build-nagios

   1 #!/usr/bin/ruby
   2
   3 # Copyright (c) 2004, 2005, 2006, 2007, 2008 Peter Palfrader <peter@palfrader.org>
   4
   5 require "yaml"
   6
   7 ORG="dsa"
   8 SHORTORG="dsa"
   9 GENERATED_PREFIX="./generated/"
  10
  11 nagios_filename = {};
  12 %w(hosts hostgroups services dependencies hostextinfo serviceextinfo servicegroups).each{
  13         |x| nagios_filename[x] = GENERATED_PREFIX+"auto-#{x}.cfg"
  14 }
  15 nagios_filename['nrpe'] = GENERATED_PREFIX+"nrpe_#{ ORG }.cfg"
  16
  17
  18 MAX_CHECK_ATTEMPTS_DEFAULT=6
  19
  20 NRPE_CHECKNAME="#{ ORG }_check_nrpe"           # check that takes one argument:  service name to be checked
  21 NRPE_CHECKNAME_HOST="#{ ORG }_check_nrpe_host" # check that takes two arguments: relay host on which to run check, service name to be checked
  22
  23 HOST_TEMPLATE_NAME='generic-host'          # host templates that all our host definitions use
  24 SERVICE_TEMPLATE_NAME='generic-service'    # host templates that all our host definitions use
  25 HOST_ALIVE_CHECK='check-host-alive'        # host alive check if server is pingable
  26 NRPE_PROCESS_SERVICE='process - nrpe'      # nrpe checks will depend on this one
  27
  28
  29 def warn (msg)
  30         STDERR.puts msg
  31 end
  32 def set_if_unset(hash, key, value)
  33         hash[key] = value unless hash.has_key?(key)
  34 end
  35 def set_complain_if_set(hash, key, value, type, name)
  36         throw "#{type} definition '#{name}' has '#{key}' already defined" if hash.has_key?(key)
  37         hash[key] = value
  38 end
  39
  40 # Make an array out of something.  If there is nothing, create an empty array
  41 # if it is just a string, make a list with just that element, if it already is
  42 # an array keep it.
  43 def ensure_array(something)
  44         if (something == nil)
  45                 result = []
  46         elsif something.kind_of?(String)
  47                 result = [ something ]
  48         elsif something.kind_of?(Array)
  49                 result = something
  50         else
  51                 throw "Do now know how to make an array out of #{something}: " + something.to_yaml
  52         end
  53         return result
  54 end
  55
  56
  57 # This class keeps track of the checks done via NRPE and makes sure
  58 # each gets a unique name.
  59 #
  60 # Unforutunately NRPE limits check names to some 30 characters, so
  61 # we need to mangle service names near the end.
  62 class Nrpe
  63         def initialize
  64                 @checks = {}
  65         end
  66
  67         def make_name( name, check )
  68                 name = name.tr_s("^a-zA-Z", "_").gsub("process", "ps")
  69
  70                 result = "#{ SHORTORG }_" + name[0,19]
  71
  72                 hash = ''
  73                 skew = ''
  74                 while (@checks.has_key?(result + hash))
  75                         # hash it, so that we don't lose uniqeness by cutting it off
  76                         hash = (check+skew).crypt("$1$")
  77                         hash = hash[-5,5]  # 5 chars are enough
  78                         hash.tr!("/", "_")
  79                         skew += ' ' # change it a bit so the hash changes
  80                 end
  81                 result += hash
  82                 return result      # max of 32 or so chars
  83         end
  84
  85         def add( name, check )
  86                 if @checks.has_value? check
  87                         @checks.each_pair{ |key, value|
  88                                 return key if value == check
  89                         }
  90                 end
  91                 key = make_name(name, check)
  92                 @checks[ key ] = check
  93                 return key
  94         end
  95
  96         def checks
  97                 return @checks
  98         end
  99 end
 100 $nrpe = Nrpe.new()
 101
 102
 103 # Prints the keys and values of hash to a file
 104 # This is the function that prints the bodies of most our
 105 # host/service/etc definitions
 106 #
 107 # It skips over such keys as are listed in exclude_keys
 108 # and also skips private keys (those starting with an underscre)
 109 def print_block(fd, kind, hash, exclude_keys)
 110         fd.puts "define #{kind} {"
 111         hash.each_pair{ |key, value|
 112                 next if key[0,1] == '_'
 113                 next if exclude_keys.include? key
 114                 fd.puts "       #{key}          #{value}"
 115         }
 116         fd.puts "}"
 117         fd.puts
 118 end
 119
 120 def merge_contacts(host, service)
 121         %w{contacts contact_groups}.each{ |k|
 122                 contacts = []
 123                 [host, service].each{ |source|
 124                         contacts.push source[k] if source.has_key?(k)
 125                 }
 126                 service[k] = contacts.join(",") unless contacts.empty?
