checks/livestatus_status

   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 output from agent:
  28 # <<<livestatus_status:sep(59)>>>
  29 # [downsite]
  30 # [mysite]
  31 # accept_passive_host_checks;accept_passive_service_checks;cached_log_messages;check_external_commands;check_host_freshness;check_service_freshness;connections;connections_rate;enable_event_handlers;enable_flap_detection;enable_notifications;execute_host_checks;execute_service_checks;external_command_buffer_max;external_command_buffer_slots;external_command_buffer_usage;external_commands;external_commands_rate;forks;forks_rate;host_checks;host_checks_rate;interval_length;last_command_check;last_log_rotation;livecheck_overflows;livecheck_overflows_rate;livechecks;livechecks_rate;livestatus_active_connections;livestatus_queued_connections;livestatus_threads;livestatus_version;log_messages;log_messages_rate;nagios_pid;neb_callbacks;neb_callbacks_rate;num_hosts;num_services;obsess_over_hosts;obsess_over_services;process_performance_data;program_start;program_version;requests;requests_rate;service_checks;service_checks_rate
  32 # 1;1;0;1;0;1;231;1.0327125668e-01;1;1;1;1;1;0;32768;0;0;0.0000000000e+00;0;0.0000000000e+00;0;0.0000000000e+00;60;1359471450;0;0;0.0000000000e+00;0;0.0000000000e+00;1;0;20;2013.01.23;0;0.0000000000e+00;15126;15263;6.5307324420e+00;0;0;0;0;1;1359469039;3.2.3;230;1.0327125668e-01;0;0.0000000000e+00
  33
  34 factory_settings["livestatus_status_default_levels"] = {
  35     "site_stopped": 2,
  36     "execute_host_checks": 2,
  37     "execute_service_checks": 2,
  38     "accept_passive_host_checks": 2,
  39     "accept_passive_service_checks": 2,
  40     "check_host_freshness": 0,  # Was in OMD the default up to now, better not warn
  41     "check_service_freshness": 1,
  42     "enable_event_handlers": 1,
  43     "enable_flap_detection": 1,
  44     "enable_notifications": 2,
  45     "process_performance_data": 1,
  46     "check_external_commands": 2,
  47     "site_cert_days": (30, 7),
  48 }
  49
  50
  51 def parse_livestatus_status(info):
  52     parsed = {}
  53     site, headers = None, None
  54     for line in info:
  55         if line and line[0][0] == "[" and line[0][-1] == "]":
  56             site = line[0][1:-1]
  57             parsed[site] = None  # Site is marked as down until overwritten later
  58
  59         elif site:
  60             if headers is None:
  61                 headers = line
  62             else:
  63                 parsed[site] = dict(zip(headers, line))
  64                 headers = None
  65
  66     return parsed
  67
  68
  69 def parse_livestatus_ssl_certs(info):
  70     if not info:
  71         return {}
  72
  73     parsed = {}
  74     site = None
  75     for line in info:
  76         if line and line[0][0] == "[" and line[0][-1] == "]":
  77             site = line[0][1:-1]
  78             parsed[site] = {}
  79
  80         elif site and len(line) == 2:
  81             pem_path, valid_until = line
  82             parsed[site][pem_path] = valid_until
  83
  84     return parsed
  85
  86
  87 def inventory_livestatus_status(sections):
  88     return [(site, {}) for (site, status) in sections[0].items() if status is not None]
  89
  90
  91 def check_livestatus_status(item, params, sections):
  92     parsed, ssl_certs = sections[0], parse_livestatus_ssl_certs(sections[1])
  93
  94     if item not in parsed:
  95         return
  96     status = parsed[item]
  97
  98     # Ignore down sites. This happens on a regular basis due to restarts
  99     # of the core. The availability of a site is monitored with 'omd_status'.
 100     if status is None:
 101         yield params["site_stopped"], "Site is currently not running"
 102         return
 103
 104     # Check Performance counters
 105     this_time = time.time()
 106     status_counters = [
 107         # factor,calc_rate,column,format,title
 108         (1, True, "%.1f/s", "host_checks", "HostChecks"),
 109         (1, True, "%.1f/s", "service_checks", "ServiceChecks"),
 110         (1, True, "%.1f/s", "forks", "ProcessCreations"),
 111         (1, True, "%.1f/s", "connections", "LivestatusConnects"),
 112         (1, True, "%.1f/s", "requests", "LivestatusRequests"),
 113         (1, True, "%.1f/s", "log_messages", "LogMessages"),
 114     ]
 115
 116     if status["program_version"].startswith("Check_MK"):
 117         # We have a CMC here.
