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 # Deprecated option since 1.6. cmk_base creates a config warning when finding rules
  75 # for this ruleset. Can be dropped with 1.7.
  76 inventory_processes = []
  77
  78 inventory_processes_perf = []
  79 ANY_USER = None
  80
  81
  82 def ps_cleanup_counters(parsed):
  83     # remove legacy key used for some kind of caching
  84     cleanup_idents = ["last.cleared.ps_"]
  85
  86     pids = ps_get_current_pids(parsed)
  87     cleanup_idents += ps_get_counters_to_delete(pids)
  88
  89     # Avoid growing up the item state with info about processes that aren't
  90     # executing anymore. Clean all information about process that are not
  91     # the ones specifically inside the current parsed agent output
  92     clear_item_states_by_full_keys(cleanup_idents)
  93
  94
  95 # Get the idents of the counters which can be deleted because the process id of
  96 # the counter is not found anymore in the process table.
  97 #
  98 # Handle these formats of idents:
  99 # Old string based keys: 'ps_stat.pcpu.669': (1448634267.875281, 1),
 100 # New magic keys:        ('ps', None, 'ps_wmic.kernel.692'): (1448633487.573496, 1092007),
 101 def ps_get_counters_to_delete(pids):
 102     counters_to_delete = []
 103     for ident in get_all_item_states():
 104         if isinstance(ident, tuple) and ident[0] in ["ps", "ps.perf"]:
 105             check_ident = ident[2]
 106         elif not isinstance(ident, tuple) and (ident.startswith("ps_stat") or
 107                                                ident.startswith("ps_wmic")):
 108             check_ident = ident
 109         else:
 110             continue
 111
 112         pid = check_ident.split(".")[-1]
 113         if pid.isdigit() and pid not in pids:
 114             counters_to_delete.append(ident)
 115
 116     return counters_to_delete
 117
 118
 119 def ps_get_current_pids(parsed):
 120     pids = set()
 121     for line in parsed:
 122         process_info = line[1]
 123         if process_info.process_id:
 124             pids.add(process_info.process_id)
 125     return pids
 126
 127
 128 # This function is only concerned with deprecated output from psperf.bat,
 129 # in case of all other output it just returns info unmodified. But if it is
 130 # a windows output it will extract the number of cpu cores
 131 def ps_merge_wmic_info(info):
 132     # Agent output version cmk>1.2.5
 133     # Assumes line = [CLUSTER, PS_INFO, COMMAND]
 134     has_wmic = False
 135     for line in info:
 136         if len(line) > 2 and line[2].lower() == "system idle process":
 137             cpu_cores = int(line[1][1:-1].split(",")[9])
 138             return cpu_cores, info
 139         if "wmic process" in line[-1]:
 140             has_wmic = True
 141             break
 142
 143     # Data from other systems than windows
 144     if not has_wmic:
 145         return 1, info
 146
 147     # Data from windows with wmic info, cmk<1.2.5
 148     return extract_wmic_info(info)
 149
 150
 151 def extract_wmic_info(info):
 152     ps_result = []
 153     lines = iter(info)
 154     wmic_info = {}
 155     is_wmic = False
 156
 157     while True:
 158         try:
 159             line = lines.next()
 160             if line[-1] == '[wmic process]':
 161                 is_wmic = True
 162                 wmic_headers = ["node"] + lines.next()[1:]
 163                 continue
 164             elif line[-1] == '[wmic process end]':
 165                 is_wmic = False
 166                 continue
 167         except StopIteration:
 168             break  # Finished with all lines
 169
 170         if is_wmic:
 171             row = dict(zip(wmic_headers, line))
 172             # Row might be damaged. I've seen this agent output:
 173             # Node - TILE-BUILDER02
 174             # ERROR:
 175             # Description = Quota violation
 176             #
 177             # Node,
 178             if "Name" in row and "ProcessId" in row:
 179                 wmic_info.setdefault((row["node"], row["Name"]), []).append(row)
 180         else:
 181             ps_result.append(line)  # plain list of process names
 182
 183     return merge_wmic(ps_result, wmic_info, wmic_headers)
 184
 185
 186 def merge_wmic(ps_result, wmic_info, wmic_headers):
 187     info = []
 188     seen_pids = set([])  # Remove duplicate entries
 189     cpu_cores = 1
 190     for line in ps_result:
 191         psinfos = wmic_info.get((line[0], line[1]), [])
 192         if psinfos:
 193             psinfo = psinfos.pop()  # each info is used only once!
