def salt_start():
lg = None
try:
lg, err = logger.get_script_logger('Conditional salt start completed',
'/var/log/integralstor/scripts.log',
level=logging.DEBUG)
logger.log_or_print('Conditional salt start initiated.',
lg,
level='info')
pog, err = grid_ops.is_part_of_grid()
logger.log_or_print('Part of grid : %s' % pog, lg, level='info')
if err:
raise Exception(err)
if not pog:
logger.log_or_print('Starting salt services.', lg, level='info')
out, err = command.get_command_output('service salt-minion start',
False)
if err:
raise Exception(err)
out, err = command.get_command_output('service salt-master start',
False)
if err:
raise Exception(err)
else:
logger.log_or_print(
'Not starting salt services as I am part of the grid.',
lg,
level='info')
except Exception, e:
str = 'Error doing a conditional salt start : %s' % e
logger.log_or_print(str, lg, level='critical')
return False, str
def check_windows_processes_status(print_to_stdout=True, separator='\n'):
ret_list = []
try:
if print_to_stdout:
ret_list.append('GRIDCell windows access processes status: ')
out, err = command.get_command_output('service smb status', False)
if err:
raise Exception(err)
ret_list.append('Windows (SMB) process status : %s' % ', '.join(out))
out, err = command.get_command_output('service winbind status', False)
if err:
raise Exception(err)
ret_list.append('Windows (winbind) process status : %s' %
', '.join(out))
out, err = command.get_command_output('service ntpd status', False)
if err:
raise Exception(err)
ret_list.append('Time sync (NTP) process status : %s' % ', '.join(out))
'''
Commenting out as we wont use CTDB for this release
out, err = command.get_command_output('service ctdb status', False)
if err:
raise Exception(err)
ret_list.append('Clustered windows access (CTDB) process status : %s'%', '.join(out))
'''
if print_to_stdout:
print separator.join(ret_list)
except Exception, e:
return None, 'Error checking windows access process status: %s' % str(
e)
def get_ad_users_or_groups(type):
"""Issue an active directory command that gets all users and groups to display in share creation/editing."""
o = None
try:
d, err = get_auth_settings()
if err:
raise Exception(err)
workgroup = d['workgroup']
if type and type == "users":
o, err = command.get_command_output('wbinfo -u --domain=%s' %
workgroup)
if err:
raise Exception(err)
elif type and type == "groups":
o, err = command.get_command_output('wbinfo -g --domain=%s' %
workgroup)
if err:
raise Exception(err)
else:
raise Exception("Unknown type specified.")
# print '%s - '%type, o
except Exception, e:
return None, 'Error retrieving Active Directory Users/Groups : %s' % str(
e)
def salt_start():
lg = None
try:
lg, err = logger.get_script_logger(
'Conditional salt start completed', '/var/log/integralstor/scripts.log', level=logging.DEBUG)
logger.log_or_print(
'Conditional salt start initiated.', lg, level='info')
pog, err = grid_ops.is_part_of_grid()
logger.log_or_print('Part of grid : %s' % pog, lg, level='info')
if err:
raise Exception(err)
if not pog:
logger.log_or_print('Starting salt services.', lg, level='info')
out, err = command.get_command_output(
'service salt-minion start', False)
if err:
raise Exception(err)
out, err = command.get_command_output(
'service salt-master start', False)
if err:
raise Exception(err)
else:
logger.log_or_print(
'Not starting salt services as I am part of the grid.', lg, level='info')
except Exception, e:
str = 'Error doing a conditional salt start : %s' % e
logger.log_or_print(str, lg, level='critical')
return False, str
def view_task_details(request, task_id):
return_dict = {}
try:
task, err = scheduler_utils.get_task(int(task_id))
if err:
raise Exception(err)
return_dict['task'] = task
subtasks, err = scheduler_utils.get_subtasks(int(task_id))
if err:
raise Exception(err)
return_dict["subtasks"] = subtasks
# print subtasks, err
task_output = ""
log_path, err = config.get_log_folder_path()
if err:
raise Exception(err)
log_dir = '%s/task_logs' % log_path
log_file_path = '%s/%d.log' % (log_dir, int(task_id))
if os.path.isfile(log_file_path):
lines, err = command.get_command_output("wc -l %s" % log_file_path)
no_of_lines = lines[0].split()[0]
# print no_of_lines
if int(no_of_lines) <= 41:
# This code always updates the 0th element of the command list.
# This is assuming that we will only have one long running
# command.
with open(log_file_path) as output:
task_output = task_output + ''.join(output.readlines())
else:
first, err = command.get_command_output("head -n 5 %s" %
log_file_path,
shell=True)
if err:
print err
last, err = command.get_command_output("tail -n 20 %s" %
log_file_path,
shell=True)
if err:
print err
# print last
task_output = task_output + '\n'.join(first)
task_output = task_output + "\n.... \n ....\n"
task_output = task_output + '\n'.join(last)
return_dict['task_output'] = task_output
return django.shortcuts.render_to_response(
"view_task_details.html",
return_dict,
context_instance=django.template.context.RequestContext(request))
except Exception, e:
return_dict['base_template'] = "scheduler_base.html"
return_dict["page_title"] = 'Background jobs'
return_dict['tab'] = 'view_background_tasks_tab'
return_dict["error"] = 'Error retriving background task details'
return_dict["error_details"] = e
return django.shortcuts.render_to_response(
"logged_in_error.html",
return_dict,
context_instance=django.template.context.RequestContext(request))
def update_manifest(request):
return_dict = {}
try:
if request.method == "GET":
from integralstor_utils import manifest_status as iu
mi, err = iu.generate_manifest_info()
# print mi, err
if err:
raise Exception(err)
if not mi:
raise Exception('Could not load new configuration')
return_dict["mi"] = mi[mi.keys()[0]] # Need the hostname here.
