checks/apc_symmetra

   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 # .1.3.6.1.4.1.318.1.1.1.2.1.1.0 2
  28 # .1.3.6.1.4.1.318.1.1.1.4.1.1.0 2
  29 # .1.3.6.1.4.1.318.1.1.1.11.1.1.0 0001010000000000001000000000000000000000000000000000000000000000
  30 # .1.3.6.1.4.1.318.1.1.1.2.2.1.0 100
  31 # .1.3.6.1.4.1.318.1.1.1.2.2.4.0 1
  32 # .1.3.6.1.4.1.318.1.1.1.2.2.6.0 0
  33 # .1.3.6.1.4.1.318.1.1.1.2.2.3.0 360000
  34 # .1.3.6.1.4.1.318.1.1.1.7.2.6.0 2
  35 # .1.3.6.1.4.1.318.1.1.1.7.2.4.0 0
  36 # .1.3.6.1.4.1.318.1.1.1.2.2.2.0 25
  37 # .1.3.6.1.4.1.318.1.1.1.2.2.9.0 0
  38
  39 # upsBasicStateOutputState:
  40 # The flags are numbered 1 to 64, read from left to right. The flags are defined as follows:
  41 # 1: Abnormal Condition Present, 2: On Battery, 3: Low Battery, 4: On Line
  42 # 5: Replace Battery, 6: Serial Communication Established, 7: AVR Boost Active
  43 # 8: AVR Trim Active, 9: Overload, 10: Runtime Calibration, 11: Batteries Discharged
  44 # 12: Manual Bypass, 13: Software Bypass, 14: In Bypass due to Internal Fault
  45 # 15: In Bypass due to Supply Failure, 16: In Bypass due to Fan Failure
  46 # 17: Sleeping on a Timer, 18: Sleeping until Utility Power Returns
  47 # 19: On, 20: Rebooting, 21: Battery Communication Lost, 22: Graceful Shutdown Initiated
  48 # 23: Smart Boost or Smart Trim Fault, 24: Bad Output Voltage, 25: Battery Charger Failure
  49 # 26: High Battery Temperature, 27: Warning Battery Temperature, 28: Critical Battery Temperature
  50 # 29: Self Test In Progress, 30: Low Battery / On Battery, 31: Graceful Shutdown Issued by Upstream Device
  51 # 32: Graceful Shutdown Issued by Downstream Device, 33: No Batteries Attached
  52 # 34: Synchronized Command is in Progress, 35: Synchronized Sleeping Command is in Progress
  53 # 36: Synchronized Rebooting Command is in Progress, 37: Inverter DC Imbalance
  54 # 38: Transfer Relay Failure, 39: Shutdown or Unable to Transfer, 40: Low Battery Shutdown
  55 # 41: Electronic Unit Fan Failure, 42: Main Relay Failure, 43: Bypass Relay Failure
  56 # 44: Temporary Bypass, 45: High Internal Temperature, 46: Battery Temperature Sensor Fault
  57 # 47: Input Out of Range for Bypass, 48: DC Bus Overvoltage, 49: PFC Failure
  58 # 50: Critical Hardware Fault, 51: Green Mode/ECO Mode, 52: Hot Standby
  59 # 53: Emergency Power Off (EPO) Activated, 54: Load Alarm Violation, 55: Bypass Phase Fault
  60 # 56: UPS Internal Communication Failure, 57-64: <Not Used>
  61
  62
  63 def parse_apc_symmetra(info):
  64     parsed = {}
  65     if not info:
  66         return parsed
  67
  68     # some numeric fields may be empty
  69     battery_status,  output_status,           battery_capacity, \
  70     battery_replace, battery_num_batt_packs,  battery_time_remain,    calib_result, \
  71     last_diag_date,  battery_temp,            battery_current, state_output_state      = info[0]
  72
  73     if state_output_state != '':
  74         # string contains a bitmask, convert to int
  75         output_state_bitmask = int(state_output_state, 2)
  76     else:
  77         output_state_bitmask = 0
  78     self_test_in_progress = output_state_bitmask & 1 << 35 != 0
  79
  80     for key, val in [
  81         ("status", battery_status),
  82         ("output", output_status),
  83         ("self_test", self_test_in_progress),
  84         ("capacity", battery_capacity),
  85         ("replace", battery_replace),
  86         ("num_packs", battery_num_batt_packs),
  87         ("time_remain", battery_time_remain),
  88         ("calib", calib_result),
  89         ("diag_date", last_diag_date),
  90     ]:
  91         if val:
  92             parsed.setdefault("status", {})
  93             parsed["status"][key] = val
  94
  95     if battery_temp:
  96         parsed["temp"] = float(battery_temp)
  97
  98     if battery_current:
  99         parsed["elphase"] = {"Battery": {"current": float(battery_current)}}
 100
 101     return parsed
 102
 103
 104 #   .--battery status------------------------------------------------------.
