nodebad.py

   1 #!/usr/bin/python
   2
   3 import os
   4 import sys
   5 import string
   6 import time
   7 from datetime import datetime,timedelta
   8
   9 from nodequery import verify,query_to_dict,node_select
  10
  11 from nodecommon import *
  12
  13 from monitor import config
  14 from monitor.wrapper import plc,plccache
  15 from monitor.const import MINUP
  16 from monitor.database.info.model import  FindbadNodeRecord, HistoryNodeRecord
  17 from monitor.database.dborm import  mon_session as session
  18
  19 from monitor.model import *
  20
  21 api = plc.getAuthAPI()
  22
  23 round = 1
  24 count = 0
  25
  26 def main(config):
  27
  28         l_plcnodes = plccache.l_nodes
  29         l_nodes = get_nodeset(config)
  30
  31         checkAndRecordState(l_nodes, l_plcnodes)
  32
  33 def checkAndRecordState(l_nodes, l_plcnodes):
  34         global count
  35
  36         for nodename in l_nodes:
  37                 d_node = None
  38                 for node in l_plcnodes:
  39                         if node['hostname'] == nodename:
  40                                 d_node = node
  41                                 break
  42                 if not d_node:
  43                         continue
  44
  45                 pf = HistoryNodeRecord.findby_or_create(hostname=nodename)
  46                 pf.last_checked = datetime.now()
  47
  48                 try:
  49                         # Find the most recent record
  50                         noderec = FindbadNodeRecord.query.filter(FindbadNodeRecord.hostname==nodename).order_by(FindbadNodeRecord.date_checked.desc()).first()
  51                         #print "NODEREC: ", noderec.date_checked
  52                 except:
  53                         print "COULD NOT FIND %s" % nodename
  54                         import traceback
  55                         print traceback.print_exc()
  56                         continue
  57
  58                 if not noderec:
  59                         print "none object for %s"% nodename
  60                         continue
  61
  62                 node_state = noderec.observed_status
  63                 if noderec.plc_node_stats:
  64                         boot_state = noderec.plc_node_stats['boot_state']
  65                 else:
  66                         boot_state = "unknown"
  67
  68                 if node_state == "BOOT":
  69                         if pf.status != "good":
  70                                 pf.last_changed = datetime.now()
  71                                 pf.status = "good"
  72                 elif node_state == "DEBUG":
  73                         if pf.status != boot_state:
  74                                 pf.last_changed = datetime.now()
  75                                 pf.status = boot_state
  76                 else:
  77                         if pf.status != "down":
  78                                 pf.last_changed = datetime.now()
  79                                 pf.status = "down"
  80
  81                 count += 1
  82                 print "%d %35s %s since(%s)" % (count, nodename, pf.status, diff_time(time.mktime(pf.last_changed.timetuple())))
  83
  84         # NOTE: this commits all pending operations to the DB.  Do not remove, or
  85         # replace with another operations that also commits all pending ops, such
  86         # as session.commit() or flush() or something
  87         print HistoryNodeRecord.query.count()
  88         session.flush()
  89
  90         return True
  91
  92 if __name__ == '__main__':
  93         from monitor import parser as parsermodule
  94         parser = parsermodule.getParser(['nodesets'])
  95         parser.set_defaults(filename=None, node=None, nodeselect=False, nodegroup=None, cachenodes=False)
  96         parser = parsermodule.getParser(['defaults'], parser)
  97         config = parsermodule.parse_args(parser)
  98
  99         try:
 100                 main(config)
 101         except Exception, err:
 102                 import traceback
 103                 print traceback.print_exc()
 104                 print "Exception: %s" % err
 105                 sys.exit(0)