checks/ps

   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 # First generation of agents output only the process command line:
  28 # /usr/sbin/xinetd -pidfile /var/run/xinetd.pid -stayalive -inetd_compat -inetd_ipv6
  29
  30 # Second generation of agents output the user in brackets in the first columns:
  31 # (root) /usr/sbin/xinetd -pidfile /var/run/xinetd.pid -stayalive -inetd_compat -inetd_ipv6
  32
  33 # Third generation (from 1.1.5) output also virtual memory, resident memory and %CPU:
  34 # (class,122376,88128,0.0) /usr/jre1.6.0_13/bin/java -Dn=Cart_16TH13 -Dmcs.node=zbgh1ca -Dmcs.mdt.redundan
  35
  36 # Forth generation (>=1.2.5), additional columns in bracket:
  37 # (user, virtual_size, resident_size, %cpu, processID, pagefile_usage, usermodetime, kernelmodetime, openHandles, threadCount) name
  38 # (\\KLAPPRECHNER\ab,29284,2948,0,3124,904,400576,901296,35,1)    NOTEPAD.EXE
  39
  40 # Sixth generation (>=1.2.7) adds an optional etime, joined by "/" with the CPU time
  41
  42 # The plugin "psperf.bat" is deprecated. As of version 1.2.5 all of this information
  43 # is reported by the windows agent itself. However, we still support sections from psperf.bat
  44 # if the agent version is lower than 1.2.5.
  45 # Windows agent now ships a plugin "psperf.bat" that adds a section from wmic
  46 # to the output:
  47 # <<<ps:sep(44)>>>
  48 # [wmic process]
  49 # ^M
  50 # Node,KernelModeTime,Name,PageFileUsage,ThreadCount,UserModeTime,VirtualSize,WorkingSetSize^M
  51 # WINDOWSXP,43478281250,System Idle Process,0,2,0,0,28672^M
  52 # WINDOWSXP,155781250,System,0,59,0,1957888,253952^M
  53 # WINDOWSXP,468750,smss.exe,176128,3,156250,3928064,442368^M
  54 # WINDOWSXP,56406250,csrss.exe,1863680,12,11406250,25780224,3956736^M
  55 # WINDOWSXP,18593750,winlogon.exe,6832128,19,4843750,59314176,2686976^M
  56 # WINDOWSXP,167500000,services.exe,1765376,16,13750000,22601728,4444160^M
  57 # WINDOWSXP,16875000,lsass.exe,3964928,21,3906250,43462656,6647808^M
  58 # WINDOWSXP,8750000,VBoxService.exe,1056768,8,468750,26652672,3342336^M
  59
  60 # New since 1.2.1i2: WATO compatible syntax
  61 #
  62 # Holds a list of rules which are matching hosts by names or tags and
  63 # where each rule holds a dictionary.
  64 #
  65 # Each of those entries defines the following options:
  66 #
  67 # 1. descr:    item name to be used for the service description
  68 # 2. match:    matching-definition
  69 # 3. user:     user definition
  70 # 5. perfdata: monitor with perfdata
  71 # 4. levels:   four numbers (thresholds)
  72 inventory_processes_rules = []
  73
  74 inventory_processes = []
  75 inventory_processes_perf = []
  76 ANY_USER = None
  77
  78
  79 def ps_cleanup_counters(parsed):
  80     # remove legacy key used for some kind of caching
  81     cleanup_idents = ["last.cleared.ps_"]
  82
  83     pids = ps_get_current_pids(parsed)
  84     cleanup_idents += ps_get_counters_to_delete(pids)
  85
  86     # Avoid growing up the item state with info about processes that aren't
  87     # executing anymore. Clean all information about process that are not
  88     # the ones specifically inside the current parsed agent output
  89     clear_item_states_by_full_keys(cleanup_idents)
  90
  91
  92 # Get the idents of the counters which can be deleted because the process id of
  93 # the counter is not found anymore in the process table.
