assign_wrapper.in 129 KB
Newer Older
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1
#!/usr/bin/perl -w
Leigh B. Stoller's avatar
Leigh B. Stoller committed
2
3
#
# EMULAB-COPYRIGHT
4
# Copyright (c) 2000-2004 University of Utah and the Flux Group.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
5
6
# All rights reserved.
#
7
8
use English;
use Getopt::Std;
9
10
use POSIX qw(setsid);
use POSIX ":sys_wait_h";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
11

12
#
Leigh B. Stoller's avatar
Leigh B. Stoller committed
13
14
15
16
17
18
# This function as the main assign loop.  It converts the virtual
# topology into a top input including LAN and delay translation.  It
# then snapshots the current testbed physical state and runs assign,
# looping a couple times if assign fails.  When assign successfully
# completes it will interpret the results.  Attempt to match any
# existing portmap entries and then update the delays and vlans table.
Chad Barb's avatar
Chad Barb committed
19
#
20
21
22
23
# XXX: Update does not work with widearea nodes.
#      Internally created nodes (jailhost,delay,sim) are not treated
#        consistently. Needs more thought.
#
24
25
26
27
28
29
30
31
32
33
34
35
36
# Return codes: -1 is an uncontrolled error (someone called die()). No
# recovery is possible so the caller has to check for this explicitly.
# The CANRECOVER bit indicates 'recoverability' (no db or physical
# state was modified by the time the error occurred). It also is used
# to signal that a retry is possible. 
#
my $WRAPPER_SUCCESS		 = 0x00;
my $WRAPPER_FAILED		 = 0x01;	# Failed (Add other values)
my  $WRAPPER_FAILED_MAXCONCURENT = 0x02;	# max_concurrent violation
my  $WRAPPER_FAILED_BANDWIDTH    = 0x04;	# Bandwidth violation
my  $WRAPPER_FAILED_LINKUSERS    = 0x08;	# Linkusers violation
my  $WRAPPER_FAILED_DESIRES      = 0x10;	# Desires violation
my  $WRAPPER_FAILED_UNASSIGNED   = 0x20;	# Unassigned violation
37
my  $WRAPPER_FAILED_CANRECOVER   = 0x40;        # Can recover, retry.
38
my $WRAPPER_DIED_A_HORRIBLE_DEATH= 0x80;	# Do not use this.
39
40
41
# Set this once we modify DB state; forces no recover in fatal().
my $NoRecover = 0;
    
42
43
sub usage ()
{
44
45
46
47
    print STDERR "Usage: $0 [-v] [-u [-f] | -n] pid eid\n";
    print STDERR " -v   - Enables verbose output\n";
    print STDERR " -u   - Enables update mode\n";
    print STDERR " -f   - Fix current resources during update mode\n";
48
49
50
51
    print STDERR " -t   - Create the TOP file and then exit\n";
    print STDERR " -n   - Run assign, but do not reserve/modify resources.\n";
    exit(-1);
}
52
my $optlist  = "vutnf";
53
my $verbose  = 0;
54
my $fixmode  = 0;
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
my $updating = 0;
my $toponly  = 0;
my $impotent = 0;

#
# Configure variables
#
my $TBROOT	  = "@prefix@";
my $DELAYCAPACITY = @DELAYCAPACITY@;
$ENV{'PATH'} = "/usr/bin:$TBROOT/libexec:$TBROOT/sbin:$TBROOT/bin";

#
# Turn off line buffering on output
#
$| = 1;

#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;
require exitonwarn;

Chad Barb's avatar
Chad Barb committed
79
#
80
81
82
83
84
# assign_wrapper Settings
#
# Maximum delay in ms above which a delay node is needed.
# (Note that the DB represents delays as floating point numbers)
my $delaythresh = 2;
Chad Barb's avatar
Chad Barb committed
85

86
87
# Maximum number of times we run assign.
my $maxrun = 20;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
88

89
90
91
# Use the switch to delay when possible. Currentlythis only works for 10mbit
# links (actually, its turned off cause it does not work; auto handshake).
my $delaywithswitch = 0;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
92

