checks/solaris_fmadm

   1 #!/usr/bin/python
   2 # -*- encoding: utf-8; py-indent-offset: 4 -*-
   3 # +------------------------------------------------------------------+
   4 # |             ____ _               _        __  __ _  __           |
   5 # |            / ___| |__   ___  ___| | __   |  \/  | |/ /           |
   6 # |           | |   | '_ \ / _ \/ __| |/ /   | |\/| | ' /            |
   7 # |           | |___| | | |  __/ (__|   <    | |  | | . \            |
   8 # |            \____|_| |_|\___|\___|_|\_\___|_|  |_|_|\_\           |
   9 # |                                                                  |
  10 # | Copyright Mathias Kettner 2015             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 # No faults:
  28 # Agent section is empty
  29
  30 # From https://docs.oracle.com/cd/E23824_01/html/821-1451/glisy.html
  31 # One fault:
  32 # <<<solaris_fmadm:sep(58)>>>
  33 # --------------- ------------------------------------  -------------- ---------
  34 # TIME            EVENT-ID                              MSG-ID         SEVERITY
  35 # --------------- ------------------------------------  -------------- ---------
  36 # Aug 24 17:56:03 7b83c87c-78f6-6a8e-fa2b-d0cf16834049  SUN4V-8001-8H  Minor
  37 #
  38 # Host        : bur419-61
  39 # Platform    : SUNW,T5440        Chassis_id  : BEL07524BN
  40 # Product_sn  : BEL07524BN
  41 #
  42 # Fault class : fault.cpu.ultraSPARC-T2plus.ireg
  43 # Affects     : cpu:///cpuid=0/serial=1F95806CD1421929
  44 #                   faulted and taken out of service
  45 # FRU         : "MB/CPU0" (hc://:product-id=SUNW,T5440:server-id=bur419-61:\
  46 #               serial=3529:part=541255304/motherboard=0/cpuboard=0)
  47 #                   faulty
  48 # Serial ID.  : 3529
  49 #               1F95806CD1421929
  50 #
  51 # Description : The number of integer register errors associated with this thread
  52 #               has exceeded acceptable levels.
  53 #
  54 # Response    : The fault manager will attempt to remove the affected thread from
  55 #               service.
  56 #
  57 # Impact      : System performance may be affected.
  58 #
  59 # Action      : Use 'fmadm faulty' to provide a more detailed view of this event.
  60 #               Please refer to the associated reference document at
  61 #               http://sun.com/msg/SUN4V-8001-8H for the latest service
  62 #               procedures and policies regarding this diagnosis.
  63
  64 # Multiple faults:
  65 # <<<solaris_fmadm:sep(58)>>>
  66 # --------------- ------------------------------------  -------------- -------
  67 # TIME            EVENT-ID                              MSG-ID         SEVERITY
  68 # --------------- ------------------------------------  -------------- -------
  69 # Sep 21 10:01:36 d482f935-5c8f-e9ab-9f25-d0aaafec1e6c  PCIEX-8000-5Y  Major
  70 #
  71 # Fault class  : fault.io.pci.device-invreq
  72 # Affects      : dev:///pci@0,0/pci1022,7458@11/pci1000,3060@0
  73 #                dev:///pci@0,0/pci1022,7458@11/pci1000,3060@1
  74 #                 ok and in service
  75 #                dev:///pci@0,0/pci1022,7458@11/pci1000,3060@2
  76 #                dev:///pci@0,0/pci1022,7458@11/pci1000,3060@3
  77 #                  faulty and taken out of service
  78 # FRU          : "SLOT 2" (hc://.../pciexrc=3/pciexbus=4/pciexdev=0)
  79 #                  repair attempted
  80 #                "SLOT 3" (hc://.../pciexrc=3/pciexbus=4/pciexdev=1)
  81 #                  acquitted
  82 #                "SLOT 4" (hc://.../pciexrc=3/pciexbus=4/pciexdev=2)
  83 #                  not present
  84 #                "SLOT 5" (hc://.../pciexrc=3/pciexbus=4/pciexdev=3)
  85 #                  faulty
  86 #
  87 #  Description  : The transmitting device sent an invalid request.
  88 #
  89 #  Response     : One or more device instances may be disabled
  90 #
  91 #  Impact       : Possible loss of services provided by the device instances
  92 #                 associated with this fault
  93 #
  94 #  Action       : Use 'fmadm faulty' to provide a more detailed view of this event.
  95 #                 Please refer to the associated reference document at
  96 #                 http://sun.com/msg/PCIEX-8000-5Y for the latest service
  97 #                 procedures and policies regarding this diagnosis.
