node_ssh/nst.py

   1 #!/usr/bin/python
   2
   3 import time, sys, urllib, os, tempfile, random
   4 import xmlrpclib
   5 from optparse import OptionParser
   6 from getpass import getpass
   7 from time import sleep
   8
   9 parser = OptionParser()
  10 parser.add_option("-c", "--config", action="store", dest="config",  help="Path to alternate config file")
  11 parser.add_option("-x", "--url", action="store", dest="url", help = "API URL")
  12 parser.add_option("-s", "--slice", action="store", dest="slice", help = "Name of slice to use")
  13 parser.add_option("-n", "--nodes", action="store", dest="nodes", help = "File that contains a list of nodes to try to access")
  14 parser.add_option("-k", "--key", action="store", dest="key", help = "Path to alternate public key")
  15 parser.add_option("-u", "--user", action="store", dest="user", help = "API user name")
  16 parser.add_option("-p", "--password", action="store", dest="password", help = "API password")
  17 parser.add_option("-v", "--verbose", action="store_true",  dest="verbose", help="Be verbose (default: %default)")
  18 (options, args) = parser.parse_args()
  19
  20 # If user is specified but password is not
  21 if options.user is not None and options.password is None:
  22     try:
  23         options.password = getpass()
  24     except (EOFError, KeyboardInterrupt):
  25         print
  26         sys.exit(0)
  27
  28 class Config:
  29
  30     def __init__(self, options):
  31
  32         # if options are specified use them
  33         # otherwise use options from config file
  34         if options.config: config_file = options.config
  35         else: config_file = '/usr/share/planetlab/tests/nst/nst_config'
  36
  37         try:
  38             execfile(config_file, self.__dict__)
  39         except:
  40             raise "Could not find nst config in " + config_file
  41
  42         if options.url: self.NST_API_SERVER = options.url
  43         if options.slice: self.NST_SLICE = options.slice
  44         if options.key: self.NST_KEY_PATH = options.key
  45         if options.user: self.NST_USER = options.user
  46         if options.password: self.NST_PASSWORD = options.password
  47         if options.nodes: self.NST_NODES = options.nodes
  48         else: self.NST_NODES = None
  49
  50         self.api = xmlrpclib.Server(self.NST_API_SERVER)
  51         self.auth = {}
  52         self.auth['Username'] = self.NST_USER
  53         self.auth['AuthString'] = self.NST_PASSWORD
  54         self.auth['AuthMethod'] = 'password'
  55         self.key = self.NST_KEY_PATH
  56         self.slice = self.NST_SLICE
  57         self.nodes = self.NST_NODES
  58         self.multiquery_path = self.NST_MULTIQUERY_PATH
  59         self.verbose = options.verbose
  60
  61         self.data_path = '/usr/share/planetlab/tests/nst/data/'
  62         self.plots_path = '/usr/share/planetlab/tests/nst/plots/'
  63
  64
  65 # get formatted tic string for gnuplot
  66 def getTimeTicString(t1, t2, step):
  67         first_hour = list(time.localtime(t1))
  68         if not first_hour[4] == first_hour[5] == 0:
  69                 first_hour[4] = 0
  70                 first_hour[5] = 0
  71
  72         first_hour_time = int(time.mktime(first_hour))
  73         first_hour_time += 3600
  74
  75         backsteps = (first_hour_time - t1)
  76         backsteps /= step
  77         start = first_hour_time - backsteps * step
  78
  79         tics = []
  80         thistime = start
  81         while thistime < t2:
  82                 tics.append("\"%s\" %d" % \
  83                         (time.strftime("%H:%M", time.localtime(thistime)), thistime))
  84                 thistime += step
  85
  86         ticstr = ", ".join(tics)
  87         return ticstr
  88
  89
  90 # count total number of nodes in PlanetLab, according to the api
  91 # count total number  of nodes in slice, according to the api
  92 def count_nodes_by_api(config):
  93
  94         # count all nodes
  95         all_nodes = [row['node_id'] for row in config.api.GetNodes(config.auth, {}, ['node_id', 'slice_ids'])]
