checks/mkeventd_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 # <<<mkeventd_status:sep(0)>>>
  29 # ["heute"]
  30 # [["status_config_load_time", "status_num_open_events", "status_messages", "status_message_rate", "status_average_message_rate", "status_connects", "status_connect_rate", "status_average_connect_rate", "status_rule_tries", "status_rule_trie_rate", "status_average_rule_trie_rate", "status_drops", "status_drop_rate", "status_average_drop_rate", "status_events", "status_event_rate", "status_average_event_rate", "status_rule_hits", "status_rule_hit_rate", "status_average_rule_hit_rate", "status_average_processing_time", "status_average_request_time", "status_average_sync_time", "status_replication_slavemode", "status_replication_last_sync", "status_replication_success", "status_event_limit_host", "status_event_limit_rule", "status_event_limit_overall", "status_event_limit_active_hosts", "status_event_limit_active_rules", "status_event_limit_active_overall"], [1474040901.678517, 19, 0, 0.0, 0.0, 2, 0.1998879393337847, 0.1998879393337847, 0, 0.0, 0.0, 0, 0.0, 0.0, 0, 0.0, 0.0, 0, 0.0, 0.0, 0.0, 0.002389192581176758, 0.0, "master", 0.0, false, 10, 5, 20, [], ["catch_w", "catch_y", "catch_x"], false]]
  31
  32
  33 def parse_mkeventd_status(info):
  34     import json
  35
  36     parsed, site = {}, None
  37     for line in info:
  38         try:
  39             data = json.loads(line[0])
  40         except ValueError:
  41             # The agent plugin asks the event console for json OutputFormat, but
  42             # older versions always provide python format - even when other format
  43             # was requested. Skipping the site. Won't eval data from other systems.
  44             continue
  45
  46         if len(data) == 1:
  47             site = data[0]
  48             parsed[site] = None  # Site is marked as down until overwritten later
  49         elif site:
  50             # strip "status_" from the column names
  51             keys = [col[7:] for col in data[0]]
  52             parsed[site] = dict(zip(keys, data[1]))
  53
  54     return parsed
  55
  56
  57 def inventory_mkeventd_status(parsed):
  58     return [(site, {}) for (site, status) in parsed.items() if status is not None]
  59
  60
  61 def check_mkeventd_status(item, params, parsed):
  62     if item not in parsed:
  63         return
  64
  65     status = parsed[item]
  66
  67     # Ignore down sites. This happens on a regular basis due to restarts
  68     # of the core. The availability of a site is monitored with 'omd_status'.
  69     if status is None:
  70         yield 0, "Currently not running"
  71         return
  72
  73     yield 0, "Current events: %d" % status["num_open_events"], \
  74         [("num_open_events", status["num_open_events"])]
  75
  76     yield 0, "Virtual memory: %s" % \
  77         get_bytes_human_readable(status["virtual_memory_size"]), \
  78         [("process_virtual_size", status["virtual_memory_size"])]
  79
  80     # Event limits
  81     if status["event_limit_active_overall"]:
  82         yield 2, "Overall event limit active"
  83     else:
  84         yield 0, "Overall event limit inactive"
  85
  86     for ty in ["hosts", "rules"]:
  87         limited = status["event_limit_active_%s" % ty]
  88         if limited:
  89             yield 1, "Event limit active for %d %s (%s)" % \
  90                 (len(limited), ty, ", ".join(limited))
  91         else:
  92             yield 0, "No %s event limit active" % ty
  93
  94     # Rates
  95     columns = [
  96         ("Received messages", "message", "%.2f/s"),
  97         ("Rule hits", "rule_hit", "%.2f/s"),
  98         ("Rule tries", "rule_trie", "%.2f/s"),
  99         ("Message drops", "drop", "%.2f/s"),
 100         ("Created events", "event", "%.2f/s"),
 101         ("Client connects", "connect", "%.2f/s"),
 102     ]
 103     rates = {}
 104     this_time = time.time()
 105     for title, col, fmt in columns:
 106         counter_value = status[col + "s"]
 107         rate = get_rate(col, this_time, counter_value)
 108         rates[col] = rate
 109         yield 0, ("%s: " + fmt) % (title, rate), \
 110             [("average_%s_rate" % col, rate)]
 111
 112     # Hit rate
 113     if rates["rule_trie"] == 0.0:
 114         hit_rate_txt = "-"
 115     else:
 116         value = rates["rule_hit"] / rates["rule_trie"] * 100
 117         hit_rate_txt = "%.2f%%" % value
 118         yield 0, '', [("average_rule_hit_ratio", value)]
 119     yield 0, "%s: %s" % ("Rule hit ratio", hit_rate_txt)
 120
 121     # Time columns
 122     time_columns = [
 123         ("Processing time per message", "processing"),
 124         ("Time per client request", "request"),
 125         ("Replication synchronization", "sync"),
 126     ]
 127     for title, name in time_columns:
 128         value = status.get("average_%s_time" % name)
 129         if value:
 130             txt = "%.2f ms" % (value * 1000)
 131             yield 0, '', [("average_%s_time" % name, value)]
 132         else:
 133             if name == "sync":
 134                 continue  # skip if not available
 135             txt = "-"
 136         yield 0, "%s: %s" % (title, txt)
 137
 138
 139 check_info["mkeventd_status"] = {
 140     "parse_function": parse_mkeventd_status,
 141     "inventory_function": inventory_mkeventd_status,
 142     "check_function": check_mkeventd_status,
 143     "service_description": "OMD %s Event Console",
 144     "has_perfdata": True,
 145 }