comon.py

   1 #
   2 # Copyright (c) 2004  The Trustees of Princeton University (Trustees).
   3 #
   4 # Faiyaz Ahmed <faiyaza@cs.princeton.edu>
   5 #
   6 # $Id: comon.py,v 1.6 2007/06/29 12:42:22 soltesz Exp $
   7 #
   8 # Get CoMon data, unsorted, in CSV, and create a huge hash.
   9 #
  10
  11
  12 import urllib2
  13 import httplib
  14 import time
  15 import Queue
  16 import logging
  17 import pickle
  18 from threading import *
  19 #httplib.HTTPConnection.debuglevel = 1
  20
  21 logger = logging.getLogger("monitor")
  22
  23 # Time between comon refresh
  24 COSLEEP=1200
  25
  26 # CoMon
  27 COMONURL = "http://summer.cs.princeton.edu/status/tabulator.cgi?table=table_nodeview"
  28
  29 # node type:
  30 # null == <not in DB?>
  31 #        0 ==
  32 #        1 == Prod
  33 #        2 == alpha
  34 #        3 == beta
  35
  36 # boot state:
  37 #       0 == new
  38 #       1 == boot
  39 #       2 == dbg
  40 #       3 == rins
  41 #       4 == ins
  42
  43
  44 class Comon(Thread):
  45         """
  46         cdb is the comon database (dictionary)
  47         all buckets is a queue of all problem nodes. This gets sent to rt to find
  48         tickets open for host.
  49         """
  50         def __init__(self, cdb, d_allplc_nodes, q_allbuckets):
  51                 self.codata = cdb
  52                 self.d_allplc_nodes = d_allplc_nodes
  53                 self.updated = time.time()
  54                 self.q_allbuckets = q_allbuckets
  55                 #self.comon_buckets = {"down" : "resptime%20==%200%20&&%20keyok==null",
  56                 #       "ssh": "sshstatus%20%3E%202h",
  57                 #       "clock_drift": "drift%20%3E%201m",
  58                 #       "dns": "dns1udp%20%3E%2080%20&&%20dns2udp%20%3E%2080",
  59                 #       "filerw": "filerw%3E0",
  60                 #       "dbg" : "keyok==0"}
  61                 self.comon_buckets = {
  62                         #"down" : "resptime==0 && keyok==null",
  63                         #"ssh": "sshstatus > 2h",
  64                         #"clock_drift": "drift > 1m",
  65                         #"dns": "dns1udp>80 && dns2udp>80",
  66                         #"filerw": "filerw > 0",
  67                         #"all" : ""
  68                         "dbg" : "keyok==0",
  69                         }
  70                 Thread.__init__(self)
  71
  72         def __tohash(self,rawdata):
  73                 # First line Comon returns is list of keys with respect to index
  74                 keys = rawdata.readline().rstrip().split(", ")
  75                 l_host = []
  76                 hash = {}
  77                 try:
  78                         i_ignored = 0
  79                         for line in rawdata.readlines():
  80                                 l_host = line.rstrip().split(", ")              # split the line on ', '
  81                                 hostname = l_host[0]
  82                                 if hostname in self.d_allplc_nodes:             # then we'll track it
  83                                         hash[hostname] = {}
  84                                         for i in range(1,len(keys)):
  85                                                 hash[hostname][keys[i]]=l_host[i]
  86                                 else:
  87                                         i_ignored += 1
  88
  89                         print "Retrieved %s hosts" % len(hash.keys())
  90                         print "Ignoring %d hosts" % i_ignored
  91
  92                         logger.debug("Retrieved %s hosts" % len(hash.keys()))
  93                         logger.debug("Ignoring %d hosts" % i_ignored)
  94                 except Exception, err:
  95                         logger.debug("No hosts retrieved")
  96                         return {}
  97                 return hash
  98
  99         # Update individual buckekts.  Hostnames only.
 100         def updatebuckets(self):
 101                 for (bucket,url) in self.comon_buckets.items():
 102                         logger.debug("COMON:  Updating bucket %s" % bucket)
 103                         tmp = self.coget(COMONURL + "&format=formatcsv&select='" + url + "'").keys()
 104                         setattr(self, bucket, tmp)
 105
 106         # Update ALL node information
 107         def updatedb(self):
 108                 # Get time of update
 109                 self.updated = time.time()
 110                 # Make a Hash, put in self.
