7 from datetime import datetime,timedelta
12 from pcucontrol import reboot
14 from pcucontrol.util import command
15 from monitor import config
17 from monitor.database.info.model import *
19 from monitor.sources import comon
20 from monitor.wrapper import plc, plccache
23 from monitor.common import nmap_port_status
25 COMON_COTOPURL= "http://summer.cs.princeton.edu/status/tabulator.cgi?" + \
26 "table=table_nodeview&" + \
27 "dumpcols='name,resptime,sshstatus,uptime,lastcotop,cpuspeed,memsize,disksize'&" + \
30 api = plc.getAuthAPI()
31 plc_lock = threading.Lock()
40 #print "GetPCU from PLC %s" % pcuname
41 l_pcu = plc.GetPCUs({'pcu_id' : pcuname})
47 #print "GetPCU from file %s" % pcuname
48 l_pcus = plccache.l_pcus
50 if i['pcu_id'] == pcuname:
59 def get_nodes(node_ids):
63 l_node = plc.getNodes(node_ids, ['hostname', 'last_contact', 'node_id', 'ports'])
66 plc_nodes = plccache.l_plcnodes
68 if n['node_id'] in node_ids:
80 def get_plc_pcu_values(pcuname):
82 Try to contact PLC to get the PCU info.
83 If that fails, try a backup copy from the last run.
84 If that fails, return None
88 l_pcu = get_pcu(pcuname)
91 site_id = l_pcu['site_id']
92 node_ids = l_pcu['node_ids']
93 l_node = get_nodes(node_ids)
95 if l_node is not None:
97 values[node['hostname']] = node['ports'][0]
99 values['nodenames'] = [node['hostname'] for node in l_node]
101 # NOTE: this is for a dry run later. It doesn't matter which node.
102 values['node_id'] = l_node[0]['node_id']
110 class ScanInterface(object):
113 primarykey = 'hostname'
115 def __init__(self, round=1):
119 def __getattr__(self, name):
120 if 'collect' in name or 'record' in name:
121 method = getattr(self, name, None)
123 raise Exception("No such method %s" % name)
126 raise Exception("No such method %s" % name)
128 def collect(self, nodename, data):
131 def record(self, request, (nodename, values) ):
138 fbnodesync = self.syncclass.findby_or_create(
139 #if_new_set={'round' : self.round},
140 **{ self.primarykey : nodename})
141 # NOTE: This code will either add a new record for the new self.round,
142 # OR it will find the previous value, and update it with new information.
143 # The data that is 'lost' is not that important, b/c older
144 # history still exists.
145 fbrec = self.recordclass.findby_or_create(
146 **{ self.primarykey:nodename})
148 fbrec.set( **values )
152 fbnodesync.round = self.round
155 print "%d %s %s" % (self.count, nodename, values)
160 print traceback.print_exc()
163 class ScanNodeInternal(ScanInterface):
164 recordclass = FindbadNodeRecord
165 #syncclass = FindbadNodeRecordSync
167 primarykey = 'hostname'
169 def collectNMAP(self, nodename, cohash):
170 #### RUN NMAP ###############################
173 print "nmap -oG - -P0 -p22,80,806 %s | grep Host:" % nodename
174 (oval,eval) = nmap.run_noexcept("nmap -oG - -P0 -p22,80,806 %s | grep Host:" % nodename)
175 # NOTE: an empty / error value for oval, will still work.
176 (values['port_status'], continue_probe) = nmap_port_status(oval)
178 values['date_checked'] = datetime.now()
180 return (nodename, values)
182 def collectInternal(self, nodename, cohash):
183 ### RUN PING ######################
185 (oval,errval) = ping.run_noexcept("ping -c 1 -q %s | grep rtt" % nodename)
192 values['ping_status'] = False
194 values['ping_status'] = True
197 for port in [22, 806]:
198 ssh = command.SSH('root', nodename, port)
200 (oval, errval) = ssh.run_noexcept2(""" <<\EOF
202 echo ' "kernel_version":"'`uname -a`'",'
203 echo ' "bmlog":"'`ls /tmp/bm.log`'",'
204 echo ' "bootcd_version":"'`cat /mnt/cdrom/bootme/ID`'",'
205 echo ' "nm_status":"'`ps ax | grep nm.py | grep -v grep`'",'
206 echo ' "fs_status":"'`touch /var/log/monitor 2>&1`'",'
207 echo ' "dns_status":"'`host boot.planet-lab.org 2>&1`'",'
208 echo ' "princeton_comon_dir":"'`ls -d /vservers/princeton_comon`'",'
210 ID=`grep princeton_comon /etc/passwd | awk -F : '{if ( $3 > 500 ) { print $3}}'`
211 echo ' "princeton_comon_running":"'`ls -d /proc/virtual/$ID`'",'
212 echo ' "princeton_comon_procs":"'`vps ax | grep $ID | grep -v grep | wc -l`'",'
216 values['ssh_error'] = errval
218 #print "OVAL: %s" % oval
219 values.update(eval(oval))
220 values['ssh_portused'] = port
223 values.update({'kernel_version': "", 'bmlog' : "", 'bootcd_version' : '',
227 'princeton_comon_dir' : "",
228 'princeton_comon_running' : "",
229 'princeton_comon_procs' : "", 'ssh_portused' : None})
231 print traceback.print_exc()
234 ### RUN SSH ######################
235 b_getbootcd_id = True
237 oval = values['kernel_version']
238 if "2.6.17" in oval or "2.6.2" in oval:
239 values['ssh_status'] = True
240 values['observed_category'] = 'PROD'
241 if "bm.log" in values['bmlog']:
242 values['observed_status'] = 'DEBUG'
244 values['observed_status'] = 'BOOT'
245 elif "2.6.12" in oval or "2.6.10" in oval:
246 values['ssh_status'] = True
247 values['observed_category'] = 'OLDPROD'
248 if "bm.log" in values['bmlog']:
249 values['observed_status'] = 'DEBUG'
251 values['observed_status'] = 'BOOT'
253 # NOTE: on 2.6.8 kernels, with 4.2 bootstrapfs, the chroot
254 # command fails. I have no idea why.
