Allow regular expressions from ignore list (-r)

Python3 compatibility

README.md

@@ -8,4 +8,4 @@ This is the public git repository for development of the plugin.
 
 Documentation + Production Ready Plugin
 -------------
-Please refer to http://www.claudiokuenzler.com/nagios-plugins/check_esxi_hardware.php 
+Please refer to https://www.claudiokuenzler.com/monitoring-plugins/check_esxi_hardware.php 

check_esxi_hardware.py

@@ -24,7 +24,7 @@
 # Copyright (c) 2008 David Ligeret
 # Copyright (c) 2009 Joshua Daniel Franklin
 # Copyright (c) 2010 Branden Schneider
-# Copyright (c) 2010-2014 Claudio Kuenzler
+# Copyright (c) 2010-2019 Claudio Kuenzler
 # Copyright (c) 2010 Samir Ibradzic
 # Copyright (c) 2010 Aaron Rogers
 # Copyright (c) 2011 Ludovic Hutin
@@ -35,6 +35,10 @@
 # Copyright (c) 2011 Ian Chard
 # Copyright (c) 2012 Craig Hart
 # Copyright (c) 2013 Carl R. Friend
+# Copyright (c) 2015 Andreas Gottwald
+# Copyright (c) 2015 Stanislav German-Evtushenko
+# Copyright (c) 2015 Stefan Roos
+# Copyright (c) 2018 Peter Newman
 #
 # The VMware 4.1 CIM API is documented here:
 #   http://www.vmware.com/support/developer/cim-sdk/4.1/smash/cim_smash_410_prog.pdf
@@ -43,8 +47,8 @@
 # The VMware 5.x CIM API is documented here:
 #   http://pubs.vmware.com/vsphere-50/index.jsp?nav=/5_1_1
 #
-# This Nagios plugin is maintained here:
-#   http://www.claudiokuenzler.com/nagios-plugins/check_esxi_hardware.php
+# This monitoring plugin is maintained and documented here:
+#   https://www.claudiokuenzler.com/monitoring-plugins/check_esxi_hardware.php
 #
 #@---------------------------------------------------
 #@ History
@@ -207,17 +211,72 @@
 #@ Author : Claudio Kuenzler (www.claudiokuenzler.com)
 #@ Reason : Another two workarounds for Dell systems (VGA Cable Pres 0, Add-in Card 4 PEM Presence 0)
 #@---------------------------------------------------
+#@ Date   : 20150109
+#@ Author : Claudio Kuenzler (www.claudiokuenzler.com)
+#@ Reason : Output serial number of chassis if a blade server is checked
+#@---------------------------------------------------
+#@ Date   : 20150119
+#@ Author : Andreas Gottwald
+#@ Reason : Fix NoneType element bug
+#@---------------------------------------------------
+#@ Date   : 20150626
+#@ Author : Claudio Kuenzler (www.claudiokuenzler.com)
+#@ Reason : Added support for patched pywbem 0.7.0 and new version 0.8.0, handle SSL error exception
+#@---------------------------------------------------
+#@ Date   : 20150710
+#@ Author : Stanislav German-Evtushenko
+#@ Reason : Exit Unknown instead of Critical for timeouts and auth errors
+#@---------------------------------------------------
+#@ Date   : 20151111
+#@ Author : Stefan Roos
+#@ Reason : Removed unused sensor_value variable and string import.
+#@ Reason : Added global hosturl variable declaration after imports.
+#@---------------------------------------------------
+#@ Date   : 20160411
+#@ Author : Claudio Kuenzler (www.claudiokuenzler.com)
+#@ Reason : Distinguish between pywbem 0.7 and 0.8 (which is now released)
+#@---------------------------------------------------
+#@ Date   : 20160531
+#@ Author : Claudio Kuenzler (www.claudiokuenzler.com)
+#@ Reason : Add parameter for variable CIM port (useful when behind NAT)
+#@---------------------------------------------------
+#@ Date   : 20161013
+#@ Author : Claudio Kuenzler (www.claudiokuenzler.com)
+#@ Reason : Added support for pywbem 0.9.x (and upcoming releases)
+#@---------------------------------------------------
+#@ Date   : 20170905
+#@ Author : Claudio Kuenzler (www.claudiokuenzler.com)
+#@ Reason : Added option to ignore LCD/Display related elements (--no-lcd)
+#@---------------------------------------------------
+#@ Date   : 20180329
+#@ Author : Claudio Kuenzler (www.claudiokuenzler.com)
+#@ Reason : Try to use internal pywbem function to determine version
+#@---------------------------------------------------
+#@ Date   : 20180411
+#@ Author : Peter Newman
+#@ Reason : Throw an unknown if we can't fetch the data for some reason
+#@---------------------------------------------------
+#@ Date   : 20181001
+#@ Author : Claudio Kuenzler
+#@ Reason : python3 compatibility
+#@---------------------------------------------------
+#@ Date   : 20190510
+#@ Author : Claudio Kuenzler
+#@ Reason : Allow regular expressions from ignore list (-r)
+#@---------------------------------------------------
 
