checks/smart

   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 # EXAMPLE DATA FROM: WDC SSC-D0128SC-2100
  28 #<<<smart>>>
  29 #/dev/sda ATA WDC_SSC-D0128SC-   1 Raw_Read_Error_Rate     0x000b   100   100   050    Pre-fail  Always       -       16777215
  30 #/dev/sda ATA WDC_SSC-D0128SC-   3 Spin_Up_Time            0x0007   100   100   050    Pre-fail  Always       -       0
  31 #/dev/sda ATA WDC_SSC-D0128SC-   5 Reallocated_Sector_Ct   0x0013   100   100   050    Pre-fail  Always       -       0
  32 #/dev/sda ATA WDC_SSC-D0128SC-   7 Seek_Error_Rate         0x000b   100   100   050    Pre-fail  Always       -       0
  33 #/dev/sda ATA WDC_SSC-D0128SC-   9 Power_On_Hours          0x0012   100   100   000    Old_age   Always       -       1408
  34 #/dev/sda ATA WDC_SSC-D0128SC-  10 Spin_Retry_Count        0x0013   100   100   050    Pre-fail  Always       -       0
  35 #/dev/sda ATA WDC_SSC-D0128SC-  12 Power_Cycle_Count       0x0012   100   100   000    Old_age   Always       -       523
  36 #/dev/sda ATA WDC_SSC-D0128SC- 168 Unknown_Attribute       0x0012   100   100   000    Old_age   Always       -       1
  37 #/dev/sda ATA WDC_SSC-D0128SC- 175 Program_Fail_Count_Chip 0x0003   100   100   010    Pre-fail  Always       -       0
  38 #/dev/sda ATA WDC_SSC-D0128SC- 192 Power-Off_Retract_Count 0x0012   100   100   000    Old_age   Always       -       0
  39 #/dev/sda ATA WDC_SSC-D0128SC- 194 Temperature_Celsius     0x0022   040   100   000    Old_age   Always       -       40 (Lifetime Min/Max 30/60)
  40 #/dev/sda ATA WDC_SSC-D0128SC- 197 Current_Pending_Sector  0x0012   100   100   000    Old_age   Always       -       0
  41 #/dev/sda ATA WDC_SSC-D0128SC- 240 Head_Flying_Hours       0x0013   100   100   050    Pre-fail  Always       -       0
  42 #/dev/sda ATA WDC_SSC-D0128SC- 170 Unknown_Attribute       0x0003   100   100   010    Pre-fail  Always       -       1769478
  43 #/dev/sda ATA WDC_SSC-D0128SC- 173 Unknown_Attribute       0x0012   100   100   000    Old_age   Always       -       4217788040605
  44
  45 # TODO: Need to completely rework smart check. Use IDs instead of changing
  46 # descriptions! But be careful: There is no standard neither for IDs nor for
  47 # descriptions. Only use those, which are common sense.
  48
  49 factory_settings["smart_temp_default_levels"] = {"levels": (35, 40)}
  50
  51 smart_stats_default_levels = {
  52     'realloc_events': (1, 1),
  53     'realloc_sectors': (1, 1),
  54     'spin_retries': (1, 1),
  55     'pending_retries': (1, 1),
  56     'pending_sectors': (1, 1),
  57     'cmd_timeouts': (5, 10),
  58     'e2e_errs': (1, 1),
  59     'uncorr_errs': (1, 1),
  60     'udma_crcs': (1, 1),
  61 }
  62
  63
  64 def parse_smart_raw_values(info):
  65     disks = {}
  66     disk_name = None
  67
  68     for line in info:
  69         if len(line) >= 13:
  70             if line[0] != disk_name:
  71                 disk_name = line[0]
  72                 disk = {}
  73                 disks[disk_name] = disk
  74
  75             field = line[4]
  76             if field != "Unknown_Attribute":
  77                 value = saveint(line[12])
  78                 disk[field] = value
  79     return disks
  80
  81
  82 def parse_smart_normalized_values(info):
  83     disks = {}
  84     disk_name = None
  85
  86     for line in info:
  87         if len(line) >= 13:
  88             if line[0] != disk_name:
  89                 disk_name = line[0]
  90                 disk = {}
  91                 disks[disk_name] = disk
  92
  93             field = line[4]
  94             if field != "Unknown_Attribute":
  95                 value = int(line[6])
  96                 if isinstance(line[8], int):
  97                     threshold = int(line[8])
  98                 else:
  99                     threshold = None
 100                 disk[field] = value, threshold
 101     return disks
 102
 103
 104 smart_stats_fields = [
 105     'Reallocated_Sector_Ct',
 106     'Spin_Retry_Count',
 107     'Reallocated_Event_Count',
 108     'Current_Pending_Sector',
 109     'Command_Timeout',
 110     'End-to-End_Error',
 111     'Reported_Uncorrect',
 112     'Uncorrectable_Error_Cnt',
 113     'UDMA_CRC_Error_Count',