255 elif "2.4" in oval or "2.6.8" in oval:
256 b_getbootcd_id = False
257 values['ssh_status'] = True
258 values['observed_category'] = 'OLDBOOTCD'
259 values['observed_status'] = 'DEBUG'
261 values['ssh_status'] = True
262 values['observed_category'] = 'UNKNOWN'
263 if "bm.log" in values['bmlog']:
264 values['observed_status'] = 'DEBUG'
266 values['observed_status'] = 'BOOT'
269 b_getbootcd_id = False
270 values['ssh_status'] = False
271 values['observed_category'] = 'ERROR'
272 values['observed_status'] = 'DOWN'
274 values['ssh_error'] = val
275 values['kernel_version'] = ""
278 # try to get BootCD for all nodes that are not 2.4 nor inaccessible
279 oval = values['bootcd_version']
281 values['bootcd_version'] = oval
282 if "v2" in oval and \
283 ( nodename is not "planetlab1.cs.unc.edu" and \
284 nodename is not "planetlab2.cs.unc.edu" ):
285 values['observed_category'] = 'OLDBOOTCD'
287 values['bootcd_version'] = ""
289 values['bootcd_version'] = ""
291 oval = values['nm_status']
293 values['nm_status'] = "Y"
295 values['nm_status'] = "N"
297 continue_slice_check = True
298 oval = values['princeton_comon_dir']
299 if "princeton_comon_dir" in oval:
300 values['princeton_comon_dir'] = True
302 values['princeton_comon_dir'] = False
303 continue_slice_check = False
305 if continue_slice_check:
306 oval = values['princeton_comon_running']
307 if len(oval) > len('/proc/virtual/'):
308 values['princeton_comon_running'] = True
310 values['princeton_comon_running'] = False
311 continue_slice_check = False
313 values['princeton_comon_running'] = False
315 if continue_slice_check:
316 oval = values['princeton_comon_procs']
317 values['princeton_comon_procs'] = int(oval)
319 values['princeton_comon_procs'] = None
322 if nodename in cohash:
323 values['comon_stats'] = cohash[nodename]
325 values['comon_stats'] = {'resptime': '-1',
332 # include output value
333 ### GET PLC NODE ######################
337 d_node = plc.getNodes({'hostname': nodename}, ['pcu_ids', 'site_id',
338 'date_created', 'last_updated',
339 'last_contact', 'boot_state', 'nodegroup_ids'])[0]
341 traceback.print_exc()
343 values['plc_node_stats'] = d_node
345 ##### NMAP ###################
346 (n, v) = self.collectNMAP(nodename, None)
349 ### GET PLC PCU ######################
353 pcu = d_node['pcu_ids']
357 site_id = d_node['site_id']
359 values['plc_pcuid'] = d_pcu
361 ### GET PLC SITE ######################
364 values['loginbase'] = ""
366 d_site = plc.getSites({'site_id': site_id},
367 ['max_slices', 'slice_ids', 'node_ids', 'login_base'])[0]
368 values['loginbase'] = d_site['login_base']
370 traceback.print_exc()
373 values['plc_site_stats'] = d_site
374 values['date_checked'] = datetime.now()
376 print traceback.print_exc()
378 return (nodename, values)
380 def internalprobe(hostname):
381 #fbsync = FindbadNodeRecordSync.findby_or_create(hostname="global",
382 # if_new_set={'round' : 1})
383 scannode = ScanNodeInternal() # fbsync.round)
385 (nodename, values) = scannode.collectInternal(hostname, {})
386 scannode.record(None, (nodename, values))
390 print traceback.print_exc()
393 def externalprobe(hostname):
394 #fbsync = FindbadNodeRecordSync.findby_or_create(hostname="global",
395 # if_new_set={'round' : 1})
396 scannode = ScanNodeInternal() # fbsync.round)
398 (nodename, values) = scannode.collectNMAP(hostname, {})
399 scannode.record(None, (nodename, values))
403 print traceback.print_exc()
406 class ScanPCU(ScanInterface):
407 recordclass = FindbadPCURecord
409 primarykey = 'plc_pcuid'
411 def collectInternal(self, pcuname, cohash):
413 continue_probe = True
415 values = {'reboot_trial_status' : 'novalue'}
