active_checks/check_bi_aggr

   1 #!/usr/bin/env 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 import getopt
  28 import os
  29 import sys
  30 import time
  31 import traceback
  32 import urllib
  33 import ast
  34
  35 import requests
  36 import urllib3  # type: ignore
  37 from pathlib2 import Path
  38
  39 import cmk.utils.password_store
  40
  41 cmk.utils.password_store.replace_passwords()
  42
  43 urllib3.disable_warnings(urllib3.exceptions.InsecureRequestWarning)
  44
  45 # tell requests not to honour "http(s)_proxy" env variables
  46 proxies = {
  47     'http': None,
  48     'https': None,
  49 }
  50
  51
  52 def usage():
  53     sys.stderr.write("""
  54 USAGE: check_bi_aggr -b <BASE_URL> -a <AGGR_NAME> -u <USER> -s <SECRET>
  55                      [-m <AUTH_MODE>] [-r] [-n <HOSTNAME>] [-t <TIMEOUT>] [-d]
  56
  57 OPTIONS:
  58   -b BASE_URL           The base URL to the monitoring environment, e.g.
  59                         http://<hostname>/<site-id>
  60   -a AGGR_NAME          Name of the aggregation, not the aggregation group.
  61                         It is possible that there are multiple aggregations
  62                         with an equal name, but you should ensure, that it
  63                         is a unique one to prevent confusions
  64   -u USER               User-ID of an automation user which is permitted to
  65                         see all contents of the aggregation
  66   -s SECRET             Automation secret of the user
  67   --use-automation-user Use credentials from the local "automation" user
  68   -m AUTH_MODE          Authentication mode, either "cookie", "basic", "digest"
  69                         or "kerberos", defaults to "cookie"
  70   -t TIMEOUT            HTTP connect timeout in seconds (Default: 60)
  71   -r                    track downtimes. This requires the hostname to be set.
  72   -n HOSTNAME           The hostname for which this check is run.
  73   --in-downtime S       S can be "ok" or "warn". Force this state if the
  74                         aggregate is in scheduled downtime. OK states will always
  75                         be unchanged.
  76   --acknowledged S      Same as --in-downtime, but for acknowledged aggregates.
  77   -d                    Enable debug mode
  78   -h, --help            Show this help message and exit
  79
  80 """)
  81
  82
  83 short_options = 'b:a:u:s:m:t:n:dhr'
  84 long_options = ["help", "in-downtime=", "acknowledged=", "use-automation-user"]
  85
  86 try:
  87     opts, args = getopt.getopt(sys.argv[1:], short_options, long_options)
  88 except getopt.GetoptError, err:
  89     sys.stderr.write("%s\n" % err)
  90     sys.exit(1)
  91
  92 base_url = None
  93 aggr_name = None
  94
  95 username = None
  96 password = None
  97 use_automation_user = False
  98
  99 auth_mode = 'cookie'
 100 timeout = 60
 101 debug = False
 102 opt_in_downtime = None
 103 opt_acknowledged = None
 104 track_downtime = False
 105 hostname = None
 106
 107 for o, a in opts:
 108     if o in ['-h', '--help']:
 109         usage()
 110         sys.exit(0)
 111     elif o == '-b':
 112         base_url = a
 113     elif o == '-a':
 114         aggr_name = a
 115     elif o == '-u':
 116         username = a
 117     elif o == '-s':
 118         password = a
 119     elif o == '-m':
 120         auth_mode = a
 121     elif o == '-t':
 122         timeout = int(a)
 123     elif o == '-r':
 124         track_downtime = True
 125     elif o == '-n':
 126         hostname = a
 127     elif o == '-d':
 128         debug = True
 129     elif o == '--in-downtime':
 130         opt_in_downtime = a
 131     elif o == '--acknowledged':
 132         opt_acknowledged = a
 133     elif o == '--use-automation-user':
 134         use_automation_user = True
 135
 136 if not base_url:
 137     sys.stderr.write('Please provide the URL to the monitoring instance.\n')
 138     usage()
 139     sys.exit(1)
 140
 141 if not aggr_name:
 142     sys.stderr.write('Please provide the name of the aggregation.\n')
 143     usage()
 144     sys.exit(1)
 145
 146 if use_automation_user:
 147     username = "automation"
 148     secret_file_path = Path(cmk.utils.paths.var_dir) / "web" / username / "automation.secret"
 149     try:
 150         with secret_file_path.open(encoding="utf-8") as f:  # pylint: disable=no-member
 151             password = f.read()
 152     except IOError:
 153         sys.stderr.write('Unable to read credentials for "automation" user.\n')
 154         sys.exit(1)
 155
 156 if not username or not password:
 157     sys.stderr.write('Please provide valid user credentials.\n')
 158     usage()
 159     sys.exit(1)
 160
 161 if track_downtime and not hostname:
 162     sys.stderr.write('Please provide a hostname when using downtime tracking.\n')
 163     usage()
 164     sys.exit(1)
 165
 166
 167 def init_auth():
 168     auth_obj = None
 169     if username and password:
 170         if auth_mode == 'kerberos':
