libplab.py.in 72.9 KB
Newer Older
1
# -*- python -*-
Kirk Webb's avatar
Kirk Webb committed
2
3
#
# EMULAB-COPYRIGHT
4
# Copyright (c) 2000-2004, 2006, 2007 University of Utah and the Flux Group.
Kirk Webb's avatar
Kirk Webb committed
5
6
7
# All rights reserved.
#

8
"""
Kirk Webb's avatar
Kirk Webb committed
9
10
11
12
13
14
15
16
17
Library for interfacing with Plab.  This abstracts out the concepts of
Plab central, slices, and nodes.  All data (except static things like
certificates) is kept in the Emulab DB.  Unlike the regular dslice
svm, this one supports dynamically changing which nodes are in a
slice.

This requires an already obtained dslice certficate and key.  By
default it expects to find these in the @prefix@/etc/plab/
subdirectory.
18
19
20
"""

import sys
21
22
23
sys.path.append("@prefix@/lib")

import os, time
24
import string
Kirk Webb's avatar
   
Kirk Webb committed
25
import traceback
Kirk Webb's avatar
   
Kirk Webb committed
26
import signal
Kirk Webb's avatar
   
Kirk Webb committed
27
import socket
Kirk Webb's avatar
Kirk Webb committed
28
29
30
import httplib
import xml.parsers.expat
import re
Kirk Webb's avatar
   
Kirk Webb committed
31
import calendar
Kirk Webb's avatar
   
Kirk Webb committed
32
import shlex
Kirk Webb's avatar
   
Kirk Webb committed
33

Kirk Webb's avatar
   
Kirk Webb committed
34
from popen2 import Popen4
Kirk Webb's avatar
   
Kirk Webb committed
35
from warnings import warn
36

37
38
39
40
41
42
43
44
45
46
47
#
# Testbed and DB access libs
#
from libtestbed import *
from libdb import *

#
# Plab modules to import
#
from mod_PLC import mod_PLC
from mod_dslice import mod_dslice
Kirk Webb's avatar
   
Kirk Webb committed
48
from mod_PLCNM import mod_PLCNM
49
from mod_PLC4 import mod_PLC4
50
51

agents = {'PLC'    : mod_PLC,
Kirk Webb's avatar
   
Kirk Webb committed
52
          'dslice' : mod_dslice,
53
54
          'PLCNM'  : mod_PLCNM,
          'PLC4'   : mod_PLC4}
55

Kirk Webb's avatar
   
Kirk Webb committed
56
57
58
59
60
61
62
#
# Initialize the AOP stuff
#
from aspects import wrap_around
from timer_advisories import initTimeAdvice, timeAdvice
initTimeAdvice("plabtiming")

Kirk Webb's avatar
Kirk Webb committed
63
64
65
66
67
68
#
# output control vars
#
verbose = 0
debug = 0

69
70
71
#
# Constants
#
72
DEF_AGENT = "PLC4";
73
74
75
76
#
# Add a compatibility agent --- libplab will make all PLC calls on this agent
# in addition to the default agent.
#
77
COMPAT_AGENT = "PLCNM";
78
# if this is set, do all PLC calls on COMPAT_AGENT as well as DEF_AGENT
79
compat_mode = 0
80

81
RENEW_TIME = 2*24*60*60  # Renew two days before lease expires
Kirk Webb's avatar
   
Kirk Webb committed
82
83
84

RENEW_TIMEOUT = 1*60     # give the node manager a minute to respond to renew
FREE_TIMEOUT  = 1*60     # give the node manager a minute to respond to free
85
NODEPROBEINT  = 30
86

Kirk Webb's avatar
   
Kirk Webb committed
87
88
USERNODE = "@USERNODE@"
TBOPS = "@TBOPSEMAIL_NOSLASH@"
89
MAILTAG = "@THISHOMEBASE@"
Kirk Webb's avatar
   
Kirk Webb committed
90
SLICE_ALIAS_DIR = "/etc/mail/plab-slice-addrs"
91

92
RESERVED_PID = "emulab-ops"
93
RESERVED_EID = "hwdown"       # start life in hwdown
94
95
MONITOR_PID  = "emulab-ops"
MONITOR_EID  = "plab-monitor"
96

Kirk Webb's avatar
   
Kirk Webb committed
97
98
MAGIC_INET2_GATEWAYS = ("205.124.237.10",  "205.124.244.18",
                        "205.124.244.178", )
Kirk Webb's avatar
   
Kirk Webb committed
99
100
101
MAGIC_INET_GATEWAYS =  ("205.124.244.150", "205.124.239.185",
                        "205.124.244.154", "205.124.244.138",
                        "205.124.244.130", )
102
LOCAL_PLAB_DOMAIN = ".flux.utah.edu"
103
LOCAL_PLAB_LINKTYPE = "pcplabinet2"
Kirk Webb's avatar
   
Kirk Webb committed
104
105
106

# allowed nil/unknown values (sentinels).
ATTR_NIL_VALUES = ('None',)
107

108
109
110
111
# 'critical' node identifiers - those that are actually used to uniquely
# identify a planetlab node
ATTR_CRIT_KEYS = ('HNAME', 'IP', 'PLABID', 'MAC',)

112
113
114
115
# The amount by which latitude and longitude are allowed to differ before we
# classify them ask changed
LATLONG_DELTA = 0.001

Kirk Webb's avatar
   
Kirk Webb committed
116
PLABNODE = "@prefix@/sbin/plabnode"
117
SSH = "@prefix@/bin/sshtb"
118
NAMED_SETUP = "@prefix@/sbin/named_setup"
Kirk Webb's avatar
   
Kirk Webb committed
119
PELAB_PUSH  = "@prefix@/sbin/pelab_opspush"
120

Kirk Webb's avatar
Kirk Webb committed
121
122
123
124
ROOTBALL_URL = "http://localhost:1492/" # ensure this ends in a slash

DEF_PLAB_URL = "www.planet-lab.org"
DEF_SITE_XML = "/xml/sites.xml"
Kirk Webb's avatar
   
Kirk Webb committed
125
126
IGNORED_NODES_FILE = "@prefix@/etc/plab/IGNOREDNODES"
ALLOWED_NODES_FILE = "@prefix@/etc/plab/ALLOWEDNODES"
127