416 ### GET PCU ######################
420 v = get_plc_pcu_values(pcuname)
421 if v['hostname'] is not None: v['hostname'] = v['hostname'].strip()
422 if v['ip'] is not None: v['ip'] = v['ip'].strip()
425 values['plc_pcu_stats'] = v
427 continue_probe = False
430 traceback.print_exc()
431 continue_probe = False
433 if b_except or not continue_probe: return (None, None, None)
435 #### RUN NMAP ###############################
438 print "nmap -oG - -P0 -p22,23,80,443,5869,9100,16992 %s | grep Host:" % reboot.pcu_name(values['plc_pcu_stats'])
439 (oval,eval) = nmap.run_noexcept("nmap -oG - -P0 -p22,23,80,443,5869,9100,16992 %s | grep Host:" % reboot.pcu_name(values['plc_pcu_stats']))
440 # NOTE: an empty / error value for oval, will still work.
441 (values['port_status'], continue_probe) = nmap_port_status(oval)
443 values['port_status'] = None
445 #### COMPLETE ENTRY #######################
447 values['entry_complete'] = []
448 #if values['protocol'] is None or values['protocol'] is "":
449 # values['entry_complete'] += ["protocol"]
450 if values['plc_pcu_stats']['model'] is None or values['plc_pcu_stats']['model'] is "":
451 values['entry_complete'] += ["model"]
452 # Cannot continue due to this condition
453 continue_probe = False
455 if values['plc_pcu_stats']['password'] is None or values['plc_pcu_stats']['password'] is "":
456 values['entry_complete'] += ["password"]
457 # Cannot continue due to this condition
458 continue_probe = False
460 if len(values['entry_complete']) > 0:
461 continue_probe = False
463 if values['plc_pcu_stats']['hostname'] is None or values['plc_pcu_stats']['hostname'] is "":
464 values['entry_complete'] += ["hostname"]
465 if values['plc_pcu_stats']['ip'] is None or values['plc_pcu_stats']['ip'] is "":
466 values['entry_complete'] += ["ip"]
468 # If there are no nodes associated with this PCU, then we cannot continue.
469 if len(values['plc_pcu_stats']['node_ids']) == 0:
470 continue_probe = False
471 values['entry_complete'] += ['nodeids']
474 #### DNS and IP MATCH #######################
475 if values['plc_pcu_stats']['hostname'] is not None and values['plc_pcu_stats']['hostname'] is not "" and \
476 values['plc_pcu_stats']['ip'] is not None and values['plc_pcu_stats']['ip'] is not "":
478 ipaddr = socket.gethostbyname(values['plc_pcu_stats']['hostname'])
479 if ipaddr == values['plc_pcu_stats']['ip']:
480 values['dns_status'] = "DNS-OK"
482 values['dns_status'] = "DNS-MISMATCH"
483 continue_probe = False
485 except Exception, err:
486 values['dns_status'] = "DNS-NOENTRY"
487 values['plc_pcu_stats']['hostname'] = values['plc_pcu_stats']['ip']
489 if values['plc_pcu_stats']['ip'] is not None and values['plc_pcu_stats']['ip'] is not "":
490 values['dns_status'] = "NOHOSTNAME"
491 values['plc_pcu_stats']['hostname'] = values['plc_pcu_stats']['ip']
493 values['dns_status'] = "NO-DNS-OR-IP"
494 values['plc_pcu_stats']['hostname'] = "No_entry_in_DB"
495 continue_probe = False
498 ###### DRY RUN ############################
499 if continue_probe and 'node_ids' in values['plc_pcu_stats'] and \
500 len(values['plc_pcu_stats']['node_ids']) > 0:
501 rb_ret = reboot.reboot_test_new(values['plc_pcu_stats']['nodenames'][0],
504 rb_ret = "Not_Run" # No nodes to test"
506 values['reboot_trial_status'] = rb_ret
509 print "____________________________________"
512 print "____________________________________"
513 errors['traceback'] = traceback.format_exc()
514 print errors['traceback']
515 values['reboot_trial_status'] = str(errors['traceback'])
518 values['entry_complete']=" ".join(values['entry_complete'])
520 values['date_checked'] = datetime.now()
521 return (pcuname, values)