monitor/common.py

   1
   2 import time
   3 import struct
   4 from monitor import reboot
   5 from monitor import util
   6 from monitor.wrapper import plc
   7
   8 from datetime import datetime, timedelta
   9 from monitor.model import Message
  10 from monitor.database.info import HistoryNodeRecord
  11
  12 esc = struct.pack('i', 27)
  13 RED     = esc + "[1;31m"
  14 GREEN   = esc + "[1;32m"
  15 YELLOW  = esc + "[1;33m"
  16 BLUE    = esc + "[1;34m"
  17 LIGHTBLUE       = esc + "[1;36m"
  18 NORMAL  = esc + "[0;39m"
  19
  20 def red(str):
  21         return RED + str + NORMAL
  22
  23 def yellow(str):
  24         return YELLOW + str + NORMAL
  25
  26 def green(str):
  27         return GREEN + str + NORMAL
  28
  29 def lightblue(str):
  30         return LIGHTBLUE + str + NORMAL
  31
  32 def blue(str):
  33         return BLUE + str + NORMAL
  34
  35 def get_current_state(fbnode):
  36         if 'observed_status' in fbnode:
  37                 state = fbnode['observed_status']
  38         else:
  39                 state = "none"
  40         l = state.lower()
  41         if l == "debug": l = 'dbg '
  42         return l
  43
  44 def color_pcu_state(fbnode):
  45
  46         if 'plcnode' in fbnode and 'pcu_ids' in fbnode['plcnode'] and len(fbnode['plcnode']['pcu_ids']) > 0 :
  47                 values = reboot.get_pcu_values(fbnode['plcnode']['pcu_ids'][0])
  48                 if values == None:
  49                         return fbnode['pcu']
  50         else:
  51                 if 'pcu' not in fbnode:
  52                         return 'NOPCU'
  53                 else:
  54                         return fbnode['pcu']
  55
  56         if 'reboot' in values:
  57                 rb = values['reboot']
  58                 if rb == 0 or rb == "0":
  59                         return fbnode['pcu'] + "OK  "
  60                 elif "NetDown" == rb  or "Not_Run" == rb:
  61                         return fbnode['pcu'] + "DOWN"
  62                 else:
  63                         return fbnode['pcu'] + "BAD "
  64         else:
  65                 return fbnode['pcu'] + "BAD "
  66
  67 def color_boot_state(l):
  68         if    l == "dbg": return yellow("debg")
  69         elif  l == "dbg ": return yellow("debg")
  70         elif  l == "failboot": return yellow("debg")
  71         elif  l == "diag": return lightblue(l)
  72         elif  l == "diagnose": return lightblue(l)
  73         elif  l == "safeboot": return lightblue(l)
  74         elif  l == "disable": return red("dsbl")
  75         elif  l == "disabled": return red("dsbl")
  76         elif  l == "down": return red(l)
  77         elif  l == "boot": return green(l)
  78         elif  l == "rins": return blue(l)
  79         elif  l == "reinstall": return blue(l)
  80         else:
  81                 return l
  82
  83 def diff_time(timestamp, abstime=True):
  84         import math
  85         now = time.time()
  86         if timestamp == None:
  87                 return "unknown"
  88         if type(timestamp) == type(datetime.now()):
  89                 timestamp = time.mktime(timestamp.timetuple())
  90         if abstime:
  91                 diff = now - timestamp
  92         else:
  93                 diff = timestamp
  94         # return the number of seconds as a difference from current time.
  95         t_str = ""
  96         if diff < 60: # sec in min.
  97                 t = diff / 1
  98                 t_str = "%s sec ago" % int(math.ceil(t))
  99         elif diff < 60*60: # sec in hour
 100                 t = diff / (60)
 101                 t_str = "%s min ago" % int(math.ceil(t))
 102         elif diff < 60*60*24: # sec in day
 103                 t = diff / (60*60)
 104                 t_str = "%s hrs ago" % int(math.ceil(t))
 105         elif diff < 60*60*24*14: # sec in week
 106                 t = diff / (60*60*24)
 107                 t_str = "%s days ago" % int(math.ceil(t))
 108         elif diff <= 60*60*24*30: # approx sec in month
 109                 t = diff / (60*60*24*7)
 110                 t_str = "%s wks ago" % int(math.ceil(t))
 111         elif diff > 60*60*24*30: # approx sec in month
 112                 t = diff / (60*60*24*30)
 113                 t_str = "%s mnths ago" % int(t)
 114         return t_str
 115
 116 def getvalue(fb, path):
 117     indexes = path.split("/")
 118     values = fb
 119     for index in indexes:
 120         if index in values:
 121             values = values[index]
 122         else:
 123             return None
 124     return values
 125
 126 def nmap_port_status(status):
 127         ps = {}
 128         l_nmap = status.split()
 129         ports = l_nmap[4:]
 130
 131         continue_probe = False
 132         for port in ports:
 133                 results = port.split('/')
 134                 ps[results[0]] = results[1]
 135                 if results[1] == "open":
 136                         continue_probe = True
 137         return (ps, continue_probe)
 138
 139
 140 def nodegroup_display(node, fbdata, conf=None):
 141         node['current'] = get_current_state(fbdata)
 142
 143         s = fbdata['kernel_version'].split()
 144         if len(s) >=3:
 145                 node['kernel_version'] = s[2]
 146         else:
 147                 node['kernel_version'] = fbdata['kernel_version']
 148
 149         if '2.6' not in node['kernel_version']: node['kernel_version'] = ""
 150         if conf and not conf.nocolor:
 151             node['boot_state']  = color_boot_state(node['boot_state'])
