def __run_sshcmd(self, int_cmd):
""" run internal used ssh command """
self.__run_ssh(int_cmd)
if self.rc != 0:
msg = 'Host [%s]: Failed to run ssh commands, check SSH password or connectivity' % self.host
self.logger.error(msg)
err_m(msg)
def __run_sshcmd(self, int_cmd):
""" run internal used ssh command """
self.__run_ssh(int_cmd)
if self.rc != 0:
msg = 'Host [%s]: Failed to run ssh commands, check SSH password or connectivity' % self.host
self.logger.error(msg)
err_m(msg)
def main():
options = get_options()
cfgs = defaultdict(str)
if options.cfgfile:
if not os.path.exists(options.cfgfile):
err_m('Cannot find config file \'%s\'' % options.cfgfile)
config_file = options.cfgfile
else:
config_file = DBCFG_FILE
if options.pwd:
pwd = getpass.getpass('Input remote host SSH Password: '******''
if os.path.exists(config_file):
cfgs = ParseInI(config_file).load()
else:
node_lists = expNumRe(
raw_input(
'Enter list of Nodes separated by comma, support numeric RE, i.e. n[01-12]: '
))
# check if node list is expanded successfully
if len([1 for node in node_lists if '[' in node]):
err('Failed to expand node list, please check your input.')
cfgs['node_list'] = ','.join(node_lists)
results = wrapper.run(cfgs, options, mode='discover', pwd=pwd)
format_output('Discover results')
if len(results) > 4:
output = output_row(results)
else:
output = output_column(results)
print output
with open('discover_result', 'w') as f:
f.write('Discover Date: %s\n' % time.strftime('%Y-%m-%d %H:%M'))
f.write(output)
def run(dbcfgs, options, mode='install', pwd=''):
""" main entry
mode: install/discover
"""
STAT_FILE = mode + '.status'
LOG_FILE = '%s/logs/%s_%s.log' % (INSTALLER_LOC, mode,
time.strftime('%Y%m%d_%H%M'))
logger = get_logger(LOG_FILE)
verbose = True if hasattr(options,
'verbose') and options.verbose else False
upgrade = True if hasattr(options,
'upgrade') and options.upgrade else False
user = options.user if hasattr(options, 'user') and options.user else ''
threshold = options.fork if hasattr(options,
'fork') and options.fork else 10
script_output = [] # script output array
conf = ParseJson(SCRCFG_FILE).load()
script_cfgs = conf[mode]
dbcfgs_json = json.dumps(dbcfgs)
hosts = dbcfgs['node_list'].split(',')
# handle skipped scripts, skip them if no need to run
skipped_scripts = []
if upgrade:
skipped_scripts += [
'hadoop_mods', 'apache_mods', 'apache_restart', 'traf_dep',
'traf_kerberos'
]
if dbcfgs['secure_hadoop'] == 'N':
skipped_scripts += ['traf_kerberos']
if dbcfgs['traf_start'].upper() == 'N':
skipped_scripts += ['traf_start']
if dbcfgs['ldap_security'].upper() == 'N':
skipped_scripts += ['traf_ldap']
if 'APACHE' in dbcfgs['distro']:
skipped_scripts += ['hadoop_mods']
else:
skipped_scripts += ['apache_mods', 'apache_restart']
# set ssh config file to avoid known hosts verify on current installer node
SSH_CFG_FILE = os.environ['HOME'] + '/.ssh/config'
ssh_cfg = 'StrictHostKeyChecking=no\nNoHostAuthenticationForLocalhost=yes\n'
with open(SSH_CFG_FILE, 'w') as f:
f.write(ssh_cfg)
run_cmd('chmod 600 %s' % SSH_CFG_FILE)
def run_local_script(script, json_string, req_pwd):
cmd = '%s/%s \'%s\'' % (INSTALLER_LOC, script, json_string)
# pass the ssh password to sub scripts which need SSH password
if req_pwd: cmd += ' ' + pwd
if verbose: print cmd
# stdout on screen
p = subprocess.Popen(cmd, stderr=subprocess.PIPE, shell=True)
stdout, stderr = p.communicate()
rc = p.returncode
if rc != 0:
msg = 'Failed to run \'%s\'' % script
if stderr:
msg += ': ' + stderr
print stderr