  96         all_nodes_output = "%d\t%d" % (round(time.time()), len(all_nodes))
  97
  98         # count all nodes in slice
  99         if config.slice == 'root':
 100             nodes_in_slice = all_nodes
 101             nodes_in_slice_output = all_nodes_output
 102         else:
 103             slice_id =config.api.GetSlices(config.auth, {'name': config.slice}, ['slice_id'])[0]['slice_id']
 104             nodes_in_slice = [row['node_id'] for row in \
 105                               all_nodes if slice_id in row['slice_ids']]
 106             nodes_in_slice_output =  "%d\t%d" % (round(time.time()), len(nodes_in_slice))
 107
 108         # write result to datafiles
 109         all_nodes_file_name = config.data_path + os.sep + "nodes"
 110         all_nodes_file = open(all_nodes_file_name, 'w')
 111         all_nodes_file.write(all_nodes_output)
 112         all_nodes_file.close()
 113
 114         nodes_in_slice_file_name = config.data_path + os.sep + "nodes_in_slice"
 115         nodes_in_slice_file = open(nodes_in_slice_file_name, 'w')
 116         nodes_in_slice_file.write(nodes_in_slice_output)
 117         nodes_in_slice_file.close()
 118
 119         if config.verbose:
 120             print "all node: " + all_nodes_output
 121             print "nodes in slice: " + nodes_in_slice_output
 122
 123
 124 # count total number of "good" nodes, according to CoMon
 125 def count_nodes_good_by_comon(config):
 126
 127
 128         comon = urllib.urlopen("http://summer.cs.princeton.edu/status/tabulator.cgi?table=table_nodeviewshort&format=nameonly&select='resptime%20%3E%200%20&&%20((drift%20%3E%201m%20||%20(dns1udp%20%3E%2080%20&&%20dns2udp%20%3E%2080)%20||%20gbfree%20%3C%205%20||%20sshstatus%20%3E%202h)%20==%200)'")
 129         good_nodes = comon.readlines()
 130
 131         comon_output =  "%d\t%d" % (round(time.time()), len(good_nodes))
 132         nodes_good_comon_file_name = config.data_path + os.sep + "nodes_good"
 133         nodes_good_comon_file = open(nodes_good_comon_file_name, 'a')
 134         nodes_good_comon_file.write(comon_output)
 135         nodes_good_comon_file.close()
 136
 137         if config.verbose:
 138             print "comon: " + comon_output
 139
 140 # count total number of nodes reachable by ssh
 141 def count_nodes_can_ssh(config):
 142
 143         api = config.api
 144         slice = config.slice
 145         key = config.key
 146         verbose = config.verbose
 147         auth = config.auth
 148         nodes = config.nodes
 149         multiquery_path = config.multiquery_path
 150
 151         if verbose:
 152             verbose_text = ""
 153             print "Creating list of nodes to ssh to"
 154         else:
 155             verbose_text = ">/dev/null 2>&1"
 156
 157         # creaet node dict
 158         all_nodes = api.GetNodes(auth, {}, ['hostname', 'boot_state', 'last_updated'])
 159         node_dict = {}
 160         for node in all_nodes:
 161             node_dict[node['hostname']] = node
 162
 163         # create node list
 164         if nodes:
 165             nodes_file = open(nodes, 'r')
 166             nodes_filename = nodes_file.name
 167             lines = nodes_file.readlines()
 168             node_list = [node.replace('\n', '') for node in lines]
 169             nodes_file.close()
 170
 171         else:
 172             node_list = node_dict.keys()
 173             nodes_filename = tempfile.mktemp()
 174             nodes_file = open(nodes_filename, 'w')
 175             nodes_file.write_lines(node_list)
 176             nodes_file.close()
 177
 178         # creaet node dict
 179         node_dict = {}
 180         for node in all_nodes:
 181             node_dict[node['hostname']] = node
 182
 183         private_key = key.split(".pub")[0]
 184
 185         # create ssh command
 186         if verbose:
 187             print "Attemptng to ssh to nodes in " + nodes_filename
 188
 189         ssh_filename = tempfile.mktemp()
 190         ssh_file = open(ssh_filename, 'w')
 191         ssh_file.write("""
 192         export PATH=$PATH:%(multiquery_path)s
