add email_exception() to all except: statements.

[monitor.git] / grouprins.py

diff --git a/grouprins.py b/grouprins.py
index d859727..97ba05b 100755 (executable)
--- a/grouprins.py
+++ b/grouprins.py
@@ -75,6 +75,7 @@ class Reboot(object):
                                        return ret
 
                                except Exception,e:
+                                       email_exception()
                                        print traceback.print_exc(); print e
 
                                        # NOTE: this failure could be an implementation issue on
@@ -97,6 +98,7 @@ class Reboot(object):
                                        return ret
 
                                except Exception,e:
+                                       email_exception()
                                        print traceback.print_exc(); print e
 
                                        # NOTE: this failure could be an implementation issue on
@@ -140,6 +142,7 @@ class Reboot(object):
                try:
                        return mailmonitor.reboot(host)
                except Exception, e:
+                       email_exception(host)
                        print traceback.print_exc(); print e
                        return False
 
@@ -228,6 +231,11 @@ if config.nodegroup:
        nodelist = api.GetNodes(ng[0]['node_ids'])
        hostnames = [ n['hostname'] for n in nodelist ]
 
+if config.site:
+       site = api.GetSites(config.site)
+       l_nodes = api.GetNodes(site[0]['node_ids'], ['hostname'])
+       hostnames = [ n['hostname'] for n in l_nodes ]
+
 if config.node or config.nodelist:
        if config.node: hostnames = [ config.node ] 
        else: hostnames = util.file.getListFromFile(config.nodelist)
@@ -257,6 +265,7 @@ for host in hostnames:
                try:
                        node = api.GetNodes(host)[0]
                except:
+                       email_exception()
                        print traceback.print_exc(); 
                        print "FAILED GETNODES for host: %s" % host
                        continue
@@ -281,6 +290,7 @@ for host in hostnames:
                                        # todo: send thank you, etc.
                                        mailmonitor.reboot(host)
                                except Exception, e:
+                                       email_exception()
                                        print traceback.print_exc(); print e
 
                                continue
@@ -339,10 +349,10 @@ for host in hostnames:
                                print "ALL METHODS OF RESTARTING %s FAILED" % host
                                args = {}
                                args['hostname'] = host
-                               m = PersistMessage(host, "ALL METHODS FAILED for %(hostname)s" % args,
-                                                                                        "CANNOT CONTACT", False, db='suspect_persistmessages')
-                               m.reset()
-                               m.send(['monitor-list@lists.planet-lab.org'])
+                               #m = PersistMessage(host, "ALL METHODS FAILED for %(hostname)s" % args,
+                               #                                                        "CANNOT CONTACT", False, db='suspect_persistmessages')
+                               #m.reset()
+                               #m.send(['monitor-list@lists.planet-lab.org'])
 
                        l = Log(host, record)
                        print l
@@ -351,6 +361,7 @@ for host in hostnames:
                print "Killed by interrupt"
                sys.exit(0)
        except:
+               email_exception()
                print traceback.print_exc();
                print "Continuing..."