checks/winperf_phydisk

   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 # <<<winperf_phydisk>>>
  29 # 1352457622.31 234
  30 # 2 instances: 0_C: _Total
  31 # -36 0 0 rawcount
  32 # -34 235822560 235822560 type(20570500)
  33 # -34 2664021277 2664021277 type(40030500)
  34 # 1166 235822560 235822560 type(550500)        ----> Average Disk Queue Length
  35 # -32 109877176 109877176 type(20570500)
  36 # -32 2664021277 2664021277 type(40030500)
  37 # 1168 109877176 109877176 type(550500)        ----> Average Disk Read Queue Length
  38 # -30 125945384 125945384 type(20570500)
  39 # -30 2664021277 2664021277 type(40030500)
  40 # 1170 125945384 125945384 type(550500)        ----> Average Disk Write Queue Length
  41 # -28 3526915777 3526915777 average_timer
  42 # -28 36283 36283 average_base
  43 # -26 1888588843 1888588843 average_timer     ----> Average Disk Seconds/Read
  44 # -26 17835 17835 average_base
  45 # -24 1638326933 1638326933 average_timer
  46 # -24 18448 18448 average_base                ----> Average Disk Seconds/Write
  47 # -22 36283 36283 counter
  48 # -20 17835 17835 counter                     ----> Disk Reads/sec
  49 # -18 18448 18448 counter                     ----> Disk Writes/sec
  50 # -16 1315437056 1315437056 bulk_count
  51 # -14 672711680 672711680 bulk_count          ----> Disk Read Bytes/sec
  52 # -12 642725376 642725376 bulk_count          ----> Disk Write Bytes/sec
  53 # -10 1315437056 1315437056 average_bulk
  54 # -10 36283 36283 average_base
  55 # -8 672711680 672711680 average_bulk
  56 # -8 17835 17835 average_base
  57 # -6 642725376 642725376 average_bulk
  58 # -6 18448 18448 average_base
  59 # 1248 769129229 769129229 type(20570500)
  60 # 1248 2664021277 2664021277 type(40030500)
  61 # 1250 1330 1330 counter
  62
  63
  64 def winperf_phydisk_convert(info):
  65     # node_info has been activated. This check simply ignores this
  66     # for now.
  67     # In case disk performance counters are not enabled, the agent sends
  68     # an almost empty section, where the second line is missing completely
  69     def disk_perfcounters_disabled(info):
  70         nodes = set([l[0] for l in info])
  71         return len(nodes) == len(info)
  72
  73     if disk_perfcounters_disabled(info):
  74         return []
  75
  76     lines = iter(info)
  77     entries = []
  78
  79     def finalize_block(nodename):
  80         # Missing columns are donted by negative values (Linux sends here latency information)
  81         return zip(
  82             [nodename for _x in current_disks],
  83             current_disks or [],
  84             current_disk_read_bytes or [],
  85             current_disk_write_bytes or [],
  86             current_disk_reads or [],
  87             current_disk_writes or [],
  88             [-1 for _x in current_disks],
  89             current_disk_readq_ctrs or [],
  90             current_disk_writeq_ctrs or [],
  91             current_disk_seconds_per_read or [],
  92             current_disk_seconds_per_read_base or [],
  93             current_disk_seconds_per_write or [],
  94             current_disk_seconds_per_write_base or [],
  95         )
  96
  97     current_node = ""
  98     try:
  99         while True:
 100             line = lines.next()
 101             if line[2] == "instances:":
 102                 if current_node != "":
 103                     entries.extend(finalize_block(current_node))
 104                 current_node = line[0]
 105                 current_disks = []
 106                 for disk_id in line[3:-1]:
 107                     disk_id = disk_id.split('_')
 108
 109                     if disk_id[-1] not in current_disks:
 110                         disk_id = disk_id[-1]
 111                     else:
 112                         disk_id = "%s_%s" % (disk_id[-1], disk_id[0])
 113                     current_disks.append(disk_id)
 114             elif line[1] == '-14':
 115                 current_disk_read_bytes = map(int, line[2:-2])
 116             elif line[1] == '-12':
 117                 current_disk_write_bytes = map(int, line[2:-2])