 127         }
 128 end
 129
 130 # Add the service definition service to hosts
 131 # f is the file for service definitions, deps the file for dependencies
 132 def addService(hosts, service, files, servers)
 133
 134         set_if_unset        service, 'use'               , SERVICE_TEMPLATE_NAME
 135         set_if_unset        service, 'max_check_attempts', MAX_CHECK_ATTEMPTS_DEFAULT
 136
 137         service['max_check_attempts'] = MAX_CHECK_ATTEMPTS_DEFAULT + service['max_check_attempts'] if service['max_check_attempts'] < 0
 138
 139         if service['nrpe']
 140                 throw "We already have a check_command (#{service['check_command']}) but we are in the NRPE block (nrpe: #{service['nrpe']})."+
 141                         "  This should have been caught much earlier" if service.has_key?('check_command');
 142
 143                 check = $nrpe.add(service['service_description'], service['nrpe'])
 144                 service['check_command'] = "#{ NRPE_CHECKNAME }!#{ check }"
 145
 146                 service['depends'] = ensure_array( service['depends'] )
 147                 service['depends'] << NRPE_PROCESS_SERVICE unless service['service_description'] == NRPE_PROCESS_SERVICE  # Depend on NRPE unless we are it
 148         end
 149
 150         hosts.each{ |host|
 151                 s = service.clone
 152                 set_complain_if_set s, 'host_name', host, 'Service', s['service_description']
 153                 merge_contacts(servers[host], s)
 154
 155                 print_block files['services'], 'service', s, %w(nrpe runfrom remotecheck
 156                                                                 depends
 157                                                                 hosts hostgroups excludehosts excludehostgroups)
 158         }
 159
 160         if service['depends']
 161                 service['depends'].each{ |prerequisite|
 162                         hosts.each{ |host|
 163                                 prerequisite_host = host
 164                                 pre = prerequisite
 165                                 # split off a hostname if there's one
 166                                 bananasplit = prerequisite.split(':')
 167                                 if bananasplit.size == 2
 168                                         prerequisite_host = bananasplit[0]
 169                                         pre = bananasplit[1]
 170                                 elsif bananasplit.size > 2
 171                                         throw "Cannot prase prerequisite #{prerequisite} for service #{service['service_description']} into host:service"
 172                                 end
 173                                 dependency = {
 174                                         'host_name'                     => prerequisite_host,
 175                                         'service_description'           => pre,
 176                                         'dependent_host_name'           => host,
 177                                         'dependent_service_description' => service['service_description'],
 178                                         'execution_failure_criteria'    => 'n',
 179                                         'notification_failure_criteria' => 'w,u,c'
 180                                 };
 181                                 print_block files['dependencies'], 'servicedependency', dependency, %w()
 182                         }
 183                 }
 184         end
 185
 186
 187         set_complain_if_set service['_extinfo'], 'service_description' , service['service_description'], 'serviceextinfo', service['service_description']
 188         set_complain_if_set service['_extinfo'], 'host_name'           , hosts.join(',')               , 'serviceextinfo', service['service_description']
 189
 190         print_block files['serviceextinfo'], 'serviceextinfo', service['_extinfo'], %w()
 191 end
 192
 193 # hostlists in services can be given as both, single hosts and hostgroups
 194 # This functinn merges hostgroups and a simple list of hosts
 195 #
 196 # it also takes a prefix so that it can be used for excludelists as well
 197 def merge_hosts_and_hostgroups(service, servers, hostgroups, prefix)
 198         hosts = []
 199         hosts = service[prefix+'hosts'].split(/,/).map{ |x| x.strip } if service[prefix+'hosts']
 200         hosts.each{ |host|
 201                 throw "host #{host} does not exist - used in service #{service['service_description']}" unless servers[host]
 202         };
 203         if service[prefix+'hostgroups']
 204                 service[prefix+'hostgroups'].split(/,/).map{ |x| x.strip }.each{ |hg|
 205                         throw "hostgroup #{hg} does not exist - used in service #{service['service_description']}" unless hostgroups[hg]
 206                         hosts = hosts.concat hostgroups[hg]['_memberlist']
 207                 }
 208         end
 209
 210         return hosts
 211 end
 212
 213 # Figure out the hosts a given service applies to
 214 #
 215 # For a given service find the list of hosts minus excluded hosts that this service runs on
 216 def find_hosts(service, servers, hostgroups)
 217         hosts        = merge_hosts_and_hostgroups service, servers, hostgroups, ''
 218         excludehosts = merge_hosts_and_hostgroups service, servers, hostgroups, 'exclude'
 219
 220         excludehosts.each{ |host|
 221                 if hosts.delete(host) == nil
 222                         throw "Cannot remove host #{host} from service #{service['service_description']}: it's not included anyway or excluded twice."
