checks/cisco_qos

   1 #!/usr/bin/python
   2 # -*- encoding: utf-8; py-indent-offset: 4 -*-
   3 # +------------------------------------------------------------------+
   4 # |             ____ _               _        __  __ _  __           |
   5 # |            / ___| |__   ___  ___| | __   |  \/  | |/ /           |
   6 # |           | |   | '_ \ / _ \/ __| |/ /   | |\/| | ' /            |
   7 # |           | |___| | | |  __/ (__|   <    | |  | | . \            |
   8 # |            \____|_| |_|\___|\___|_|\_\___|_|  |_|_|\_\           |
   9 # |                                                                  |
  10 # | Copyright Mathias Kettner 2014             mk@mathias-kettner.de |
  11 # +------------------------------------------------------------------+
  12 #
  13 # This file is part of Check_MK.
  14 # The official homepage is at http://mathias-kettner.de/check_mk.
  15 #
  16 # check_mk is free software;  you can redistribute it and/or modify it
  17 # under the  terms of the  GNU General Public License  as published by
  18 # the Free Software Foundation in version 2.  check_mk is  distributed
  19 # in the hope that it will be useful, but WITHOUT ANY WARRANTY;  with-
  20 # out even the implied warranty of  MERCHANTABILITY  or  FITNESS FOR A
  21 # PARTICULAR PURPOSE. See the  GNU General Public License for more de-
  22 # tails. You should have  received  a copy of the  GNU  General Public
  23 # License along with GNU Make; see the file  COPYING.  If  not,  write
  24 # to the Free Software Foundation, Inc., 51 Franklin St,  Fifth Floor,
  25 # Boston, MA 02110-1301 USA.
  26
  27 # Author: Lars Michelsen <lm@mathias-kettner.de>
  28
  29 # Relevant SNMP OIDs:
  30 # .1.3.6.1.4.1.9.9.166.1.1.1.1.4.144 9
  31 # .1.3.6.1.4.1.9.9.166.1.1.1.1.4.258 16
  32 # .1.3.6.1.4.1.9.9.166.1.1.1.1.4.400 25
  33 #
  34 # .1.3.6.1.4.1.9.9.166.1.6.1.1.1.3974704 "6cos"
  35 # .1.3.6.1.4.1.9.9.166.1.6.1.1.1.6208592 "ingress-map"
  36 #
  37 # .1.3.6.1.4.1.9.9.166.1.7.1.1.1.1593 "class-default"
  38 # .1.3.6.1.4.1.9.9.166.1.7.1.1.1.18785 "EF"
  39 # .1.3.6.1.4.1.9.9.166.1.7.1.1.1.284945 "AF1"
  40 # .1.3.6.1.4.1.9.9.166.1.7.1.1.1.284961 "AF2"
  41 # .1.3.6.1.4.1.9.9.166.1.7.1.1.1.284977 "AF3"
  42
  43 # http://www.oidview.com/mibs/9/CISCO-CLASS-BASED-QOS-MIB.html
  44
  45 # TEST:
  46 #
  47 # search class table:
  48 # .1.3.6.1.4.1.9.9.166.1.7.1.1.1.284945 (cbQosCMName) "AF1"
  49 # class_id = 284945 (cbQosConfigIndex)
  50 #
  51 # search config table for matching value
  52 # .1.3.6.1.4.1.9.9.166.1.5.1.1.2.144.5256 284945
  53 # key = 144.5256 (cbQosPolicyIndex: 144, cbQosObjectsIndex: 5256)
  54 #
  55 # search if table for matchin if_id: 144
  56 # .1.3.6.1.4.1.9.9.166.1.1.1.1.4.144 (cbQosIfIndex) 9
  57 # if_policy = 9 (ifIndex -> standard mib)
  58 #
  59 # get config_id from config table using if_id.if_id 144.144
  60 # .1.3.6.1.4.1.9.9.166.1.5.1.1.2.144.144 (cbQosConfigIndex) 6208592
  61 # config_index = 6208592
  62 #
  63 # get policy name using the policy_index
  64 # .1.3.6.1.4.1.9.9.166.1.6.1.1.1.6208592 "ingress-map"
  65 # policy_name = "ingress-map"
  66 #
  67 # get post bytes using the key
  68 # .1.3.6.1.4.1.9.9.166.1.15.1.1.9.144.5256 0
  69 # post_bytes = 0
  70 #
  71 # get dropped bytes using the key
  72 # .1.3.6.1.4.1.9.9.166.1.15.1.1.16.144.5256 0
  73 # dropped_bytes = 0
  74 #
  75 # get if_name using the if_policy: 9
  76 # .1.3.6.1.2.1.31.1.1.1.1.9 Vl1
  77 # if_name = Vl1
  78 #
  79 # get if_speed using the if_policy: 9
  80 # .1.3.6.1.2.1.2.2.1.5.9 100000000
  81 # if_speed = 100000000
  82 #
  83 ###
  84 # Test to find the badwidth of the classes. Not finished...