+from __future__ import print_function
 import sys
 import time
 import pywbem
 import re
-import string
+import pkg_resources
 from optparse import OptionParser,OptionGroup
 
-version = '20140319'
+version = '20190510'
 
 NS = 'root/cimv2'
+hosturl = ''
 
 # define classes to check 'OperationStatus' instance
 ClassesToCheck = [
@@ -278,6 +337,9 @@ perf_Prefix = {
 # host name
 hostname=''
 
+# cim port
+cimport=''
+
 # user
 user=''
 
@@ -298,6 +360,8 @@ timeout = 0
 
 # elements to ignore (full SEL, broken BIOS, etc)
 ignore_list=[]
+regex_ignore_list=[]
+regex=False
 
 # urlise model and tag numbers (currently only Dell supported, but the code does the right thing for other vendors)
 urlise_country=''
@@ -308,6 +372,7 @@ get_volts   = True
 get_current = True
 get_temp    = True
 get_fan     = True
+get_lcd     = True
 
 # define exit codes
 ExitOK = 0
@@ -315,6 +380,9 @@ ExitWarning = 1
 ExitCritical = 2
 ExitUnknown = 3
 
+# Special handling for blade servers
+isblade = "no"
+
 def dell_country(country):
   if country == 'at':  # Austria
     return 'at/de/'
@@ -424,19 +492,16 @@ def urlised_serialnumber(vendor,country,SerialNumber):
 
 def verboseoutput(message) :
   if verbose:
-    print "%s %s" % (time.strftime("%Y%m%d %H:%M:%S"), message)
+    print(time.strftime("%Y%m%d %H:%M:%S"), message)
 
 # ----------------------------------------------------------------------
 
 def getopts() :
-  global hosturl,user,password,vendor,verbose,perfdata,urlise_country,timeout,ignore_list,get_power,get_volts,get_current,get_temp,get_fan
-  usage = "usage: %prog  https://hostname user password system [verbose]\n" \
-    "example: %prog https://my-shiny-new-vmware-server root fakepassword dell\n\n" \
-    "or, using new style options:\n\n" \
-    "usage: %prog -H hostname -U username -P password [-V system -v -p -I XX]\n" \
-    "example: %prog -H my-shiny-new-vmware-server -U root -P fakepassword -V auto -I uk\n\n" \
+  global hosturl,cimport,user,password,vendor,verbose,perfdata,urlise_country,timeout,ignore_list,regex,get_power,get_volts,get_current,get_temp,get_fan,get_lcd
+  usage = "usage: %prog -H hostname -U username -P password [-C port -V vendor -v -p -I XX -i list,list -r]\n" \
+    "example: %prog -H hostname -U root -P password -C 5989 -V auto -I uk\n\n" \
     "or, verbosely:\n\n" \
-    "usage: %prog --host=hostname --user=username --pass=password [--vendor=system --verbose --perfdata --html=XX]\n"
+    "usage: %prog --host=hostname --user=username --pass=password [--cimport=port --vendor=system --verbose --perfdata --html=XX]\n"
 
   parser = OptionParser(usage=usage, version="%prog "+version)
   group1 = OptionGroup(parser, 'Mandatory parameters')
@@ -447,6 +512,7 @@ def getopts() :
   group1.add_option("-P", "--pass", dest="password", \
       help="password, if password matches file:<path>, first line of given file will be used as password", metavar="PASS")
 