 193         export MQ_SLICE="%(slice)s"
 194         export MQ_NODES="%(nodes_filename)s"
 195
 196         eval `ssh-agent` >/dev/null 2>&1
 197         trap "kill $SSH_AGENT_PID" 0
 198         ssh-add %(private_key)s >/dev/null 2>&1
 199
 200         multiquery 'hostname' 2>/dev/null |
 201         grep "bytes" |
 202         grep -v ": 0 bytes"
 203         """ % locals())
 204         ssh_file.close()
 205         ssh_results = os.popen("python /tmp/test").readlines()
 206         ssh_result = eval(ssh_results[0].replace('\\n', ''))
 207
 208         # create a list of hostname out of results that are not empty
 209         good_nodes = []
 210         for result in ssh_result:
 211             if result.find("bytes") > -1:
 212                 result_parts = result.split(":")
 213                 hostname = result_parts[0]
 214                 good_nodes.append(hostname)
 215
 216         # count number of node we can ssh into
 217         ssh_count = len(good_nodes)
 218
 219         # determine whince nodes are dead:
 220         dead_nodes = set(node_list).difference(good_nodes)
 221
 222         # write dead nodes to file
 223         curr_time = round(time.time())
 224         dead_node_count_output = "%d\t%d" % (curr_time, len(dead_nodes))
 225         dead_nodes_file_name = config.data_path + os.sep + "dead_nodes"
 226         dead_nodes_file = open(dead_nodes_file_name, 'a')
 227         for hostname in dead_nodes:
 228             boot_state = node_dict[hostname]['boot_state']
 229             last_updated = 0
 230             if node_dict[hostname]['last_updated']:
 231                 last_updated = node_dict[hostname]['last_updated']
 232             dead_nodes_file.write("%(curr_time)d\t%(hostname)s\t%(boot_state)s\t%(last_updated)d\n" % \
 233                                   locals())
 234         dead_nodes_file.close()
 235
 236         # write good node count
 237         ssh_result_output =  "%d\t%d" % (round(time.time()), ssh_count)
 238         nodes_can_ssh_file_name = config.data_path + os.sep + "nodes_can_ssh"
 239         nodes_can_ssh_file = open(nodes_can_ssh_file_name, 'a')
 240         nodes_can_ssh_file.write(ssh_result_output)
 241         nodes_can_ssh_file.close()
 242
 243         if verbose:
 244             print "nodes that can ssh: " + ssh_result_output
 245             print "dead nodes: " + dead_node_count_output
 246
 247
 248 # remove all nodes from a slice
 249 def empty_slice(config):
 250
 251         if config.verbose:
 252             print "Removing %s from all nodes" % config.slice
 253
 254         all_nodes = [row['node_id'] for row in config.api.GetNodes(config.auth, {}, ['node_id'])]
 255         config.api.DeleteSliceFromNodes(config.auth, slice, all_nodes)
 256
 257
 258 # add slice to all nodes.
 259 # make sure users key is up to date
 260 def init_slice(config):
 261
 262     # make sure slice exists
 263     slices = config.api.GetSlices(config.auth, [config.slice], \
 264                                   ['slice_id', 'name', 'person_ids'])
 265     if not slices:
 266         raise "No such slice %s" % config.slice
 267     slice = slices[0]
 268
 269     # make sure user is in slice
 270     person = config.api.GetPersons(config.auth, config.auth['Username'], \
 271                                    ['person_id', 'email', 'slice_ids', 'key_ids'])[0]
 272     if slice['slice_id'] not in person['slice_ids']:
 273         raise "%s not in %s slice. Must be added first" % \
 274               (person['email'], slice['name'])
 275
 276     # make sure user key is up to date
 277     current_key = open(key_path, 'r').readline().strip()
 278     if len(current_key) == 0:
 279         raise "Key cannot be empty"
 280
 281     keys = config.api.GetKeys(auth, person['key_ids'])
 282     if not keys:
 283         if config.verbose:
 284             print "Adding new key " + key_path
 285         config.api.AddPersonKey(config.auth, person['person_id'], \
 286                                 {'key_type': 'ssh', 'key': current_key})
 287
