zabbix/zabbixsite.py

   1 #!/usr/bin/python
   2
   3 from os import getcwd
   4 from os.path import dirname, exists, join
   5 import sys
   6 import md5
   7 import glob
   8
   9 from monitor import config
  10 from monitor.database.dborm import zab_session as session
  11 from monitor.database.zabbixapi.model import *
  12 from monitor.database.zabbixapi.emailZabbix import *
  13 from monitor.database.zabbixapi import defines
  14 from monitor.util.file import *
  15
  16
  17
  18 HOSTGROUP_NAME="%s_hostgroup"
  19 USERGROUP_NAME="%s_usergroup"
  20
  21 DISCOVERY_RULE_NAME="discovery rule for %s"
  22 DISCOVERY_ACTION_NAME="Auto-discover %s action"
  23 ESCALATION_ACTION_NAME="Escalation Action for %s"
  24
  25 def delete_site(loginbase):
  26
  27         # get host group, usrgrp
  28         # get all users in usrgrp, delete each
  29         usergroupname = USERGROUP_NAME % loginbase
  30         hostgroupname = HOSTGROUP_NAME % loginbase
  31         discovery_action_name = DISCOVERY_ACTION_NAME % loginbase
  32         discovery_rule_name = DISCOVERY_RULE_NAME % loginbase
  33         escalation_action_name = ESCALATION_ACTION_NAME % loginbase
  34
  35         ug = UsrGrp.get_by(name=usergroupname)
  36         if ug:
  37                 for user in ug.user_list:
  38                         # remove user from group, if a member of no other groups,
  39                         # delete user.
  40                         #user.delete()
  41                         pass
  42                 ug.delete()
  43
  44         hg = HostGroup.get_by(name=hostgroupname)
  45         if hg:
  46                 # figure out how to delete all the hosts...
  47                 # NOTE: hosts are listed in hg.host_list
  48                 for host in hg.host_list:
  49                         host.delete()
  50                 hg.delete()
  51
  52         # delete dr
  53         dr = DiscoveryRule.get_by(name=discovery_rule_name)
  54         if dr: dr.delete()
  55
  56         da = Action.get_by(name=discovery_action_name)
  57         if da: da.delete()
  58
  59         ea = Action.get_by(name=escalation_action_name)
  60         if ea: ea.delete()
  61
  62         return
  63
  64
  65 def setup_global():
  66         # GLOBAL:
  67         #       update mediatype for email.
  68         ############################### MAIL
  69         print "checking for MediaType Email"
  70         mediatype = MediaType.get_by(description="Email")
  71         if not mediatype:
  72                 print "ERROR:  There is no defined media type for 'Email'"
  73                 raise Exception("No Email Media type in Zabbix db")
  74
  75         print "checking for correct configuration"
  76         mediatype = MediaType.get_by(smtp_email=config.from_email)
  77         if not mediatype:
  78                 mediatype = MediaType.get_by(description="Email")
  79                 # NOTE: assumes smtp server is local to this machine.
  80                 print "updating email server configuration"
  81                 mediatype.smtp_server='localhost'
  82                 mediatype.smtp_helo=".".join(config.MONITOR_HOSTNAME.split('.')[1:])
  83                 mediatype.smtp_email=config.from_email
  84
  85         ############################# EMAIL
  86         mailtxt.reformat({'hostname' : config.MONITOR_HOSTNAME,
  87                                           'support_email' : config.support_email})
  88
  89         ############################### CENTRAL SERVER
  90         print "checking zabbix server host info"
  91         zabbixserver = Host.get_by(host="ZABBIX Server")
  92         if zabbixserver:
  93                 # TODO: verify that this works.  it has failed once on fresh
  94                 # install... not sure why.
  95                 print "Removing default Zabbix server entry"
  96                 zabbixserver.delete()
  97
  98                 # NOTE: creating a host and assigning a template cannot work
  99                 #       due to the crazy item, trigger, action
 100                 #               copying that the php code does during a host add.
 101                 # NOTE: Instead, reformat any *xml.in templates and import those
 102                 #               during /etc/plc.d/monitor sync
 103                 for file in glob.glob("%s/zabbix/templates/*.xml.in" config.MONITOR_SCRIPT_ROOT):
 104                         if 'zabbix_server' in file:
 105                                 buf = loadFile(file)
 106                                 args = {'hostname' : config.MONITOR_HOSTNAME, 'ip' : config.MONITOR_IP}
 107                                 dumpFile(file[:-3], buf % args)
 108
 109
 110         ##################### SCRIPTS
 111         ## TODO: add calls to check/reset the boot states.
