pcubad.py

   1 #!/usr/bin/python
   2
   3 import os
   4 import sys
   5 import string
   6 import time
   7
   8 from reboot import pcu_name
   9
  10 import soltesz
  11 import comon
  12 import threadpool
  13 import syncplcdb
  14 from nodequery import verify,query_to_dict,node_select
  15
  16 import plc
  17 import auth
  18 api = plc.PLC(auth.auth, auth.plc)
  19 from unified_model import *
  20 from monitor_policy import MINUP
  21
  22 round = 1
  23 externalState = {'round': round, 'nodes': {}}
  24 count = 0
  25
  26 def main(config):
  27         global externalState
  28         externalState = soltesz.if_cached_else(1, config.dbname, lambda : externalState)
  29         if config.increment:
  30                 # update global round number to force refreshes across all pcus
  31                 externalState['round'] += 1
  32
  33         l_plcpcus = soltesz.if_cached_else_refresh(1, 1, "pculist", lambda : plc.GetPCUs())
  34
  35         l_pcu = None
  36         if config.pcu:
  37                 for pcu in l_plcpcus:
  38                         if pcu['hostname'] == config.pcu  or pcu['ip'] == config.pcu:
  39                                 l_pcus = [pcu['pcu_id']]
  40                 if not l_pcu:
  41                         print "ERROR: could not find pcu %s" % config.pcu
  42                         sys.exit(1)
  43         else:
  44                 l_pcus = [pcu['pcu_id'] for pcu in l_plcpcus]
  45
  46         checkAndRecordState(l_pcus, l_plcpcus)
  47
  48 def checkAndRecordState(l_pcus, l_plcpcus):
  49         global externalState
  50         global count
  51         global_round = externalState['round']
  52
  53         for pcuname in l_pcus:
  54                 if pcuname not in externalState['nodes']:
  55                         externalState['nodes'][pcuname] = {'round': 0, 'values': []}
  56
  57                 pcu_round   = externalState['nodes'][pcuname]['round']
  58                 if pcu_round < global_round:
  59                         # do work
  60                         values = collectStatusAndState(pcuname, l_plcpcus)
  61                         global_round = externalState['round']
  62                         externalState['nodes'][pcuname]['values'] = values
  63                         externalState['nodes'][pcuname]['round'] = global_round
  64                 else:
  65                         count += 1
  66
  67                 if count % 20 == 0:
  68                         soltesz.dbDump(config.dbname, externalState)
  69
  70         soltesz.dbDump(config.dbname, externalState)
  71
  72 fbpcu = soltesz.dbLoad('findbadpcus')
  73 hn2lb = soltesz.dbLoad("plcdb_hn2lb")
  74
  75 def get(fb, path):
  76         indexes = path.split("/")
  77         values = fb
  78         for index in indexes:
  79                 if index in values:
  80                         values = values[index]
  81                 else:
  82                         return None
  83         return values
  84
  85 def collectStatusAndState(pcuname, l_plcpcus):
  86         global count
  87
  88         d_pcu = None
  89         for pcu in l_plcpcus:
  90                 if pcu['pcu_id'] == pcuname:
  91                         d_pcu = pcu
  92                         break
  93         if not d_pcu:
  94                 return None
  95
  96         pf = PersistFlags(pcuname, 1, db='pcu_persistflags')
  97
  98         if not pf.checkattr('last_changed'):
  99                 pf.last_changed = time.time()
 100
 101         pf.last_checked = time.time()
 102
 103         if not pf.checkattr('valid'):
 104                 pf.valid = "unknown"
 105                 pf.last_valid = 0
 106
 107         if not pf.checkattr('status'):
 108                 pf.status = "unknown"
 109
 110         state_path     = "nodes/id_" + str(pcuname) + "/values/reboot"
 111         bootstate_path = "nodes/id_" + str(pcuname) + "/values/plcpcu/boot_state"
 112
 113         current_state = get(fbpcu, state_path)
 114         if current_state == 0:
 115                 if pf.status != "good": pf.last_changed = time.time()
 116                 pf.status = "good"
 117         elif current_state == 'NetDown':
 118                 if pf.status != "netdown": pf.last_changed = time.time()
 119                 pf.status = "netdown"
 120         elif current_state == 'Not_Run':
 121                 if pf.status != "badconfig": pf.last_changed = time.time()
 122                 pf.status = "badconfig"
 123         else:
 124                 if pf.status != "error": pf.last_changed = time.time()
 125                 pf.status = "error"
 126
 127         count += 1
 128         print "%d %35s %s since(%s)" % (count, pcu_name(d_pcu), pf.status, diff_time(pf.last_changed))
 129         # updated by other modules
 130         #pf.enabled =
 131         #pf.suspended =
 132
 133         pf.save()
 134
 135         return True
 136
 137 if __name__ == '__main__':
 138         from config import config
 139         from optparse import OptionParser
 140         parser = OptionParser()
 141         parser.set_defaults(filename=None, pcu=None, pcuselect=False, pcugroup=None,
 142                                                 increment=False, dbname="pcubad", cachepcus=False)
 143         parser.add_option("", "--pcu", dest="pcu", metavar="hostname",
 144                                                 help="Provide a single pcu to operate on")
 145         parser.add_option("", "--pculist", dest="pculist", metavar="file.list",
 146                                                 help="Provide a list of files to operate on")
 147
 148         parser.add_option("", "--dbname", dest="dbname", metavar="FILE",
 149                                                 help="Specify the name of the database to which the information is saved")
 150         parser.add_option("-i", "--increment", action="store_true", dest="increment",
 151                                                 help="Increment round number to force refresh or retry")
 152         config = config(parser)
 153         config.parse_args()
 154
 155         try:
 156                 main(config)
 157         except Exception, err:
 158                 import traceback
 159                 print traceback.print_exc()
 160                 print "Exception: %s" % err
 161                 print "Saving data... exitting."
 162                 soltesz.dbDump(config.dbname, externalState)
 163                 sys.exit(0)