logger.error(msg)
state_fail('localhost: Script [%s]' % script)
exit(rc)
else:
state_ok('Script [%s]' % script)
logger.info('Script [%s] ran successfully!' % script)
return stdout
# run sub scripts
try:
remote_instances = []
if mode == 'discover':
remote_instances = [
RemoteRun(host, logger, user=user, pwd=pwd, quiet=True)
for host in hosts
]
else:
remote_instances = [
RemoteRun(host, logger, user=user, pwd=pwd) for host in hosts
]
first_instance = remote_instances[0]
for instance in remote_instances:
if instance.host == dbcfgs['first_rsnode']:
first_rs_instance = instance
break
logger.info(' ***** %s Start *****' % mode)
for cfg in script_cfgs:
script = cfg['script']
node = cfg['node']
desc = cfg['desc']
run_user = ''
if not 'run_as_traf' in cfg.keys():
pass
elif cfg['run_as_traf'] == 'yes':
run_user = dbcfgs['traf_user']
if not 'req_pwd' in cfg.keys():
req_pwd = False
elif cfg['req_pwd'] == 'yes':
req_pwd = True
status = Status(STAT_FILE, script)
if status.get_status():
msg = 'Script [%s] had already been executed' % script
state_skip(msg)
logger.info(msg)
continue
if script.split('.')[0] in skipped_scripts:
continue
else:
print '\nTASK: %s %s' % (desc, (83 - len(desc)) * '*')
#TODO: timeout exit
if node == 'local':
run_local_script(script, dbcfgs_json, req_pwd)
elif node == 'first':
first_instance.run_script(script,
run_user,
dbcfgs_json,
verbose=verbose)
elif node == 'first_rs':
first_rs_instance.run_script(script,
run_user,
dbcfgs_json,
verbose=verbose)
elif node == 'all':
l = len(remote_instances)
if l > threshold:
piece = (l - (l % threshold)) / threshold
parted_remote_instances = [
remote_instances[threshold * i:threshold * (i + 1)]
for i in range(piece)
]
parted_remote_instances.append(remote_instances[threshold *
piece:])
else:
parted_remote_instances = [remote_instances]
for parted_remote_inst in parted_remote_instances:
threads = [
Thread(target=r.run_script,
args=(script, run_user, dbcfgs_json, verbose))
for r in parted_remote_inst
]
for t in threads:
t.start()
for t in threads:
t.join()
if sum([r.rc for r in parted_remote_inst]) != 0:
err_m(
'Script failed to run on one or more nodes, exiting ...\nCheck log file %s for details.'
% LOG_FILE)
script_output += [{
r.host: r.stdout.strip()
} for r in parted_remote_inst]
else:
# should not go to here
err_m('Invalid configuration for %s' % SCRCFG_FILE)
status.set_status()
except KeyboardInterrupt:
err_m('User quit')
# remove status file if all scripts run successfully
os.remove(STAT_FILE)
return script_output
def run(dbcfgs, options, mode='install', pwd=''):
""" main entry
mode: install/discover
"""
STAT_FILE = mode + '.status'
LOG_FILE = '%s/logs/%s_%s.log' % (INSTALLER_LOC, mode, time.strftime('%Y%m%d_%H%M'))
logger = get_logger(LOG_FILE)
verbose = True if hasattr(options, 'verbose') and options.verbose else False
upgrade = True if hasattr(options, 'upgrade') and options.upgrade else False
user = options.user if hasattr(options, 'user') and options.user else ''
threshold = options.fork if hasattr(options, 'fork') and options.fork else 10
script_output = [] # script output array
conf = ParseJson(SCRCFG_FILE).load()
script_cfgs = conf[mode]
dbcfgs_json = json.dumps(dbcfgs)
hosts = dbcfgs['node_list'].split(',')
# handle skipped scripts, skip them if no need to run
skipped_scripts = []
if upgrade:
skipped_scripts += ['hadoop_mods', 'apache_mods', 'apache_restart', 'traf_dep', 'traf_kerberos']
if dbcfgs['secure_hadoop'] == 'N':
skipped_scripts += ['traf_kerberos']