 112         print "checking scripts"
 113         script1 = Script.find_or_create(name="RebootNode",
 114                                                                         set_if_new = {
 115                                                                                 'command':"%s/reboot.py {HOST.CONN}" % config.MONITOR_SCRIPT_ROOT,
 116                                                                                 'host_access':3 # r/w)
 117                                                                         })
 118         script2 = Script.find_or_create(name="NMap",
 119                                                         set_if_new = {
 120                                                                 'command':"/usr/bin/nmap -A {HOST.CONN}",
 121                                                                 'host_access':2 # r/o)
 122                                                 })
 123         return
 124
 125 def merge_iplist(iplist):
 126         # TODO:  rewrite addresses as x.x.x.y-z rather than x.x.x.y,x.x.x.z if y-z==1
 127         ips = iplist.split(',')
 128         ips.sort()
 129         prev=None
 130         newlist=""
 131         for ip in ips:
 132                 fields = ip.split('.')
 133                 first = ".".join(fields[:2])
 134                 last  = int(fields[3])
 135                 if prev:
 136                         if last - prev == 1:
 137                                 pass
 138                 prev=last
 139                 newlist += "%s,"
 140         return newlist[:-1]
 141
 142 def setup_site(loginbase, techemail, piemail, iplist):
 143
 144         # TODO: send a message when host is discovered.
 145
 146         # TODO: update 'discovered' hosts with dns name.
 147         # TODO: remove old nodes that are no longer in the plcdb.
 148         # TODO: remove old users that are no longer in the plcdb.
 149         # TODO: consider creating two user groups for Tech & PI emails
 150
 151         # NOTE: setup default valus for EMAIL
 152         mailtxt.reformat({'hostname' : config.MONITOR_HOSTNAME,
 153                                           'support_email' : config.support_email})
 154
 155         # NOTE: verify arguments
 156         if len(iplist) > 255:
 157                 raise Exception("iplist length is too long!")
 158
 159         BI_WEEKLY_ESC_PERIOD = int(60*60*24)
 160         #BI_WEEKLY_ESC_PERIOD = int(60) # testing...
 161
 162         # User Group
 163         site_user_group = UsrGrp.find_or_create(name=USERGROUP_NAME % loginbase)
 164         for user in set(techemail + piemail + [config.cc_email]):
 165                 if not user: continue
 166                 # USER
 167                 u = User.find_or_create(alias=user, type=1,
 168                                                                 set_if_new={'passwd' : md5.md5(user).hexdigest()},
 169                                                                 # exec_if_new avoids creating a Media object that
 170                                                                 # will not actually be used, if the user already exists
 171                                                                 exec_if_new=lambda obj: \
 172                                                                 obj.media_list.append( Media(mediatypeid=1, sendto=user)))
 173
 174                 if site_user_group not in u.usrgrp_list:
 175                         u.append_group(site_user_group)
 176
 177         # HOST GROUP
 178         plc_host_group = HostGroup.find_or_create(name="MyPLC Hosts")
 179         site_host_group = HostGroup.find_or_create(name=HOSTGROUP_NAME % loginbase)
 180         plctemplate = Host.get_by(host="Template_Linux_PLC_Host")
 181         escalation_action_name = ESCALATION_ACTION_NAME % loginbase
 182         discovery_action_name = DISCOVERY_ACTION_NAME % loginbase
 183         discovery_rule_name = DISCOVERY_RULE_NAME % loginbase
 184
 185         # ADD hg to ug
 186         if site_host_group not in site_user_group.hostgroup_list:
 187                 site_user_group.append_hostgroup(site_host_group)
 188
 189         # DISCOVERY RULE & CHECK
 190         dr = DiscoveryRule.find_or_create(name=discovery_rule_name,
 191                           delay=3600,
 192                           proxy_hostid=0,
 193                           set_if_new = {'iprange':iplist},
 194                           exec_if_new=lambda obj: \
 195                                 obj.discoverycheck_list.append( DiscoveryCheck(type=9,
 196                                                                                 key_="system.uname", ports=10050) )
 197                         )
 198         if dr.iprange != iplist:
 199                 dr.iprange = iplist
 200
 201
 202         # DISCOVERY ACTION for these servers
 203         a = Action.find_or_create(name=discovery_action_name,
 204                         eventsource=defines.EVENT_SOURCE_DISCOVERY,
 205                         status=defines.DRULE_STATUS_ACTIVE,
 206                         evaltype=defines.ACTION_EVAL_TYPE_AND_OR)
 207         if len(a.actioncondition_list) == 0:
 208                 a.actioncondition_list=[
 209                                         # Host IP Matches
 210                                         ActionCondition(
 211                                                 conditiontype=defines.CONDITION_TYPE_DHOST_IP,
 212                                                 operator=defines.CONDITION_OPERATOR_EQUAL,
 213                                                 value=iplist),
 214                                         # AND, Service type is Zabbix agent
