X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=grouprins.py;h=b85bbadd55c3dae8fb86c04c796ba2a588d70950;hb=refs%2Fheads%2F1.0;hp=97ba05b687297539d9c0e2fb30c8a72b762cc40d;hpb=c9b0045bba8ab66adf5036f9cac7f37f476b9a69;p=monitor.git

diff --git a/grouprins.py b/grouprins.py
index 97ba05b..b85bbad 100755
--- a/grouprins.py
+++ b/grouprins.py
@@ -67,12 +67,20 @@ class Reboot(object):
 			#pflags.resetRecentFlag('pcutried')
 			if not pflags.getRecentFlag('pcutried'):
 				try:
-					print "CALLING REBOOT!!!"
-					ret = reboot.reboot(host)
+					node_pf = PersistFlags(host, 1, db='node_persistflags')
+					if  node_pf.checkattr('last_change') and \
+						node_pf.last_change < time.time() - 60*60*24 and \
+						node_pf.checkattr('status') and \
+						node_pf.status != "good":
 
-					pflags.setRecentFlag('pcutried')
-					pflags.save()
-					return ret
+						print "CALLING REBOOT!!!"
+						ret = reboot.reboot(host)
+
+						pflags.setRecentFlag('pcutried')
+						pflags.save()
+						return ret
+					else:
+						return True
 
 				except Exception,e:
 					email_exception()
@@ -88,14 +96,26 @@ class Reboot(object):
 
 			elif not pflags.getRecentFlag('pcu_rins_tried'):
 				try:
-					# set node to 'rins' boot state.
-					print "CALLING REBOOT +++ RINS"
-					plc.nodeBootState(host, 'rins')
-					ret = reboot.reboot(host)
-
-					pflags.setRecentFlag('pcu_rins_tried')
-					pflags.save()
-					return ret
+					# NOTE: check that the node has been down for at least a
+					# day before rebooting it.  this avoids false-reboots/rins
+					# from failed node detections. circa 03-12-09
+					node_pf = PersistFlags(host, 1, db='node_persistflags')
+					if  node_pf.checkattr('last_change') and \
+						node_pf.last_change < time.time() - 60*60*24 and \
+						node_pf.checkattr('status') and \
+						node_pf.status != "good":
+
+						# set node to 'rins' boot state.
+						print "CALLING REBOOT +++ RINS"
+						plc.nodeBootState(host, 'reinstall')
+						ret = reboot.reboot(host)
+
+						pflags.setRecentFlag('pcu_rins_tried')
+						pflags.save()
+						return ret
+
+					else:
+						return True
 
 				except Exception,e:
 					email_exception()
@@ -169,11 +189,11 @@ def set_node_to_rins(host, fb):
 	node = api.GetNodes(host, ['boot_state', 'last_contact', 'last_updated', 'date_created'])
 	record = {'observation' : node[0], 
 			  'model' : 'USER_REQUEST', 
-			  'action' : 'api.UpdateNode(%s, {"boot_state" : "rins"})' % host, 
+			  'action' : 'api.UpdateNode(%s, {"boot_state" : "reinstall"})' % host, 
 			  'time' : time.time()}
 	l = Log(host, record)
 
-	ret = api.UpdateNode(host, {'boot_state' : 'rins'})
+	ret = api.UpdateNode(host, {'boot_state' : 'reinstall'})
 	if ret:
 		# it's nice to see the current status rather than the previous status on the console
 		node = api.GetNodes(host)[0]
@@ -227,7 +247,7 @@ config = parsermodule.parse_args(parser)
 
 # COLLECT nodegroups, nodes and node lists
 if config.nodegroup:
-	ng = api.GetNodeGroups({'name' : config.nodegroup})
+	ng = api.GetNodeGroups({'groupname' : config.nodegroup})
 	nodelist = api.GetNodes(ng[0]['node_ids'])
 	hostnames = [ n['hostname'] for n in nodelist ]