X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=policy.py;h=fe54863c8d41a6a4198d378cd232ce7b9eed6ce3;hb=32e64e33bc81735e22024c5a44510848bb3c88df;hp=7525c0d4e62933797e5022dc9011b4a8242dc966;hpb=7028fdb5e122089f2e1fb1c6a9178ab8f8aa5fe1;p=monitor.git

diff --git a/policy.py b/policy.py
index 7525c0d..fe54863 100755
--- a/policy.py
+++ b/policy.py
@@ -76,7 +76,7 @@ def main(hostnames, sitenames):
 				# NOTE: there is a narrow window in which this command must be
 				# 		evaluated, otherwise the notice will not go out.  
 				#		this is not ideal.
-				sitehist.sendMessage('online_notice', hostname=host, viart=False)
+				sitehist.sendMessage('online_notice', hostname=host, viart=False, saveact=True)
 				print "send message for host %s online" % host
 
 
@@ -119,12 +119,17 @@ def main(hostnames, sitenames):
 			#	sitehist.sendMessage('retry_bootman', hostname=host)
 
 		if nodehist.status == 'down' and \
-			changed_greaterthan(nodehist.last_changed, 2) and \
-			not found_within(recent_actions, 'down_notice', 3.5):
-				# send down node notice
-
-				sitehist.sendMessage('down_notice', hostname=host)
-				print "send message for host %s down" % host
+			changed_greaterthan(nodehist.last_changed, 2):
+				if not nodehist.firewall and not found_within(recent_actions, 'down_notice', 3.5):
+					# send down node notice
+					sitehist.sendMessage('down_notice', hostname=host)
+					print "send message for host %s down" % host
+
+				if nodehist.firewall and not found_within(recent_actions, 'firewall_notice', 3.5):
+					# send down node notice
+					email_exception(host, "firewall_notice")
+					sitehist.sendMessage('firewall_notice', hostname=host)
+					print "send message for host %s down" % host
 
 		node_count = node_count + 1
 		print "time: ", time.strftime('%Y-%m-%d %H:%M:%S')
@@ -134,9 +139,13 @@ def main(hostnames, sitenames):
 	for i,site in enumerate(sitenames):
 		sitehist = SiteInterface.get_or_make(loginbase=site)
 		siteblack = BlacklistRecord.get_by(loginbase=site)
+		skip_due_to_blacklist=False
 
 		if siteblack and not siteblack.expired():
 			print "skipping %s due to blacklist.  will expire %s" % (site, siteblack.willExpire() )
+			skip_due_to_blacklist=True
+			sitehist.clearPenalty()
+			sitehist.applyPenalty()
 			continue
 
 		# TODO: make query only return records within a certin time range,
@@ -144,33 +153,7 @@ def main(hostnames, sitenames):
 		recent_actions = sitehist.getRecentActions(loginbase=site)
 
 		print "%s %s %s" % (i, sitehist.db.loginbase, sitehist.db.status)
-		if sitehist.db.status == 'down':
-			if  not found_within(recent_actions, 'pause_penalty', 30) and \
-				not found_within(recent_actions, 'increase_penalty', 7) and \
-				changed_greaterthan(sitehist.db.last_changed, 7):
-
-				# TODO: catch errors
-				sitehist.increasePenalty()
-				#sitehist.applyPenalty()
-				sitehist.sendMessage('increase_penalty')
-
-				print "send message for site %s penalty increase" % site
-
-		if sitehist.db.status == 'good':
-			# clear penalty
-			# NOTE: because 'all clear' should have an indefinite status, we
-			# 		have a boolean value rather than a 'recent action'
-			if sitehist.db.penalty_applied:
-				# send message that penalties are cleared.
-
-				sitehist.clearPenalty()
-				#sitehist.applyPenalty()
-				sitehist.sendMessage('clear_penalty')
-				sitehist.closeTicket()
-
-				print "send message for site %s penalty cleared" % site
 
-		# find all ticket ids for site ( could be on the site record? )
 		# determine if there are penalties within the last 30 days?
 		# if so, add a 'pause_penalty' action.
 		if sitehist.db.message_id != 0 and sitehist.db.message_status == 'open' and \
@@ -178,6 +161,34 @@ def main(hostnames, sitenames):
 			#	pause escalation
 			print "Pausing penalties for %s" % site
 			sitehist.pausePenalty()
+		else:
+
+			if sitehist.db.status == 'down':
+				if  not found_within(recent_actions, 'pause_penalty', 30) and \
+					not found_within(recent_actions, 'increase_penalty', 7) and \
+					changed_greaterthan(sitehist.db.last_changed, 7):
+
+					# TODO: catch errors
+					sitehist.increasePenalty()
+					sitehist.applyPenalty()
+					sitehist.sendMessage('increase_penalty')
+
+					print "send message for site %s penalty increase" % site
+
+			if sitehist.db.status == 'good':
+				# clear penalty
+				# NOTE: because 'all clear' should have an indefinite status, we
+				# 		have a boolean value rather than a 'recent action'
+				if sitehist.db.penalty_applied:
+					# send message that penalties are cleared.
+
+					sitehist.clearPenalty()
+					sitehist.applyPenalty()
+					sitehist.sendMessage('clear_penalty')
+					sitehist.closeTicket()
+
+					print "send message for site %s penalty cleared" % site
+
 
 		site_count = site_count + 1
 
@@ -251,6 +262,6 @@ if __name__ == "__main__":
 		session.flush()
 		sys.exit(0)
 	except:
-		#email_exception()
+		email_exception()
 		print traceback.print_exc();
 		print "fail all..."