 215                                         ActionCondition(
 216                                                 conditiontype=defines.CONDITION_TYPE_DSERVICE_TYPE,
 217                                                 operator=defines.CONDITION_OPERATOR_EQUAL,
 218                                                 value=defines.SVC_AGENT),
 219                                         # AND, Received system.uname value like 'Linux'
 220                                         ActionCondition(
 221                                                 conditiontype=defines.CONDITION_TYPE_DVALUE,
 222                                                 operator=defines.CONDITION_OPERATOR_LIKE,
 223                                                 value="Linux"),
 224                                         # AND, Discovery status is Discover
 225                                         ActionCondition(
 226                                                 conditiontype=defines.CONDITION_TYPE_DSTATUS,
 227                                                 operator=defines.CONDITION_OPERATOR_EQUAL,
 228                                                 value=defines.DOBJECT_STATUS_DISCOVER),
 229                                 ]
 230                                 # THEN
 231                 a.actionoperation_list=[
 232                                         # Add Host
 233                                         ActionOperation(
 234                                                 operationtype=defines.OPERATION_TYPE_HOST_ADD,
 235                                                 object=0, objectid=0,
 236                                                 esc_period=0, esc_step_from=1, esc_step_to=1),
 237                                         # Add To Group PLC Hosts
 238                                         ActionOperation(
 239                                                 operationtype=defines.OPERATION_TYPE_GROUP_ADD,
 240                                                 object=0, objectid=plc_host_group.groupid,
 241                                                 esc_period=0, esc_step_from=1, esc_step_to=1),
 242                                         # Add To Group LoginbaseSiteGroup
 243                                         ActionOperation(
 244                                                 operationtype=defines.OPERATION_TYPE_GROUP_ADD,
 245                                                 object=0, objectid=site_host_group.groupid,
 246                                                 esc_period=0, esc_step_from=1, esc_step_to=1),
 247                                         # Link to Template 'Template_Linux_Minimal'
 248                                         ActionOperation(
 249                                                 operationtype=defines.OPERATION_TYPE_TEMPLATE_ADD,
 250                                                 object=0, objectid=plctemplate.hostid,
 251                                                 esc_period=0, esc_step_from=1, esc_step_to=1),
 252                                 ]
 253         else:
 254                 # TODO: verify iplist is up-to-date
 255                 # NOTE: len(a.actioncondition_list) > 0
 256                 ip_condition  = a.actioncondition_list[0]
 257                 assert ip_condition.conditiontype == defines.CONDITION_TYPE_DHOST_IP
 258                 if ip_condition.value != iplist:
 259                         ip_condition.value = iplist
 260
 261         # ESCALATION ACTION for these servers
 262         ea = Action.find_or_create(name=escalation_action_name,
 263                         eventsource=defines.EVENT_SOURCE_TRIGGERS,
 264                         status=defines.ACTION_STATUS_ENABLED,
 265                         evaltype=defines.ACTION_EVAL_TYPE_AND_OR,
 266                         esc_period=BI_WEEKLY_ESC_PERIOD,        # three days
 267                         recovery_msg=1,
 268                         set_if_new={
 269                                 'r_shortdata':"Thank you for maintaining {HOSTNAME}!",
 270                                 'r_longdata': mailtxt.thankyou_nodeup, }
 271                         )
 272         if len(ea.actioncondition_list) == 0:
 273                         # THEN this is a new entry
 274                 print "SETTING UP ESCALATION ACTION"
 275                 ea.actioncondition_list=[
 276                                 ActionCondition(conditiontype=defines.CONDITION_TYPE_TRIGGER_VALUE,
 277                                                                 operator=defines.CONDITION_OPERATOR_EQUAL,
 278                                                                 value=defines.TRIGGER_VALUE_TRUE),
 279                                 ActionCondition(conditiontype=defines.CONDITION_TYPE_TRIGGER_NAME,
 280                                                                 operator=defines.CONDITION_OPERATOR_LIKE,
 281                                                                 value="is unreachable"),
 282                                 ActionCondition(conditiontype=defines.CONDITION_TYPE_HOST_GROUP,
 283                                                                 operator=defines.CONDITION_OPERATOR_EQUAL,
 284                                                                 value=site_host_group.groupid),
 285                         ]
 286                 ea.actionoperation_list=[
 287                                 # STAGE 1
 288                                 ActionOperation(operationtype=defines.OPERATION_TYPE_MESSAGE,
 289                                         shortdata=mailtxt.nodedown_one_subject,
 290                                         longdata=mailtxt.nodedown_one,