 105 #   |   _           _   _                        _        _                |
 106 #   |  | |__   __ _| |_| |_ ___ _ __ _   _   ___| |_ __ _| |_ _   _ ___    |
 107 #   |  | '_ \ / _` | __| __/ _ \ '__| | | | / __| __/ _` | __| | | / __|   |
 108 #   |  | |_) | (_| | |_| ||  __/ |  | |_| | \__ \ || (_| | |_| |_| \__ \   |
 109 #   |  |_.__/ \__,_|\__|\__\___|_|   \__, | |___/\__\__,_|\__|\__,_|___/   |
 110 #   |                                |___/                                 |
 111 #   '----------------------------------------------------------------------'
 112
 113 # old format:
 114 # apc_default_levels = ( 95, 40, 1, 220 )  or  { "levels" : ( 95, 40, 1, 220 ) }
 115 # crit_capacity, crit_sys_temp, crit_batt_curr, crit_voltage = levels
 116 # Temperature default now 60C: regadring to a apc technician a temperature up tp 70C is possible
 117 factory_settings["apc_default_levels"] = {
 118     "capacity": (95, 80),
 119     "calibration_state": 0,
 120     "battery_replace_state": 1,
 121 }
 122
 123
 124 def inventory_apc_symmetra(parsed):
 125     return [(None, {})]
 126
 127
 128 def check_apc_symmetra(_no_item, params, parsed):
 129     data = parsed["status"]
 130     battery_status = data["status"]
 131     output_status = data.get("output")
 132     self_test_in_progress = data.get("self_test")
 133     battery_capacity = data.get("capacity")
 134     battery_replace = data.get("replace")
 135     battery_num_batt_packs = data.get("num_packs")
 136     battery_time_remain = data.get("time_remain")
 137     calib_result = data.get("calib")
 138     last_diag_date = data.get("diag_date")
 139
 140     # convert old format tuple to dict, new format with up to 6 params in dict
 141     if isinstance(params, tuple):
 142         params = {"levels": params}
 143
 144     if "levels" in params:
 145         crit_cap = params["levels"][0]
 146         params["capacity"] = (crit_cap, crit_cap)
 147
 148     alt_crit_capacity = None
 149     warn_cap, crit_cap = params['capacity']
 150     # the last_diag_date is reported as %m/%d/%Y or %y
 151     if params.get("post_calibration_levels") and \
 152        last_diag_date not in [ None, 'Unknown' ] and \
 153        len(last_diag_date) in [8, 10]:
 154         year_format = '%y' if len(last_diag_date) == 8 else '%Y'
 155         last_ts = time.mktime(time.strptime(last_diag_date, '%m/%d/' + year_format))
 156         diff_sec = time.time() - last_ts
 157         allowed_delay_sec = 86400 + params['post_calibration_levels']['additional_time_span']
 158         alt_crit_capacity = params['post_calibration_levels']['altcapacity']
 159
 160     state, state_readable = {
 161         "1": (3, "unknown"),
 162         "2": (0, "normal"),
 163         "3": (2, "low"),
 164         "4": (2, "in fault condition"),
 165     }.get(battery_status, (3, "unexpected(%s)" % battery_status))
 166     yield state, "Battery status: %s" % state_readable
 167
 168     if battery_replace:
 169         state, state_readable = {
 170             "1": (0, "no battery needs replacing"),
 171             "2": (params.get("battery_replace_state", 1), "battery needs replacing"),
 172         }.get(battery_replace, (3, "battery needs replacing: unknown"))
 173         if battery_num_batt_packs and int(battery_num_batt_packs) > 1:
 174             yield 2, "%i batteries need replacement" % int(battery_num_batt_packs)
 175         elif state:
 176             yield state, state_readable
 177
 178     if output_status:
 179         output_status_txts = {
 180             "1": "unknown",
 181             "2": "on line",
 182             "3": "on battery",
 183             "4": "on smart boost",
 184             "5": "timed sleeping",
 185             "6": "software bypass",
 186             "7": "off",
 187             "8": "rebooting",
 188             "9": "switched bypass",
 189             "10": "hardware failure bypass",