128
DEF_ROOTBALL_NAME = "@PLAB_ROOTBALL@"
129
SLICEPREFIX = "@PLAB_SLICEPREFIX@"
Kirk Webb's avatar
Kirk Webb committed
130
131
132
133
NODEPREFIX  = "plab"

BADSITECHARS = re.compile(r"\W+")
PLABBASEPRIO = 20000
Kirk Webb's avatar
   
Kirk Webb committed
134
PLAB_SVC_SLICENAME = "utah_svc_slice"
Kirk Webb's avatar
   
Kirk Webb committed
135
136
137
138
PLAB_SVC_SLICEDESC = "Emulab management service slice. Performs periodic " \
                     "checkins with Emulab central, and routes events for " \
                     "other Emulab slices. Slivers in this slice should " \
                     "only interact with other PlanetLab machines, and Emulab."
Kirk Webb's avatar
   
Kirk Webb committed
139
140
PLABMON_PID = "emulab-ops"
PLABMON_EID = "plab-monitor"
Kirk Webb's avatar
   
Kirk Webb committed
141
DEF_SLICE_DESC = "Slice created by Emulab"
Kirk Webb's avatar
   
Kirk Webb committed
142

Kirk Webb's avatar
   
Kirk Webb committed
143
144
PLABEXPIREWARN = 1*WEEK        # one week advance warning for slice expiration.
NODEEXPIREWARN = 2*WEEK+2*DAY  # about two weeks advance warning for slivers.
145

146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
#
# This is a trigger table for dist'ing out multiple rootballs depending on
# which NM is running on the node.  Since we don't know which version is
# running on a node until we call into (DEF|COMPAT)_AGENT and actually connect
# to the node, it's based off classnames.
#
# The table specifies a prefix to the default rootball name; this rootball
# should be placed in the normal location.
#
# If the plcagent class and nodeagent class are both not in the trigger table,
# we simply push the default rootball.
#
# Note that we could store version info in the database, but that doesn't
# really have long-term value.  Plus, we have to be checking sites.xml all the
# time to catch version changes during rollout.  This way, plabmonitord and
# the web interface know exactly what is the version whenever a node is
# contacted for setup.
#
# For now, only NM4agent has a custom tarball, since v4 nodes are in the
# distinct minority right now.
#
from mod_PLC4 import NM4agent

rootball_triggers = { mod_PLC4 : { NM4agent : 'NM4-' } }

171
172
173
174
175
#
# var to track failed renewals
#
failedrenew = []

176
177
178
179
180
#
# Disable line buffering
#
sys.stdout = os.fdopen(sys.stdout.fileno(), sys.stdout.mode, 0)

Kirk Webb's avatar
   
Kirk Webb committed
181
182
183
184
185
#
# Ensure SIGPIPE doesn't bite us:
#
signal.signal(signal.SIGPIPE, signal.SIG_IGN)

186

187
188
189
#
# Plab abstraction
#
Kirk Webb's avatar
Kirk Webb committed
190

Kirk Webb's avatar
   
Kirk Webb committed
191
192
193
194
195
196
197
198
199
200
#
# Multiple attribute change exception
#
class MultiChangeError(Exception):
    def __init__(self, nodeid, chattrs={}):
        self.nodeid = nodeid
        self.chattrs = chattrs
        pass
    pass

Kirk Webb's avatar
   
Kirk Webb committed
201

Kirk Webb's avatar
Kirk Webb committed
202
203
204
#
# Class responsible for parsing planetlab sites file
#
Kirk Webb's avatar
Kirk Webb committed
205
206
207
208
209
210
211
212
class siteParser:

    def __init__(self):
        self.parser = xml.parsers.expat.ParserCreate()
        self.parser.StartElementHandler = self.__site_start_elt
        self.parser.EndElementHandler = self.__site_end_elt
        self.__hosts = []
        self.__sitename = ""
213
214
        self.__latitude = 0
        self.__longitude = 0
Kirk Webb's avatar
Kirk Webb committed
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
        
    def getPlabNodeInfo(self):
                
        conn = httplib.HTTPSConnection(DEF_PLAB_URL)
        conn.request("GET", DEF_SITE_XML)
        res = conn.getresponse()
        if res.status != 200:
            raise RuntimeError, "HTTP Error getting site list:\n" \
                  "Code: %d Reason: %s" % \
                  (res.status, res.reason)
        try:
            self.parser.ParseFile(res)
            pass
        except xml.parsers.expat.ExpatError, e:
            print "Error parsing XML file, lineno: %d, offset: %d:\n%s" % \
                  (e.lineno, e.offset, xml.parsers.expat.ErrorString(e.code))
            raise

        return self.__hosts

    def __site_start_elt(self, name, attrs):
        
        if name == "PLANETLAB_SITES":
            pass
        
        elif name == "SITE":
            self.__sitename = attrs['SHORT_SITE_NAME']
242
243
244
            if attrs.has_key('LATITUDE'):
                self.__latitude = attrs['LATITUDE']
            else:
Kirk Webb's avatar
   
Kirk Webb committed
245
                self.__latitude = 0
246
247
248
249
            if attrs.has_key('LONGITUDE'):
                self.__longitude = attrs['LONGITUDE']
            else:
                self.__longitude = 0
Kirk Webb's avatar
Kirk Webb committed
250
251
252
            pass
        
        elif name == "HOST":
Kirk Webb's avatar
   
Kirk Webb committed
253
254
255
            if not attrs.has_key('MAC'):
                attrs['MAC'] = "None"
                pass
Kirk Webb's avatar
   
Kirk Webb committed
256
257
258
            if not attrs.has_key('BWLIMIT'):
                attrs['BWLIMIT'] = "-1"
                pass
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
            if not attrs.has_key('IP'):
                print "node %s did not have IP!" % attrs['NAME']
                pass
            else:
                self.__hosts.append({
                    'HNAME'     : attrs['NAME'],
                    'IP'        : attrs['IP'],
                    'PLABID'    : attrs['NODE_ID'],
                    'MAC'       : attrs['MAC'],
                    'BWLIMIT'   : attrs['BWLIMIT'],
                    'SITE'      : self.__sitename,
                    'LATITUDE'  : self.__latitude,
                    'LONGITUDE' : self.__longitude,
                    'STATUS'    : attrs['STATUS']
                    })
                pass
Kirk Webb's avatar
Kirk Webb committed
275
276
277
278
279
280
281
282
283
284
285
286
            pass
        
        else:
            print "Unknown element in site file: %s: %s" % (name, attrs)
            pass
        
        return

    def __site_end_elt(self, name):
        
        if name == "SITE":
            self.__sitename = "Unknown"
287
288
            self.__latitude = 0
            self.__longitude = 0
Kirk Webb's avatar
Kirk Webb committed
289
290
291
            pass
        return