return django.shortcuts.render_to_response(
"update_manifest.html",
return_dict,
context_instance=django.template.context.RequestContext(
request))
elif request.method == "POST":
common_python_scripts_path, err = config.get_common_python_scripts_path(
)
if err:
raise Exception(err)
ss_path, err = config.get_system_status_path()
if err:
raise Exception(err)
#(ret,rc), err = command.execute_with_rc("python %s/generate_manifest.py %s"%(common_python_scripts_path, ss_path))
ret, err = command.get_command_output(
"python %s/generate_manifest.py %s" %
(common_python_scripts_path, ss_path))
# print 'mani', ret, err
if err:
raise Exception(err)
#(ret,rc), err = command.execute_with_rc("python %s/generate_status.py %s"%(config.get_python_scripts_path(),config.get_system_status_path()))
ret, err = command.get_command_output(
"python %s/generate_status.py %s" %
(common_python_scripts_path, ss_path))
# print 'stat', ret, err
if err:
raise Exception(err)
return django.http.HttpResponseRedirect("/view_system_info/")
except Exception, e:
return_dict['base_template'] = "system_base.html"
return_dict["page_title"] = 'Reload system configuration'
return_dict['tab'] = 'node_info_tab'
return_dict["error"] = 'Error reloading system configuration'
return_dict["error_details"] = str(e)
return django.shortcuts.render_to_response(
"logged_in_error.html",
return_dict,
context_instance=django.template.context.RequestContext(request))
def update_date_time(system_date, system_time, system_timezone):
return_dict = {}
try:
if system_time == None:
pass
else:
cmd = 'date -s %s' % system_time
output, err = command.get_command_output(cmd)
if err:
raise Exception(err)
return_dict['time_set'] = True
return_dict['time_set_to'] = system_time
if system_date == None:
pass
else:
date_component = system_date.split("/")
if len(date_component) != 3:
raise Exception(
"The input date format is not correct (Ex: MM/DD/YYYY)")
system_date = '%s-%s-%s' % (date_component[2], date_component[0],
date_component[1])
output_time, err = command.get_command_output("date +%T")
if err:
raise Exception(err)
cmd = 'date -s "%s %s"' % (system_date, output_time[0])
output, err = command.get_command_output(cmd)
if err:
raise Exception(err)
return_dict['date_set'] = True
return_dict['date_set_to'] = system_date
if system_timezone == None:
pass
else:
cmd = 'timedatectl set-timezone %s' % system_timezone
output, err = command.get_command_output(cmd)
if err:
raise Exception(err)
return_dict['timezone_set'] = True
return_dict['timezone_set_to'] = system_timezone
cmd = "hwclock -w"
output, err = command.get_command_output(cmd)
if err:
raise Exception(err)
except Exception, e:
return None, str(e)
def update_date_time(system_date, system_time, system_timezone):
return_dict = {}
try:
if system_time == None:
pass
else:
cmd = 'date -s %s' % system_time
output, err = command.get_command_output(cmd)
if err:
raise Exception(err)
return_dict['time_set'] = True
return_dict['time_set_to'] = system_time
if system_date == None:
pass
else:
date_component = system_date.split("/")
if len(date_component) != 3:
raise Exception(
"The input date format is not correct (Ex: MM/DD/YYYY)")
system_date = '%s-%s-%s' % (date_component[2],
date_component[0], date_component[1])
output_time, err = command.get_command_output("date +%T")
if err:
raise Exception(err)
cmd = 'date -s "%s %s"' % (system_date, output_time[0])
output, err = command.get_command_output(cmd)
if err:
raise Exception(err)
return_dict['date_set'] = True
return_dict['date_set_to'] = system_date
if system_timezone == None:
pass
else:
cmd = 'timedatectl set-timezone %s' % system_timezone
output, err = command.get_command_output(cmd)
if err:
raise Exception(err)
return_dict['timezone_set'] = True
return_dict['timezone_set_to'] = system_timezone
cmd = "hwclock -w"
output, err = command.get_command_output(cmd)
if err:
raise Exception(err)
except Exception, e:
return None, str(e)
def create_ramdisk(size, path, pool):
try:
if not size:
raise Exception('Size not specified')
mem, err = get_mem_stats('MB')
if err:
raise Exception(err)
if not mem:
raise Exception('Error getting memory information : %s' % err)
used, err = get_total_ramdisks_size()
if err:
raise Exception(err)
if used < 0:
raise Exception('Error checking RAM size')
if (used + size) / mem['total'] > 0.25:
raise Exception('Insufficient memory ')
if not os.path.exists(path):
os.makedirs(path)
cmd = 'mount -t tmpfs -o size=%dm tmpfs %s' % (size, path)
lines, err = command.get_command_output(cmd)
if err:
raise Exception('Error creating ramdisk : %s' % err)
cmd = 'dd if=/dev/zero of=%s/ramfile bs=1024 count=%dK' % (path, size)
lines, err = command.get_command_output(cmd)
if err:
raise Exception('Error initializing ramdisk : %s' % err)
res, err = insert_into_ramdisks_config(pool, path, size)
if err or not res:
cmd = 'umount %s' % path
lines, err1 = command.get_command_output(cmd)
if err1:
if err:
err += err1
raise Exception(err)
else:
raise Exception(err1)
else:
raise Exception(err)
except Exception, e:
return False, 'Error creating ramdisk : %s' % str(e)
def parse_ssl_certificate(cert_location):
cert_info = {}
try:
if not os.path.exists(cert_location):
raise Exception('Could not locate certificate')
lines, err = command.get_command_output(
'openssl x509 -in %s -text -noout' % cert_location)
if err:
raise Exception(err)
look_for = [{'srch_for': 'Not Before', 'field_name': 'expiry_start'}, {'srch_for': 'Issuer', 'field_name': 'issuer'}, {'srch_for': 'Not After', 'field_name': 'expiry_end'}, {'srch_for': 'Subject', 'field_name': 'dn'}, {
'srch_for': 'Public Key Algorithm', 'field_name': 'key_algorithm'}, {'srch_for': 'Serial Number', 'field_name': 'serial_number'}, {'srch_for': 'Public-Key', 'field_name': 'key_length'}]
for line in lines:
for l in look_for:
pattern = '[\s]*%s[\s]*:[\s]*([\S\s]*)$' % l['srch_for']
ret = re.match(pattern, line)
if ret:
grps = ret.groups()
if grps:
cert_info[l['field_name']] = grps[0]
if 'key_length' in cert_info:
ret = re.match('[\s]*\(([\sa-zA-Z0-9]*)\)',
cert_info['key_length'])
if ret and ret.groups():
cert_info['key_length'] = ret.groups()[0]
except Exception, e:
return None, 'Error parsing certificate : %s' % str(e)
def sync_ntp():
return_dict = {}
try:
ntp_servers, err = get_ntp_servers()
if err:
raise Exception(err)
if len(ntp_servers) < 1:
raise Exception(
"NTP servers are not configured, Please configure at least one server to sync"
)
output, err = services_management.update_service_status('ntpd', 'stop')
if err:
raise Exception(err)
for server in ntp_servers:
cmd = "ntpdate -b %s" % server
output, err = command.get_command_output(cmd, shell=True)
if err:
continue
else:
return_dict['ntp_sync'] = True
return_dict['server_used'] = server
break
output, err = services_management.update_service_status(
'ntpd', 'start')
if err:
raise Exception(err)
except Exception, e:
return None, 'Sync failed'
def get_mem_stats(units):
mem = None
try:
if units not in ['MB', 'GB']:
raise Exception('Incorrect units requested')
if not units:
raise Exception('Units not specified in request')
if units == 'MB':
flag = '-m'
elif units == 'GB':
flag = '-g'
cmd = 'free %s' % flag
lines = None
lines, err = command.get_command_output(cmd)
if err:
raise Exception('Error getting free memory: %s' % err)
if lines:
for line in lines:
if 'Mem:' not in line:
continue
components = line.split()
mem = {}
mem['total'] = float(components[1])
mem['used'] = float(components[2])
mem['free'] = float(components[3])
except Exception, e:
return None, 'Error retriving free memory : %s' % str(e)
def get_os_partition_stats():
ret_list = []
try:
lines, err = command.get_command_output(
'df -HT --exclude-type=devtmpfs --exclude-type=tmpfs --exclude-type=zfs'
)
if err:
raise Exception(err)
#print lines, err
ret_list = []
for line in lines[1:]:
components = line.split()
tmp_dict = {}
#print components
used_percentage_str = components[5]
percentage_pos = used_percentage_str.find('%')
used_percentage = int(used_percentage_str[:percentage_pos])
free_percentage = 100 - used_percentage
tmp_dict['percentage_free_str'] = '%d%%' % free_percentage
tmp_dict['percentage_free'] = free_percentage
tmp_dict['fs_name'] = components[6]
tmp_dict['used'] = components[3]
tmp_dict['capacity'] = components[2]
tmp_dict['free'] = components[4]
ret_list.append(tmp_dict)
except Exception, e:
return None, "Error retrieving OS partition statistics : %s" % str(e)
def net_ads_join(user, pswd, password_server):
try:
cmd_to_run = "net ads join -U %s%%%s" % (user, pswd)
# print 'Running %s'%cmd_to_run
use_salt, err = config.use_salt()
if err:
raise Exception(err)
if use_salt:
import salt.client
errors = []
client = salt.client.LocalClient()
# print 'Running %s'%cmd_to_run
#assert False
r1 = client.cmd('*', 'cmd.run_all', [cmd_to_run])
# print r1
if r1:
for node, ret in r1.items():
# print ret
if ret["retcode"] != 0:
e = "Error joining AD on node %s" % node
if "stderr" in ret:
e += " : %s" % ret["stderr"]
errors.append(e)
print errors
# print r1
if errors:
raise Exception(' '.join(errors))
else:
lines, err = command.get_command_output(cmd_to_run)
if err:
raise Exception(err)
except Exception, e:
return False, 'Error joining AD : %s' % str(e)
def get_os_partition_stats():
ret_list = []
try:
lines, err = command.get_command_output('df -HT --exclude-type=devtmpfs --exclude-type=tmpfs --exclude-type=zfs')
if err:
raise Exception(err)
#print lines, err
ret_list = []
for line in lines[1:]:
components = line.split()
tmp_dict = {}
#print components
used_percentage_str = components[5]
percentage_pos = used_percentage_str.find('%')
used_percentage = int(used_percentage_str[:percentage_pos])
free_percentage = 100-used_percentage
tmp_dict['percentage_free_str'] = '%d%%'%free_percentage
tmp_dict['percentage_free'] = free_percentage
tmp_dict['fs_name'] = components[6]
tmp_dict['used'] = components[3]
tmp_dict['capacity'] = components[2]
tmp_dict['free'] = components[4]
ret_list.append(tmp_dict)
except Exception, e:
return None, "Error retrieving OS partition statistics : %s" % str(e)
def reload_configuration():
try:
use_salt, err = config.use_salt()
if err:
raise Exception(err)
if use_salt:
import salt.client
errors = ''
client = salt.client.LocalClient()
r1 = client.cmd('*', 'cmd.run_all',
['smbcontrol all reload-config'])
if r1:
for node, ret in r1.items():
# print ret
if ret["retcode"] != 0:
errors += "Error reloading samba on node %s " % node
if errors:
raise Exception(errors)
else:
cmd_to_run = 'smbcontrol all reload-config'
lines, err = command.get_command_output(cmd_to_run)
if err:
raise Exception(err)
ret, err = services_management.update_service_status(
'winbind', 'restart')
if err:
raise Exception(err)
except Exception, e:
return False, 'Error reloading CIFS configuration : %s' % str(e)
def kinit(user, pswd, realm):
try:
cmd_to_run = 'echo "%s" | kinit %s@%s' % (pswd, user, realm.upper())
# print cmd_to_run
use_salt, err = config.use_salt()
if err:
raise Exception(err)
if use_salt:
import salt.client
errors = []
client = salt.client.LocalClient()
# print 'Running %s'%cmd_to_run
#assert False
r1 = client.cmd('*', 'cmd.run_all', [cmd_to_run])
if r1:
for node, ret in r1.items():
# print ret
if ret["retcode"] != 0:
e = "Error initiating kerberos on GRIDCell %s" % node
if "stderr" in ret:
e += " : %s" % ret["stderr"]
errors.append(e)
print errors
# print r1
if errors:
raise Exception(' '.join(errors))
else:
lines, err = command.get_command_output(cmd_to_run, shell=True)
# print lines, err
if err:
raise Exception(err)
except Exception, e:
return False, 'Error initializing kerberos : %s' % str(e)
def delete_local_group(grpname):
try:
if not grpname:
raise Exception('No username specified')
d, err = get_local_group(grpname)
if not d:
if err:
raise Exception('Error locating group : %s' % err)
else:
raise Exception('Error locating group')
use_salt, err = config.use_salt()
if err:
raise Exception(err)
if use_salt:
import salt.client
client = salt.client.LocalClient()
rc = client.cmd('*', 'group.delete', [grpname])
# print rc
if rc:
for hostname, status in rc.items():
if not status:
raise Exception("Error deleting the system group")
else:
raise Exception("Error deleting the system group")
else:
cmd_to_run = 'groupdel %s' % (grpname)
lines, err = command.get_command_output(cmd_to_run)
if err:
raise Exception(err)
except Exception, e:
return False, 'Error deleting local group : %s' % str(e)
def get_certificate(cert_location):
"""Parse a certificate file and return a dict with required components"""
cert_info = {}
try:
if not os.path.exists(cert_location):
raise Exception('Could not locate certificate')
lines, err = command.get_command_output(
'openssl x509 -in %s -text -noout' % cert_location)
if err:
raise Exception(err)
look_for = [{'srch_for': 'Not Before', 'field_name': 'expiry_start'}, {'srch_for': 'Issuer', 'field_name': 'issuer'}, {'srch_for': 'Not After', 'field_name': 'expiry_end'}, {'srch_for': 'Subject', 'field_name': 'dn'}, {
'srch_for': 'Public Key Algorithm', 'field_name': 'key_algorithm'}, {'srch_for': 'Serial Number', 'field_name': 'serial_number'}, {'srch_for': 'Public-Key', 'field_name': 'key_length'}]
for line in lines:
for l in look_for:
pattern = '[\s]*%s[\s]*:[\s]*([\S\s]*)$' % l['srch_for']
ret = re.match(pattern, line)
if ret:
grps = ret.groups()
if grps:
cert_info[l['field_name']] = grps[0]
if 'key_length' in cert_info:
ret = re.match('[\s]*\(([\sa-zA-Z0-9]*)\)',
cert_info['key_length'])
if ret and ret.groups():
cert_info['key_length'] = ret.groups()[0]
except Exception, e:
return None, 'Error parsing certificate : %s' % str(e)
def flag_node(request):
try:
return_dict = {}
if "node" not in request.GET:
raise Exception("Error flagging node. No node specified")
node_name = request.GET["node"]
blink_time = 255
use_salt, err = config.use_salt()
if use_salt:
import salt.client
client = salt.client.LocalClient()
ret = client.cmd(node_name, 'cmd.run',
['ipmitool chassis identify %s' % (blink_time)])
print ret
if ret[node_name] == 'Chassis identify interval: %s seconds' % (
blink_time):
return django.http.HttpResponse("Success")
else:
raise Exception("")
else:
out_list, err = command.get_command_output(
'service winbind status')
if err:
raise Exception(err)
if 'Chassis identify interval: %s seconds' % (
blink_time) in out_list[0]:
return django.http.HttpResponse("Success")
else:
raise Exception("")
except Exception, e:
return django.http.HttpResponse("Error")
def volume_start_stop(vol_name, operation):
try:
cmd = 'gluster volume %s %s' % (operation, vol_name)
print cmd
if operation == 'stop':
(ret, rc), err = command.execute_with_conf_and_rc(cmd)
if rc == 0:
lines, er = command.get_output_list(ret)
if er:
raise Exception(er)
else:
err = ''
tl, er = command.get_output_list(ret)
if er:
raise Exception(er)
if tl:
err = ''.join(tl)
tl, er = command.get_error_list(ret)
if er:
raise Exception(er)
if tl:
err = err + ''.join(tl)
raise Exception(err)
else:
lines, err = command.get_command_output(cmd)
# print 'a', ret, err
if err:
raise Exception(err)
if lines:
print '\n'.join(lines)
except Exception, e:
return False, 'Error performing volume %s : %s' % (operation, e)
def check_salt_status(print_to_stdout=True, separator='\n'):
ret_list = []
try:
if print_to_stdout:
ret_list.append(
'IntegralStor Monitoring agents and processes status')
out, err = command.get_command_output('service salt-master status',
False)
if err:
raise Exception(err)
ret_list.append('Monitoring service master status : %s' %
','.join(out))
out, err = command.get_command_output('service salt-minion status',
False)
if err:
raise Exception(err)
ret_list.append('Monitoring service agent status : %s' % ','.join(out))
keys, err = grid_ops.get_all_salt_keys()
if err:
raise Exception(err)
if keys:
ret_list.append(separator +
'Admin agents that are part of the grid : ')
ret_list.append(separator.join(keys['minions']))
ret_list.append(
separator +
'Admin agents that are pending to be accepted into grid : ')
ret_list.append(separator.join(keys['minions_pre']))
results, err = grid_ops.salt_ping_all()
if err:
raise Exception(err)
if results:
ret_list.append(separator + 'Admin agent reachability :')
for minion, result in results.items():
if result:
st = 'Reachable'
else:
st = '**Not reachable**'
ret_list.append('%s : %s' % (minion, st))
if print_to_stdout:
print separator.join(ret_list)
except Exception, e:
return None, 'Error checking admin service status : %s' % str(e)
def unmount():
try:
ret, err = _header()
if err:
print err
mount_points, err = command.get_command_output(
"cat /var/log/usb-mount.track | cut -d ':' -f 1", shell=True)
if err:
raise Exception(err)
if not mount_points:
raise Exception(' Nothing to unmount!')