 171             from requests_kerberos import HTTPKerberosAuth  # type: ignore
 172
 173             from subprocess import Popen, PIPE
 174             kinit = Popen(["kinit", username], stdin=PIPE, stdout=PIPE, stderr=PIPE)
 175             output, errors = kinit.communicate("%s\n" % password)
 176             kinit.wait()
 177             if kinit.returncode or errors:
 178                 sys.stderr.write("Error getting Kerberos Ticket:\n")
 179                 sys.stderr.write(
 180                     "stdout: %s\nstderr: %s\nrc: %s" % (output, errors, kinit.returncode))
 181                 sys.exit(1)
 182
 183             auth_obj = HTTPKerberosAuth(principal=username)
 184         elif auth_mode == 'digest':
 185             auth_obj = requests.auth.HTTPDigestAuth(username, password)
 186         else:
 187             auth_obj = requests.auth.HTTPBasicAuth(username, password)
 188     return auth_obj
 189
 190 url = "%s/check_mk/view.py" \
 191       "?view_name=aggr_single_api" \
 192       "&aggr_name=%s&output_format=python" % \
 193       (base_url.rstrip('/'), urllib.quote(aggr_name))
 194
 195 auth = None
 196 if auth_mode in ['basic', 'digest', 'kerberos']:
 197     auth = init_auth()
 198 else:
 199     url += "&_username=%s&_secret=%s" % \
 200         (username, password)
 201
 202 if debug:
 203     sys.stderr.write('URL: %s\n' % url)
 204
 205 try:
 206     r = requests.get(url, timeout=timeout, auth=auth, proxies=proxies)
 207     r.raise_for_status()
 208     raw_response = r.text
 209 except requests.Timeout:
 210     sys.stdout.write('ERROR: Socket timeout while opening URL: %s\n' % (url))
 211     sys.exit(3)
 212 except requests.URLRequired, e:
 213     sys.stdout.write("UNKNOWN: %s\n" % e)
 214     sys.exit(3)
 215 except Exception, e:
 216     sys.stdout.write(
 217         'ERROR: Exception while opening URL: %s - %s\n%s' % (url, e, traceback.format_exc()))
 218     sys.exit(3)
 219
 220 if raw_response.startswith("ERROR:"):
 221     sys.stdout.write(raw_response.rstrip() + "\n")
 222     sys.exit(3)
 223
 224 try:
 225     obj = ast.literal_eval(raw_response)
 226 except Exception, e:
 227     sys.stdout.write('ERROR: Invalid response (%s): %s\n' % (e, raw_response))
 228     sys.exit(3)
 229
 230 if len(obj) == 1:
 231     sys.stdout.write(
 232         'ERROR: Aggregation "%s" does not exist or user is not permitted\n' % aggr_name)
 233     sys.exit(3)
 234
 235 if not isinstance(obj, list):
 236     sys.stdout.write('ERROR: Invalid response: %s\n' % (raw_response.replace("\n", "")))
 237     sys.exit(3)
 238
 239 headers = obj[0]
 240 row = dict(zip(headers, obj[1]))
 241
 242 aggr_output = row["aggr_output"]
 243 aggr_state = int(row["aggr_state_num"])
 244
 245 if aggr_state == -1:
 246     aggr_state = 3
 247
 248 if aggr_output == '':
 249     aggr_output = 'Aggregation state is %s' % ['OK', 'WARN', 'CRIT', 'UNKNOWN'][aggr_state]
 250
 251 # Handle downtimes and acknowledgements
 252 if opt_in_downtime and row["aggr_in_downtime"] == '1':
 253     aggr_output += ", currently in downtime"
 254     if opt_in_downtime == "ok":
 255         aggr_state = 0
 256     else:  # "warn"
 257         aggr_state = min(aggr_state, 1)
 258
 259 if track_downtime:
 260     # connect to livestatus
 261     try:
 262         import livestatus
 263     except ImportError:
 264         sys.stderr.write('The python livestatus api module is missing. Please install from\n'
 265                          'Check_MK livestatus sources to a python import path.\n')
 266         sys.exit(1)
 267
 268     socket_path = os.environ['OMD_ROOT'] + '/tmp/run/live'
 269
 270     conn = livestatus.SingleSiteConnection('unix:' + socket_path)
 271
 272     now = time.time()
 273     # find out if, according to previous tracking, there already is a downtime
 274     ids = conn.query_table(("GET downtimes\n"
 275                             "Columns: id\n"
 276                             "Filter: service_description = Aggr Host %s\n"
 277                             "Filter: author = tracking\n"
 278                             "Filter: end_time > %d") % (hostname, now))
 279     downtime_tracked = len(ids) > 0
 280     if downtime_tracked != (row["aggr_in_downtime"] == '1'):
 281         # there is a discrepance between tracked downtime state and the real state
 282         if row["aggr_in_downtime"] == '1':
 283             # need to track downtime
 284             conn.command("[%d] SCHEDULE_SVC_DOWNTIME;%s;Aggr Host %s;%d;%d;1;0;0;"
 285                          "tracking;Automatic downtime" % (now, hostname, hostname, now, 2147483647))
 286         else:
 287             for dt_id in ids:
 288                 conn.command("[%d] DEL_SVC_DOWNTIME;%d" % (now, dt_id[0]))
 289
 290 if opt_acknowledged and row["aggr_acknowledged"] == '1':
 291     aggr_output += ", is acknowledged"
 292     if opt_acknowledged == "ok":
 293         aggr_state = 0
 294     else:  # "warn"
 295         aggr_state = min(aggr_state, 1)
 296
 297 sys.stdout.write('%s\n' % aggr_output)
 298 sys.exit(aggr_state)