Kirk Webb's avatar
   
Kirk Webb committed
292

293
class Plab:
Kirk Webb's avatar
Kirk Webb committed
294
    def __init__(self, agent=None):
295
296
297
        if not agent:
            self.agent = agents[DEF_AGENT]()
            pass
298
299
        if compat_mode:
            self.compat_agent = agents[COMPAT_AGENT]()
Kirk Webb's avatar
Kirk Webb committed
300
        if debug:
301
            print "Using module: %s" % self.agent.modname
302
303
            if compat_mode:
                print "COMPAT: Using module: %s" % self.compat_agent.modname
304
305
            pass
        pass
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322

    def createSlice(self, pid, eid):
        """
        Slice factory function
        """
        slice = Slice(self, pid, eid)
        slice._create()
        return slice

    def loadSlice(self, pid, eid):
        """
        Slice factory function
        """
        slice = Slice(self, pid, eid)
        slice._load()
        return slice

Kirk Webb's avatar
Kirk Webb committed
323
    def updateNodeEntries(self, ignorenew = False):
324
        """
Kirk Webb's avatar
Kirk Webb committed
325
        Finds out which Plab nodes are available, and
326
327
328
329
330
        update the DB accordingly.  If ignorenew is True, this will only
        make sure that the data in the DB is correct, and not complete.
        If ignorenew is False (the default), this will do a complete
        update of the DB.  However, this can take some time, as
        information about new nodes (such as link type) must be
Kirk Webb's avatar
Kirk Webb committed
331
        discovered.
332
333
334
335
336
337

        Note that this seemingly innocent funciton actually does a lot of
        magic.  This is the main/only way that Plab nodes get into the
        nodes DB, and this list is updated dynamically.  It also gathers
        static data about new nodes.
        """
Kirk Webb's avatar
Kirk Webb committed
338
339
340
        
        print "Getting available Plab nodes ..."

341
        avail = []
342
        try:
Kirk Webb's avatar
Kirk Webb committed
343
344
345
            parser = siteParser()
            avail = parser.getPlabNodeInfo()
            pass
346
        # XXX: rewrite to use more elegant exception info gathering.
347
348
        except:
            extype, exval, extrace = sys.exc_info()
349
            print "Error talking to agent: %s: %s" % (extype, exval)
Kirk Webb's avatar
Kirk Webb committed
350
            if debug:
351
352
353
354
                #print extrace
                traceback.print_exc()
                pass
            
355
356
            print "Going back to sleep until next scheduled poll"
            return
Kirk Webb's avatar
   
Kirk Webb committed
357

Kirk Webb's avatar
Kirk Webb committed
358
        if debug:
359
360
            print "Got advertisement list:"
            print avail
Kirk Webb's avatar
Kirk Webb committed
361
            pass
Kirk Webb's avatar
   
Kirk Webb committed
362

Kirk Webb's avatar
   
Kirk Webb committed
363
364
365
        ignored_nodes = self.__readNodeFile(IGNORED_NODES_FILE)
        allowed_nodes = self.__readNodeFile(ALLOWED_NODES_FILE)

Kirk Webb's avatar
   
Kirk Webb committed
366
367
368
369
370
371
        # Enforce node limitations, if any.
        # XXX: This is ugly - maybe move to a separate function
        #      that takes a list of filter functions.  I know!!
        #      Create a generator out of a set of filter functions
        #      and the initial node list! :-)  Python geek points to me if
        #      I ever get around to it...  KRW
Kirk Webb's avatar
   
Kirk Webb committed
372
        if len(allowed_nodes) or len(ignored_nodes):
Kirk Webb's avatar
   
Kirk Webb committed
373
            allowed = []
Kirk Webb's avatar
Kirk Webb committed
374
            for nodeent in avail:
Kirk Webb's avatar
   
Kirk Webb committed
375
                if nodeent['PLABID'] in ignored_nodes:
Kirk Webb's avatar
   
Kirk Webb committed
376
                    continue
Kirk Webb's avatar
   
Kirk Webb committed
377
378
                elif len(allowed_nodes):
                    if nodeent['IP'] in allowed_nodes:
Kirk Webb's avatar
   
Kirk Webb committed
379
380
381
382
383
                        allowed.append(nodeent)
                        pass
                    pass
                else:
                    allowed.append(nodeent)
Kirk Webb's avatar
Kirk Webb committed
384
385
                    pass
                pass
Kirk Webb's avatar
   
Kirk Webb committed
386
387
388
389
            if verbose:
                print "Advertisements in allowed nodes list:\n%s" % allowed
                pass
            avail = allowed
Kirk Webb's avatar
Kirk Webb committed
390
            pass
391

Kirk Webb's avatar
   
Kirk Webb committed
392
393
394
395
396
397
398
399
400
401
402
        # Check for duplicate node attributes (sanity check)
        availdups = self.__findDuplicateAttrs(avail)
        if len(availdups):
            SENDMAIL(TBOPS, "Duplicates in plab advertised node list",
                     "Duplicate attributes:\n"
                     "%s\n\n"
                     "Let plab support know!" % availdups,
                     TBOPS)
            raise RuntimeError, \
                  "Duplicate attributes in plab node listing:\n%s" % availdups

403
        # Get node info we already have.
404
        known = self.__getKnownPnodes()
Kirk Webb's avatar
Kirk Webb committed
405
        if debug:
406
407
            print "Got known pnodes:"
            print known
Kirk Webb's avatar
Kirk Webb committed
408
            pass
409

Kirk Webb's avatar
Kirk Webb committed
410
        # Create list of nodes to add or update
Kirk Webb's avatar
   