  94 #
  95 # Handle these formats of idents:
  96 # Old string based keys: 'ps_stat.pcpu.669': (1448634267.875281, 1),
  97 # New magic keys:        ('ps', None, 'ps_wmic.kernel.692'): (1448633487.573496, 1092007),
  98 def ps_get_counters_to_delete(pids):
  99     counters_to_delete = []
 100     for ident in get_all_item_states():
 101         if isinstance(ident, tuple) and ident[0] == "ps":
 102             check_ident = ident[2]
 103         elif not isinstance(ident, tuple) and (ident.startswith("ps_stat") or
 104                                                ident.startswith("ps_wmic")):
 105             check_ident = ident
 106         else:
 107             continue
 108
 109         pid = check_ident.split(".")[-1]
 110         if pid.isdigit() and pid not in pids:
 111             counters_to_delete.append(ident)
 112
 113     return counters_to_delete
 114
 115
 116 def ps_get_current_pids(parsed):
 117     pids = set()
 118     for line in parsed:
 119         process_info = line[1]
 120         if process_info.process_id:
 121             pids.add(process_info.process_id)
 122     return pids
 123
 124
 125 # This function is only concerned with deprecated output from psperf.bat,
 126 # in case of all other output it just returns info unmodified. But if it is
 127 # a windows output it will extract the number of cpu cores
 128 def ps_merge_wmic_info(info):
 129     # Agent output version cmk>1.2.5
 130     # Assumes line = [CLUSTER, PS_INFO, COMMAND]
 131     has_wmic = False
 132     for line in info:
 133         if len(line) > 2 and line[2].lower() == "system idle process":
 134             cpu_cores = int(line[1][1:-1].split(",")[9])
 135             return cpu_cores, info
 136         if "wmic process" in line[-1]:
 137             has_wmic = True
 138             break
 139
 140     # Data from other systems than windows
 141     if not has_wmic:
 142         return 1, info
 143
 144     # Data from windows with wmic info, cmk<1.2.5
 145     return extract_wmic_info(info)
 146
 147
 148 def extract_wmic_info(info):
 149     ps_result = []
 150     lines = iter(info)
 151     wmic_info = {}
 152     is_wmic = False
 153
 154     while True:
 155         try:
 156             line = lines.next()
 157             if line[-1] == '[wmic process]':
 158                 is_wmic = True
 159                 wmic_headers = ["node"] + lines.next()[1:]
 160                 continue
 161             elif line[-1] == '[wmic process end]':
 162                 is_wmic = False
 163                 continue
 164         except StopIteration:
 165             break  # Finished with all lines
 166
 167         if is_wmic:
 168             row = dict(zip(wmic_headers, line))
 169             # Row might be damaged. I've seen this agent output:
 170             # Node - TILE-BUILDER02
 171             # ERROR:
 172             # Description = Quota violation
 173             #
 174             # Node,
 175             if "Name" in row and "ProcessId" in row:
 176                 wmic_info.setdefault((row["node"], row["Name"]), []).append(row)
 177         else:
 178             ps_result.append(line)  # plain list of process names
 179
 180     return merge_wmic(ps_result, wmic_info, wmic_headers)
 181
 182
 183 def merge_wmic(ps_result, wmic_info, wmic_headers):
 184     info = []
 185     seen_pids = set([])  # Remove duplicate entries
 186     cpu_cores = 1
 187     for line in ps_result:
 188         psinfos = wmic_info.get((line[0], line[1]), [])
 189         if psinfos:
 190             psinfo = psinfos.pop()  # each info is used only once!
 191             # Get number of CPU cores from system idle process
 192             if "ThreadCount" in wmic_headers and psinfo["Name"].lower() == "system idle process":
 193                 cpu_cores = int(psinfo["ThreadCount"])
 194             pid = int(psinfo["ProcessId"])
 195             if pid not in seen_pids:
 196                 seen_pids.add(pid)
 197                 virt = int(psinfo["VirtualSize"]) / 1024  # Bytes -> KB
 198                 resi = int(psinfo["WorkingSetSize"]) / 1024  # Bytes -> KB
 199                 pagefile = int(psinfo["PageFileUsage"]) / 1024  # Bytes -> KB
 200                 userc = int(psinfo["UserModeTime"])  # do not resolve counter here!
 201                 kernelc = int(psinfo["KernelModeTime"])  # do not resolve counter here!
 202                 handlec = int(psinfo.get("HandleCount", 0))  # Only in newer psperf.bat versions
 203                 threadc = int(psinfo["ThreadCount"])  # do not resolve counter here!