93
94
95
96
97
98
99
#
# Some handy constants. Speed in Mbits/sec and Kbits/sec units.
#
# Its probably a good idea to leave portbw (current_speed) in Mbs, since
# those numbers are used at the switch and the interfaces, which really
# only think in Mbps.
#
100
101
102
103
104
105
my $S10Mbs   = 10;
my $S100Mbs  = 100;
my $S1000Mbs = 1000;
my $S10Kbs   = 10000;
my $S100Kbs  = 100000;
my $S1000Kbs = 1000000;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
106

107
#
108
109
# Parse command arguments. Once we return from getopts, all that should be
# left are the required arguments.
110
#
111
112
113
%options = ();
if (! getopts($optlist, \%options)) {
    usage();
114
}
115
116
if (@ARGV != 2) {
    usage();
Chad Barb's avatar
Chad Barb committed
117
}
118
if (defined($options{"v"})) {
Chad Barb's avatar
   
Chad Barb committed
119
120
    $verbose = 1;
}
121
if (defined($options{"u"})) {
Chad Barb's avatar
   
Chad Barb committed
122
    $updating = 1;
123
}
124
125
126
127
128
if (defined($options{"t"})) {
    $toponly = 1;
}
if (defined($options{"n"})) {
    $impotent = 1;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
129
}
130
131
132
if (defined($options{"f"})) {
    $fixmode = 1;
}
133
134
135
my $pid = $ARGV[0];
my $eid = $ARGV[1];
my $ptopfile = "$pid-$eid-$$.ptop";
136
137
138
139
140
# Since the topfile could change across
# swapins and modifies, it makes sense
# to store all of them. Helps in
# degugging.
my $topfile  = "$pid-$eid-$$.top";
141

142
143
144
sub fatal ($$)
{
    my($exitcode, $message) = @_;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
145

146
147
148
149
150
151
152
    print STDERR "*** $0:\n".
	         "    $message\n";

    # Remove recover bit.
    $exitcode = $exitcode & ~$WRAPPER_FAILED_CANRECOVER
	if ($NoRecover);
    
153
154
    exit($exitcode);
}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
155

156
157
sub printdb ($)
{
Leigh B. Stoller's avatar
Leigh B. Stoller committed
158
159
160
    if ($verbose) {
	print $_[0];
    }
161
}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
162

163
print "assign_wrapper improved started\n";
164
165
TBDebugTimeStamp("assign_wrapper started");

Leigh B. Stoller's avatar
Leigh B. Stoller committed
166
#
167
# The main data structures:
Leigh B. Stoller's avatar
Leigh B. Stoller committed
168
#
169
170
171
172
173
# virt_nodes: The virtual nodes, indexed by vname. Each entry is a
# hash reference, initially of just the DB info, but possibly
# augmented as we proceed through assign.  Do not confuse these
# virtual nodes with the other virtual nodes! These are the ones from
# the actual topology, the virt_nodes table in the DB.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
174
#
175
176
my %virt_nodes = ();

Leigh B. Stoller's avatar
Leigh B. Stoller committed
177
#
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
# virt_lans: The equivalent of virt_nodes; the virt_lans table in the DB.
# Since there are multiple rows per lan (one for each node), this is a
# multilevel structure. The first slot is another hash, one for each node.
# The rest of the slots store other random things associated with the lan.
# So, looks something like:
#
#   %virt_lans = (link0 => {members    => member0 => { db row ref },
#                                         member1 => { db row ref }}
#                           mustdelay    => 0,
#                           emulated     => 0,
#                           uselinkdelay => 0,
#                           nobwshaping  => 0,
#                           useveth      => 0,
#                           trivok       => 0
#                          }
#                 link1 => ...
#                )
#
my %virt_lans = ();

#
# virt_vtypes: The virt_vtypes table in the DB, indexed by the vtype
# name (user chosen name).
#
my %virt_vtypes = ();

#
# node_types: The node_types table from the DB, indexed by the type name.
#
my %node_types = ();

Leigh B. Stoller's avatar
Leigh B. Stoller committed
209
# 
210
211
212
213
# interface_types: We need this to find out the bandwidths of the devices
# we actually have on the testbed. Index by interface type name.
#
my %interface_types = ();
Leigh B. Stoller's avatar
Leigh B. Stoller committed
214

215
216
217
218
219
220
221
222
#
# XXX Hack table for determining if a delay node is required. We store
# the native link speeds for each type,class. Indexed by type and class,
# the value is a list of link hardware speeds for which no delay node is
# required. This is an awful way to do this, and quite wrong.
#
my %node_type_linkbw = ();