Kirk Webb committed
411
412
        toadd    = []  # List of node entries to add to DB
        toupdate = []  # List of node entries to update in the DB
Kirk Webb's avatar
Kirk Webb committed
413
        for nodeent in avail:
Kirk Webb's avatar
Kirk Webb committed
414
415
            # Replace sequences of bad chars in the site entity with
            # a single "-".
Kirk Webb's avatar
Kirk Webb committed
416
            nodeent['SITE'] = BADSITECHARS.sub("-", nodeent['SITE'])
Kirk Webb's avatar
   
Kirk Webb committed
417
418
419
            # Determine if we already know about this node.
            matchres = self.__matchPlabNode(nodeent, known)
            if not matchres:
Kirk Webb's avatar
   
Kirk Webb committed
420
                toadd.append(nodeent)
Kirk Webb's avatar
Kirk Webb committed
421
                pass
Kirk Webb's avatar
   
Kirk Webb committed
422
423
            elif len(matchres[1]):
                toupdate.append((nodeent,matchres))
Kirk Webb's avatar
Kirk Webb committed
424
                pass
Kirk Webb's avatar
   
Kirk Webb committed
425
            pass
Kirk Webb's avatar
Kirk Webb committed
426

Kirk Webb's avatar
   
Kirk Webb committed
427
428
        # Process the list of nodes to add
        addstr = ""
429
        if len(toadd):
Kirk Webb's avatar
Kirk Webb committed
430
            # Are we ignoring new entries?
431
            if ignorenew:
Kirk Webb's avatar
Kirk Webb committed
432
                if verbose:
433
                    print "%d new Plab nodes, but ignored for now" % len(toadd)
Kirk Webb's avatar
Kirk Webb committed
434
435
                    pass
                pass
Kirk Webb's avatar
Kirk Webb committed
436
            # If not ignoring, do the addition/update.
437
            else:
Kirk Webb's avatar
   
Kirk Webb committed
438
439
                print "There are %d new Plab nodes." % len(toadd)
                for nodeent in toadd:
Kirk Webb's avatar
Kirk Webb committed
440
                    # Get the linktype here so we can report it in email.
Kirk Webb's avatar
Kirk Webb committed
441
                    self.__findLinkType(nodeent)
Kirk Webb's avatar
Kirk Webb committed
442
                    if debug:
Kirk Webb's avatar
Kirk Webb committed
443
444
                        print "Found linktype %s for node %s" % \
                              (nodeent['LINKTYPE'], nodeent['IP'])
Kirk Webb's avatar
Kirk Webb committed
445
                        pass
Kirk Webb's avatar
   
Kirk Webb committed
446
447
448
                    # Add the node.
                    self.__addNode(nodeent)
                    # Add a line for the add/update message.
Kirk Webb's avatar
Kirk Webb committed
449
                    nodestr = "%s\t\t%s\t\t%s\t\t%s\t\t%s\n" % \
Kirk Webb's avatar
   
Kirk Webb committed
450
                              (nodeent['PLABID'],
Kirk Webb's avatar
Kirk Webb committed
451
452
453
454
                               nodeent['IP'],
                               nodeent['HNAME'],
                               nodeent['SITE'],
                               nodeent['LINKTYPE'])
Kirk Webb's avatar
   
Kirk Webb committed
455
                    addstr += nodestr
Kirk Webb's avatar
Kirk Webb committed
456
                    pass
Kirk Webb's avatar
   
Kirk Webb committed
457
458
                pass
            pass
459

Kirk Webb's avatar
   
Kirk Webb committed
460
        # Process node updates.
Kirk Webb's avatar
   
Kirk Webb committed
461
        updstr = ""
Kirk Webb's avatar
   
Kirk Webb committed
462
        chgerrstr = ""
Kirk Webb's avatar
   
Kirk Webb committed
463
464
        if len(toupdate):
            print "There are %d plab node updates." % len(toupdate)
Kirk Webb's avatar
   
Kirk Webb committed
465
            for (nodeent,(nodeid,diffattrs)) in toupdate:
Kirk Webb's avatar
   
Kirk Webb committed
466
                try:
Kirk Webb's avatar
   
Kirk Webb committed
467
                    self.__updateNodeMapping(nodeid, diffattrs)
Kirk Webb's avatar
   
Kirk Webb committed
468
469
470
471
472
                    pass
                except MultiChangeError, e:
                    print "%s not updated: Too many attribute changes." % \
                          e.nodeid
                    chgerrstr += "%s:\n" % e.nodeid
Kirk Webb's avatar
   
Kirk Webb committed
473
474
                    for (attr,(old,new)) in e.chattrs.items():
                        chgerrstr += "\t%s:\t%s => %s\n" % (attr,old,new)
Kirk Webb's avatar
   
Kirk Webb committed
475
476
477
                        pass
                    chgerrstr += "\n"
                    continue
Kirk Webb's avatar
   
Kirk Webb committed
478
                self.__updateNode(nodeid, nodeent)
Kirk Webb's avatar
   
Kirk Webb committed
479
                # Add a line for the add/update message.
Kirk Webb's avatar
   
Kirk Webb committed
480
481
482
                nodestr = nodeid + "\n"
                for (attr,(old,new)) in diffattrs.items():
                    nodestr += "\t%s:\t%s => %s\n" % (attr,old,new)
Kirk Webb's avatar
   
Kirk Webb committed
483
484
                    pass
                updstr += nodestr + "\n"
Kirk Webb's avatar
Kirk Webb committed
485
486
                pass
            pass
Kirk Webb's avatar
   
Kirk Webb committed
487

488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
        # Do node features updates separately since very few nodes are usually
        # updated, whereas we must do status separately from other fields.
        # XXX: munge this in with other fields later.
        upfeatures = []
        for nodeent in avail:
            # Determine if we already know about this node.
            try:
                matchres = self.__matchPlabNode(nodeent, known)
                if matchres:
                    upfeatures.append((nodeent,matchres))
                    pass
                pass
            except:
                pass
            pass
        
        for (nodeent,(nodeid,other)) in upfeatures:
            self.__updateNodeFeatures(nodeid,nodeent)
            pass
        