 152             node['current']     = color_boot_state(node['current'])
 153
 154         if type(fbdata['plc_node_stats']['pcu_ids']) == type([]):
 155                 node['pcu'] = "PCU"
 156         node['lastupdate'] = diff_time(node['last_contact'])
 157
 158         pf = HistoryNodeRecord.get_by(hostname=node['hostname'])
 159         try:
 160                 node['lc'] = diff_time(pf.last_changed)
 161         except:
 162                 node['lc'] = "err"
 163
 164         ut = fbdata['comon_stats']['uptime']
 165         if ut != "null":
 166                 ut = diff_time(float(fbdata['comon_stats']['uptime']), False)
 167         node['uptime'] = ut
 168
 169         return "%(hostname)-42s %(boot_state)8s %(current)5s %(pcu)6s %(key)10.10s... %(kernel_version)35.35s %(lastupdate)12s, %(lc)s, %(uptime)s" % node
 170
 171 def datetime_fromstr(str):
 172         if '-' in str:
 173                 try:
 174                         tup = time.strptime(str, "%Y-%m-%d")
 175                 except:
 176                         tup = time.strptime(str, "%Y-%m-%d-%H:%M")
 177         elif '/' in str:
 178                 tup = time.strptime(str, "%m/%d/%Y")
 179         else:
 180                 tup = time.strptime(str, "%m/%d/%Y")
 181         ret = datetime.fromtimestamp(time.mktime(tup))
 182         return ret
 183
 184 def get_nodeset(config):
 185         """
 186                 Given the config values passed in, return the set of hostnames that it
 187                 evaluates to.
 188         """
 189         from monitor.wrapper import plccache
 190         api = plc.getAuthAPI()
 191         l_nodes = plccache.l_nodes
 192
 193         if config.nodelist:
 194                 f_nodes = util.file.getListFromFile(config.nodelist)
 195                 l_nodes = filter(lambda x: x['hostname'] in f_nodes, l_nodes)
 196         elif config.node:
 197                 f_nodes = [config.node]
 198                 l_nodes = filter(lambda x: x['hostname'] in f_nodes, l_nodes)
 199         elif config.nodegroup:
 200                 ng = api.GetNodeGroups({'name' : config.nodegroup})
 201                 l_nodes = api.GetNodes(ng[0]['node_ids'], ['hostname'])
 202         elif config.site:
 203                 site = api.GetSites(config.site)
 204                 l_nodes = api.GetNodes(site[0]['node_ids'], ['hostname'])
 205
 206         l_nodes = [node['hostname'] for node in l_nodes]
 207
 208         # perform this query after the above options, so that the filter above
 209         # does not break.
 210         if config.nodeselect:
 211                 fbquery = FindbadNodeRecord.get_all_latest()
 212                 node_list = [ n.hostname for n in fbquery ]
 213                 l_nodes = node_select(config.nodeselect, node_list, None)
 214
 215         return l_nodes
 216
 217 def email_exception(content=None, title=None):
 218     import config
 219     from monitor.model import Message
 220     import traceback
 221     msg=traceback.format_exc()
 222     if content:
 223         msg = content + "\n" + msg
 224
 225     full_title = "exception running monitor"
 226     if title:
 227         full_title = "exception running monitor %s" % title
 228
 229     m=Message(full_title, msg, False)
 230     m.send([config.exception_email])
 231     return
 232
 233 def changed_lessthan(last_changed, days):
 234         if datetime.now() - last_changed <= timedelta(days):
 235                 #print "last changed less than %s" % timedelta(days)
 236                 return True
 237         else:
 238                 #print "last changed more than %s" % timedelta(days)
 239                 return False
 240
 241 def changed_greaterthan(last_changed, days):
 242         if datetime.now() - last_changed > timedelta(days):
 243                 #print "last changed more than %s" % timedelta(days)
 244                 return True
 245         else:
 246                 #print "last changed less than %s" % timedelta(days)
 247                 return False
 248
 249 def found_between(recent_actions, action_type, lower, upper):
 250         return found_before(recent_actions, action_type, upper) and found_within(recent_actions, action_type, lower)
 251
 252 def found_before(recent_actions, action_type, within):
 253         for action in recent_actions:
 254                 if action_type == action.action_type and \
 255                                 action.date_created < (datetime.now() - timedelta(within)):
 256                         return True
 257         return False
 258
 259 def found_within(recent_actions, action_type, within):
 260         for action in recent_actions:
 261                 #print "%s - %s %s > %s - %s (%s) ==> %s" % (action.loginbase, action.action_type, action.date_created, datetime.now(), timedelta(within), datetime.now()-timedelta(within), action.date_created > (datetime.now() - timedelta(within)) )
 262                 if action_type == action.action_type and \
 263                                 action.date_created > (datetime.now() - timedelta(within)):
 264                                 #datetime.now() - action.date_created < timedelta(within):
 265                         # recent action of given type.
 266                         #print "%s found_within %s in recent_actions from %s" % (action_type, timedelta(within), action.date_created)
 267                         return True
 268
 269         print "%s NOT found_within %s in recent_actions" % (action_type, timedelta(within) )
 270         return False
 271