+  group2.add_option("-C", "--cimport", dest="cimport", help="CIM port (default 5989)", metavar="CIMPORT")
   group2.add_option("-V", "--vendor", dest="vendor", help="Vendor code: auto, dell, hp, ibm, intel, or unknown (default)", \
       metavar="VENDOR", type='choice', choices=['auto','dell','hp','ibm','intel','unknown'],default="unknown")
   group2.add_option("-v", "--verbose", action="store_true", dest="verbose", default=False, \
@@ -459,6 +525,8 @@ def getopts() :
       help="timeout in seconds - no effect on Windows (default = no timeout)")
   group2.add_option("-i", "--ignore", action="store", type="string", dest="ignore", default="", \
       help="comma-separated list of elements to ignore")
+  group2.add_option("-r", "--regex", action="store_true", dest="regex", default=False, \
+      help="allow regular expression lookup of ignore list")
   group2.add_option("--no-power", action="store_false", dest="get_power", default=True, \
       help="don't collect power performance data")
   group2.add_option("--no-volts", action="store_false", dest="get_volts", default=True, \
@@ -469,20 +537,22 @@ def getopts() :
       help="don't collect temperature performance data")
   group2.add_option("--no-fan", action="store_false", dest="get_fan", default=True, \
       help="don't collect fan performance data")
+  group2.add_option("--no-lcd", action="store_false", dest="get_lcd", default=True, \
+      help="don't collect lcd/front display status")
 
   parser.add_option_group(group1)
   parser.add_option_group(group2)
 
   # check input arguments
   if len(sys.argv) < 2:
-    print "no parameters specified\n"
+    print("no parameters specified\n")
     parser.print_help()
     sys.exit(-1)
   # if first argument starts with 'https://' we have old-style parameters, so handle in old way
   if re.match("https://",sys.argv[1]):
     # check input arguments
     if len(sys.argv) < 5:
-      print "too few parameters\n"
+      print("too few parameters\n")
       parser.print_help()
       sys.exit(-1)
     if len(sys.argv) > 5 :
@@ -500,7 +570,7 @@ def getopts() :
     mandatories = ['host', 'user', 'password']
     for m in mandatories:
       if not options.__dict__[m]:
-        print "mandatory parameter '--" + m + "' is missing\n"
+        print("mandatory parameter '--" + m + "' is missing\n")
         parser.print_help()
         sys.exit(-1)
 
@@ -514,17 +584,20 @@ def getopts() :
 
     user=options.user
     password=options.password
+    cimport=options.cimport
     vendor=options.vendor.lower()
     verbose=options.verbose
     perfdata=options.perfdata
     urlise_country=options.urlise_country.lower()
     timeout=options.timeout
     ignore_list=options.ignore.split(',')
+    regex=options.regex
     get_power=options.get_power
     get_volts=options.get_volts
     get_current=options.get_current
     get_temp=options.get_temp
     get_fan=options.get_fan
+    get_lcd=options.get_lcd
 
   # if user or password starts with 'file:', use the first string in file as user, second as password
   if (re.match('^file:', user) or re.match('^file:', password)):
@@ -553,12 +626,46 @@ if os_platform != "win32":
   on_windows = False
   import signal
   def handler(signum, frame):
-    print 'CRITICAL: Execution time too long!'
-    sys.exit(ExitCritical)
+    print('UNKNOWN: Execution time too long!')
+    sys.exit(ExitUnknown)
+
+if cimport:
+  verboseoutput("Using manually defined CIM port "+cimport)
+  hosturl += ':'+cimport 
+  
+# Append lcd related elements to ignore list if --no-lcd was used
+verboseoutput("LCD Status: %s" % get_lcd)
+if not get_lcd:
+  ignore_list.append("System Board 1 LCD Cable Pres 0: Connected")
+  ignore_list.append("System Board 1 VGA Cable Pres 0: Connected")
+  ignore_list.append("Front Panel Board 1 FP LCD Cable 0: Connected")
+  ignore_list.append("Front Panel Board 1 FP LCD Cable 0: Config Error")
 
