address issues found on TP myplc. default values in getsshkeys, generalize
authorStephen Soltesz <soltesz@cs.princeton.edu>
Tue, 25 Nov 2008 22:22:13 +0000 (22:22 +0000)
committerStephen Soltesz <soltesz@cs.princeton.edu>
Tue, 25 Nov 2008 22:22:13 +0000 (22:22 +0000)
email addresses in bootman, and add extra param to monitor-default.conf

bootman.py
getsshkeys.py
monitor-default.conf
showlatlon.py

index 8a1baf5..4bd503b 100755 (executable)
@@ -324,7 +324,8 @@ def reboot(hostname, config=None, forced_action=None):
                                                        mailtxt.newbootcd_one[1] % args, True, db='bootcd_persistmessages')
 
                loginbase = plc.siteId(hostname)
-               m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
+               emails = plc.getTechEmails(loginbase)
+               m.send(emails) 
 
                print "\tDisabling %s due to out-of-date BOOTCD" % hostname
                api.UpdateNode(hostname, {'boot_state' : 'disable'})
@@ -455,7 +456,8 @@ def reboot(hostname, config=None, forced_action=None):
                                                                                 mailtxt.baddisk[1] % args, True, db='hardware_persistmessages')
 
                        loginbase = plc.siteId(hostname)
-                       m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
+                       emails = plc.getTechEmails(loginbase)
+                       m.send(emails) 
                        conn.set_nodestate('disable')
                        return False
 
@@ -674,7 +676,7 @@ def reboot(hostname, config=None, forced_action=None):
                m = PersistMessage(hostname, mailtxt.unknownsequence[0] % args,
                                                                         mailtxt.unknownsequence[1] % args, False, db='unknown_persistmessages')
                m.reset()
-               m.send(['monitor-list@lists.planet-lab.org'])
+               m.send([config.cc_email]) 
 
                conn.restart_bootmanager('boot')
 
@@ -712,7 +714,7 @@ def reboot(hostname, config=None, forced_action=None):
                        m = PersistMessage(hostname, "Suspicous error from BootManager on %s" % args,
                                                                                 mailtxt.unknownsequence[1] % args, False, db='suspect_persistmessages')
                        m.reset()
-                       m.send(['monitor-list@lists.planet-lab.org'])
+                       m.send([config.cc_email]) 
 
                        conn.restart_bootmanager('boot')
 
@@ -723,7 +725,8 @@ def reboot(hostname, config=None, forced_action=None):
                        m = PersistMessage(hostname,  mailtxt.plnode_cfg[0] % args,  mailtxt.plnode_cfg[1] % args, 
                                                                True, db='nodeid_persistmessages')
                        loginbase = plc.siteId(hostname)
-                       m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
+                       emails = plc.getTechEmails(loginbase)
+                       m.send(emails) 
                        conn.dump_plconf_file()
                        conn.set_nodestate('disable')
 
@@ -735,7 +738,8 @@ def reboot(hostname, config=None, forced_action=None):
                        m = PersistMessage(hostname,  mailtxt.plnode_network[0] % args,  mailtxt.plnode_cfg[1] % args, 
                                                                True, db='nodenet_persistmessages')
                        loginbase = plc.siteId(hostname)
-                       m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
+                       emails = plc.getTechEmails(loginbase)
+                       m.send(emails) 
                        conn.dump_plconf_file()
                        conn.set_nodestate('disable')
 
@@ -750,7 +754,8 @@ def reboot(hostname, config=None, forced_action=None):
                                                                mailtxt.newalphacd_one[1] % args, True, db='bootcd_persistmessages')
 
                        loginbase = plc.siteId(hostname)
-                       m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
+                       emails = plc.getTechEmails(loginbase)
+                       m.send(emails) 
 
                        print "\tDisabling %s due to out-of-date BOOTCD" % hostname
                        conn.set_nodestate('disable')
@@ -768,7 +773,8 @@ def reboot(hostname, config=None, forced_action=None):
                                                                                 mailtxt.baddisk[1] % args, True, db='hardware_persistmessages')
 
                        loginbase = plc.siteId(hostname)
-                       m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
+                       emails = plc.getTechEmails(loginbase)
+                       m.send(emails) 
                        conn.set_nodestate('disable')
 
                elif sequences[s] == "update_hardware_email":
@@ -780,7 +786,8 @@ def reboot(hostname, config=None, forced_action=None):
                                                                                 mailtxt.minimalhardware[1] % args, True, db='minhardware_persistmessages')
 
                        loginbase = plc.siteId(hostname)
-                       m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
+                       emails = plc.getTechEmails(loginbase)
+                       m.send(emails) 
                        conn.set_nodestate('disable')
 
                elif sequences[s] == "bad_dns_email":
@@ -803,7 +810,8 @@ def reboot(hostname, config=None, forced_action=None):
                                                                                 mailtxt.baddns[1] % args, True, db='baddns_persistmessages')
 
                        loginbase = plc.siteId(hostname)
-                       m.send([const.PIEMAIL % loginbase, const.TECHEMAIL % loginbase])
+                       emails = plc.getTechEmails(loginbase)
+                       m.send(emails) 
                        conn.set_nodestate('disable')
 
        if flag_set:
index fbfc65e..9f1d835 100755 (executable)
@@ -17,7 +17,12 @@ except:
 
 args = {}
 args['known_hosts'] =  os.environ['HOME'] + os.sep + ".ssh" + os.sep + "known_hosts"
-args['XMLRPC_SERVER'] = 'https://boot.planet-lab.org/PLCAPI/'
+try:
+       import config
+       args['XMLRPC_SERVER'] = config.API_SERVER
+except:
+       args['XMLRPC_SERVER'] = 'https://boot.planet-lab.org/PLCAPI/'
+       print "Using default API server %s"  args['XMLRPC_SERVER']
 
 class SSHKnownHosts:
        def __init__(self, args = args):
index d525196..86cefe5 100644 (file)
@@ -24,6 +24,8 @@ MONITOR_DATA_ROOT=/var/lib/monitor-server
 MONITOR_ARCHIVE_ROOT=/var/lib/monitor-server/archive-pdb
 
 email=monitor@another-lab.org
+# used by bootman for failed notices
+cc_email=monitor-list@lists.planet-lab.org
 
 [commandline]
 debug=0
index af01bd7..4289e3d 100755 (executable)
@@ -83,8 +83,8 @@ def main():
                begin = time.strftime(format)
 
        d = datetime_fromstr(begin)
-       fbstr = get_filefromglob(d, "production.findbad")
-       fbpcustr = get_filefromglob(d, "production.findbadpcus")
+       fbstr = get_filefromglob(d, "production.findbad")[0]
+       fbpcustr = get_filefromglob(d, "production.findbadpcus")[0]
 
        l_plcnodes = database.dbLoad("l_plcnodes")
        l_plcsites = database.dbLoad("l_plcsites")