  98
  99 # Newer outputs:
 100 # <<<solaris_fmadm:sep(58)>>>
 101 # --------------- ------------------------------------  -------------- ---------
 102 # TIME            EVENT-ID                              MSG-ID         SEVERITY
 103 # --------------- ------------------------------------  -------------- ---------
 104 # Jun 05 22:11:06 66b65c7f-33c4-4625-94dc-b35e1926ab47  PCIEX-8000-YJ  Major
 105 #
 106 # Problem Status    : isolated
 107 # Diag Engine       : eft / 1.16
 108 # System
 109 #     Manufacturer  : unknown
 110 #     Name          : SPARC-Enterprise-M4000-Server
 111 #     Part_Number   : unknown
 112 #     Serial_Number : BCF1030049
 113 #     Host_ID       : 856ec4fa
 114 #
 115 # ----------------------------------------
 116 # Suspect 1 of 3 :
 117 #    Problem class : fault.io.pciex.device-pcie-ce
 118 #    Certainty   : 40%
 119 #    Affects     : dev:////pci@0,600000/pci@0/pci@9/SUNW,emlxs@0
 120 #    Status      : faulted and taken out of service
 121 #
 122 #    FRU
 123 #      Status           : faulty
 124 #      Location         : "/SYS/iou#0-pci#1"
 125 #      Manufacturer     : unknown
 126 #      Name             : unknown
 127 #      Part_Number      : unknown
 128 #      Revision         : unknown
 129 #      Serial_Number    : unknown
 130 #      Chassis
 131 #         Manufacturer  : unknown
 132 #         Name          : SPARC-Enterprise-M4000-Server
 133 #         Part_Number   : unknown
 134 #         Serial_Number : BCF1030049
 135 # ----------------------------------------
 136 # Suspect 2 of 3 :
 137 #    Problem class : fault.io.pciex.device-pcie-ce
 138 #    Certainty   : 40%
 139 #    Affects     : dev:////pci@0,600000/pci@0/pci@9
 140 #    Status      : faulted and taken out of service
 141 #
 142 #    FRU
 143 #      Status           : faulty
 144 #      FMRI             : "hc:///compon"nt=iou#0"
 145 #      Manufacturer     : unknown
 146 #      Name             : unknown
 147 #      Part_Number      : unknown
 148 #      Revision         : unknown
 149 #      Serial_Number    : unknown
 150 #      Chassis
 151 #         Manufacturer  : unknown
 152 #         Name          : unknown
 153 #         Part_Number   : unknown
 154 #         Serial_Number : unknown
 155 #    Resource
 156 #      Location         : "iou#0-pci#1"
 157 # ...
 158
 159
 160 def parse_solaris_fmadm(info):
 161     if len(info) < 4:
 162         return {}
 163
 164     event = []
 165     for entry in ":".join(info[3]).split():
 166         if entry:
 167             event.append(entry)
 168
 169     parsed = {
 170         "event": {
 171             "time": " ".join(event[:-3]),
 172             "id": event[-3],
 173             "msg": event[-2],
 174             "severity": event[-1].lower(),
 175         },
 176         "problems": [],
 177     }
 178
 179     # Skip header
 180     for line in info[4:]:
 181         stripped = [x.strip() for x in line]
 182         if stripped[0] in ["Problem class", "Fault class"]:
 183             parsed["problems"].append(":".join(stripped[1:]))
 184
 185     return parsed
 186
 187
 188 def inventory_solaris_fmadm(parsed):
 189     return [(None, None)]
 190
 191
 192 def check_solaris_fmadm(_no_item, params, parsed):
 193     if not parsed:
 194         yield 0, "No faults detected"
 195         return
 196
 197     map_state = {
 198         "minor": (1, "minor"),
 199         "major": (2, "major"),
 200         "critical": (2, "critical"),
 201     }
 202
 203     event = parsed["event"]
 204     state, state_readable = map_state.get(event["severity"], (3, "unknown"))
 205     yield state, "Severity: %s (%s)" % (state_readable, event["time"])
 206
 207     problems = parsed["problems"]
 208     if problems:
 209         yield 0, "Problems: %s" % ", ".join(problems)
 210
 211
 212 check_info['solaris_fmadm'] = {
 213     'parse_function': parse_solaris_fmadm,
 214     'inventory_function': inventory_solaris_fmadm,
 215     'check_function': check_solaris_fmadm,
 216     'service_description': 'FMD Status',
 217 }