dev_names, err = command.get_command_output(
"cat /var/log/usb-mount.track | cut -d ':' -f 2", shell=True)
if err:
raise Exception(err)
count = []
for i, mount in enumerate(mount_points, start=1):
count.append(str(i))
print ' %s. %s of /dev/%s' % (i, mount, dev_names[i - 1])
print
ch = ''
while True:
ch = raw_input(
'Provide a number to unmount its corresponding device(0 to exit): ')
if ch != '0' and ch not in count:
print '\t- Provide a valid number'
else:
break
shell_scripts_path, err = config.get_shell_scripts_path()
if err:
raise Exception(err)
if ch != '0':
ret, err = command.get_command_output(
"/bin/bash %s/usb-mount.sh remove %s" % (shell_scripts_path, dev_names[int(ch) - 1]))
if err:
raise Exception(err)
for r in ret:
print '\n- %s' % r
except Exception, e:
print str(e)
raw_input('\nPress any key to continue')
def get_hardware_raid_hard_drives(controller_number, device_id):
disk_list = []
try:
lines, err = command.get_command_output(
'/opt/dell/srvadmin/sbin/omreport storage pdisk controller=%s vdisk=%d -fmt xml'
% (controller_number, device_id))
if err:
raise Exception(err)
if lines:
root = ElementTree.fromstring(''.join(lines))
sl = root.findall('ArrayDisks/DCStorageObject')
for s in sl:
disk_dict = {}
node = s.find('DeviceSerialNumber')
if node is not None:
disk_dict['serial_number'] = node.text
node = s.find('DiskProductVendor')
if node is not None:
disk_dict['vendor'] = node.text
node = s.find('ProductID')
if node is not None:
disk_dict['product_id'] = node.text
node = s.find('ManufactureDay')
if node is not None:
disk_dict['manufacture_day'] = node.text
node = s.find('ManufactureWeek')
if node is not None:
disk_dict['manufacture_week'] = node.text
node = s.find('ManufactureYear')
if node is not None:
disk_dict['manufacture_year'] = node.text
node = s.find('ObjStatus')
if node is not None:
try:
status = int(node.text)
if status == 2:
disk_dict['status'] = 'Ok'
elif status == 4:
disk_dict['status'] = 'Critical'
else:
disk_dict['status'] = 'Unknown'
except Exception, e:
pass
node = s.find('ObjState')
if node is not None:
try:
state = int(node.text)
if state == 4:
disk_dict['state'] = 'Online'
elif state == 8388608:
disk_dict['state'] = 'Rebuilding'
elif state == 1024:
disk_dict['state'] = 'Removed'
else:
disk_dict['state'] = 'Unknown'
except Exception, e:
pass
disk_list.append(disk_dict)
def check_integralview_processes_status(print_to_stdout=True, separator='\n'):
ret_list = []
try:
if print_to_stdout:
ret_list.append('GRIDCell IntegralView processes status: ')
out, err = command.get_command_output('service nginx status', False)
if err:
raise Exception(err)
ret_list.append('Web server (NGINX) status : %s' % ', '.join(out))
out, err = command.get_command_output('service uwsgi status', False)
if err:
raise Exception(err)
ret_list.append('Web application server (uwsgi) status : %s' %
', '.join(out))
if print_to_stdout:
print separator.join(ret_list)
except Exception, e:
return None, 'Error checking IntegralView process status: %s' % str(e)
def get_hardware_raid_hard_drives(controller_number, device_id):
disk_list = []
try:
lines, err = command.get_command_output(
'/opt/dell/srvadmin/sbin/omreport storage pdisk controller=%s vdisk=%d -fmt xml' % (controller_number, device_id))
if err:
raise Exception(err)
if lines:
root = ElementTree.fromstring(''.join(lines))
sl = root.findall('ArrayDisks/DCStorageObject')
for s in sl:
disk_dict = {}
node = s.find('DeviceSerialNumber')
if node is not None:
disk_dict['serial_number'] = node.text
node = s.find('DiskProductVendor')
if node is not None:
disk_dict['vendor'] = node.text
node = s.find('ProductID')
if node is not None:
disk_dict['product_id'] = node.text
node = s.find('ManufactureDay')
if node is not None:
disk_dict['manufacture_day'] = node.text
node = s.find('ManufactureWeek')
if node is not None:
disk_dict['manufacture_week'] = node.text
node = s.find('ManufactureYear')
if node is not None:
disk_dict['manufacture_year'] = node.text
node = s.find('ObjStatus')
if node is not None:
try:
status = int(node.text)
if status == 2:
disk_dict['status'] = 'Ok'
elif status == 4:
disk_dict['status'] = 'Critical'
else:
disk_dict['status'] = 'Unknown'
except Exception, e:
pass
node = s.find('ObjState')
if node is not None:
try:
state = int(node.text)
if state == 4:
disk_dict['state'] = 'Online'
elif state == 8388608:
disk_dict['state'] = 'Rebuilding'
elif state == 1024:
disk_dict['state'] = 'Removed'
else:
disk_dict['state'] = 'Unknown'
except Exception, e:
pass
disk_list.append(disk_dict)
def get_local_users(get_system_users=False):
user_list = []
try:
sys_ul = []
smb_ul = []
all = pwd.getpwall()
# print all
min_uid, err = config.get_minimum_user_uid_gid('user')
if err:
raise Exception(err)
for user in all:
if not get_system_users:
if user.pw_uid < min_uid or user.pw_uid >= 65534:
continue
sys_ul.append(user.pw_name)
d = {}
d['uid'] = user.pw_uid
d['gid'] = user.pw_gid
d['home_dir'] = user.pw_dir
d['username'] = user.pw_name
d['comment'] = user.pw_gecos
d['shell'] = user.pw_shell
g = grp.getgrgid(d['gid'])
if g:
d['grpname'] = g.gr_name
groups = [
g.gr_name for g in grp.getgrall()
if d['username'] in g.gr_mem and g.gr_gid != d['gid']
]
gid = pwd.getpwnam(d['username']).pw_gid
d['other_groups'] = groups
user_list.append(d)
ul, err = command.get_command_output("/usr/bin/pdbedit -d 1 -L")
if err:
raise Exception(err)
# print '1'
if ul:
smb_ul = []
smb_dict = {}
for u in ul:
l = u.split(':')
if l:
if len(l) > 1:
smb_dict['name'] = l[2]
smb_ul.append(l[0])
if user_list:
for ud in user_list:
if ud['username'] in smb_ul:
ud['smb_user'] = True
if ud['username'] in smb_dict:
ud['smb_comment'] = smb_dict[ud['username']]
else:
ud['smb_user'] = True
ud['smb_user'] = False
except Exception, e:
return None, 'Error retrieving local users : %s' % str(e)
def get_services_status():
"""The status of various services."""
