executed_command) = execute_on_resource(ssh_command,
False, resource_config, logger)
except Exception, e:
logger.error('Exception executing remote SSH from requestinteractivejob.py: %s '
% e)
o.reply_and_exit(o.ERROR)
if exit_code != 0:
logger.error('Error executing interactive job script on resource: %s command: %s exit code: %s'
% (unique_resource_name, executed_command, exit_code))
o.reply_and_exit(o.ERROR)
o.out('requestinteractivejob OK. The job was started on the resource: %s exe: %s remote addr: %s'
% (unique_resource_name, exe, os.getenv('REMOTE_ADDR')))
o.reply_and_exit(o.OK)
ssh_command = \
'ssh -X %s@%s \\"cd %s; mv -f %s job-dir_%s; cd job-dir_%s; chmod +x %s; bash -c \'./%s\'\\"'\
% (
exe_conf['execution_user'],
exe_conf['execution_node'],
exe_conf['execution_dir'],
dest_filename,
localjobname,
localjobname,
dest_filename,
dest_filename,
)
logger.info('execute_on_resource: %s' % ssh_command)
exit_code = -1
try:
(exit_code, executed_command) = execute_on_resource(ssh_command,
False, resource_conf, logger)
except Exception, e:
logger.error('Exception executing remote SSH from requestinteractivejob.py: %s '
% e)
o.reply_and_exit(o.ERROR)
if exit_code != 0:
logger.error('Error executing interactive job script on resource: %s command: %s exit code: %s'
% (unique_resource_name, executed_command, exit_code))
o.reply_and_exit(o.ERROR)
o.out('requestinteractivejob OK. The job was started on the resource: %s exe: %s remote addr: %s'
% (unique_resource_name, exe_conf, os.getenv('REMOTE_ADDR')))
o.reply_and_exit(o.OK)
