Tagging module Monitor - Monitor-2.0-6

[monitor.git] / bootman.py

diff --git a/bootman.py b/bootman.py
index e8dc7b8..640f9ee 100755 (executable)
--- a/bootman.py
+++ b/bootman.py
@@ -2,29 +2,29 @@
 
 # Attempt to reboot a node in debug state.
 
-import plc
+from monitor import const
+from monitor.database.info.model import *
+from monitor.wrapper import plc
 api = plc.getAuthAPI()
 
 import sys
 import os
-import const
 
 from getsshkeys import SSHKnownHosts
 
 import subprocess
 import time
-import database
-import moncommands
+from pcucontrol.util import command as moncommands
 from sets import Set
 
-import ssh.pxssh as pxssh
-import ssh.fdpexpect as fdpexpect
-import ssh.pexpect as pexpect
-from unified_model import *
-from emailTxt import mailtxt
+from pcucontrol.transports.ssh import pxssh as pxssh
+from pcucontrol.transports.ssh import fdpexpect as fdpexpect
+from pcucontrol.transports.ssh import pexpect as pexpect
+from monitor.model import *
+from monitor.wrapper.emailTxt import mailtxt
 from nodeconfig import network_config_to_str
 import traceback
-import config
+from monitor import config
 
 import signal
 class Sopen(subprocess.Popen):
@@ -317,7 +317,8 @@ 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                print "\tDisabling %s due to out-of-date BOOTCD" % hostname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'})
@@ -330,6 +331,8 @@ def reboot(hostname, config=None, forced_action=None):
        try:
                k = SSHKnownHosts(); k.update(node); k.write(); del k
        except:
+               from monitor.common import email_exception
+               email_exception()
                print traceback.print_exc()
                return False
 
@@ -339,8 +342,11 @@ def reboot(hostname, config=None, forced_action=None):
                else:
                        session = PlanetLabSession(node, config.nosetup, config.verbose)
        except Exception, e:
-               print "ERROR setting up session for %s" % hostname
+               msg = "ERROR setting up session for %s" % hostname
+               print msg
                print traceback.print_exc()
+               from monitor.common import email_exception
+               email_exception(msg)
                print e
                return False
 
@@ -354,6 +360,8 @@ def reboot(hostname, config=None, forced_action=None):
                        conn = session.get_connection(config)
                except:
                        print traceback.print_exc()
+                       from monitor.common import email_exception
+                       email_exception()
                        return False
 
        if forced_action == "reboot":
@@ -448,7 +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
 
@@ -582,6 +591,9 @@ def reboot(hostname, config=None, forced_action=None):
                        "bminit-cfg-auth-getplc-update-hardware-installinit-installdisk-installbootfs-installcfg-installstop-update-installinit-validate-rebuildinitrd-netcfg-disk-update4-update3-update3-kernelcopyfail-exception-update-debug-done",
                        "bminit-cfg-auth-getplc-hardware-installinit-installdisk-installbootfs-installcfg-installstop-update-installinit-validate-rebuildinitrd-netcfg-disk-update4-update3-update3-kernelcopyfail-exception-update-debug-done",
                        "bminit-cfg-auth-getplc-installinit-validate-exception-noinstall-update-debug-done",
+                       # actual solution appears to involve removing the bad files, and
+                       # continually trying to boot the node.
+                       "bminit-cfg-auth-getplc-update-installinit-validate-rebuildinitrd-netcfg-disk-update4-update3-update3-implementerror-update-debug-done",
                        ]:
                sequences.update({n : "restart_bootmanager_rins"})
 
@@ -610,6 +622,7 @@ def reboot(hostname, config=None, forced_action=None):
                         "bminit-cfg-auth-implementerror-bootcheckfail-update-implementerror-bootupdatefail-done",
                         "bminit-cfg-auth-getplc-update-installinit-validate-rebuildinitrd-netcfg-update3-implementerror-nospace-update-debug-done",
                         "bminit-cfg-auth-getplc-hardware-installinit-installdisk-installbootfs-exception-downloadfail-update-debug-done",
+                        "bminit-cfg-auth-getplc-update-installinit-validate-implementerror-update-debug-done",
                         ]:
                sequences.update({n: "restart_node_boot"})
 
@@ -666,7 +679,7 @@ 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')
 
@@ -704,7 +717,7 @@ 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')
 
@@ -715,7 +728,8 @@ 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')
 
@@ -724,10 +738,11 @@ def reboot(hostname, config=None, forced_action=None):
                        args = {}
                        args['hostname'] = hostname
                        args['bmlog'] = conn.get_bootmanager_log().read()
-                       m = PersistMessage(hostname,  mailtxt.plnode_network[0] % args,  mailtxt.plnode_cfg[1] % args, 
+                       m = PersistMessage(hostname,  mailtxt.plnode_cfg[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')
 
@@ -742,7 +757,8 @@ 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')
@@ -760,7 +776,8 @@ 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":
@@ -772,7 +789,8 @@ 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":
@@ -782,6 +800,8 @@ def reboot(hostname, config=None, forced_action=None):
                                node = api.GetNodes(hostname)[0]
                                net = api.GetNodeNetworks(node['nodenetwork_ids'])[0]
                        except:
+                               from monitor.common import email_exception
+                               email_exception()
                                print traceback.print_exc()
                                # TODO: api error. skip email, b/c all info is not available,
                                # flag_set will not be recorded.
@@ -795,7 +815,8 @@ 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:
@@ -808,7 +829,7 @@ def reboot(hostname, config=None, forced_action=None):
 # MAIN -------------------------------------------------------------------
 
 def main():
-       import parser as parsermodule
+       from monitor import parser as parsermodule
        parser = parsermodule.getParser()
 
        parser.set_defaults(child=False, collect=False, nosetup=False, verbose=False,