 223                 end
 224         }
 225
 226         return hosts
 227 end
 228
 229 # Move all elements that have a key that starts with "extinfo-"
 230 # into the _extinfo subhash
 231 def split_away_extinfo(hash)
 232         hash['_extinfo'] = {}
 233         hash.keys.each{ |key|
 234                 if key[0, 8] == 'extinfo-'
 235                         hash['_extinfo'][ key[8, key.length-8] ] = hash[key]
 236                         hash.delete(key);
 237                 end
 238         }
 239 end
 240
 241
 242 #############################################################################################
 243 #############################################################################################
 244 #############################################################################################
 245
 246 # Load the config
 247 config = YAML::load( File.open( 'nagios-master.cfg' ) )
 248
 249 files = {}
 250 # Remove old created files
 251 nagios_filename.each_pair{ |name, filename|
 252         files[name] = File.new(filename, "w")
 253 }
 254
 255 #################################
 256 # create a few hostgroups
 257 #################################
 258 # create the "all" and "pingable" hostgroups
 259 config['hostgroups']['all'] = {}
 260 config['hostgroups']['all']['alias'] = "all servers"
 261 config['hostgroups']['all']['private'] = true
 262 config['hostgroups']['pingable'] = {}
 263 config['hostgroups']['pingable']['alias'] = "pingable servers"
 264 config['hostgroups']['pingable']['private'] = true
 265
 266 config['hostgroups'].each_pair{ |name, hg|
 267         throw "Empty hostgroup or hostgroup #{name} not a hash" unless hg.kind_of?(Hash)
 268         split_away_extinfo hg
 269
 270         hg['_memberlist'] = []
 271 }
 272
 273 config['servers'].each_pair{ |name, server|
 274         throw "Empty server or server #{name} not a hash" unless server.kind_of?(Hash)
 275
 276         split_away_extinfo server
 277
 278         throw "No hostgroups defined for #{name}" unless server['hostgroups']
 279         server['_hostgroups'] = server['hostgroups'].split(/,/).map{ |x| x.strip };
 280         server['_hostgroups'] << 'all'
 281         server['_hostgroups'] << 'pingable' unless server['pingable'] == false
 282
 283         server['_hostgroups'].each{ |hg|
 284                 throw "Hostgroup #{hg} is not defined" unless config['hostgroups'].has_key?(hg)
 285                 config['hostgroups'][hg]['_memberlist'] << name
 286         };
 287 }
 288
 289 config['servicegroups'] = {} unless config.has_key? 'servicegroups'
 290
 291 ##############
 292 # HOSTS
 293 ##############
 294 config['servers'].each_pair{ |name, server|
 295         # Formerly we used 'ip' instead of 'address' in our source file
 296         # Handle this change but warn                                   XXX
 297         if server.has_key?('ip')
 298                 STDERR.puts("Host definition for #{name} has an 'ip' field.  Please use 'address' instead");
 299                 server['address'] = server['ip'];
 300                 server.delete('ip');
 301         end
 302
 303         set_complain_if_set server, 'host_name'    , name, 'Host', name
 304         set_if_unset        server, 'alias'        , name
 305         set_if_unset        server, 'use'          , HOST_TEMPLATE_NAME
 306         set_if_unset        server, 'check_command', HOST_ALIVE_CHECK    unless server['pingable'] == false
 307
 308         print_block files['hosts']      , 'host'       , server            , %w(hostgroups pingable)
 309
 310
 311
 312         # Handle hostextinfo
 313         #config['hostgroups'][  server['_hostgroups'].first  ]['_extinfo'].each_pair{ |k, v|
 314         # find the first hostgroup that has extinfo
 315         extinfo = server['_hostgroups'].collect{ |hgname | config['hostgroups'][hgname]['_extinfo'] }.delete_if{ |ei| ei.size == 0 }.first
 316         if extinfo then
 317                 extinfo.each_pair do |k, v|
 318                         # substitute hostname into the notes_url
 319                         v = sprintf(v,name) if k == 'notes_url'
 320
 321                         set_if_unset server['_extinfo'], k ,v
 322                 end
 323         end
 324
 325         set_complain_if_set server['_extinfo'], 'host_name'       , name, 'hostextinfo', name
 326         set_if_unset        server['_extinfo'], 'vrml_image'      , server['_extinfo']['icon_image'] if server['_extinfo'].has_key?('icon_image')
 327         set_if_unset        server['_extinfo'], 'statusmap_image' , server['_extinfo']['icon_image'] if server['_extinfo'].has_key?('icon_image')
 328
 329         print_block files['hostextinfo'], 'hostextinfo', server['_extinfo'], %w()
 330 }
 331
 332
 333
 334 ##############
 335 # HOSTGROUPS
 336 ##############
 337 config['hostgroups'].each_pair{ |name, hg|
 338         next if hg['private']
 339