 118             elif line[1] == '-20':
 119                 current_disk_reads = map(int, line[2:-2])
 120             elif line[1] == '-18':
 121                 current_disk_writes = map(int, line[2:-2])
 122             elif line[1] == '1168':  # Average Disk Read Queue Length
 123                 current_disk_readq_ctrs = map(int, line[2:-2])
 124             elif line[1] == '1170':  # Average Disk Read Queue Length
 125                 current_disk_writeq_ctrs = map(int, line[2:-2])
 126             elif line[1] == '-24':
 127                 if line[-1] == 'average_base':
 128                     current_disk_seconds_per_write_base = map(int, line[2:-2])
 129                 else:
 130                     current_disk_seconds_per_write = map(int, line[2:-2])
 131             elif line[1] == '-26':
 132                 if line[-1] == 'average_base':
 133                     current_disk_seconds_per_read_base = map(int, line[2:-2])
 134                 else:
 135                     current_disk_seconds_per_read = map(int, line[2:-2])
 136     except StopIteration:
 137         if current_node != "":
 138             entries.extend(finalize_block(current_node))
 139
 140     return entries
 141
 142
 143 def inventory_winperf_phydisk(info):
 144     return inventory_diskstat_generic(winperf_phydisk_convert(info))
 145
 146
 147 def check_winperf_phydisk(item, params, info):
 148     init_line = info[0]
 149     if len(init_line) >= 4:
 150         frequency = int(init_line[3])
 151     else:
 152         frequency = None
 153
 154     this_time = float(init_line[1])
 155
 156     # convert input data to a dictionary. This also calculates rates, the non-dict
 157     # version of check_diskstat calculates those itself. Also, this means we don't create
 158     # useless counters during inventory which is neater
 159     def dictify(converted_info):
 160         result = {}
 161
 162         for disk in converted_info:
 163             disk_id = "%s_%s" % (disk[0], disk[1])
 164
 165             result[disk[1]] = {
 166                 'node': disk[0],
 167                 'read_ios': get_rate("readios_" + disk_id, this_time, disk[4]),
 168                 'read_throughput': get_rate("readtp_" + disk_id, this_time, disk[2]),
 169                 'write_ios': get_rate("writeios_" + disk_id, this_time, disk[5]),
 170                 'write_throughput': get_rate("writetp_" + disk_id, this_time, disk[3]),
 171             }
 172             if frequency:
 173                 # using 1 for the base if the counter didn't increase. This makes little to no sense
 174                 sec_per_read_counter = get_rate("spr_counter_" + disk_id, this_time, disk[9])
 175                 sec_per_read_base = get_rate("spr_base_" + disk_id, this_time, disk[10]) or 1
 176                 sec_per_write_counter = get_rate("spw_counter_" + disk_id, this_time, disk[11])
 177                 sec_per_write_base = get_rate("spw_base_" + disk_id, this_time, disk[12]) or 1
 178                 result[disk[1]].update({
 179                     'average_read_wait': (sec_per_read_counter / frequency) / sec_per_read_base,
 180                     'average_write_wait': (sec_per_write_counter / frequency) / sec_per_write_base,
 181                 })
 182
 183             # Queue Lengths (currently only Windows). Windows uses counters here.
 184             # I have not understood, why....
 185             for what, ctr in [("read_ql", disk[7]), ("write_ql", disk[8])]:
 186                 qlx = get_rate("%s_%s" % (what, disk_id), this_time, int(ctr))
 187                 ql = qlx / 10000000.0
 188                 result[disk[1]].update({what: ql})
 189
 190         return result
 191
 192     return check_diskstat_dict(item, params, dictify(winperf_phydisk_convert(info)))
 193
 194
 195 check_info["winperf_phydisk"] = {
 196     'check_function': check_winperf_phydisk,
 197     'inventory_function': inventory_winperf_phydisk,
 198     'service_description': 'Disk IO %s',
 199     'node_info': True,
 200     'has_perfdata': True,
 201     'includes': ["diskstat.include"],
 202     'group': 'disk_io',
 203 }