223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
#
# phys_nodes: The equiv of virt_nodes above, except that these are pulled
# from the DB once the physical resources have been allocated. Indexed
# by physname, but there is a pointer from the virt_nodes table entry
# to the the corresponding physnode entry. 
# 
my %phys_nodes = ();

#
# More physical side data structures.
# v2pmap is indexed by virtual and contains the physical node.
my %v2pmap = ();
# p2vmap is indexed by physical and contains one or more virtual nodes.
my %p2vmap = ();
# plinks is indexed by virtual name and contains
#  (pnodeportA,pnodeportB) .  If one is a delay node it is always
#  the second.
my %plinks = ();
# virtnodes is the list of subnodes on physnodes.
my %virtnodes = ();
my %v2vmap = ();

#
# Support for experiment modify. We create v2p and v2v mappings of the
# current topology so we can figure out how its changed after assign
# runs. These correspond to v2pmap and v2vmap mentioned above.
# 
my %reserved_v2pmap  = ();
my %reserved_v2vmap  = ();
252
253
254
255
my %oldreservednodes = ();
# reserved_p2vmap is indexed by physical and contains one or more virtual
# nodes
my %reserved_p2vmap  = ();
256

Leigh B. Stoller's avatar
Leigh B. Stoller committed
257
#
258
259
260
261
262
263
264
265
266
267
268
269
270
# Experiment wide options. See below. They come from the experiments table.
# Defining these will override experiment table setting. 
#
# Set this when forcing linkdelays instead of delay nodes. Set in the NS
# file with a tb-compat directive. The force directive says to set up a
# link delay, even when no delay would otherwise be inserted.
# usewatunnels is also set in the NS, and can be used to turn them off. 
# The multiplex_factor is to override node_types table for virtnode.
my $uselinkdelays;
my $forcelinkdelays;
my $usewatunnels;
my $multiplex_factor;

271
272
273
274
# For admission control. Not well defined yet.
my $cpu_usage;
my $mem_usage;

275
276
277
278
# Allow override of jail/delay osids.
my $jail_osid;
my $delay_osid;

279
280
281
282
283
284
# Flag that tells us whether to fix-node current
# resources or not during a swap modify. This is
# useful when vnode weights are tweaked by the experimenter
# before doing a swapmod
my $fix_current_resources;

285
######################################################################
Leigh B. Stoller's avatar
Leigh B. Stoller committed
286

287
288
289
290
291
# ips is indexed by node:port and contains the IP address for the port.
my %ips	      = ();

# memberof is indexed by node:port and holds the lan it is a member of.
my %memberof = ();
292
293

# delaylinks stores the actual link delay info, converted from the
294
295
296
297
298
# virt delay params above. It is indexed by link name and contains a
# [delay,bw,loss] array for each direction. The final member of the
# array is an indicator of whether the info is there strictly for a 
# trivial link (linkdelay inserted only when assign makes link trivial).
my %delaylinks     = ();
299
300
301
302
303
304
305
306
307
308
309

# delaynodes stores the names of delaynodes that we create on the
# fly using delayid. This is useful for doing isdelay? tests.
my %delaynodes = ();
my $delayid    = 0;

# nodedelays and linkdelays are the final (!) representation. Indexed by
# integer id, they store the physical node info and the delay info. 
my %nodedelays = ();
my %linkdelays = ();

310
311
312
313
# Virtual nodes that the user has requested be "fixed" to a specific
# physical node.
my %fixed_nodes     = ();

314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
# portbw is indexed by virtual nodeport and contains the bandwidth of
# that port. Note that port bandwidth in the interfaces table is left
# in Mbps units for now. Thats inconsistent, I know. For LANs with
# other bandwidths the port speed will be 100 and a delay node will be
# inserted.
my %portbw = ();

# lannodes is indexed by physical name and is the set of fake lan nodes.
# lan nodes are named lan/<virtual lan>. We need to know these so that
# when they come back from assign, we can ignore them.
my %lannodes = ();