 190             "11": "sleeping until power return",
 191             "12": "on smart trim",
 192             "13": "eco mode",
 193             "14": "hot standby",
 194             "15": "on battery test",
 195             "16": "emergency static bypass",
 196             "17": "static bypass standby",
 197             "18": "power saving mode",
 198             "19": "spot mode",
 199             "20": "e conversion",
 200         }
 201         state_readable = output_status_txts.get(output_status, "unexpected(%s)" % output_status)
 202
 203         if output_status not in output_status_txts:
 204             state = 3
 205         elif output_status not in ["2", "4", "12"] and \
 206              calib_result != "3" and not self_test_in_progress:
 207             state = 2
 208         elif output_status in ["2", "4", "12"] and \
 209              calib_result == "2" and not self_test_in_progress:
 210             state = params.get("calibration_state")
 211         else:
 212             state = 0
 213
 214         calib_text = {
 215             "1": "",
 216             "2": " (calibration invalid)",
 217             "3": " (calibration in progress)",
 218         }.get(calib_result, " (calibration unexpected(%s))" % calib_result)
 219
 220         yield state, "Output status: %s%s%s" % (
 221             state_readable, calib_text, self_test_in_progress and " (self-test running)" or "")
 222
 223     if battery_capacity:
 224         battery_capacity = int(battery_capacity)
 225         state = 0
 226         levelstxt = ""
 227         if alt_crit_capacity is not None and diff_sec < allowed_delay_sec:
 228             if battery_capacity < alt_crit_capacity:
 229                 state = 2
 230                 levelstxt = " (crit below %d%% in delay after calibration)" % alt_crit_capacity
 231         else:
 232             if battery_capacity < crit_cap:
 233                 state = 2
 234                 levelstxt = " (warn/crit below %.1f%%/%.1f%%)" % (warn_cap, crit_cap)
 235             elif battery_capacity < warn_cap:
 236                 state = 1
 237                 levelstxt = " (warn/crit below %.1f%%/%.1f%%)" % (warn_cap, crit_cap)
 238
 239         yield state, "Capacity: %d%%%s" % (battery_capacity, levelstxt), \
 240               [("capacity", battery_capacity, warn_cap, crit_cap, 0, 100)]
 241
 242     if battery_time_remain:
 243         battery_time_remain = float(battery_time_remain) / 100
 244         battery_time_remain_readable = get_age_human_readable(battery_time_remain)
 245         state = 0
 246         levelstxt = ""
 247         battery_time_warn, battery_time_crit = None, None
 248         if params.get('battime'):
 249             battery_time_warn, battery_time_crit = params['battime']
 250             if battery_time_remain < battery_time_crit:
 251                 state = 2
 252             elif battery_time_remain < battery_time_warn:
 253                 state = 1
 254             perfdata = [("runtime", battery_time_remain / 60, battery_time_warn / 60,
 255                          battery_time_crit / 60)]
 256         else:
 257             perfdata = [("runtime", battery_time_remain / 60)]
 258
 259         if state:
 260             levelstxt = " (warn/crit below %s/%s)" % \
 261                           (get_age_human_readable(battery_time_warn),
 262                            get_age_human_readable(battery_time_crit))
 263
 264         yield state, "Time remaining: %s%s" % (battery_time_remain_readable, levelstxt), perfdata
 265
 266
 267 check_info['apc_symmetra'] = {
 268     "parse_function": parse_apc_symmetra,
 269     "inventory_function": inventory_apc_symmetra,
 270     "check_function": check_apc_symmetra,
 271     "service_description": "APC Symmetra status",
 272     # A Note on the order of OIDs: If the 11.1.1.0 is not the last to be polled,
 273     # this leads to bogus values for some other OIDs on some devices.
