From: Stephen Soltesz <soltesz@cs.princeton.edu>
Date: Thu, 3 Jun 2010 17:35:30 +0000 (+0000)
Subject: add some service escalation templates
X-Git-Tag: monitor-3.1-1~41
X-Git-Url: http://git.onelab.eu/?p=monitor.git;a=commitdiff_plain;h=bd560be0fb9d54e343b67b705499073874fa1b5d

add some service escalation templates
---

diff --git a/tools/plc2nagios.py b/tools/plc2nagios.py
index de74d42..37e15a3 100755
--- a/tools/plc2nagios.py
+++ b/tools/plc2nagios.py
@@ -1,7 +1,5 @@
 #!/usr/bin/python
 
-import database
-
 class NagiosObject(object):
 	trans = {'d2_coords': '2d_coords'}
 
@@ -30,6 +28,18 @@ class HostGroup(NagiosObject):
 	def __init__(self, **kwargs):	
 		NagiosObject.__init__(self, "hostgroup", **kwargs)
 
+class HostEscalation(NagiosObject):
+	def __init__(self, **kwargs):	
+		NagiosObject.__init__(self, "hostescalation", **kwargs)
+
+class Contact(NagiosObject):
+	def __init__(self, **kwargs):	
+		NagiosObject.__init__(self, "contact", **kwargs)
+
+class ContactGroup(NagiosObject):
+	def __init__(self, **kwargs):	
+		NagiosObject.__init__(self, "contactgroup", **kwargs)
+
 class Service(NagiosObject):
 	def __init__(self, **kwargs):	
 		NagiosObject.__init__(self, "service", **kwargs)
@@ -38,10 +48,38 @@ class ServiceDependency(NagiosObject):
 	def __init__(self, **kwargs):	
 		NagiosObject.__init__(self, "servicedependency", **kwargs)
 
+class ServiceEscalation(NagiosObject):
+	def __init__(self, **kwargs):	
+		NagiosObject.__init__(self, "serviceescalation", **kwargs)
+
 class ServiceGroup(NagiosObject):
 	def __init__(self, **kwargs):	
 		NagiosObject.__init__(self, "servicegroup", **kwargs)
 
+def getContactsAndContactGroupsFor(lb, type, email_list):
+
+	contact_list = []
+	for person in email_list:
+		c1 = Contact(contact_name=person,
+						host_notifications_enabled=1,
+						service_notifications_enabled=1,
+						host_notification_period="24x7",
+						service_notification_period="24x7",
+						host_notification_options="d,r,s",
+						service_notification_options="c,r",
+						host_notification_commands="notify-by-email",
+						service_notification_commands="notify-by-email",
+						email=person)
+		contact_list.append(c1)
+
+	cg1 = ContactGroup(contactgroup_name="%s-%s" % (lb,type),
+						alias="%s-%s" % (lb,type),
+						members=",".join(email_list))
+
+	contact_list.append(cg1)
+
+	return contact_list
+
 
 globalservices = []
 for service in [('NET', "Network Services"),
@@ -58,9 +96,9 @@ for service in [('NET', "Network Services"),
 globalhost = [Host(	name="planetlab-host",
 					use="generic-host",
 					check_period="24x7",
-					check_interval="60",
-					retry_interval="5",
-					max_check_attempts="10",
+					check_interval="120",
+					retry_interval="10",
+					max_check_attempts="6",
 					check_command="check-host-alive",
 					contact_groups="admins",
 					register="0")]
@@ -68,11 +106,17 @@ globalhost = [Host(	name="planetlab-host",
 for obj in globalhost + globalservices:
 	print obj.toString()
 
-plcdb = database.dbLoad("l_plcsites")
-netid2ip = database.dbLoad("plcdb_netid2ip")
-lb2hn = database.dbLoad("plcdb_lb2hn")
+from monitor.wrapper import plccache
+
+plcdb = plccache.l_sites
+netid2ip = plccache.d_from_l(plccache.plc.api.GetInterfaces(), 'interface_id')
+lb2hn = plccache.plcdb_lb2hn
 
-for site in plcdb:
+sites = plccache.plc.api.GetSites([10243, 22, 10247, 138, 139, 10050, 10257, 18, 20, 
+							21, 10134, 24, 10138, 10141, 30, 31, 33, 10279, 41, 29, 10193, 10064, 81,
+							10194, 10067, 87, 10208, 10001, 233, 157, 10100, 10107])
+
+for site in sites:
 	shortname = site['abbreviated_name']
 	lb = site['login_base']
 	hg = HostGroup(hostgroup_name=lb, alias=shortname)
@@ -95,12 +139,22 @@ for site in plcdb:
 
 	print hg.toString()
 
+	# NOTE: do duplcate groups create duplicate emails?
+	cl1 = getContactsAndContactGroupsFor(lb, "techs", plccache.plc.getTechEmails(lb))
+	cl2 = getContactsAndContactGroupsFor(lb, "pis", plccache.plc.getPIEmails(lb))
+	# NOTE: slice users will change often.
+	cl3 = getContactsAndContactGroupsFor(lb, "sliceusers", plccache.plc.getSliceUserEmails(lb))
+
+	for c in [cl1,cl2,cl3]:
+		for i in c:
+			print i.toString()
+
 	for node in nodes:
 		hn = node['hostname']
 		if len(node['interface_ids']) == 0:
 			continue
 
-		ip = netid2ip[node['interface_ids'][0]]
+		ip = netid2ip[str(node['interface_ids'][0])]['ip']
 
 		if lon_x is not -1 and lat_y is not -1:
 			coords="%s,%s" % (lon_x, lat_y)
@@ -142,6 +196,30 @@ for site in plcdb:
 					servicegroups="NET,COTOP",
 					check_command="check_http!-p 3120 -t 120")
 
+		se1 = ServiceEscalation( host_name=hn,
+						service_description='aSSH',
+						first_notification=0,
+						last_notification=2,
+						notification_interval=24*60*3.5,
+						escalation_options="r,c",
+						contact_groups="%s-techs" % lb)
+
+		se2 = ServiceEscalation( host_name=hn,
+						service_description='aSSH',
+						first_notification=2,
+						last_notification=4,
+						notification_interval=24*60*3.5,
+						escalation_options="r,c",
+						contact_groups="%s-techs,%s-pis" % (lb,lb))
+
+		se3 = ServiceEscalation( host_name=hn,
+						service_description='aSSH',
+						first_notification=4,
+						last_notification=0,
+						notification_interval=24*60*3.5,
+						escalation_options="r,c",
+						contact_groups="%s-techs,%s-pis,%s-sliceusers" % (lb,lb,lb))
+
 		sd1 = ServiceDependency(host_name=hn,
 								service_description="aSSH",
 								dependent_host_name=hn,
@@ -160,6 +238,6 @@ for site in plcdb:
 								dependent_service_description="dCOTOP",
 								execution_failure_criteria="w,u,c,p",)
 
-		for service in [s1,s2,s3,s4,sd1,sd2,sd3]:
+		for service in [s1,s2,s3,s4,se1,se2,se3,sd1,sd2,sd3]:
 			print service.toString()