 # connection to host
 verboseoutput("Connection to "+hosturl)
-wbemclient = pywbem.WBEMConnection(hosturl, (user,password), NS)
+# pywbem 0.7.0 handling is special, some patched 0.7.0 installations work differently
+try:
+  pywbemversion = pywbem.__version__
+except:
+  pywbemversion = pkg_resources.get_distribution("pywbem").version
+else:
+  pywbemversion = pywbem.__version__
+verboseoutput("Found pywbem version "+pywbemversion)
+
+if '0.7.' in pywbemversion:
+  try:
+    conntest = pywbem.WBEMConnection(hosturl, (user,password))
+    c = conntest.EnumerateInstances('CIM_Card')
+  except:
+    #raise
+    verboseoutput("Connection error, disable SSL certificate verification (probably patched pywbem)")
+    wbemclient = pywbem.WBEMConnection(hosturl, (user,password), no_verification=True)
+  else:
+    verboseoutput("Connection worked")
+    wbemclient = pywbem.WBEMConnection(hosturl, (user,password))
+# pywbem 0.8.0 and later
+else:
+  wbemclient = pywbem.WBEMConnection(hosturl, (user,password), NS, no_verification=True)
 
 # Add a timeout for the script. When using with Nagios, the Nagios timeout cannot be < than plugin timeout.
 if on_windows == False and timeout > 0:
@@ -577,16 +684,19 @@ ExitMsg = ""
 if vendor=='auto':
   try:
     c=wbemclient.EnumerateInstances('CIM_Chassis')
-  except pywbem.cim_operations.CIMError,args:
+  except pywbem.cim_operations.CIMError as args:
     if ( args[1].find('Socket error') >= 0 ):
-      print "UNKNOWN: %s" %args
+      print("UNKNOWN: {}".format(args))
+      sys.exit (ExitUnknown)
+    elif ( args[1].find('ThreadPool --- Failed to enqueue request') >= 0 ):
+      print("UNKNOWN: {}".format(args))
       sys.exit (ExitUnknown)
     else:
       verboseoutput("Unknown CIM Error: %s" % args)
-  except pywbem.cim_http.AuthError,arg:
+  except pywbem.cim_http.AuthError as arg:
     verboseoutput("Global exit set to UNKNOWN")
     GlobalStatus = ExitUnknown
-    print "UNKNOWN: Authentication Error"
+    print("UNKNOWN: Authentication Error")
     sys.exit (GlobalStatus)
   else:
     man=c[0][u'Manufacturer']
@@ -605,27 +715,37 @@ for classe in ClassesToCheck :
   verboseoutput("Check classe "+classe)
   try:
     instance_list = wbemclient.EnumerateInstances(classe)
-  except pywbem.cim_operations.CIMError,args:
+  except pywbem.cim_operations.CIMError as args:
     if ( args[1].find('Socket error') >= 0 ):
-      print "CRITICAL: %s" %args
-      sys.exit (ExitCritical)
+      print("UNKNOWN: {}".format(args))
+      sys.exit (ExitUnknown)
+    elif ( args[1].find('ThreadPool --- Failed to enqueue request') >= 0 ):
+      print("UNKNOWN: {}".format(args))
+      sys.exit (ExitUnknown)
     else:
       verboseoutput("Unknown CIM Error: %s" % args)
-  except pywbem.cim_http.AuthError,arg:
+  except pywbem.cim_http.AuthError as arg:
     verboseoutput("Global exit set to UNKNOWN")
-    GlobalStatus = ExitCritical
-    print "UNKNOWN: Authentication Error"
+    GlobalStatus = ExitUnknown
+    print("UNKNOWN: Authentication Error")
     sys.exit (GlobalStatus)
   else:
     # GlobalStatus = ExitOK #ARR
     for instance in instance_list :
-      sensor_value = ""
       elementName = instance['ElementName']
+      if elementName is None :
+        elementName = 'Unknown'
       elementNameValue = elementName
       verboseoutput("  Element Name = "+elementName)
 
       # Ignore element if we don't want it
-      if elementName in ignore_list :
+      if (regex == True) and (len(ignore_list) > 0) :
+        for ignore in ignore_list :
+          if re.search(ignore, elementName, re.IGNORECASE) :
+            verboseoutput("    (ignored through regex)")
+            regex_ignore_list.append(elementName)
+
+      if (elementName in ignore_list) or (elementName in regex_ignore_list) :
         verboseoutput("    (ignored)")
         continue
 