Kirk Webb's avatar
   
Kirk Webb committed
508
509
510
511
512
513
514
515
        if chgerrstr:
            SENDMAIL(TBOPS,
                     "Two or more changes detected for some plab nodes",
                     "Two or more distinguishing attributes have changed "
                     "on the following planetlab nodes:\n\n%s\n" % chgerrstr,
                     TBOPS)
            pass

Kirk Webb's avatar
   
Kirk Webb committed
516
517
518
519
        if len(toadd) or len(toupdate):
            # We need to update DNS since we've added hosts..
            print "Forcing a named map update ..."
            os.spawnl(os.P_WAIT, NAMED_SETUP, NAMED_SETUP)
Kirk Webb's avatar
   
Kirk Webb committed
520
521
            print "Pushing out site_mapping ..."
            os.spawnl(os.P_WAIT, PELAB_PUSH, PELAB_PUSH)
Kirk Webb's avatar
   
Kirk Webb committed
522
523
524
525
526
527
528
529
530
531
532
533
            # Now announce that we've added/updated nodes.
            SENDMAIL(TBOPS,
                     "Plab nodes have been added/updated in the DB.",
                     "The following plab nodes have been added to the DB:\n"
                     "PlabID\t\tIP\t\tHostname\t\tSite\t\tLinktype\n\n"
                     "%s\n\n"
                     "The following plab nodes have been updated in the DB:\n"
                     "\n%s\n\n" % \
                     (addstr, updstr),
                     TBOPS)
            print "Done adding new Plab nodes."
            pass
Kirk Webb's avatar
Kirk Webb committed
534
        return
535

Kirk Webb's avatar
   
Kirk Webb committed
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
    def __matchPlabNode(self, plabent, knownents):
        """
        Helper function.  Returns a two-element tuple or null.
        Null is returned when the node does not match any in the
        knownents list (none of it's attributes match those of any
        in the list).  If a match (or partial match) is found, a two
        element tuple is returned.  The first element is the emulab
        node id that matched, and the second is a dictionary containing
        thos elements that differed between the two (in the case of a
        partial match).
        """
        for nid in knownents:
            ent = knownents[nid]
            same = {}
            diff = {}
            for attr in ent:
                if ent[attr] in ATTR_NIL_VALUES:
                    continue
554
555
556
                elif (attr == "LATITUDE") or (attr == "LONGITUDE"):
                    # Special rules for latitude and longitude to avoid
                    # FP errors
557
558
559
560
561
562
563
564
565
                    nasty = False
                    try:
                        x = float(ent[attr])
                        x = float(plabent[attr])
                        pass
                    except:
                        nasty = True
                        pass
                    if (not nasty and ent[attr] != None and plabent[attr] != None) \
566
567
568
569
570
                           and (ent[attr] != "" and plabent[attr] != "") \
                           and ((float(ent[attr]) > \
                                 (float(plabent[attr]) + LATLONG_DELTA)) \
                                or (float(ent[attr]) < \
                                    (float(plabent[attr]) - LATLONG_DELTA))):
571
572
573
574
                        diff[attr] = (ent[attr], plabent[attr])
                    else:
                        same[attr] = ent[attr]
                        pass
Kirk Webb's avatar
   
Kirk Webb committed
575
576
577
578
579
580
581
                elif ent[attr] == plabent[attr]:
                    same[attr] = ent[attr]
                    pass
                else:
                    diff[attr] = (ent[attr], plabent[attr])
                    pass
                pass
582
583
            # Only consider these to be the same if at least one 'critical'
            # attr is the same
Kirk Webb's avatar
   
Kirk Webb committed
584
            if len(same):
585
586
587
                for attr in same:
                    if attr in ATTR_CRIT_KEYS:
                        return (nid, diff)
Kirk Webb's avatar
   
Kirk Webb committed
588
589
590
            pass
        return ()

591
592
593
594
595
    def __getKnownPnodes(self):
        """
        getFree helper function.  Returns a dict of IP:node_id pairs
        for the Plab nodes that currently exist in the DB.
        """
596
597
        res = DBQueryFatal("select plab_mapping.node_id,plab_id,"
                           "plab_mapping.hostname,IP,mac,site,latitude,"
Kirk Webb's avatar
   
Kirk Webb committed
598
                           "longitude,bwlimit"
599
600
601
602
                           " from plab_mapping"
                           " left join widearea_nodeinfo on"
                           "    plab_mapping.node_id = "
                           "    widearea_nodeinfo.node_id")
Kirk Webb's avatar
   
Kirk Webb committed
603
        
604
        ret = {}
Kirk Webb's avatar
   
Kirk Webb committed
605
606
        for (nodeid, plabid, hostname, ip, mac, site,
             latitude, longitude, bwlimit) in res:
607
608
609
610
611
612
            ret[nodeid] = {'PLABID'    : plabid,
                           'HNAME'     : hostname,
                           'IP'        : ip,
                           'MAC'       : mac,
                           'SITE'      : site,
                           'LATITUDE'  : latitude,
Kirk Webb's avatar
   
Kirk Webb committed
613
614
                           'LONGITUDE' : longitude,
                           'BWLIMIT'   : bwlimit}
Kirk Webb's avatar
Kirk Webb committed
615
            pass
Kirk Webb's avatar
   
Kirk Webb committed
616
617
618
619
620
621
622
623
624
625
        # Check for duplicate node attributes: report any that are found.
        dups = self.__findDuplicateAttrs(ret.values())
        if len(dups):
            SENDMAIL(TBOPS, "Duplicate plab node attributes in the DB!",
                     "Duplicate node attrs:\n"
                     "%s\n\n"
                     "Fix up please!" % dups,
                     TBOPS)
            raise RuntimeError, \
                  "Duplicate node attributes in DB:\n%s" % dups            
626
        return ret
Kirk Webb's avatar
   
Kirk Webb committed
627
628
629
630
631
632
633
634
635

    def __findDuplicateAttrs(self, nodelist):
        """
        Find duplicate node attributes in the node list passed in.
        """
        attrs = {}
        dups = {}
        
        for ent in nodelist:
Kirk Webb's avatar
   
Kirk Webb committed
636
            for attr in ATTR_CRIT_KEYS:
Kirk Webb's avatar
   
Kirk Webb committed
637
638
639
640
641
642
643
644
645
646
647
648
649
650
                entry = "%s:%s" % (attr, ent[attr])
                if attrs.has_key(entry) and \
                   ent[attr] not in ATTR_NIL_VALUES:
                    print "Duplicate node attribute: %s" % entry
                    if not dups.has_key(entry):
                        dups[entry] = [attrs[entry],]
                        pass
                    dups[entry].append(ent['PLABID'])
                else:
                    attrs[entry] = ent['PLABID']
                    pass
                pass
            pass
        return dups
Kirk Webb's avatar
   
Kirk Webb committed
651
        
Kirk Webb's avatar
Kirk Webb committed
652
    def __findLinkType(self, nodeent):
653
654
655
656
657
658
659
        """
        getFree helper function.  Figures out the link type of the given
        host.  This first performs a traceroute and checks for the U of
        U's I2 gateway to classify Internet2 hosts.  If this test fails,
        it checks if the hostname is international.  If this test fails,
        this simply specifies an inet link type.