 114     'CRC_Error_Count',
 115 ]
 116
 117
 118 def inventory_smart_stats(info):
 119     disks = parse_smart_raw_values(info)
 120     inventory = []
 121     for disk_name, disk in disks.items():
 122         for field in disk.keys():
 123             if field in smart_stats_fields:  # found at least one interesting field
 124                 cleaned = dict([(f, disk[f]) for f in smart_stats_fields if f in disk])
 125                 inventory.append((disk_name, cleaned))
 126                 break
 127     return inventory
 128
 129
 130 def check_smart_stats(item, params, info):
 131     # params is a snapshot of all counters at the point of time of inventory
 132
 133     disks = parse_smart_raw_values(info)
 134     normalized = parse_smart_normalized_values(info)
 135
 136     if item not in disks:
 137         return 3, "Disk not found"
 138     disk = disks[item]
 139
 140     state = 0
 141     infos = []
 142     perfdata = []
 143
 144     for unit, field, descr in [
 145         (' hours', 'Power_On_Hours', 'Powered on'),
 146         ('', 'Power_Cycle_Count', 'Power cycles'),
 147         ('', 'Reallocated_Sector_Ct', 'Reallocated sectors'),
 148         ('', 'Reallocated_Event_Count', 'Reallocated events'),
 149         ('', 'Spin_Retry_Count', 'Spin retries'),
 150         ('', 'Current_Pending_Sector', 'Pending sectors'),
 151         ('', 'Command_Timeout', 'Command timeouts'),
 152         ('', 'End-to-End_Error', 'End-to-End errors'),
 153         ('', 'Reported_Uncorrect', 'Uncorrectable errors'),
 154         ('', 'Uncorrectable_Error_Cnt', 'Uncorrectable errors'),
 155         ('', 'UDMA_CRC_Error_Count', 'UDMA CRC errors'),
 156         ('', 'CRC_Error_Count', 'UDMA CRC errors'),
 157     ]:
 158         if field in disk:
 159             value = disk[field]
 160             infos.append("%s: %d%s" % (descr, value, unit))
 161             perfdata.append((field, value))
 162
 163             if field in params:
 164                 ref_value = params[field]
 165
 166                 # For reallocated event counts we experienced to many reported errors for disks
 167                 # which still seem to be OK. The raw value increased by a small amount but the
 168                 # aggregated value remained at it's initial/ok state. So we use the aggregated
 169                 # value now. Only for this field.
 170                 if field == "Reallocated_Event_Count":
 171                     infos[-1] += " (was %d during discovery; normalized value looks OK)" % ref_value
 172                     norm_value, norm_threshold = normalized[item][field]
 173                     if norm_value <= norm_threshold:
 174                         state = 2
 175
 176                 elif value > ref_value:
 177                     state = 2
 178                     infos[-1] += "(!!) (was %d during discovery)" % ref_value
 179
 180     return state, ", ".join(infos), perfdata
 181
 182
 183 check_info["smart.stats"] = {
 184     'check_function': check_smart_stats,
 185     'inventory_function': inventory_smart_stats,
 186     'has_perfdata': True,
 187     'service_description': 'SMART %s Stats',
 188 }
 189
 190
 191 def inventory_smart_temp(info):
 192     disks = parse_smart_raw_values(info)
 193     for disk_name, disk in disks.iteritems():
 194         if "Temperature_Celsius" in disk or \
 195            "Temperature_Internal" in disk:
 196             yield disk_name, {}
 197
 198
 199 def check_smart_temp(item, params, info):
 200     disks = parse_smart_raw_values(info)
 201
 202     if "Temperature_Celsius" in disks[item]:
 203         temperature = disks[item]["Temperature_Celsius"]
 204     elif "Temperature_Internal" in disks[item]:
 205         temperature = disks[item]["Temperature_Internal"]
 206     else:
 207         return
 208
 209     return check_temperature(temperature, params, "smart_%s" % item)
 210
 211
 212 check_info["smart.temp"] = {
 213     'check_function': check_smart_temp,
 214     'inventory_function': inventory_smart_temp,
 215     'service_description': 'Temperature SMART %s',
 216     'group': 'temperature',
 217     'has_perfdata': True,
 218     'includes': ['temperature.include'],
 219     'default_levels_variable': "smart_temp_default_levels"
 220 }