s = {}
try:
platform, err = config.get_platform()
if err:
raise Exception(err)
if platform == 'gridcell':
# Commenting out ctdb for now as we wont use it for this release!
#services = ['smb', 'winbind', 'ctdb', 'glusterd']
services = ['smb', 'winbind', 'glusterd']
for service_name in services:
stat, err = command.get_command_output(
'/sbin/service %s status' % service_name)
ret = None
rc = -1
tup, err = command.execute_with_rc('/sbin/service %s status' %
service_name)
if tup:
(ret, rc) = tup
if err:
raise Exception(err)
if rc == 0:
lines, er = command.get_output_list(ret)
if er:
raise Exception(er)
s[service_name] = [0, ','.join(lines)]
else:
err = ''
tl, er = command.get_output_list(ret)
if er:
raise Exception(er)
if tl:
err = ','.join(tl)
tl, er = command.get_error_list(ret)
if er:
raise Exception(er)
if tl:
err = err + ','.join(tl)
s[service_name] = [-1, err]
else:
service_dict, err = services_management.get_sysd_services_status()
if err:
raise Exception(err)
for service_name, service_info in service_dict.items():
if service_info['info']['status']['status_str'] in [
'Failed', 'Unknown State'
]:
s[service_name] = [
-1, service_info['info']['status']['output_str']
]
else:
s[service_name] = [
0, service_info['info']['status']['output_str']
]
except Exception, e:
return None, 'Error retrieving services status: %s' % str(e)
def get_routes():
route_list = []
for c in command.get_command_output('route -n')[0][2:]:
if re.match('0.0.0.0', c):
# A simple hack to actually remove all the unnecessary white spaces
# and return the atual content as a list
route_list.append(" ".join(c.split()).split())
else:
pass
return route_list
def blink_unblink_disk(action, controller_id, channel_id, enclosure_id, target_id):
try:
cmd = '/opt/dell/srvadmin/sbin/omconfig storage pdisk action=%s controller=%s pdisk=%s:%s:%s' % (
action, controller_id, channel_id, enclosure_id, target_id)
# print cmd
lines, err = command.get_command_output(cmd)
if err:
raise Exception(err)
except Exception, e:
return False, 'Error blinking/un-blonking disk : %s' % e
def heal_info(vol_name):
try:
ret, err = command.get_command_output(
'gluster volume heal %s info split-brain' % vol_name)
if err:
raise Exception(err)
if ret:
print '\n'.join(ret)
except Exception, e:
return False, 'Error getting heal info : %s' % e
def get_hardware_raid_to_unix_mapping():
mapping_dict = {}
try:
lines, err = command.get_command_output(
'/opt/dell/srvadmin/sbin/omreport storage vdisk -fmt xml')
if err:
raise Exception(err)
if lines:
root = ElementTree.fromstring(''.join(lines))
sl = root.findall('VirtualDisks/DCStorageObject')
for s in sl:
vd_dict = {}
# print s
node = s.find('DeviceName')
if node is not None:
device_name = node.text
else:
print 'No device name for this virtual device'
continue
node = s.find('Layout')
if node is not None:
try:
layout = int(node.text)
if layout == 4:
vd_dict['layout'] = 'RAID-1'
except Exception, e:
pass
node = s.find('ObjStatus')
if node is not None:
try:
status = int(node.text)
if status == 2:
vd_dict['status'] = 'Ok'
elif status == 3:
vd_dict['status'] = 'Non-Critical'
except Exception, e:
pass
node = s.find('ObjState')
if node is not None:
try:
state = int(node.text)
if state == 1:
vd_dict['state'] = 'Ready'
elif state == 32:
vd_dict['state'] = 'Degraded'
except Exception, e:
pass
node = s.find('ControllerNum')
if node is not None:
try:
controller_number = int(node.text)
vd_dict['controller_number'] = controller_number
except Exception, e:
pass
def get_normal_users():
"""Gets all normal users, i.e. normal to the OS, and not the normal users that are specific to Integralstor.
"""
try:
ret, err = command.get_command_output(
"awk -F'[/:]' '{if ($3 >= 1000 && $3 < 65534) print $1}' /etc/passwd")
if err:
raise Exception(err)
except Exception, e:
return None, 'Could not fetch the list of normal users: %s' % str(e)
def blink_unblink_disk(action, controller_id, channel_id, enclosure_id,
target_id):
try:
cmd = '/opt/dell/srvadmin/sbin/omconfig storage pdisk action=%s controller=%s pdisk=%s:%s:%s' % (
action, controller_id, channel_id, enclosure_id, target_id)
# print cmd
lines, err = command.get_command_output(cmd)
if err:
raise Exception(err)
except Exception, e:
return False, 'Error blinking/un-blonking disk : %s' % e
def get_hardware_raid_to_unix_mapping():
mapping_dict = {}
try:
lines, err = command.get_command_output(
'/opt/dell/srvadmin/sbin/omreport storage vdisk -fmt xml')
if err:
raise Exception(err)
if lines:
root = ElementTree.fromstring(''.join(lines))
sl = root.findall('VirtualDisks/DCStorageObject')
for s in sl:
vd_dict = {}
# print s
node = s.find('DeviceName')
if node is not None:
device_name = node.text
else:
print 'No device name for this virtual device'
continue
node = s.find('Layout')
if node is not None:
try:
layout = int(node.text)
if layout == 4:
vd_dict['layout'] = 'RAID-1'
except Exception, e:
pass
node = s.find('ObjStatus')
if node is not None:
try:
status = int(node.text)
if status == 2:
vd_dict['status'] = 'Ok'
elif status == 3:
vd_dict['status'] = 'Non-Critical'
except Exception, e:
pass
node = s.find('ObjState')
if node is not None:
try:
state = int(node.text)
if state == 1:
vd_dict['state'] = 'Ready'
elif state == 32:
vd_dict['state'] = 'Degraded'
except Exception, e:
pass
node = s.find('ControllerNum')
if node is not None:
try:
controller_number = int(node.text)
vd_dict['controller_number'] = controller_number
except Exception, e:
pass
def delete_all_targets():
try:
cmd = 'tgt-admin --delete ALL --force'
lines, err = command.get_command_output(cmd)
if err:
ret_str = err
ret_str += '. '
if lines:
ret_str += '. '.join(lines)
raise Exception(ret_str)
except Exception, e:
return False, 'Error deleting all targets : %s' % str(e)
def get_services_status():
"""The status of various services."""
s = {}
try:
platform, err = config.get_platform()
if err:
raise Exception(err)
if platform == 'gridcell':
# Commenting out ctdb for now as we wont use it for this release!
#services = ['smb', 'winbind', 'ctdb', 'glusterd']
services = ['smb', 'winbind', 'glusterd']
for service_name in services:
stat, err = command.get_command_output(
'/sbin/service %s status' % service_name)
ret = None
rc = -1
tup, err = command.execute_with_rc(
'/sbin/service %s status' % service_name)
if tup:
(ret, rc) = tup
if err:
raise Exception(err)
if rc == 0:
lines, er = command.get_output_list(ret)
if er:
raise Exception(er)
s[service_name] = [0, ','.join(lines)]
else:
err = ''
tl, er = command.get_output_list(ret)
if er:
raise Exception(er)
if tl:
err = ','.join(tl)
tl, er = command.get_error_list(ret)
if er:
raise Exception(er)
if tl:
err = err + ','.join(tl)
s[service_name] = [-1, err]
else:
service_dict, err = services_management.get_sysd_services_status()
if err:
raise Exception(err)
for service_name, service_info in service_dict.items():
if service_info['info']['status']['status_str'] in ['Failed', 'Unknown State']:
s[service_name] = [-1, service_info['info']
['status']['output_str']]
else:
s[service_name] = [
0, service_info['info']['status']['output_str']]
except Exception, e:
return None, 'Error retrieving services status: %s' % str(e)
def get_normal_users():
"""Gets all normal users, i.e. normal to the OS, and not the normal users that are specific to Integralstor.