Kirk Webb's avatar
Kirk Webb committed
660
        This can't detect DSL links..
661
        """
662
        # Is host international (or flux/emulab local)?
663
        from socket import gethostbyaddr, getfqdn, herror
Kirk Webb's avatar
Kirk Webb committed
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
        
        if not nodeent.has_key('HNAME'):
            try:
                (hname, ) = gethostbyaddr(ip)
                nodeent['HNAME'] = getfqdn(hname)
                pass
            except herror:
                nodeent['HNAME'] = nodeent['IP']
                print "WARNING: Failed to get hostname for %s" % nodeent['IP']
                pass
            pass
        
        tld = nodeent['HNAME'].split(".")[-1].lower()
        if not tld in ("edu", "org", "net", "com", "gov", "us", "ca"):
            nodeent['LINKTYPE'] = "pcplabintl"
            return
        
        # Is it us?
        if nodeent['HNAME'].endswith(LOCAL_PLAB_DOMAIN):
            nodeent['LINKTYPE'] = LOCAL_PLAB_LINKTYPE
            return
        
686
        # Is host on I2?
Kirk Webb's avatar
Kirk Webb committed
687
        traceroute = os.popen("traceroute -nm 10 -q 1 %s" % nodeent['IP'])
688
689
690
691
692
        trace = traceroute.read()
        traceroute.close()

        for gw in MAGIC_INET2_GATEWAYS:
            if trace.find(gw) != -1:
Kirk Webb's avatar
Kirk Webb committed
693
694
                nodeent['LINKTYPE'] = "pcplabinet2"
                return
695

696
697
        for gw in MAGIC_INET_GATEWAYS:
            if trace.find(gw) != -1:
Kirk Webb's avatar
   
Kirk Webb committed
698
699
                nodeent['LINKTYPE'] = "pcplabinet"
                return
700
        else:
Kirk Webb's avatar
Kirk Webb committed
701
            print "WARNING: Unknown gateway for host %s" % nodeent['IP']
702

Kirk Webb's avatar
   
Kirk Webb committed
703
704
        # We don't know - must manually classify.
        nodeent['LINKTYPE'] = "*Unknown*"
Kirk Webb's avatar
Kirk Webb committed
705
        return
706

Kirk Webb's avatar
   
Kirk Webb committed
707
    def __addNode(self, nodeent):
708
        """
Kirk Webb's avatar
   
Kirk Webb committed
709
710
711
        updateNodeEntries() helper function.  Adds a new Plab pnode and
        associated vnode to the DB.  The argument is a dictionary containing
        the new node's attributes.
712
        """
Kirk Webb's avatar
   
Kirk Webb committed
713
        # Generate/grab variables to be used when creating the node.
714
        defosid, controliface = self.__getNodetypeInfo()
Kirk Webb's avatar
Kirk Webb committed
715
        hostonly = nodeent['HNAME'].replace(".", "-")
Kirk Webb's avatar
   
Kirk Webb committed
716
717
718
719
720
721
722
723
724
725
726
727
        nidnum, priority = self.__nextFreeNodeid()
        nodeid = "%s%d" % (NODEPREFIX, nidnum)
        vnodeprefix = "%svm%d" % (NODEPREFIX, nidnum)
        print "Creating pnode %s as %s, priority %d." % \
              (nodeent['IP'], nodeid, priority)

        # Do the stuff common to both node addition and update first
        # Note that if this fails, we want the exception generated to
        # percolate up to the caller immediately, so don't catch it.
        self.__updateNode(nodeid, nodeent)

        # Now perform stuff specific to node addition
Kirk Webb's avatar
   
Kirk Webb committed
728
        try:
729
730
731
            res_exptidx = TBExptIDX(RESERVED_PID, RESERVED_EID)
            mon_exptidx = TBExptIDX(MONITOR_PID, MONITOR_EID)
            
Kirk Webb's avatar
Kirk Webb committed
732
            DBQueryFatal("replace into nodes"
733
                         " (node_id, type, phys_nodeid, role, priority,"
Kirk Webb's avatar
   
Kirk Webb committed
734
                         "  op_mode, def_boot_osid,"
735
                         "  allocstate, allocstate_timestamp,"
Kirk Webb's avatar
   
Kirk Webb committed
736
                         "  eventstate, state_timestamp)"
737
                         " values (%s, %s, %s, %s, %s,"
Kirk Webb's avatar
   
Kirk Webb committed
738
739
740
741
                         "  %s, %s, %s, now(), %s, now())",
                         (nodeid, 'pcplabphys', nodeid,
                          'testnode', priority*100,
                          'ALWAYSUP', defosid,
742
                          'FREE_CLEAN',
Kirk Webb's avatar
   
Kirk Webb committed
743
                          'ISUP'))
744

745
746
747
            DBQueryFatal("replace into node_hostkeys"
                         " (node_id)"
                         " values (%s)",
Kirk Webb's avatar
Kirk Webb committed
748
                         (nodeid))
749

Kirk Webb's avatar
Kirk Webb committed
750
            DBQueryFatal("replace into reserved"
751
                         " (node_id, exptidx, pid, eid, rsrv_time, vname)"
752
                         " values (%s, %s, %s, %s, now(), %s)",
753
754
                         (nodeid, res_exptidx,
                          RESERVED_PID, RESERVED_EID, hostonly))
Kirk Webb's avatar
   
