checks/services

   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
  28 # Output of old agent (< 1.1.10i2):
  29 # AeLookupSvc        running  Application Experience Lookup Service
  30 # Alerter            stopped  Alerter
  31 # ALG                stopped  Application Layer Gateway Service
  32 # AppMgmt            stopped  Application Management
  33 # appmgr             running  Remote Server Manager
  34
  35 # Output of new agent (>= 1.1.10i2):
  36 # Alerter stopped/disabled Warndienst
  37 # ALG running/demand Gatewaydienst auf Anwendungsebene
  38 # Apple_Mobile_Device running/auto Apple Mobile Device
  39 # AppMgmt stopped/demand Anwendungsverwaltung
  40 # AudioSrv running/auto Windows Audio
  41 # BITS running/demand Intelligenter Hintergrund<FC>bertragungsdienst
  42 # Bonjour_Service running/auto Dienst "Bonjour"
  43
  44 inventory_services = []
  45
  46 # Examples for inventory_services:
  47 # inventory_services = [
  48 #   "HirnTest",               # add service, if currently running
  49 #   "TapiSrv running",        # the same
  50 #   "TermService auto",       # add service, if start type is auto (regardless if running)
  51 #   "BackupSrv running/auto", # add, if start type is auto and it's running
  52 #   "~Backup.* running/auto",  # same, but add all services matching a regex
  53 #   ( [ "termserver" ] , ALL_HOSTS, [ "HirnTest running", "Sppoller auto" ] ), # same with tags..
  54 #   ( ALL_HOSTS, [ "!Backup.*", "FooBar auto" ] ),
  55 # ]
  56
  57 # Implemented in 1.2.1i2:
  58 # New rule-style (WATO compatible) notation:
  59 #   [({'start_mode': 'demand', 'service': ['Netman']}, [], ['@all'], {'docu_url': ''})]
  60 #
  61 # <services> is list of regexes matching the service name
  62 # <state> is the expected state to inventorize services of (running, stopped, ...)
  63 # <start_mode> is the expected state to inventorize services of (auto, manual, ...)
  64 #
  65 # All above attributes can be set to None or not set to disable this filter option for the entry.
  66 inventory_services_rules = []
  67
  68 def inventory_windows_services(info):
  69     # Handle single entries (type str)
  70     def add_matching_services(name, description, state, start_type, entry):
  71         if type(entry) == tuple:
  72             # New wato rule handling
  73             svc  = entry[0]
  74             statespec = entry[1:]
  75
  76         elif ' ' in entry and len(entry.split()) == 2:
  77             svc, statespec = entry.split()
  78
  79         else:
  80             svc = entry
  81             statespec = "running"
  82
  83         # First match name or description (optional since rule based config option available)
  84         if svc:
  85             if svc.startswith("~"):
  86                 r = regex(svc[1:])
  87                 if not r.match(name) and not r.match(description):
  88                     return []
  89             elif svc != name and svc != description:
  90                 return []
  91
  92         if type(statespec) == tuple:
  93             # New wato rule handling (always given as tuple of two)
  94             if (statespec[0] and statespec[0] != state) \
  95                or (statespec[1] and statespec[1] != start_type):
  96                 return []
  97
  98         else:
  99             for n in statespec.split("/"):
 100                 if n not in [ state, start_type ]:
 101                     return []
 102
 103         return [(name, {})]
 104
 105     # Handle entries like ( [ "term" ], ALL_HOSTS, [ "FooBar auto", ".*TEST running" ] )
 106     def add_services_with_tags(name, description, state, start_type, entry):
 107         matching = []
 108         if len(entry) == 2:
 109             entry = ( [], ) + entry
 110         taglist, hostlist, svclist = entry
 111         if hosttags_match_taglist(tags_of_host(host_name()), taglist):
 112             if in_extraconf_hostlist(hostlist, host_name()):
 113                 for svc in svclist:
 114                     matching += add_matching_services(name, description, state, start_type, svc)
 115         return matching
 116
 117     # Filter WATO compatible rules by tags/hostlist
 118     rules = []
 119     for rule in inventory_services_rules:
 120         if len(rule) >= 4:
 121             options = rule[3]
 122             if options.get("disabled"):
 123                 continue
 124
 125         # 1. Get all rules matching the current host
 126         taglist, hostlist = rule[1:3]
 127         if not hosttags_match_taglist(tags_of_host(host_name()), taglist) \
 128            or not in_extraconf_hostlist(hostlist, host_name()):
 129             continue
 130
 131         # 2. Now extract the list of service regexes
 132         value = rule[0]
 133         svcs  = value.get('services', [])
 134         state = value.get('state', None)
 135         start_mode  = value.get('start_mode', None)
 136         if svcs:
 137             for svc in svcs:
 138                 rules.append(('~' + svc, state, start_mode))
 139         else:
 140             rules.append((None, state, start_mode))
 141
 142     inventory = []
 143     for line in info:
 144         name = line[1]
 145         description = " ".join(line[3:])
 146         if '/' in line[2]:
 147             state, start_type = line[2].split('/')
 148         else:
 149             state = line[2]
 150             start_type = "unknown"
 151
 152         #
 153         # Handle "old" inventory_services notation
 154         #
 155         for entry in inventory_services:
 156             if type(entry) == str:
 157                 inventory += add_matching_services(name, description, state, start_type, entry)
 158             elif type(entry) == tuple:
 159                 inventory += add_services_with_tags(name, description, state, start_type, entry)
 160             else:
 161                 raise MKGeneralException("Invalid entry %r in inventory_services" % entry)
 162
 163         #
 164         # New WATO compatible rule matching
 165         #
 166
 167         for rule in rules:
 168             inventory += add_matching_services(name, description, state, start_type, rule)
 169
 170     return inventory
 171
 172
 173 # Format of parameters
 174 # {
 175 #    "states" : [ ( "running", "demand", 1 ),
 176 #                  ( "stopped", None, 2 ) ],
 177 #    "else" : 2,
 178 # }
 179
 180 def check_windows_services(item, params, info):
 181     # Hack for old manually defined checks:
 182     if params is None:
 183         params = factory_settings["services_default_levels"]
 184
 185     # A service may appear more than once (due to clusters).