"""
try:
ret, err = command.get_command_output(
"awk -F'[/:]' '{if ($3 >= 1000 && $3 < 65534) print $1}' /etc/passwd"
)
if err:
raise Exception(err)
except Exception, e:
return None, 'Could not fetch the list of normal users: %s' % str(e)
def delete_local_user(username):
try:
if not username:
raise Exception('No username specified')
d, err = get_local_user(username)
if not d:
if err:
raise Exception('Error locating user : %s' % err)
else:
raise Exception('Error locating user')
use_salt, err = config.use_salt()
if err:
raise Exception(err)
if use_salt:
import salt.client
client = salt.client.LocalClient()
rc = client.cmd('*', 'user.delete', [username])
# print rc
if rc:
for hostname, status in rc.items():
if not status:
raise Exception("Error deleting the system user")
else:
raise Exception("Error deleting the system user")
else:
cmd_to_run = 'userdel %s' % (username)
lines, err = command.get_command_output(cmd_to_run)
if err:
raise Exception(err)
if d['smb_user']:
lines, err = command.get_command_output(r'pdbedit -d 1 -x %s' %
username)
if err:
raise Exception(err)
except Exception, e:
return False, 'Error deleting local user : %s' % str(e)
def display_status():
try:
hostname = socket.gethostname()
status, err = command.get_command_output('service salt-master status')
if err:
raise Exception(err)
print '\n'.join(status)
status, err = command.get_command_output('service salt-minion status')
if err:
raise Exception(err)
print '\n'.join(status)
status, err = command.get_command_output('service smb status')
if err:
raise Exception(err)
print '\n'.join(status)
status, err = command.get_command_output('service winbind status')
if err:
raise Exception(err)
print '\n'.join(status)
'''
#Commenting out as we wont support CTDB for this release
status, err = command.get_command_output('service ctdb status')
if err:
raise Exception(err)
print '\n'.join(status)
'''
status, err = command.get_command_output('service glusterd status')
if err:
raise Exception(err)
print '\n'.join(status)
'''
#Commenting out as we wont support CTDB for this release
status, err = command.get_command_output('ctdb status')
if err:
raise Exception(err)
print '\n'.join(status)
'''
except Exception, e:
return False, "Error displaying GRIDCell status : %s" % e
def generate_self_signed_certificate(d):
"""Generate a self signed cert with the parameters specified in the passed dict."""
try:
pki_dir, err = config.get_pki_dir()
if err:
raise Exception(err)
path = '%s/%s' % (pki_dir, d['name'])
if os.path.exists(path):
raise Exception('A key of that name already exists')
cmd = 'openssl req -new -newkey rsa:'
if 'key_length' in d:
key_length = int(d['key_length'])
else:
key_length = 1024
cmd = '%s%d' % (cmd, key_length)
if 'days' in d:
cmd = '%s -days %d' % (cmd, int(d['days']))
subj = ''
if 'country' in d:
subj = '%s/C=%s' % (subj, d['country'])
if 'state' in d:
subj = '%s/ST=%s' % (subj, d['state'])
if 'location' in d:
subj = '%s/L=%s' % (subj, d['location'])
if 'o' in d:
subj = '%s/O=%s' % (subj, d['o'])
if 'ou' in d:
subj = '%s/OU=%s' % (subj, d['ou'])
if 'cn' in d:
subj = '%s/CN=%s' % (subj, d['cn'])
if 'email' in d:
subj = '%s/emailAddress=%s' % (subj, d['email'])
cmd += ' -nodes -x509 -subj %s -keyout %s/%s.cert -out %s/%s.cert' % (
subj, path, d['name'], path, d['name'])
# print cmd
os.mkdir(path)
lines, err = command.get_command_output(cmd)
if err:
if os.path.exists(path):
shutil.rmtree(path)
raise Exception(err)
except Exception, e:
return False, 'Error generating self signed certificate : %s' % str(e)
def get_zfs_version():
zfs_ver = ""
try:
#lines, err = command.get_command_output('dmesg | grep ZFS', shell=True)
lines, err = command.get_command_output(
"modinfo zfs | grep -iw -e version | cut -d ':' -f 2", shell=True)
if err:
zfs_ver = 'Could not determine..'
if lines and lines[0]:
zfs_ver = 'ZFS build ' + lines[0].strip()
else:
zfs_ver = 'Could not determine..'
except Exception, e:
return None, 'Error retrieving ZFS information : %s' % str(e)
def get_idrac_addr():
url = None
try:
lines, err = command.get_command_output(
'/opt/dell/srvadmin/sbin/omreport system summary -fmt xml')
if err:
raise Exception(err)
if lines:
root = ElementTree.fromstring(''.join(lines))
sl = root.findall('OMA/EMPObjSummary/EMPObj/Url')
# print sl
if sl:
url = '%s/login.html' % sl[0].text
except Exception, e:
return None, 'Error retrieving IDRAC URL : %s' % e
def delete_local_user(userid):
"""Deletes a local user account in the linux system as well as in samba.
userid -- The username of the account
"""
try:
error_list = []
# First check if samba user exists. if so kick out
ul, err = get_local_users()
if err:
raise Exception(err)
found = False
if ul:
for ud in ul:
if ud["userid"] == userid:
found = True
if not found:
raise Exception("The user \"%s\" does not exist. " % userid)
# Now check if system user exists. If so and is created by integralstor
# then delete..
delete_system_user = False
try:
d = pwd.getpwnam(userid)
name = d[4]
if name.find("integralstor_user") == 0:
delete_system_user = True
except KeyError:
pass
if delete_system_user:
client = salt.client.LocalClient()
rc = client.cmd('*', 'user.delete', [userid])
for hostname, status in rc.items():
if not status:
error_list.append(
"Error deleting the userid on GRIDCell %s" % hostname)
if error_list:
raise Exception(','.join(error_list))
# Now delete from samba's database
lines, err = command.get_command_output(r'pdbedit -d 1 -x %s' % userid)
if err:
raise Exception(err)
except Exception, e:
return False, 'Error deleting local user :%s' % str(e)
def get_serial_number(disk_name):
"""Given a disk by name, get its serial number."""
serial_number = None
try:
"""
ids, err = get_all_disk_ids(disk_name)
if err:
raise Exception(err)
id1 = None
id = None
if ids:
for id1 in ids:
if (id1.startswith('scsi')) or (id1.startswith('ata')):
id = id1
break
if id:
parts = id.split('_')
if len(parts) >= 3:
serial_number = parts[2]
"""
#cmd = 'udevadm info -q property -n %s | grep -ie ID_SERIAL_SHORT | cut -d "=" -f2' % disk_name
cmd = 'udevadm info -q property -n %s | grep -ie ID_SCSI_SERIAL | cut -d "=" -f2' % disk_name
ret, err = command.get_command_output(cmd, shell=True)
if err:
raise Exception(err)
if ret:
serial_number = ret[0]
if not serial_number:
# Cant get it this way for some SAS drives so try smartctl
cmd_disk = "/usr/sbin/smartctl -H -i /dev/%s" % disk_name
dl_output = os.popen(cmd_disk).read()
lines = re.split("\r?\n", dl_output)
for line in lines:
# In case of a SAS drive, status comes with a different string
# so ..
res = re.match('Serial Number:[\s]*([\S]*)', line)
if not res:
# For SAS drives, we get a lower case number, *&^@*&^
res = re.match('Serial number:[\s]*([\S]*)', line)
if res:
tup = res.groups()
if tup:
serial_number = tup[0]
if serial_number:
serial_number = serial_number.upper()
except Exception, e:
return None, "Error retrieving serial number : %s" % str(e)
def get_psu_status():
return_dict = {}
try:
lines, err = command.get_command_output(
'/opt/dell/srvadmin/sbin/omreport chassis pwrsupplies -fmt xml')
if err:
raise Exception(err)
if lines:
# print lines
root = ElementTree.fromstring(''.join(lines))
sl = root.findall('Chassis/Redundancy/RedunStatus')
for s in sl:
redundancy_str = s.text
if redundancy_str == '4':
return_dict['redundancy'] = True
else:
return_dict['redundancy'] = False
psus_info = root.find('Chassis/PowerSupplyList')
if psus_info is not None:
if 'count' in psus_info.attrib:
return_dict['psu_count'] = int(psus_info.attrib['count'])
psus = root.findall('Chassis/PowerSupplyList/PowerSupply')
# print psus
psu_list = []
for psu in psus:
switch_dict = {}
switch_on = psu.find('PSSwitchOn')
if switch_on is not None:
if switch_on.text == 'true':
switch_dict['switch_on'] = True
else:
switch_dict['switch_on'] = False
prescence = psu.find('PSState/PSPresenceDetected')
if prescence is not None:
if prescence.text == 'true':
switch_dict['prescence'] = True
else:
switch_dict['prescence'] = False
failure = psu.find('PSState/PSFailureDetected')
if failure is not None:
if failure.text == 'true':
switch_dict['failure'] = True
else:
switch_dict['failure'] = False
psu_list.append(switch_dict)
return_dict['psu_list'] = psu_list
except Exception, e:
return None, 'Error retrieving PSU information : %s' % e
def get_alert_logs():
alerts_odict = None
try:
alerts_dict = {}
lines, err = command.get_command_output(
'/opt/dell/srvadmin/sbin/omreport system alertlog -fmt xml')
if err:
raise Exception(err)
if lines:
root = ElementTree.fromstring(''.join(lines))
sl = root.findall('LogEntry')
for s in sl:
# print s
alert_dict = {}
timestamp = None
node = s.find('TimeStamp')
if node is not None:
timestamp = int(node.text)
alert_dict['timestamp'] = timestamp
node = s.find('DateTime')
if node is not None:
alert_dict['date_time'] = node.text
node = s.find('Description')
if node is not None:
alert_dict['description'] = node.text
node = s.find('Category')
if node is not None:
alert_dict['category'] = node.text
node = s.find('ID')
if node is not None:
id = int(node.text)
alert_dict['id'] = id
node = s.find('Type')
if node is not None:
type = int(node.text)
if type == 2:
alert_dict['Severity'] = 'Non-Critical'
elif type == 1:
alert_dict['Severity'] = 'Critical'
elif type == 4:
alert_dict['Severity'] = 'Ok'
if not timestamp in alerts_dict:
alerts_dict[timestamp] = []
alerts_dict[timestamp].append(alert_dict)
alerts_odict = collections.OrderedDict(
sorted(alerts_dict.items(), reverse=True))
except Exception, e:
return None, 'Error retrieving alerts : %s' % e
def get_current_timezone():
return_dict = {}
try:
cmd = "timedatectl | grep 'Time' | grep -oE '(\S+/.+$)'"
system_timezone, err = command.get_command_output(cmd, shell=True)
if err:
raise Exception(err)
return_dict['system_timezone'] = system_timezone[0]
path = os.path.realpath('/etc/localtime')
components = path.split('/')
if components:
return_dict['timezone_str'] = '%s/%s' % (
components[-2], components[-1])
else:
raise Exception('Unknown timezone')
except Exception, e:
return None, str(e)
def delete_ace(path, name, type, recursive=False):
"""Remove the ACE with the given name and type for the given path."""
try:
if not path or not name or not type:
raise Exception('Required information not passed')
if type not in ['user', 'group']:
raise Exception('Invalid type passed')
if recursive:
cmd = 'setfacl -R -x %s:"%s",default:%s:"%s" "%s"' % (
type, name, type, name, path)
else:
cmd = 'setfacl -x %s:"%s",default:%s:"%s" "%s"' % (
type, name, type, name, path)
# print cmd
lines, err = command.get_command_output(cmd)
if err:
raise Exception(err)
except Exception, e:
return False, 'Error removing ACL : %s' % str(e)
def get_status():
"""Return the CTDB status as a dict with the node name as the key and the status as the value."""
status_dict = {}
try:
lines, err = command.get_command_output('ctdb -Y status')
if err:
raise Exception(err)
i = 0
for line in lines:
if i == 0:
i += 1
continue
# print line
components = line.split(':')
# print components
ok = True
status = ''
if components:
if components[3] != '0':
status += '| Banned |'
ok = False
if components[4] != '0':
status += '| Disabled |'
ok = False
if components[5] != '0':
status += '| Unhealthy |'
ok = False
if components[6] != '0':
status += '| Stopped |'
ok = False
if components[7] != '0':
status += '| Inactive |'
ok = False
if components[8] != '0':
status += '| PartiallyOnline |'
ok = False
if ok:
status += 'OK'
status_dict[components[2]] = status
i += 1
except Exception, e:
return None, "Error retrieving CTDB status: %s" % str(e)
def get_mdraid_info(device):
"""Get the devices that are part of the mdraids."""
raid_list = []
try:
lines, err = command.get_command_output(
'mdadm --detail /dev/%s' % device)
if not err:
for line in lines:
# print line
if line.startswith('/dev/%s' % device):
continue
if not line.strip():
continue
if not ':' in line:
# print line
comps = line.split()
if len(comps) > 5:
raid_list.append(comps[-1][5:])
except Exception, e:
return None, "Error getting mdraid information: %s" % str(e)
def create_ace_entries(path, users, groups, recursive):
"""Create an ACE entry for the given path with the supplied users and groups."""
try:
if recursive:
cmd = r'setfacl -R -m'
else:
cmd = r'setfacl -m '
for user in users:
cmd += 'u:"%s":---,d:u:"%s":---,' % (user, user)
for group in groups:
cmd += 'g:"%s":---,d:g:"%s":---,' % (group, group)
if cmd[-1] == ',':
cmd = cmd[:-1]
cmd += ' "%s"' % path
# print cmd
lines, err = command.get_command_output(cmd)
if err:
raise Exception(err)
except Exception, e:
return False, 'Error adding ACL entries : %s' % str(e)
def get_all_aces(path):
"""Get all the access control entries for the given path."""
aces = []
try:
lines, err = command.get_command_output('getfacl "%s"' % path)
if err:
raise Exception(err)
for line in lines:
if not line.strip():
# print 'blank line so skipping'
continue
if line.strip()[0] == '#':
# print 'comment line so skipping : ',line
continue
# print 'normal line : ', line
if line.find('::'):
fields = line.strip().split(':')
aces.append(fields)
# print aces
except Exception, e:
return None, 'Error retrieving all ACLs : %s' % str(e)
def net_ads_join(user, pswd, password_server):
"""Join all gridcells to the AD realm
user -- AD's admin username
pswd -- AD's admin password
password_server -- The AD server
Returns True on success and False otherwise
"""
try:
# print 'aaa'
errors = []
client = salt.client.LocalClient()
#cmd_to_run = "net ads join -S %s -U %s%%%s"%(password_server, user, pswd)
cmd_to_run = "net ads join -U %s%%%s" % (user, pswd)
print 'Running %s' % cmd_to_run
output, err = command.get_command_output(cmd_to_run)
print output, err
if err:
raise Exception(err)
'''
r1 = client.cmd('gridcell-pri.integralstor.lan', 'cmd.run_all', [cmd_to_run])
#print r1
if r1:
for node, ret in r1.items():
print ret
if ret["retcode"] != 0:
e = "Error joining AD on GRIDCell %s"%node
if "stderr" in ret:
e += " : %s"%ret["stderr"]
errors.append(e)
#print errors
#print r1
if errors:
raise Exception(','.join(errors))
'''
except Exception, e:
return False, 'Error joining AD : %s' % str(e)
def ret_fs_usage_alert(node_name, threshold_percent=85):
ret = []
try:
threshold_percent = int(threshold_percent)
lines, err = command.get_command_output("salt '%s' disk.percent" % node_name)
if err:
raise Exception(err)
if lines:
data = [line.strip() for line in lines[2:]]
s = ''
for idx, line in enumerate(data):
if ':' in line:
s = 'Used percentage of %s' % line
if data[idx+1] and int(data[idx+1].split('%')[0]) >= threshold_percent:
s = '%s %s ' % (s, data[idx+1])
ret.append(s)
else:
s = ''
else:
continue
except Exception, e:
return None, 'Could not determine filesystem usage: %s' % e
def get_local_users():
"""Get the username, real name of all the local users in the system
userid -- The username of the account
pswd -- The password for the username
"""
user_list = None
try:
ul, err = command.get_command_output("/usr/bin/pdbedit -d 1 -L")
if err:
raise Exception(err)
user_list = []
for u in ul:
l = u.split(':')
if l:
d = {}
d["userid"] = l[0]
if len(l) > 1:
d["name"] = l[2]
user_list.append(d)
except Exception, e:
return None, 'Error retrieving local user list :%s' % str(e)
def sync_ctdb_files():
"""
Syncs CTDB files on the localhost with the mounted admin vol.