Kirk Webb committed
755

Kirk Webb's avatar
   
Kirk Webb committed
756
757
            # XXX: This should probably be checked and updated if necessary
            #      when updating.
Kirk Webb's avatar
Kirk Webb committed
758
            DBQueryFatal("replace into node_auxtypes"
Kirk Webb's avatar
   
Kirk Webb committed
759
760
                         " (node_id, type, count)"
                         " values (%s, %s, %s)",
Kirk Webb's avatar
Kirk Webb committed
761
                         (nodeid, nodeent['LINKTYPE'], 1))
Kirk Webb's avatar
   
Kirk Webb committed
762
            
Kirk Webb's avatar
Kirk Webb committed
763
            DBQueryFatal("replace into node_auxtypes"
764
765
766
767
                         " (node_id, type, count)"
                         " values (%s, %s, %s)",
                         (nodeid, 'pcplab', 1))
            
Kirk Webb's avatar
Kirk Webb committed
768
            DBQueryFatal("replace into node_status"
769
770
                         " (node_id, status, status_timestamp)"
                         " values (%s, %s, now())",
Kirk Webb's avatar
Kirk Webb committed
771
                         (nodeid, 'down'))
Kirk Webb's avatar
   
Kirk Webb committed
772

Kirk Webb's avatar
   
Kirk Webb committed
773
774
775
776
777
            DBQueryFatal("insert into plab_mapping"
                         " (node_id, plab_id, hostname, IP, mac, create_time)"
                         " values (%s, %s, %s, %s, %s, now())",
                         (nodeid, nodeent['PLABID'], nodeent['HNAME'],
                          nodeent['IP'], nodeent['MAC']))
778

Kirk Webb's avatar
   
Kirk Webb committed
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
            #
            # NowAdd the site_mapping entry for this node.
            #
            
            # See if we know about the associated site - grab idx if so
            siteidx = 0
            nodeidx = 1
            siteres = DBQueryFatal("select site_idx, node_idx from "
                                   " plab_site_mapping where site_name=%s",
                                   nodeent['SITE']);
            if len(siteres):
                # There are already nodes listed for this site, so get
                # the next node id.
                siteidx = siteres[0][0]
                for (foo, idx) in siteres:
                    if idx > nodeidx: nodeidx = idx
                    pass
                nodeidx += 1
                pass
            else:
                # No nodes listed for site, so get the largest site_idx
                # in the DB so far, and increment cuz we're going to add
                # a new one.
                maxres = DBQueryFatal("select MAX(site_idx) from "
                                      " plab_site_mapping")
                try:
                    siteidx = int(maxres[0][0]) + 1
                    pass
                except ValueError:
                    siteidx = 1
                    pass
                pass
            # Create site_mapping entry, optionally creating new site idx
            # via not specifying the site_idx field (field is auto_increment)
            DBQueryFatal("insert into plab_site_mapping "
                         " values (%s, %s, %s, %s)",
                         (nodeent['SITE'], siteidx, nodeid, nodeidx))

817
            # Create a single reserved plab vnode for the managment sliver.
Kirk Webb's avatar
   
Kirk Webb committed
818
819
820
821
822
            n = 1
            vprio = (priority * 100) + n
            sshdport = 38000 + n
            vnodeid = "%s-%d" % (vnodeprefix, n)
            vnodetype = "pcplab"
823
824
            if verbose:
                print "Creating vnode %s, priority %d" % (vnodeid, vprio)
Kirk Webb's avatar
Kirk Webb committed
825
                pass
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
                    
            DBQueryFatal("insert into nodes"
                         " (node_id, type, phys_nodeid, role, priority,"
                         "  op_mode, def_boot_osid, update_accounts,"
                         "  allocstate, allocstate_timestamp,"
                         "  eventstate, state_timestamp, sshdport)"
                         " values (%s, %s, %s, %s, %s,"
                         "  %s, %s, %s, %s, now(), %s, now(), %s)",
                         (vnodeid, vnodetype, nodeid, 'virtnode', vprio,
                          'PCVM', defosid, 1,
                          'FREE_CLEAN',
                          'SHUTDOWN', sshdport))

            DBQueryFatal("insert into node_hostkeys"
                         " (node_id)"
                         " values (%s)",
                         (vnodeid))
            
            DBQueryFatal("insert into node_status"
                         " (node_id, status, status_timestamp)"
                         " values (%s, %s, now())",
                         (vnodeid, 'up'))
            
Kirk Webb's avatar
Kirk Webb committed
849
            # Put the last vnode created into the special monitoring expt.
Kirk Webb's avatar
   
Kirk Webb committed
850
            DBQueryFatal("insert into reserved"
851
                         " (node_id, exptidx, pid, eid, rsrv_time, vname)"
852
                         " values (%s, %s, %s, %s, now(), %s)",
853
854
                         (vnodeid, mon_exptidx,
                          MONITOR_PID, MONITOR_EID, vnodeid))
Kirk Webb's avatar
Kirk Webb committed
855
856
            pass
        
Kirk Webb's avatar
   
Kirk Webb committed
857
858
859
860
861
862
        except:
            print "Error adding PLAB node to DB: someone needs to clean up!"
            tbmsg = "".join(traceback.format_exception(*sys.exc_info()))
            SENDMAIL(TBOPS, "Error adding new plab node to DB: %s\n" %
                     nodeid, "Some operation failed while trying to add a"
                     " newly discovered plab node to the DB:\n %s"
863
                     "\n Please clean up!\n" % tbmsg, TBOPS)
Kirk Webb's avatar
   