 118
 119         status_counters += [
 120             # factor, calc_rate, format, column, title
 121             (1, False, "%.3fs", "average_latency_generic", "Average check latency"),
 122             (1, False, "%.3fs", "average_latency_cmk", "Average Check_MK latency"),
 123             (100, False, "%.1f%%", "helper_usage_generic", "Check helper usage"),
 124             (100, False, "%.1f%%", "helper_usage_cmk", "Check_MK helper usage"),
 125             (100, False, "%.1f%%", "livestatus_usage", "Livestatus usage"),
 126             (1, False, "%.1f/s", "livestatus_overflows_rate", "Livestatus overflow rate"),
 127         ]
 128
 129     for factor, calc_rate, fmt, counter, title in status_counters:
 130         value = factor * float(status[counter])
 131
 132         if calc_rate:
 133             value = get_rate("livestatus_status.%s.%s" % (item, counter), this_time, value)
 134
 135         yield 0, ("%s: %s" % (title, fmt % value)), [(counter, value)]
 136
 137     yield 0, "%d Hosts" % int(status["num_hosts"]), [("monitored_hosts", int(status["num_hosts"]))]
 138     yield 0, "%d Services" % int(status["num_services"]), [("monitored_services",
 139                                                             int(status["num_services"]))]
 140
 141     # Output some general information
 142     yield 0, "Core version: %s" % status["program_version"]
 143     yield 0, "Livestatus version: %s" % status["livestatus_version"]
 144
 145     pem_path = "/omd/sites/%s/etc/ssl/sites/%s.pem" % (item, item)
 146     cert_valid_until = ssl_certs.get(item, {}).get(pem_path)
 147     if cert_valid_until is not None:
 148         days_left = (int(cert_valid_until) - time.time()) / 86400.0
 149         valid_until_formatted = time.strftime("%Y-%m-%d %H:%M:%S",
 150                                               time.localtime(int(cert_valid_until)))
 151         ssl_state, ssl_infotext, _ssl_perfdata = check_levels(
 152             value=days_left,
 153             dsname="site_cert_days",
 154             infoname="Site certificate validity (until %s)" % valid_until_formatted,
 155             unit="days",
 156             params=(None, None, params["site_cert_days"][0], params["site_cert_days"][1]),
 157         )
 158         yield ssl_state, ssl_infotext
 159
 160     settings = [
 161         ("execute_host_checks", "Active host checks are disabled"),
 162         ("execute_service_checks", "Active service checks are disabled"),
 163         ("accept_passive_host_checks", "Passive host check are disabled"),
 164         ("accept_passive_service_checks", "Passive service checks are disabled"),
 165         ("check_host_freshness", "Host freshness checking is disabled"),
 166         ("check_service_freshness", "Service freshness checking is disabled"),
 167         #   ("enable_event_handlers",         "Alert handlers are disabled"), # special case below
 168         ("enable_flap_detection", "Flap detection is disabled"),
 169         ("enable_notifications", "Notifications are disabled"),
 170         ("process_performance_data", "Performance data is disabled"),
 171         ("check_external_commands", "External commands are disabled"),
 172     ]
 173     # Check settings of enablings. Here we are quiet unless a non-OK state is found
 174     for settingname, title in settings:
 175         if status[settingname] != '1' and params[settingname] != 0:
 176             yield params[settingname], title
 177
 178     # special considerations for enable_event_handlers
 179     if status["program_version"].startswith("Check_MK 1.2.6"):
 180         # In CMC <= 1.2.6 event handlers cannot be enabled. So never warn.
 181         return
 182     if status.get("has_event_handlers", '1') == '0':
 183         # After update from < 1.2.7 the check would warn about disabled alert
 184         # handlers since they are disabled in this case. But the user has no alert
 185         # handlers defined, so this is nothing to warn about. Start warn when the
 186         # user defines his first alert handlers.
 187         return
 188     else:
 189         if status["enable_event_handlers"] != '1' and params["enable_event_handlers"] != 0:
 190             yield params["enable_event_handlers"], "Alert handlers are disabled"
 191
 192
 193 check_info['livestatus_status'] = {
 194     "parse_function": parse_livestatus_status,
 195     "inventory_function": inventory_livestatus_status,
 196     "check_function": check_livestatus_status,
 197     "service_description": "OMD %s performance",
 198     "has_perfdata": True,
 199     "group": "livestatus_status",
 200     "default_levels_variable": "livestatus_status_default_levels",
 201     "extra_sections": ["livestatus_ssl_certs"],
 202 }