Input:
['None']
Output:
Returns three variables [ret1,ret2,ret3]:
ret1 -- True if sync was sucessful, else False
ret2 -- True if there was a difference, False if they were already in sync
ret3 -- None if there were no errors/exceptions, 'Error string' otherwise
"""
is_change = False
try:
config_dir, err = config.get_config_dir()
if err:
raise Exception(err)
out, err = command.get_command_output(
'diff "/etc/sysconfig/ctdb" "%s/lock/ctdb"' % str(config_dir), True, True)
if err:
shutil.copyfile('%s/lock/ctdb' %
str(config_dir), '/etc/sysconfig/ctdb')
is_change = True
out, err = command.get_command_output(
'diff "/etc/ctdb/nodes" "%s/lock/nodes"' % str(config_dir), True, True)
if err:
shutil.copyfile('%s/lock/nodes' %
str(config_dir), '/etc/ctdb/nodes')
is_change = True
out, err = command.get_command_output(
'diff "/etc/ctdb/public_addresses" "%s/lock/public_addresses"' % str(config_dir), True, True)
if err:
shutil.copyfile('%s/lock/public_addresses' %
str(config_dir), '/etc/ctdb/public_addresses')
is_change = True
lg, err = logger.get_script_logger(
'Admin volume mounter: Sync CTDB config files', '/var/log/integralstor/scripts.log', level=logging.DEBUG)
if is_change == True:
logger.log_or_print(
'ctdb related files were synced.', lg, level='info')
logger.log_or_print(
'Restarting ctdb.', lg, level='debug')
out, err = command.get_command_output(
'service ctdb restart', False, True)
if not err and out:
logger.log_or_print(
'Service ctdb: %s' % out, lg, level='debug')
else:
logger.log_or_print(
'Service ctdb error: %s' % err, lg, level='error')
elif is_change == False:
logger.log_or_print(
'ctdb related files are in sync.', lg, level='info')
except Exception, e:
return False, is_change, "Couldn't sync ctdb files: %s" % str(e)
def mount_and_configure():
lg = None
try:
lg, err = logger.get_script_logger(
'Admin volume mounter', '/var/log/integralstor/scripts.log', level=logging.DEBUG)
logger.log_or_print(
'Admin volume mounter initiated.', lg, level='info')
pog, err = grid_ops.is_part_of_grid()
if err:
raise Exception(err)
# If the localhost is part of the gridcell, proceed
if pog:
logger.log_or_print('Checking glusterd service', lg, level='debug')
service = 'glusterd'
status, err = services_management.get_service_status([service])
if err:
raise Exception(err)
logger.log_or_print('Service %s status is %s' % (
service, status['status_code']), lg, level='debug')
if status['status_code'] != 0:
logger.log_or_print(
'Service %s not started so restarting' % service, lg, level='error')
out, err = command.get_command_output(
'service %s restart' % service, False, True)
if not err and out:
logger.log_or_print('Service %s: %s' %
(service, out), lg, level='debug')
else:
logger.log_or_print('Service %s error : %s' % (
service, err), lg, level='error')
admin_vol_name, err = config.get_admin_vol_name()
if err:
raise Exception(err)
# Get the config dir - the mount point.
config_dir, err = config.get_config_dir()
if err:
raise Exception(err)
ag, err = grid_ops.is_admin_gridcell()
if err:
raise Exception(err)
admin_gridcells, err = grid_ops.get_admin_gridcells()
if err:
raise Exception(err)
is_pooled = False
peer_list, err = gluster_trusted_pools.get_peer_list()
if peer_list:
is_pooled = True
is_mounted = False
# mount only if the localhost is pooled
if is_pooled:
is_mounted, err = grid_ops.is_admin_vol_mounted_local()
if not is_mounted:
str = 'Admin volume is not mounted. Will attempt to mount now.'
logger.log_or_print(str, lg, level='error')
# Try to mount
(ret, rc), err = command.execute_with_rc(
'mount -t glusterfs localhost:/%s %s' % (admin_vol_name, config_dir))
if err:
str = 'Mount from localhost failed.'
logger.log_or_print(str, lg, level='error')
elif (not err) and (rc == 0):
is_access, err = assert_admin_vol_mount()
if err:
raise Exception(err)
sync, is_change, error = sync_ctdb_files()
if error:
# It's only a best-effort, it will try next
# minute again.
pass
if sync == False:
#raise Exception (err)
pass
# Restart nginx
out, err = command.get_command_output(
'service nginx restart', False, True)
if not err and out:
logger.log_or_print(
'Service nginx: %s' % out, lg, level='debug')
else:
logger.log_or_print(
'Service nginx error : %s' % err, lg, level='error')
# Restart uwsgi
out, err = command.get_command_output(
'service uwsgi restart', False, True)
if not err and out:
logger.log_or_print(
'Service uwsgi: %s' % out, lg, level='debug')
else:
logger.log_or_print(
'Service uwsgi error : %s' % err, lg, level='error')
if ag:
# Restart salt-master
out, err = command.get_command_output(
'service salt-master restart', False, True)
if not err and out:
logger.log_or_print(
'Service salt-master: %s' % out, lg, level='debug')
else:
logger.log_or_print(
'Service salt-master error : %s' % err, lg, level='error')
# Restart salt-minion
out, err = command.get_command_output(
'service salt-minion restart', False, True)
if not err and out:
logger.log_or_print(
'Service salt-minion: %s' % out, lg, level='debug')
else:
logger.log_or_print(
'Service salt-minion error : %s' % err, lg, level='error')
str = 'Admin vol is mounted'
logger.log_or_print(str, lg, level='info')
# Admin volume is mounted, perform required checks
else:
sync, is_change, err = sync_ctdb_files()
if err:
raise Exception(err)
if sync == False:
raise Exception(err)
logger.log_or_print('Checking services', lg, level='debug')
service_list = ['nginx','ctdb','salt-minion']
if ag:
service_list.append('salt-master')
for service in service_list:
status, err = services_management.get_service_status([
service])
if err:
raise Exception(err)
logger.log_or_print('Service %s status is %s' % (
service, status['status_code']), lg, level='debug')
if status['status_code'] != 0:
logger.log_or_print(
'Service %s is not active, restarting' % service, lg, level='error')
out, err = command.get_command_output(
'service %s restart' % service, False, True)
if not err and out:
logger.log_or_print('Service %s: %s' % (
service, out), lg, level='debug')
else:
logger.log_or_print('Service %s error : %s' % (
service, err), lg, level='error')
# UWSGI service config not complete so need to check
# against the actual process name
(ret, rc), err = command.execute_with_rc(
'pidof uwsgi', shell=True)
if rc != 0:
logger.log_or_print(
'Service uwsgi is not active, restarting', lg, level='error')
out, err = command.get_command_output(
'service uwsgi restart', False, True)
if not err and out:
logger.log_or_print(
'Service uwsgi: %s' % out, lg, level='debug')
else:
logger.log_or_print(
'Service uwsgi error : %s' % err, lg, level='error')
str = 'Admin volume is already mounted'
logger.log_or_print(str, lg, level='info')
except Exception, e:
st = 'Error mounting admin volume : %s' % e
logger.log_or_print(st, lg, level='critical')
return False, st