 340         set_complain_if_set hg, 'hostgroup_name', name                       , 'Hostgroup', name
 341         set_complain_if_set hg, 'members'       , hg['_memberlist'].join(","), 'Hostgroup', name
 342
 343         print_block files['hostgroups'], 'hostgroup', hg, %w()
 344 }
 345
 346
 347 ##############
 348 # SERVICES and DEPENDENCIES
 349 ##############
 350 config['services'].each{ |service|
 351         throw "Empty service or service not a hash" unless service.kind_of?(Hash)
 352
 353         split_away_extinfo service
 354
 355
 356         # Both 'name' and 'service_description' are valid for a service's name
 357         # Internally we only use service_description as that's nagios' official term
 358         if service.has_key?('name')
 359                 throw "Service definition has both a name (#{service['name']})" +
 360                       "and a service_description (#{service['service_description']})" if service.has_key?('service_description')
 361                 #STDERR.puts("Service definition #{service['name']} has a 'name' field.  Please use 'service_description' instead");
 362                 service['service_description'] = service['name'];
 363                 service.delete('name');
 364         end
 365         # Both 'check' and 'check_command' are valid for a service's check command
 366         # Internally we only use check_command as that's nagios' official term
 367         if service.has_key?('check')
 368                 throw "Service definition has both a check (#{service['check']})" +
 369                       "and a check_command (#{service['check_command']})" if service.has_key?('check_command')
 370                 #STDERR.puts("Service definition #{service['service_description']} has a 'check' field.  Please use 'check_command' instead");
 371                 service['check_command'] = service['check'];
 372                 service.delete('check');
 373         end
 374
 375
 376         hosts = find_hosts service, config['servers'], config['hostgroups']
 377         throw "no hosts for service #{service['service_description']}" if hosts.empty?
 378
 379         throw "nrpe, check, and remotecheck are mutually exclusive in service #{service['service_description']}" if
 380                 (service['nrpe'] ? 1 : 0) +
 381                 (service['check_command'] ? 1 : 0) +
 382                 (service['remotecheck'] ? 1 : 0)  >= 2
 383
 384         if service['runfrom'] && service['remotecheck']
 385                 # If the service check is to be run from a remote monitor server ("relay")
 386                 # add that as an NRPE check to be run on the relay and make this
 387                 # service also depend on NRPE on the relay
 388                 relay = service['runfrom']
 389
 390                 hosts.each{ |host|
 391                         # how to recursively copy this thing?
 392                         hostservice = YAML::load( service.to_yaml )
 393                         host_ip = config['servers'][host]['address']
 394                         throw "For some reason I do not have an address for #{host}.  This shouldn't be." unless host_ip
 395
 396                         check = $nrpe.add("#{host}_#{hostservice['service_description']}", hostservice['remotecheck'].gsub(/\$HOSTADDRESS\$/, host_ip))
 397                         hostservice['check_command'] = "#{NRPE_CHECKNAME_HOST}!#{ config['servers'][ relay ]['address'] }!#{ check }"
 398
 399                         # Make sure dependencies are an array.  If there are none, create an empty array
 400                         # if depends is just a string, make a list with just that element
 401                         hostservice['depends'] = ensure_array( hostservice['depends'] )
 402                         # And append this new dependency
 403                         hostservice['depends'] << "#{ relay }:#{ NRPE_PROCESS_SERVICE }";
 404
 405                         addService( [ host ], hostservice, files, config['servers'])
 406                 }
 407         elsif service['runfrom'] || service['remotecheck']
 408                 throw "runfrom and remotecheck must either appear both or not at all in service #{service['service_description']}"
 409                 throw "must not remotecheck without runfrom" if service['remotecheck']
 410         else
 411                 addService(hosts, service, files, config['servers'])
 412         end
 413 }
 414
 415
 416
 417 ##############
 418 # SERVICEGROUPS
 419 ##############
 420 config['servicegroups'].each_pair{ |name, sg|
 421         set_complain_if_set sg, 'servicegroup_name', name                       , 'Servicegroup', name
 422
 423         print_block files['servicegroups'], 'servicegroup', sg, %w()
 424 }
 425
 426
 427 ##############
 428 # NRPE config file
 429 ##############
 430 $nrpe.checks.each_pair{ |name, check|
 431         files['nrpe'].puts "command[#{ name }]=#{ check }"
 432 }
 433
 434