Kirk Webb committed
864
            raise
Kirk Webb's avatar
   
Kirk Webb committed
865
866
        return

867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
    def __updateNodeFeatures(self,nodeid,nodeent):
        """
        Record the status of this node in the node_features
        table.
        """
        # XXX Make this atomic
        #
        try:
            # Note that we have to pass '%' as an arg to DBQuery, sigh
            DBQueryFatal("delete from node_features where node_id=%s" \
                         " and feature like %s",
                         (nodeid,'plabstatus-%'))
            
            if nodeent.has_key('STATUS'):
                # Kind of a hack - we assume most people will want Production
                # nodes
                if nodeent['STATUS'] == "Production" :
                    weight = 0.0
                    pass
                else:
                    weight = 1.0
                    pass
                DBQueryFatal("insert into node_features" \
                             " (node_id, feature, weight)" \
                             " values (%s,%s,%s)",
                             (nodeid,
                              'plabstatus-%s' % nodeent['STATUS'],
                              weight))
                pass
            pass
        except:
            print "Error updating plab node STATUS feature " \
                  "for node %s!" % nodeid
            traceback.print_exc()
            
        
        return None
    
Kirk Webb's avatar
   
Kirk Webb committed
905
906
907
908
909
910
911
912
913
914
915
916
    def __updateNode(self, nodeid, nodeent):
        """
        updateNodeEntries() helper function.  Updates attributes for plab
        nodes passed in via the nodeent argument.
        """
        # Get the name of the control interface for plab nodes.
        defosid, controliface = self.__getNodetypeInfo()

        haslatlong = (('LATITUDE' in nodeent and 'LONGITUDE' in nodeent) and
            (nodeent['LATITUDE'] != 0 or nodeent['LONGITUDE'] != 0))
        try:
            DBQueryFatal("replace into widearea_nodeinfo"
917
918
919
920
921
                         " (node_id, contact_uid, contact_idx, hostname, site,"
                         "  latitude, longitude, bwlimit)"
                         " values (%s, %s, %s, %s, %s, %s, %s, %s)",
                         (nodeid, 'nobody', '0', nodeent['HNAME'],
                          nodeent['SITE'],
Kirk Webb's avatar
   
Kirk Webb committed
922
923
924
925
                          # Poor man's ternary operator
                          haslatlong and nodeent['LATITUDE'] or "NULL",
                          haslatlong and nodeent['LONGITUDE'] or "NULL",
                          nodeent['BWLIMIT']))
926

Kirk Webb's avatar
   
Kirk Webb committed
927
928
929
930
931
932
            DBQueryFatal("replace into interfaces"
                         " (node_id, card, port, IP, interface_type,"
                         " iface, role)"
                         " values (%s, %s, %s, %s, %s, %s, %s)",
                         (nodeid, 0, 1, nodeent['IP'], 'plab_fake',
                          controliface, 'ctrl'))
933

Kirk Webb's avatar
   
Kirk Webb committed
934
935
936
937
938
939
940
941
942
            pass
        except:
            print "Error updating PLAB node in DB: someone needs to clean up!"
            tbmsg = "".join(traceback.format_exception(*sys.exc_info()))
            SENDMAIL(TBOPS, "Error updating plab node in DB: %s\n" % nodeid,
                     "Some operation failed while trying to update"
                     " plab node %s in the DB:\n\n%s"
                     "\nPlease clean up!\n" % (nodeid, tbmsg), TBOPS)
            raise
Kirk Webb's avatar
Kirk Webb committed
943
        return
944

Kirk Webb's avatar
   
Kirk Webb committed
945

Kirk Webb's avatar
   
Kirk Webb committed
946
    def __updateNodeMapping(self, nodeid, chattrs):
Kirk Webb's avatar
   
Kirk Webb committed
947
        """
Kirk Webb's avatar
   
Kirk Webb committed
948
        Updates changed node attributes in the plab mapping table.
Kirk Webb's avatar
   
Kirk Webb committed
949
950
951
        """
        uid = os.getuid()
        dbuid = uid == 0 and "root" or UNIX2DBUID(uid)
Kirk Webb's avatar
   
Kirk Webb committed
952

Kirk Webb's avatar
   
Kirk Webb committed
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
        # mapping from attrs to column names
        attrmap = {'PLABID' : 'plab_id',
                   'HNAME'  : 'hostname',
                   'IP'     : 'IP',
                   'MAC'    : 'mac'}

        # Get the intersection of mapping (critical) keys with those that
        # have changed.
        changedcritkeys = set(ATTR_CRIT_KEYS) & set(chattrs.keys())
        # nothing to do if none of the mapping attributes have changed.
        if not changedcritkeys:
            return
        # If the node has more than two critical attrs that have changed,
        # then move it to hwdown and raise an exception.
        if len(changedcritkeys) > 2:
            crattrs = {}
            for chkey in changedcritkeys:
                crattrs[chkey] = chattrs[chkey]
Kirk Webb's avatar
   
Kirk Webb committed
971
972
                pass
            errmsg = "More than 2 plab node attrs have changed!\n\n%s\n\n" \
Kirk Webb's avatar
   
Kirk Webb committed
973
                     "%s has been moved to hwdown." % (crattrs, nodeid)
Kirk Webb's avatar
   
Kirk Webb committed
974
975
            MarkPhysNodeDown(nodeid)
            TBSetNodeLogEntry(nodeid, dbuid, TB_NODELOGTYPE_MISC, errmsg)
Kirk Webb's avatar
   
Kirk Webb committed
976
977
978
979
980
981
982
            raise MultiChangeError(nodeid, crattrs)

        # Update mapping table entry.
        updstr = ",".join(map(lambda x: "%s='%s'" %
                              (attrmap[x],chattrs[x][1]), changedcritkeys))
        DBQueryFatal("update plab_mapping set %s where node_id='%s'" %
                     (updstr, nodeid))
Kirk Webb's avatar
   
Kirk Webb committed
983
984
        updmsg = "Plab node %s attributes updated:\n\n%s" % (nodeid, chattrs)
        TBSetNodeLogEntry(nodeid, dbuid, TB_NODELOGTYPE_MISC, updmsg)
985

Kirk Webb's avatar
   
Kirk Webb committed
986
987
988
989
990
        # updateNodeEtries() already sends mail.
        #SENDMAIL(TBOPS,
        #         "Plab node %s attributes updated." % nodeid, updmsg, TBOPS)
        return

991
992
    def __getNodetypeInfo(self):
        """
993
994
        addNode helper function.  Returns a (defosid, controliface) 
        tuple for the Plab pnode type.  Caches the result since
995
        it doesn't change.
996
997
        """
        if not hasattr(self, "__getNodetypeInfoCache"):
Kirk Webb's avatar
Kirk Webb committed
998
            if debug:
999
                print "Getting node type info"
Kirk Webb's avatar
Kirk Webb committed
1000
                pass
For faster browsing, not all history is shown. View entire blame