@@ -642,6 +762,7 @@ for classe in ClassesToCheck :
           man = 'Unknown Manufacturer'
         verboseoutput("    Manufacturer = "+man)
         SerialNumber = instance[u'SerialNumber']
+        SerialChassis = instance[u'SerialNumber']
         if SerialNumber:
           verboseoutput("    SerialNumber = "+SerialNumber)
         server_info = man + ' '
@@ -655,6 +776,7 @@ for classe in ClassesToCheck :
         SerialNumber = instance[u'SerialNumber']
         if SerialNumber:
           verboseoutput("    SerialNumber = "+SerialNumber)
+          isblade = "yes"
 
       # Report detail of Numeric Sensors and generate nagios perfdata
 
@@ -723,7 +845,6 @@ for classe in ClassesToCheck :
         verboseoutput("    Family = %d" % instance['Family'])
         verboseoutput("    CurrentClockSpeed = %dMHz" % instance['CurrentClockSpeed'])
 
-
       # HP Check
       if vendor == "hp" :
         if instance['HealthState'] is not None :
@@ -739,11 +860,11 @@ for classe in ClassesToCheck :
             30 : ExitCritical,  # Non-recoverable Error
           }[elementStatus]
           if (interpretStatus == ExitCritical) :
-            verboseoutput("GLobal exit set to CRITICAL")
+            verboseoutput("Global exit set to CRITICAL")
             GlobalStatus = ExitCritical
             ExitMsg += " CRITICAL : %s " % elementNameValue
           if (interpretStatus == ExitWarning and GlobalStatus != ExitCritical) :
-            verboseoutput("GLobal exit set to WARNING")
+            verboseoutput("Global exit set to WARNING")
             GlobalStatus = ExitWarning
             ExitMsg += " WARNING : %s " % elementNameValue
           # Added the following for when GlobalStatus is ExitCritical and a warning is detected
@@ -760,9 +881,7 @@ for classe in ClassesToCheck :
       elif (vendor == "dell" or vendor == "intel" or vendor == "ibm" or vendor=="unknown") :
         # Added 20121027 As long as Dell doesnt correct these CIM elements return code we have to ignore it
         ignore_list.append("System Board 1 Riser Config Err 0: Connected")
-        ignore_list.append("System Board 1 LCD Cable Pres 0: Connected")
-	ignore_list.append("System Board 1 VGA Cable Pres 0: Connected")
-	ignore_list.append("Add-in Card 4 PEM Presence 0: Connected")
+        ignore_list.append("Add-in Card 4 PEM Presence 0: Connected")
         if instance['OperationalStatus'] is not None :
           elementStatus = instance['OperationalStatus'][0]
           verboseoutput("    Element Op Status = %d" % elementStatus)
@@ -794,7 +913,7 @@ for classe in ClassesToCheck :
             GlobalStatus = ExitCritical
             ExitMsg += " CRITICAL : %s " % elementNameValue
           if (interpretStatus == ExitWarning and GlobalStatus != ExitCritical) :
-            verboseoutput("GLobal exit set to WARNING")
+            verboseoutput("Global exit set to WARNING")
             GlobalStatus = ExitWarning
             ExitMsg += " WARNING : %s " % elementNameValue
           # Added same logic as in 20100702 here, otherwise Dell servers would return UNKNOWN instead of OK
@@ -813,6 +932,10 @@ if (urlise_country != '') :
   SerialNumber = urlised_serialnumber(vendor,urlise_country,SerialNumber)
   server_info = urlised_server_info(vendor,urlise_country,server_info)
 
+# If this is a blade server, also output chassis serial number as additional info
+if (isblade == "yes") :
+  SerialNumber += " Chassis S/N: %s " % (SerialChassis)
+
 # Output performance data
 perf = '|'
 if perfdata:
@@ -834,12 +957,12 @@ if perf == '|':
   perf = ''
 
 if GlobalStatus == ExitOK :
-  print "OK - Server: %s %s %s%s" % (server_info, SerialNumber, bios_info, perf)
+  print("OK - Server: %s %s %s%s" % (server_info, SerialNumber, bios_info, perf))
 
 elif GlobalStatus == ExitUnknown :
-  print "UNKNOWN: %s" % (ExitMsg) #ARR
+  print("UNKNOWN: %s" % (ExitMsg)) #ARR
 
 else:
-  print "%s- Server: %s %s %s%s" % (ExitMsg, server_info, SerialNumber, bios_info, perf)
+  print("%s - Server:  %s %s %s%s" % (ExitMsg, server_info, SerialNumber, bios_info, perf))
 
 sys.exit (GlobalStatus)