if dbcfgs['traf_start'].upper() == 'N':
skipped_scripts += ['traf_start']
if dbcfgs['ldap_security'].upper() == 'N':
skipped_scripts += ['traf_ldap']
if 'APACHE' in dbcfgs['distro']:
skipped_scripts += ['hadoop_mods']
else:
skipped_scripts += ['apache_mods', 'apache_restart']
# set ssh config file to avoid known hosts verify on current installer node
SSH_CFG_FILE = os.environ['HOME'] + '/.ssh/config'
ssh_cfg = 'StrictHostKeyChecking=no\nNoHostAuthenticationForLocalhost=yes\n'
with open(SSH_CFG_FILE, 'w') as f:
f.write(ssh_cfg)
run_cmd('chmod 600 %s' % SSH_CFG_FILE)
def run_local_script(script, json_string, req_pwd):
cmd = '%s/%s \'%s\'' % (INSTALLER_LOC, script, json_string)
# pass the ssh password to sub scripts which need SSH password
if req_pwd: cmd += ' ' + pwd
if verbose: print cmd
# stdout on screen
p = subprocess.Popen(cmd, stderr=subprocess.PIPE, shell=True)
stdout, stderr = p.communicate()
rc = p.returncode
if rc != 0:
msg = 'Failed to run \'%s\'' % script
if stderr:
msg += ': ' + stderr
print stderr
logger.error(msg)
state_fail('localhost: Script [%s]' % script)
exit(rc)
else:
state_ok('Script [%s]' % script)
logger.info('Script [%s] ran successfully!' % script)
return stdout
# run sub scripts
try:
remote_instances = []
if mode == 'discover':
remote_instances = [RemoteRun(host, logger, user=user, pwd=pwd, quiet=True) for host in hosts]
else:
remote_instances = [RemoteRun(host, logger, user=user, pwd=pwd) for host in hosts]
first_instance = remote_instances[0]
for instance in remote_instances:
if instance.host == dbcfgs['first_rsnode']:
first_rs_instance = instance
break
logger.info(' ***** %s Start *****' % mode)
for cfg in script_cfgs:
script = cfg['script']
node = cfg['node']
desc = cfg['desc']
run_user = ''
if not 'run_as_traf' in cfg.keys():
pass
elif cfg['run_as_traf'] == 'yes':
run_user = dbcfgs['traf_user']
if not 'req_pwd' in cfg.keys():
req_pwd = False
elif cfg['req_pwd'] == 'yes':
req_pwd = True
status = Status(STAT_FILE, script)
if status.get_status():
msg = 'Script [%s] had already been executed' % script
state_skip(msg)
logger.info(msg)
continue
if script.split('.')[0] in skipped_scripts:
continue
else:
print '\nTASK: %s %s' % (desc, (83 - len(desc))*'*')
#TODO: timeout exit
if node == 'local':
run_local_script(script, dbcfgs_json, req_pwd)
elif node == 'first':
first_instance.run_script(script, run_user, dbcfgs_json, verbose=verbose)
elif node == 'first_rs':
first_rs_instance.run_script(script, run_user, dbcfgs_json, verbose=verbose)
elif node == 'all':
l = len(remote_instances)
if l > threshold:
piece = (l - (l % threshold)) / threshold
parted_remote_instances = [remote_instances[threshold*i:threshold*(i+1)] for i in range(piece)]
parted_remote_instances.append(remote_instances[threshold*piece:])
else:
parted_remote_instances = [remote_instances]
for parted_remote_inst in parted_remote_instances:
threads = [Thread(target=r.run_script, args=(script, run_user, dbcfgs_json, verbose)) for r in parted_remote_inst]
for t in threads: t.start()
for t in threads: t.join()
if sum([r.rc for r in parted_remote_inst]) != 0:
err_m('Script failed to run on one or more nodes, exiting ...\nCheck log file %s for details.' % LOG_FILE)
script_output += [{r.host:r.stdout.strip()} for r in parted_remote_inst]
else:
# should not go to here
err_m('Invalid configuration for %s' % SCRCFG_FILE)
status.set_status()
except KeyboardInterrupt:
err_m('User quit')
# remove status file if all scripts run successfully
os.remove(STAT_FILE)
return script_output