 288     elif not filter(lambda k: k['key'] == current_key, keys):
 289         if config.verbose:
 290             print "%s was modified or is new. Updating PLC"
 291         old_key = keys[0]
 292         config.api.UpdateKey(config.auth, old_key['key_id'], \
 293                              {'key': current_key})
 294
 295     # add slice to all nodes
 296     if config.verbose:
 297         print "Adding %s to all nodes" + slice
 298     all_nodes = [row['node_id'] for row in \
 299                  config.api.GetNodes(config.auth, {}, ['node_id'])]
 300     config.api.AddSliceToNodes(config.auth, config.slice, all_nodes)
 301
 302
 303 # create the fill/empty plot
 304 def plot_fill_empty():
 305         #ticstep = 3600 # 1 hour
 306         #plotlength = 36000 # 10 hours
 307         ticstep = 1800
 308         plotlength = 10800
 309
 310         all_nodes_file_name = config.data_path + os.sep + "nodes"
 311         nodes_in_slice_file_name = config.data_path + os.sep + "nodes_in_slice"
 312         nodes_can_ssh_file_name = config.data_path + os.sep + "nodes_can_ssh"
 313         nodes_good_comon_file_name = config.data_path + os.sep + "nodes_good"
 314
 315         tmpfilename = tempfile.mktemp()
 316         tmpfile = open(tmpfilename, 'w')
 317
 318         starttime = -1
 319         stoptime = -1
 320         for datafilename in [all_nodes_file_name,
 321                              nodes_in_slice_file_name, \
 322                              nodes_can_ssh_file_name, \
 323                              nodes_good_comon_file_name]:
 324                 datafile = open(datafilename, 'r')
 325                 line1 = datafile.readline()
 326                 datafile.seek(-32,2)
 327                 line2 = datafile.readlines().pop()
 328                 thisstarttime = int(line1.split("\t")[0])
 329                 if starttime == -1 or thisstarttime < starttime:
 330                         starttime = thisstarttime
 331                 thisstoptime = int(line2.split("\t")[0])
 332                 if stoptime == -1 or thisstoptime > stoptime:
 333                         stoptime = thisstoptime
 334
 335         stopx = stoptime
 336         startx = max(starttime, stopx - plotlength)
 337         starttime = startx
 338
 339         tics = getTimeTicString(starttime, stoptime, ticstep)
 340
 341         startdate = time.strftime("%b %m, %Y - %H:%M", time.localtime(startx))
 342         stopdate = time.strftime("%H:%M", time.localtime(stopx))
 343
 344         tmpfile.write("""
 345         set term png
 346         set output "%(plots_path)s/fill_empty.png"
 347
 348         set title "Number of Nodes / Time - %(startdate)s to %(stopdate)s"
 349         set xlabel "Time"
 350         set ylabel "Number of Nodes"
 351
 352         set xtics (%(tics)s)
 353         set xrange[%(startx)d:%(stopx)d]
 354         set yrange[0:950]
 355
 356         plot "%(all_nodes_file_name)s" u 1:2 w lines title "Total Nodes", \
 357                 "%(nodes_in_slice_file_name)s" u 1:2 w lines title "Nodes in Slice", \
 358                 "%(nodes_good_comon_file_name)s" u 1:2 w lines title \
 359                         "Healthy Nodes (according to CoMon)", \
 360                 "%(nodes_can_ssh_file_name)s" u 1:2 w lines title "Nodes Reachable by SSH"
 361
 362         """ % locals())
 363
 364         tmpfile.close()
 365
 366         os.system("%s %s" % (gnuplot_path, tmpfilename))
 367
 368         if os.path.exists(tmpfilename):
 369                 os.unlink(tmpfilename)
 370
 371
 372
 373 config = Config(options)
 374
 375
 376 if config.slice == 'root':
 377
 378     if config.verbose:
 379         print "Logging in as root"
 380 else:
 381     # set up slice and add it to nodes
 382     init_slice(config)
 383
 384     # wait 15 mins for nodes to get the data
 385     sleep(900)
 386
 387 # gather data
 388 count_nodes_can_ssh(config)
 389 count_nodes_by_api(config)
 390 count_nodes_good_by_comon(config)
 391
 392 # update plots
 393 plot_fill_empty()
 394 #os.system("cp plots/*.png ~/public_html/planetlab/tests")
 395
 396 # clean up
 397 empty_slice(config)
 398