 194             # Get number of CPU cores from system idle process
 195             if "ThreadCount" in wmic_headers and psinfo["Name"].lower() == "system idle process":
 196                 cpu_cores = int(psinfo["ThreadCount"])
 197             pid = int(psinfo["ProcessId"])
 198             if pid not in seen_pids:
 199                 seen_pids.add(pid)
 200                 virt = int(psinfo["VirtualSize"]) / 1024  # Bytes -> KB
 201                 resi = int(psinfo["WorkingSetSize"]) / 1024  # Bytes -> KB
 202                 pagefile = int(psinfo["PageFileUsage"]) / 1024  # Bytes -> KB
 203                 userc = int(psinfo["UserModeTime"])  # do not resolve counter here!
 204                 kernelc = int(psinfo["KernelModeTime"])  # do not resolve counter here!
 205                 handlec = int(psinfo.get("HandleCount", 0))  # Only in newer psperf.bat versions
 206                 threadc = int(psinfo["ThreadCount"])  # do not resolve counter here!
 207                 line[1:1] = [
 208                     "(unknown,%d,%d,0,%d,%d,%d,%d,%d,%d,)" % (virt, resi, pid, pagefile, userc,
 209                                                               kernelc, handlec, threadc)
 210                 ]
 211         info.append(line)
 212
 213     return cpu_cores, info
 214
 215
 216 # This mainly formats the line[1] element which contains the process info (user,...)
 217 def ps_parse_process_entries(parsed):
 218     # line[0] = node
 219     # line[1] = process_info OR (if no process info available) = process name
 220     for line in parsed:
 221         process_info = ps_info_tuple(line[1])
 222         if process_info:
 223             line[1] = process_info
 224         else:
 225             # Make number of columns in line consistent for discovery/check
 226             line.insert(1, ps_info())
 227
 228     # Filter out any lines where no process command line is available, e.g.
 229     # [None, u'(<defunct>,,,)']
 230     # [None, u'(<defunct>,,,)', u'']
 231     parsed = [x for x in parsed if len(x) > 2 and x[2]]
 232     return parsed
 233
 234
 235 # Produces a list of lists where each sub list is built as follows:
 236 # [
 237 #     [None, (u'root', u'35156', u'4372', u'00:00:05/2-14:14:49', u'1'), u'/sbin/init'],
 238 # ]
 239 # First element:  The node the data comes from in a cluster or None
 240 # Second element: The process info tuple (see ps.include: check_ps_common() for details on the elements)
 241 # Third element:  The process command line
 242 def parse_ps(info):
 243     cpu_cores, parsed = ps_merge_wmic_info(info)
 244
 245     parsed = ps_parse_process_entries(parsed)
 246
 247     # Cleanup counters of processes which do not exist anymore
 248     ps_cleanup_counters(parsed)
 249
 250     return cpu_cores, parsed
 251
 252
 253 def inventory_ps(info):
 254     _cpu_cores, parsed = info[0]
 255     return inventory_ps_common(inventory_processes_rules, parsed)
 256
 257
 258 def check_ps(item, params, info):
 259     (cpu_cores,
 260      parsed), mem_info, solaris_mem_info, statgrab_mem_info, aix_memory_info, cpu_info = info
 261
 262     # cpu_info for non windows systems
 263     if cpu_info and len(cpu_info[0]) == 6:
 264         cpu_cores = int(cpu_info[0][5])
 265
 266     if mem_info:
 267         total_ram = parse_proc_meminfo_bytes(mem_info).get("MemTotal")
 268     elif solaris_mem_info:
 269         total_ram = solaris_mem_info.get("MemTotal") * 1024
 270     elif statgrab_mem_info:
 271         total_ram = statgrab_mem_info.get("MemTotal") * 1024
 272     elif aix_memory_info:
 273         total_ram = int(aix_memory_info[0][0]) * 4 * 1024
 274     else:
 275         total_ram = None
 276
 277     return check_ps_common(item, params, parsed, cpu_cores=cpu_cores, total_ram=total_ram)
 278
 279
 280 check_info['ps'] = {
 281     "parse_function": parse_ps,
 282     "inventory_function": inventory_ps,
 283     "check_function": check_ps,
 284     "service_description": "Process %s",
 285     "includes": ["ps.include", "mem.include"],
 286     "has_perfdata": True,
 287     "node_info": True,  # add first column with actual host name
 288     "group": "ps",
 289     "default_levels_variable": "ps_default_levels",
 290     "extra_sections": ["mem", "solaris_mem", "statgrab_mem", "aix_memory", "cpu"],
 291 }
 292
 293 # NOTE: This check is deprecated and will be removed any decade now. ps now
 294 # does always performance data.
 295 check_info['ps.perf'] = {
 296     "check_function": check_ps,
 297     "service_description": "Process %s",
 298     "includes": ["ps.include", "mem.include"],
 299     "has_perfdata": True,
 300     "node_info": True,  # add first column with actual host name
 301     "group": "ps",
 302     "default_levels_variable": "ps_default_levels",
 303     "extra_sections": ["mem", "solaris_mem", "statgrab_mem", "aix_memory", "cpu"],
 304 }