 111                 self.codata.update(self.coget(COMONURL + "&format=formatcsv"))
 112
 113         def coget(self,url):
 114                 rawdata = None
 115                 print "Getting: %s" % url
 116                 try:
 117                         coserv = urllib2.Request(url)
 118                         coserv.add_header('User-Agent',
 119                                 'PL_Monitor +http://monitor.planet-lab.org/')
 120                         opener = urllib2.build_opener()
 121                         # Initial web get from summer.cs in CSV
 122                         rawdata = opener.open(coserv)
 123                 except urllib2.URLError, (err):
 124                         print "Attempting %s" %COMONURL
 125                         print "URL error (%s)" % (err)
 126                         rawdata = None
 127                 return self.__tohash(rawdata)
 128
 129         # Push nodes that are bad (in *a* bucket) into q(q_allbuckets)
 130         def push(self):
 131                 #buckets_per_node = []
 132                 #for bucket in self.comon.comon_buckets.keys():
 133                 #       if (hostname in getattr(self.comon, bucket)):
 134                 #               buckets_per_node.append(bucket)
 135
 136                 #loginbase = self.plcdb_hn2lb[hostname] # plc.siteId(node)
 137
 138                 #if not loginbase in self.sickdb:
 139                 #       self.sickdb[loginbase] = [{hostname: buckets_per_node}]
 140                 #else:
 141                 #       self.sickdb[loginbase].append({hostname: buckets_per_node})
 142
 143
 144                 print "calling Comon.push()"
 145                 for bucket in self.comon_buckets.keys():
 146                         #print "bucket: %s" % bucket
 147                         for host in getattr(self,bucket):
 148                                 diag_node = {}
 149                                 diag_node['nodename'] = host
 150                                 diag_node['message'] = None
 151                                 diag_node['bucket'] = [bucket]
 152                                 diag_node['stage'] = ""
 153                                 diag_node['args'] = None
 154                                 diag_node['info'] = None
 155                                 diag_node['time'] = time.time()
 156                                 #print "host: %s" % host
 157                                 self.q_allbuckets.put(diag_node)
 158
 159         def run(self):
 160                 self.updatedb()
 161                 self.updatebuckets()
 162                 self.push()
 163                 # insert signal that this is the final host
 164                 self.q_allbuckets.put("None")
 165
 166         def __repr__(self):
 167             return self
 168
 169 def main():
 170         logger.setLevel(logging.DEBUG)
 171         ch = logging.StreamHandler()
 172         ch.setLevel(logging.DEBUG)
 173         formatter = logging.Formatter('%(message)s')
 174         ch.setFormatter(formatter)
 175         logger.addHandler(ch)
 176
 177
 178         t = Queue.Queue()
 179         cdb = {}
 180         a = Comon(cdb,t)
 181         #for i in a.comon_buckets: print "%s : %s" % ( i, a.comon_buckets[i])
 182         a.start()
 183
 184         time.sleep(5)
 185         #for i in a.down: print i
 186
 187         time.sleep(5)
 188         #print cdb
 189         for host in cdb.keys():
 190                 #if cdb[host]['keyok'] == "0":
 191                 # null implies that it may not be in PL DB.
 192                 if  cdb[host]['bootstate'] != "null" and \
 193                         cdb[host]['bootstate'] == "2" and \
 194                         cdb[host]['keyok'] == "0":
 195                         print("%-40s \t Bootstate %s nodetype %s kernver %s keyok %s" % (
 196                                 host, cdb[host]['bootstate'], cdb[host]['nodetype'],
 197                                 cdb[host]['kernver'], cdb[host]['keyok']))
 198                         #ssh = soltesz.SSH('root', host)
 199                         #try:
 200                         #       val = ssh.run("uname -r")
 201                         #       print "%s == %s" % (host, val),
 202                         #except:
 203                         #       pass
 204         #       else:
 205         #               print("key mismatch at: %s" % host)
 206         #print a.codata['michelangelo.ani.univie.ac.at']
 207         #time.sleep(3)
 208         #a.push()
 209         #print a.filerw
 210         #print a.coget(COMONURL + "&format=formatcsv&select='" + a.comon_buckets['filerw'])
 211
 212         #os._exit(0)
 213 if __name__ == '__main__':
 214         import os
 215         try:
 216                 main()
 217         except KeyboardInterrupt:
 218                 print "Killed.  Exitting."
 219                 logger.info('Monitor Killed')
 220                 os._exit(0)