  85 #
  86 # 'cbQosObjectsType' => {
  87 #         1 => 'policymap',
  88 #         2 => 'classmap',
  89 #         3 => 'matchStatement',
  90 #         4 => 'queueing',
  91 #         5 => 'randomDetect',
  92 #         6 => 'trafficShaping',
  93 #         7 => 'police',
  94 #         8 => 'set' },
  95 #
  96 # Index:
  97 # .1.3.6.1.4.1.9.9.166.1.5.1.1.2.258.1244739 1608
  98 #
  99 # Type:
 100 # .1.3.6.1.4.1.9.9.166.1.5.1.1.3.258.1244739 4
 101 #
 102 # Parent ID:
 103 # .1.3.6.1.4.1.9.9.166.1.5.1.1.4.258.1244739 6184
 104 #
 105 # cbQosQueueingStatsEntry:
 106 # .1.3.6.1.4.1.9.9.166.1.18.1.1.2.258.1244739 64
 107 # ...
 108
 109 # Index:
 110 # .1.3.6.1.4.1.9.9.166.1.5.1.1.2.258.6184 18785
 111 # Type:
 112 # .1.3.6.1.4.1.9.9.166.1.5.1.1.3.258.6184 2
 113 # Parent ID:
 114 # .1.3.6.1.4.1.9.9.166.1.5.1.1.4.258.6184 258
 115
 116 # get cbQosQueueingCfgBandwidth
 117 # .1.3.6.1.4.1.9.9.166.1.9.1.1.1.1608 3094
 118
 119 cisco_qos_default_levels = (None, None, 0.01, 0.01)
 120
 121 #factory_settings["cisco_qos_default_values"] = {
 122 #    "drop" : (0.01, 0.01)
 123 #}
 124
 125
 126 def cisco_qos_get_config_entries_by_class_id(config, class_id):
 127     return [if_index.split('.') for if_index, value in config.iteritems() if value == class_id]
 128
 129
 130 def inventory_cisco_qos(info):
 131     if len(info) == 12:
 132         ifs = dict(info[0])
 133         config = dict([('.'.join(oid.split('.')[-2:]), value) for oid, value in info[3]])
 134         if_names = dict(info[6])
 135
 136         # Find all interfaces for each class and create one service for each pair
 137         items = []
 138         for class_id, class_name in info[2]:
 139             # Get interface ids which use this qos class
 140             for policy_id, _objects_id in cisco_qos_get_config_entries_by_class_id(
 141                     config, class_id):
 142                 if ifs.get(policy_id) in if_names:
 143                     if_name = if_names[ifs[policy_id]]
 144                     items += [('%s: %s' % (if_name, class_name), {})]
 145
 146         return items
 147
 148
 149 def check_cisco_qos(item, params, info):
 150     # Convert old params definitions
 151     # Note: the float values of the post levels are converted to int
 152     had_legacy_params = False
 153     if isinstance(params, tuple):
 154         params = {
 155             "post": tuple(int(x) if x else None for x in params[0:2]),
 156             "drop": params[2:4],
 157         }
 158         had_legacy_params = True
 159
 160     unit = params.get("unit", "bit")
 161     average = params.get("average")
 162     post_warn, post_crit = params.get("post", (None, None))
 163     drop_warn, drop_crit = params.get("drop", (None, None))
 164
 165     # Load values and format them
 166     ifs = dict(info[0])
 167     policies = dict(info[1])
 168     classes = dict(info[2])
 169     config = dict([('.'.join(oid.split('.')[-2:]), value) for oid, value in info[3]])
 170     post_bytes = dict([('.'.join(oid.split('.')[-2:]), value) for oid, value in info[4]])
 171     drop_bytes = dict([('.'.join(oid.split('.')[-2:]), value) for oid, value in info[5]])
 172     if_names = dict(info[6])
 173     if_speeds = dict(info[7])
 174     parents = dict(info[8])
 175     if_qos_bandwidth = dict(info[9])
 176     if_qos_bandwidth_units = dict(info[10])
 177     object_types = dict(info[11])
 178
 179     if_name, class_name = item.split(': ')
 180
 181     # Gather the class id by class_name
 182     class_id = None
 183     for cid, cname in classes.iteritems():
 184         if class_name == cname:
 185             class_id = cid
 186             break
 187
 188     # Gather the interface id by class_name
 189     if_id = None
 190     for iid2 in ifs.itervalues():
 191         if if_name == if_names.get(iid2):
 192             if_id = iid2
 193             break
 194
 195     if not if_id or not class_id:
 196         return (3, "QoS class not found for that interface")
 197
 198     policy_id, objects_id, policy_map_id, policy_name = None, None, None, None
 199     for this_policy_id, this_objects_id in cisco_qos_get_config_entries_by_class_id(
 200             config, class_id):
 201         if if_id != ifs[this_policy_id]:
 202             continue  # skip the ones of other interfaces
 203
 204         # Get the policy_map_id. To retrieve this get one of the config entries
 205         # of type "policy map" from the config table. All of this type should have
 206         # the same value, which is then the policy_map_id.