 204                 line[1:1] = [
 205                     "(unknown,%d,%d,0,%d,%d,%d,%d,%d,%d,)" % (virt, resi, pid, pagefile, userc,
 206                                                               kernelc, handlec, threadc)
 207                 ]
 208         info.append(line)
 209
 210     return cpu_cores, info
 211
 212
 213 # This mainly formats the line[1] element which contains the process info (user,...)
 214 def ps_parse_process_entries(parsed):
 215     # line[0] = node
 216     # line[1] = process_info OR (if no process info available) = process name
 217     for line in parsed:
 218         process_info = ps_info_tuple(line[1])
 219         if process_info:
 220             line[1] = process_info
 221         else:
 222             # Make number of columns in line consistent for discovery/check
 223             line.insert(1, ps_info())
 224
 225     # Filter out any lines where no process command line is available, e.g.
 226     # [None, u'(<defunct>,,,)']
 227     # [None, u'(<defunct>,,,)', u'']
 228     parsed = [x for x in parsed if len(x) > 2 and x[2]]
 229     return parsed
 230
 231
 232 # Produces a list of lists where each sub list is built as follows:
 233 # [
 234 #     [None, (u'root', u'35156', u'4372', u'00:00:05/2-14:14:49', u'1'), u'/sbin/init'],
 235 # ]
 236 # First element:  The node the data comes from in a cluster or None
 237 # Second element: The process info tuple (see ps.include: check_ps_common() for details on the elements)
 238 # Third element:  The process command line
 239 def parse_ps(info):
 240     cpu_cores, parsed = ps_merge_wmic_info(info)
 241
 242     parsed = ps_parse_process_entries(parsed)
 243
 244     # Cleanup counters of processes which do not exist anymore
 245     ps_cleanup_counters(parsed)
 246
 247     return cpu_cores, parsed
 248
 249
 250 def inventory_ps(info):
 251     _cpu_cores, parsed = info[0]
 252     return inventory_ps_common(inventory_processes, inventory_processes_rules, parsed)
 253
 254
 255 def check_ps(item, params, info):
 256     (cpu_cores,
 257      parsed), mem_info, solaris_mem_info, statgrab_mem_info, aix_memory_info, cpu_info = info
 258
 259     # cpu_info for non windows systems
 260     if cpu_info and len(cpu_info[0]) == 6:
 261         cpu_cores = int(cpu_info[0][5])
 262
 263     if mem_info:
 264         total_ram = parse_proc_meminfo_bytes(mem_info).get("MemTotal")
 265     elif solaris_mem_info:
 266         total_ram = solaris_mem_info.get("MemTotal") * 1024
 267     elif statgrab_mem_info:
 268         total_ram = statgrab_mem_info.get("MemTotal") * 1024
 269     elif aix_memory_info:
 270         total_ram = int(aix_memory_info[0][0]) * 4 * 1024
 271     else:
 272         total_ram = None
 273
 274     return check_ps_common(item, params, parsed, cpu_cores=cpu_cores, total_ram=total_ram)
 275
 276
 277 check_info['ps'] = {
 278     "parse_function": parse_ps,
 279     "inventory_function": inventory_ps,
 280     "check_function": check_ps,
 281     "service_description": "Process %s",
 282     "includes": ["ps.include", "mem.include"],
 283     "has_perfdata": True,
 284     "node_info": True,  # add first column with actual host name
 285     "group": "ps",
 286     "default_levels_variable": "ps_default_levels",
 287     "extra_sections": ["mem", "solaris_mem", "statgrab_mem", "aix_memory", "cpu"],
 288 }
 289
 290 # NOTE: This check is deprecated and will be removed any decade now. ps now
 291 # does always performance data.
 292 check_info['ps.perf'] = {
 293     "check_function": check_ps,
 294     "service_description": "Process %s",
 295     "includes": ["ps.include", "mem.include"],
 296     "has_perfdata": True,
 297     "node_info": True,  # add first column with actual host name
 298     "group": "ps",
 299     "default_levels_variable": "ps_default_levels",
 300     "extra_sections": ["mem", "solaris_mem", "statgrab_mem", "aix_memory", "cpu"],
 301 }