 274     "snmp_info": (
 275         ".1.3.6.1.4.1.318.1.1.1",
 276         [
 277             "2.1.1.0",  # PowerNet-MIB::upsBasicBatteryStatus,
 278             "4.1.1.0",  # PowerNet-MIB::upsBasicOutputStatus,
 279             "2.2.1.0",  # PowerNet-MIB::upsAdvBatteryCapacity,
 280             "2.2.4.0",  # PowerNet-MIB::upsAdvBatteryReplaceIndicator,
 281             "2.2.6.0",  # PowerNet-MIB::upsAdvBatteryNumOfBadBattPacks,
 282             "2.2.3.0",  # PowerNet-MIB::upsAdvBatteryRunTimeRemaining,
 283             "7.2.6.0",  # PowerNet-MIB::upsAdvTestCalibrationResults
 284             "7.2.4.0",  # PowerNet-MIB::upsLastDiagnosticsDate
 285             "2.2.2.0",  # PowerNet-MIB::upsAdvBatteryTemperature,
 286             "2.2.9.0",  # PowerNet-MIB::upsAdvBatteryCurrent,
 287             "11.1.1.0",  # PowerNet-MIB::upsBasicStateOutputState
 288         ]),
 289     "snmp_scan_function": lambda oid: oid(".1.3.6.1.2.1.1.2.0").startswith(".1.3.6.1.4.1.318.1.3"),
 290     "has_perfdata": True,
 291     "group": "apc_symentra",
 292     "default_levels_variable": "apc_default_levels",
 293 }
 294
 295 #.
 296 #   .--temperature---------------------------------------------------------.
 297 #   |      _                                      _                        |
 298 #   |     | |_ ___ _ __ ___  _ __   ___ _ __ __ _| |_ _   _ _ __ ___       |
 299 #   |     | __/ _ \ '_ ` _ \| '_ \ / _ \ '__/ _` | __| | | | '__/ _ \      |
 300 #   |     | ||  __/ | | | | | |_) |  __/ | | (_| | |_| |_| | | |  __/      |
 301 #   |      \__\___|_| |_| |_| .__/ \___|_|  \__,_|\__|\__,_|_|  \___|      |
 302 #   |                       |_|                                            |
 303 #   '----------------------------------------------------------------------'
 304
 305 # Temperature default now 60C: regadring to a apc technician a temperature up tp 70C is possible
 306 factory_settings["apc_symmetra_temp_default_levels"] = {"levels": (50, 60)}
 307
 308
 309 def inventory_apc_symmetra_temp(parsed):
 310     if "temp" in parsed:
 311         return [("Battery", {})]
 312
 313
 314 def check_apc_symmetra_temp(item, params, parsed):
 315     return check_temperature(parsed["temp"], params, "check_apc_symmetra_temp.%s" % item)
 316
 317
 318 check_info['apc_symmetra.temp'] = {
 319     'inventory_function': inventory_apc_symmetra_temp,
 320     'check_function': check_apc_symmetra_temp,
 321     'service_description': "Temperature %s",
 322     'default_levels_variable': 'apc_symmetra_temp_default_levels',
 323     'has_perfdata': True,
 324     'group': 'temperature',
 325     'includes': ['temperature.include'],
 326 }
 327
 328 #.
 329 #   .--el phase------------------------------------------------------------.
 330 #   |                      _         _                                     |
 331 #   |                  ___| |  _ __ | |__   __ _ ___  ___                  |
 332 #   |                 / _ \ | | '_ \| '_ \ / _` / __|/ _ \                 |
 333 #   |                |  __/ | | |_) | | | | (_| \__ \  __/                 |
 334 #   |                 \___|_| | .__/|_| |_|\__,_|___/\___|                 |
 335 #   |                         |_|                                          |
 336 #   '----------------------------------------------------------------------'
 337
 338 factory_settings["apc_symmetra_elphase_default_levels"] = {"current": (1, 1)}
 339
 340
 341 def inventory_apc_symmetra_elphase(parsed):
 342     for phase in parsed.get("elphase", {}):
 343         yield phase, {}
 344
 345
 346 def check_apc_symmetra_elphase(item, params, parsed):
 347     return check_elphase(item, params, parsed.get("elphase", {}))
 348
 349
 350 check_info['apc_symmetra.elphase'] = {
 351     'inventory_function': inventory_apc_symmetra_elphase,
 352     'check_function': check_apc_symmetra_elphase,
 353     'service_description': "Phase %s",
 354     'has_perfdata': True,
 355     'default_levels_variable': 'apc_symmetra_elphase_default_levels',
 356     'group': 'ups_outphase',
 357     'includes': ['elphase.include'],
 358 }