 207         for key in object_types.keys():
 208             if key.startswith(this_policy_id + '.') and object_types[key] == '1':
 209                 policy_map_id = config[key]
 210                 break
 211
 212         if policy_map_id is None:
 213             return 3, 'Invalid policy map id'
 214
 215         policy_name = policies.get(policy_map_id)
 216         policy_id = this_policy_id
 217         objects_id = this_objects_id
 218
 219     if policy_id is None or objects_id is None:
 220         return 3, 'Could not find policy_id or objects_id'
 221
 222     post_b = post_bytes.get(policy_id + '.' + objects_id, 0)
 223     drop_b = drop_bytes.get(policy_id + '.' + objects_id, 0)
 224     speed = saveint(if_speeds[if_id])
 225
 226     parent_value_cache = {}
 227     for a_key, a_value in config.items():
 228         parent_value_cache.update({a_value: a_key.split(".")[1]})
 229
 230     # if a_value == class_id:
 231     #     parent_value = a_key.split(".")[1]
 232     for b_key, b_value in parents.items():
 233         if parent_value_cache[class_id] == b_value:
 234             if object_types[b_key] == "4":
 235                 try:
 236                     # 1 kbps
 237                     # 2 percentage
 238                     # 3 percentageRemaining
 239                     # 4 ratioRemaining
 240                     # 5 perThousand
 241                     # 6 perMillion
 242                     qos_unit = int(if_qos_bandwidth_units[config[b_key]])
 243                     qos_bandwidth = saveint(if_qos_bandwidth[config[b_key]])
 244                     if qos_unit == 1:
 245                         speed = qos_bandwidth * 1000
 246
 247                     elif qos_unit == 2:
 248                         speed = speed * (qos_bandwidth / 100.0)
 249
 250                     elif qos_unit == 3:
 251                         remaining = (speed / 100) * qos_bandwidth
 252                         speed = speed - remaining
 253                     break
 254                 except KeyError:
 255                     pass
 256
 257     # Bandwidth needs to be in bytes for later calculations
 258     bw = speed / 8.0
 259
 260     # Determine post warn/crit levels
 261     if isinstance(post_warn, float):
 262         post_warn = bw / 100.0 * post_warn
 263         post_crit = bw / 100.0 * post_crit
 264     elif isinstance(post_warn, int):
 265         if unit == 'bit':
 266             post_warn = post_warn / 8
 267             post_crit = post_crit / 8
 268
 269     # Determine drop warn/crit levels
 270     if isinstance(drop_warn, float) and not had_legacy_params:
 271         drop_warn = bw / 100.0 * drop_warn
 272         drop_crit = bw / 100.0 * drop_crit
 273     # Convert the drop levels to byte
 274     elif unit == "bit":
 275         # But only if our params where already provided in the new format
 276         if not had_legacy_params:
 277             if isinstance(drop_warn, int):
 278                 drop_warn = drop_warn / 8.0
 279             if isinstance(drop_crit, int):
 280                 drop_crit = drop_crit / 8.0
 281
 282     # Handle counter values
 283     state = 0
 284     infotext = ''
 285     this_time = time.time()
 286     rates = []
 287     perfdata = []
 288     perfdata_avg = []
 289
 290     min_value = "0.0" if unit == "bit" else "0"
 291     for name, counter, warn, crit, min_val, max_val in [