 291                                         object=defines.OPERATION_OBJECT_GROUP,
 292                                         objectid=site_user_group.usrgrpid,
 293                                         esc_period=0, esc_step_to=3, esc_step_from=3,
 294                                         operationcondition_list=[ OperationConditionNotAck() ] ),
 295                                 ActionOperation(operationtype=defines.OPERATION_TYPE_MESSAGE,
 296                                         shortdata=mailtxt.nodedown_one_subject,
 297                                         longdata=mailtxt.nodedown_one,
 298                                         object=defines.OPERATION_OBJECT_GROUP,
 299                                         objectid=site_user_group.usrgrpid,
 300                                         esc_period=0, esc_step_to=7, esc_step_from=7,
 301                                         operationcondition_list=[ OperationConditionNotAck() ] ),
 302                                 # STAGE 2
 303                                 ActionOperation(operationtype=defines.OPERATION_TYPE_COMMAND,
 304                                         esc_step_from=10, esc_step_to=10,
 305                                         esc_period=0,
 306                                         shortdata="",
 307                                         longdata="%s:%s/checkslices.py {HOSTNAME} disablesite" % ( config.MONITOR_HOSTNAME, config.MONITOR_SCRIPT_ROOT ),
 308                                         operationcondition_list=[ OperationConditionNotAck() ]),
 309                                 ActionOperation(operationtype=defines.OPERATION_TYPE_MESSAGE,
 310                                         shortdata=mailtxt.nodedown_two_subject,
 311                                         longdata=mailtxt.nodedown_two,
 312                                         esc_step_from=10, esc_step_to=10,
 313                                         esc_period=0,
 314                                         object=defines.OPERATION_OBJECT_GROUP,
 315                                         objectid=site_user_group.usrgrpid,
 316                                         operationcondition_list=[ OperationConditionNotAck() ] ),
 317                                 ActionOperation(operationtype=defines.OPERATION_TYPE_MESSAGE,
 318                                         shortdata=mailtxt.nodedown_two_subject,
 319                                         longdata=mailtxt.nodedown_two,
 320                                         esc_step_from=14, esc_step_to=14,
 321                                         esc_period=0,
 322                                         object=defines.OPERATION_OBJECT_GROUP,
 323                                         objectid=site_user_group.usrgrpid,
 324                                         operationcondition_list=[ OperationConditionNotAck() ] ),
 325
 326                                 # STAGE 3
 327                                 ActionOperation(operationtype=defines.OPERATION_TYPE_COMMAND,
 328                                         esc_step_from=17, esc_step_to=17,
 329                                         esc_period=0,
 330                                         shortdata="",
 331                                         longdata="%s:%s/checkslices.py {HOSTNAME} disableslices" % ( config.MONITOR_HOSTNAME, config.MONITOR_SCRIPT_ROOT ),
 332                                         # TODO: send notice to users of slices
 333                                         operationcondition_list=[ OperationConditionNotAck() ]),
 334                                 ActionOperation(operationtype=defines.OPERATION_TYPE_MESSAGE,
 335                                         shortdata=mailtxt.nodedown_three_subject,
 336                                         longdata=mailtxt.nodedown_three,
 337                                         esc_step_from=17, esc_step_to=17,
 338                                         esc_period=0,
 339                                         object=defines.OPERATION_OBJECT_GROUP,
 340                                         objectid=site_user_group.usrgrpid,
 341                                         operationcondition_list=[ OperationConditionNotAck() ] ),
 342                                 # STAGE 4++
 343                                 ActionOperation(operationtype=defines.OPERATION_TYPE_COMMAND,
 344                                         esc_step_from=21, esc_step_to=0,
 345                                         esc_period=int(BI_WEEKLY_ESC_PERIOD*3.5),
 346                                         shortdata="",
 347                                         longdata="%s:%s/checkslices.py {HOSTNAME} forever" % ( config.MONITOR_HOSTNAME, config.MONITOR_SCRIPT_ROOT),
 348                                         operationcondition_list=[ OperationConditionNotAck() ]),
 349                                 ActionOperation(operationtype=defines.OPERATION_TYPE_MESSAGE,
 350                                         shortdata=mailtxt.nodedown_four_subject,
 351                                         longdata=mailtxt.nodedown_four,
 352                                         esc_step_from=21, esc_step_to=0,
 353                                         esc_period=int(BI_WEEKLY_ESC_PERIOD*3.5),
 354                                         object=defines.OPERATION_OBJECT_GROUP,
 355                                         objectid=site_user_group.usrgrpid,
 356                                         operationcondition_list=[ OperationConditionNotAck() ] ),
 357                         ]
 358
 359 if __name__ == "__main__":
 360         setup_global()
 361         session.flush()