 186     # First make a list of all matching entries with their
 187     # states
 188     found = []
 189     for line in info:
 190         # allow to match agains the internal name or agains the display name
 191         # of the service
 192         display_name = " ".join(line[3:])
 193         if item == line[1] or item == display_name \
 194            or line[1] in params['additional_servicenames'] \
 195            or display_name in params['additional_servicenames']:
 196             # newer agents also send start type as part of state,
 197             # e.g. running/auto
 198             if '/' in line[2]:
 199                 state, start_type = line[2].split('/')
 200             else:
 201                 state = line[2]
 202                 start_type = "unknown"
 203             found.append((line[0], state, start_type, " ".join(line[3:])))
 204
 205     if not found:
 206         return params["else"], "service not found"
 207
 208     # We take the best found state (neccessary for clusters)
 209     best_desc, best_state, best_info, best_running_on = None, None, None, None
 210     for running_on, state, start_type, desc in found:
 211         for t_state, t_start_type, mon_state in params["states"]:
 212             if (t_state is None or t_state == state) \
 213              and (t_start_type is None or t_start_type == start_type):
 214                 this_state = mon_state
 215                 break
 216         else:
 217             this_state = params["else"]
 218
 219         if best_state is None or this_state < best_state:
 220             best_state      = this_state
 221             best_info       = state, start_type
 222             best_running_on = running_on
 223             best_desc       = desc
 224
 225     infotext = best_desc + ": %s (start type is %s)" % best_info
 226     if best_running_on and best_state != 2: #if best state ist critical, there should no message "running on"
 227         infotext += " (running on: %s)" % best_running_on
 228
 229     return (best_state, infotext)
 230
 231
 232 factory_settings["services_default_levels"] = {
 233     "states"                    : [ ( "running", None, 0 ) ],
 234     "else"                      : 2,
 235     "additional_servicenames"   : [],
 236 }
 237
 238 check_info['services'] = {
 239     "check_function"          : check_windows_services,
 240     "inventory_function"      : inventory_windows_services,
 241     "service_description"     : "Service %s",
 242     "node_info"               : True,
 243     "group"                   : "services",
 244     "default_levels_variable" : "services_default_levels",
 245 }
 246
 247 factory_settings["services_summary_default_levels"] = {
 248         "ignored": [],
 249         "state_if_stopped": 0
 250 }
 251
 252 def inventory_services_summary(info):
 253     if info[0]:
 254         return [ (None, "services_summary_default_levels") ]
 255
 256 def check_services_summary(item, params, info):
 257     blacklist = params.get("ignored", () )
 258     stoplist = []
 259     num_blacklist = 0
 260     num_auto = 0
 261     for line in info:
 262         # newer agents also send start type as part of state,
 263         # e.g. running/auto
 264         if '/' in line[2]:
 265             startstop, auto = line[2].split('/')
 266         else:
 267             startstop = line[2]
 268             auto      = "unknown"
 269         srv_name = line[1]
 270         if auto == "auto":
 271             num_auto += 1
 272             if startstop == "stopped":
 273                 match = False
 274                 for srv in blacklist:
 275                     if  re.match(srv,srv_name):
 276                         match = True
 277                 if match is False:
 278                     stoplist.append(srv_name)
 279                 else:
 280                     num_blacklist += 1
 281
 282     num_stoplist = len(stoplist)
 283     num_srv = len(info)
 284
 285     if len(stoplist) > 0:
 286         stopped_srvs = " ("+", ".join(stoplist)+")"
 287         state = params.get("state_if_stopped")
 288         if state == 1:
 289             sym = "(!)"
 290         elif state == 2:
 291             sym="(!!)"
 292         else:
 293             sym = ""
 294     else:
 295         stopped_srvs = ""
 296         state = 0
 297         sym = ""
 298
 299
 300     infotext = "%d services, %d services in autostart - of which %d services are stopped%s%s, %d services stopped but ignored" % \
 301         ( num_srv, num_auto, num_stoplist, sym, stopped_srvs,  num_blacklist )
 302
 303     return state, infotext
 304
 305 check_info['services.summary'] = {
 306     "check_function"          : check_services_summary,
 307     "inventory_function"      : inventory_services_summary,
 308     "default_levels_variable" : "services_summary_default_levels",
 309     "service_description"     : "Services Summary",
 310     "node_info"               : True,
 311     "group"                   : "services_summary",
 312 }