 292         ("post", post_b, post_warn, post_crit, min_value, bw),
 293         ("drop", drop_b, drop_warn, drop_crit, min_value, bw),
 294     ]:
 295         rate = get_rate("cisco_qos.%s.%s" % (name, item), this_time, saveint(counter))
 296         rates.append(rate)
 297         perfdata.append((name, rate, warn, crit, min_val, max_val))
 298
 299         if average:
 300             avg_value = get_average("cisco_qos.%s.%s.avg" % (name, item), this_time, rate, average)
 301             rates.append(avg_value)
 302             perfdata_avg.append(("%s_avg_%d" % (name, average), avg_value, warn, crit, min_val,
 303                                  max_val))
 304
 305     perfdata.extend(perfdata_avg)
 306
 307     def format_value(value):
 308         if unit == "bit":
 309             value = value * 8
 310             return get_nic_speed_human_readable(value)
 311         return "%s/s" % get_bytes_human_readable(value)
 312
 313     if average:
 314         post_rate = rates[1]
 315         drop_rate = rates[3]
 316     else:
 317         post_rate = rates[0]
 318         drop_rate = rates[1]
 319
 320     for what, rate, warn, crit in [("post", post_rate, post_warn, post_crit),
 321                                    ("drop", drop_rate, drop_warn, drop_crit)]:
 322         infotext += ', %s: %s' % (what, format_value(rate))
 323         if crit is not None and rate >= crit:
 324             state = max(2, state)
 325             infotext += '(!!)'
 326         elif warn is not None and rate >= warn:
 327             state = max(1, state)
 328             infotext += '(!)'
 329
 330     if policy_name:
 331         infotext += ', Policy-Name: %s, Int-Bandwidth: %s' % (policy_name, format_value(bw))
 332     else:
 333         infotext += ', Policy-Map-ID: %s, Int-Bandwidth: %s' % (policy_map_id, format_value(bw))
 334     return (state, infotext.lstrip(', '), perfdata)
 335
 336
 337 check_info["cisco_qos"] = {
 338     "service_description": "QoS %s",
 339     "check_function": check_cisco_qos,
 340     "inventory_function": inventory_cisco_qos,
 341     "has_perfdata": True,
 342     "group": "cisco_qos",
 343     "default_levels_variable": "cisco_qos_default_"
 344 }
 345
 346 snmp_info['cisco_qos'] = [
 347     ('.1.3.6.1.4.1.9.9.166.1', [OID_END, '1.1.1.4']),  # qosIfIndex
 348     ('.1.3.6.1.4.1.9.9.166.1', [OID_END, '6.1.1.1']),  # qosPolicies
 349     ('.1.3.6.1.4.1.9.9.166.1', [OID_END, '7.1.1.1']),  # qosClasses
 350     ('.1.3.6.1.4.1.9.9.166.1', [OID_STRING, '5.1.1.2']),  # qosConfig
 351     ('.1.3.6.1.4.1.9.9.166.1', [OID_STRING, '15.1.1.9']),  # qosPostBytes
 352     ('.1.3.6.1.4.1.9.9.166.1', [OID_STRING, '15.1.1.16']),  # qosDropBytes
 353     ('.1.3.6.1.2.1.2.2.1', [OID_END, '2']),  # ifNames
 354     ('.1.3.6.1.2.1.2.2.1', [OID_END, '5']),  # ifSpeeds
 355     ('.1.3.6.1.4.1.9.9.166.1', [OID_END, '5.1.1.4']),  # cbQosParentObjectsIndex
 356     ('.1.3.6.1.4.1.9.9.166.1', [OID_END, '9.1.1.1']),  # qosQueueingConfigBandwidth
 357     ('.1.3.6.1.4.1.9.9.166.1', [OID_END, '9.1.1.2']),  # cbQosQueueingCfgBandwidthUnits
 358     ('.1.3.6.1.4.1.9.9.166.1', [OID_END, '5.1.1.3']),  # cbQosObjectsType
 359 ]
 360 snmp_scan_functions['cisco_qos'] = lambda oid: "cisco" in oid(".1.3.6.1.2.1.1.1.0").lower() and \
 361                                    oid(".1.3.6.1.4.1.9.9.166.1.1.1.1.4.*")