# Node estimates and counts. Avoid running assign if there is no way to
# satisfy the estimates for physical nodes.
my $minimum_nodes;
my $maximum_nodes;
330
331
my $reserved_pcount  = 0;
my $reserved_vcount  = 0;
332
my $reserved_simcount= 0;
333
334
335
336
my $remotecount      = 0;
my $virtcount        = 0;
my $plabcount        = 0;
my $needwanassign    = 0;
337
my $simcount         = 0;
338

339
340
341
342
343
344
345
346
347
348
349
350
#
# This is for stats gathering. It might duplicate other stuff, but
# thats okay.
#
my %expt_stats = (# pnodes include jailnodes and delaynodes.
		  # We let the wrapper determine pnodes once the
		  # experiment is fully swapped in so that the record
		  # is not "committed" until successful swapin.
		  jailnodes   => 0,
		  vnodes      => 0,
                  # vnodes include wanodes.
		  wanodes     => 0,
351
352
		  # wanodes includes plabnodes.
		  plabnodes   => 0,
353
354
355
356
357
358
359
360
361
362
363
364
365
		  simnodes    => 0,
		  delaynodes  => 0,
		  linkdelays  => 0,
		  links       => 0,
		  walinks     => 0,
		  lans        => 0,
		  shapedlinks => 0,
		  shapedlans  => 0,
		  minlinks    => 100000,
		  # includes emulated links. Maybe thats wrong.
		  maxlinks    => 0,
);

366
367
368
my $simhost_id     = 0;
my %pnode2simhostid;
my %simhostid2pnode;
Shashi Guruprasad's avatar
Shashi Guruprasad committed
369

370
# Counters for generating IDs.
371
my $virtnode_id  = 0;
372
my $veth_id      = 0;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
373

374
375
376
377
378
379
380
381
382
383
#
# Every vnode on a pnode gets its own routing
# table, thus an rtabid. In the case of simulated
# nodes, we need rtabids only for border nodes
# i.e. nodes that have links going out of the pnode.
# Either way, new rtabids are requested via getrtabid(<vnode>,<pnode>)
# when the interfaces table is updated or new veth_interfaces
# are inserted. This hash maintains the rtabids per vnode
my %vnode2rtabid = ();

384
385
######################################################################
# Step 1 - Setup virtual topology
386
#
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
# Here we need to read the virtual topology in from the virt_nodes
# and virt_lans table.  We then need to add delay and lan nodes as
# necessary.
#
# Conversion details:
#   Let L be a LAN with N members.
#   If N == 2 
#      Let N1 be node 1
#      Let N2 be node 2
#      If L is delayed
#         Generate delay node D
#         Link N1 to D
#         Link N2 to D
#      Else
#         Link N1 to N2
#   Else
#      Generate lan node A
#      If L is delayed
#        Foreach node N in L
#           Generate delay node DN
#           Link A to DN
#           Link N to DN
#      Else
#        Foreach node N in L
#           Link N to A
#
# Delay node names:
#  delay nodes are named tbdelayXX N > 2
#   and tbsdelayXX for N == 2.
#
########################################################################
Chad Barb's avatar
   
Chad Barb committed
418

419
420
printdb "Generating TOP file.\n";
TBDebugTimeStamp("TOP started");
421

422
423
424
425
426
427
#
# vtypes are a funny mix beteween physical and virtual state, so we have to
# load them before getting the PhysInfo.
#
LoadVirtTypes();

428
429
430
431
432
#
# Load phys info. Interface types, node types, etc. Its any physical stuff
# we need.
#
LoadPhysInfo();
433

434
435
436
437
#
# Load the Experiment info and virt topology.
#
LoadExperiment();
438

439
440
441
442
#
# If updating, load current experiment resources. We have to be careful
# of how this is merged in with the (new) desired topology. See below.
#
443
444
445
446
447
if ($updating) {
    LoadCurrent();
    print STDERR "Resetting DB before updating.\n";
    TBExptRemovePhysicalState( $pid, $eid );
}
Chad Barb's avatar
   
Chad Barb committed
448

449
450
451
452
#
# Check Max Concurrent for OSID violations.
#
CheckMaxConcurrent();
453

454
455
456
457
#
# Create the TOP file.
#
CreateTopFile();
458

459
460
461
462
463
464
TBDebugTimeStamp("TOP finished");

# Stop here ...
if ($toponly) {
    print "Stopping after creating the TOP file, as directed.\n";
    exit(0);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
465
}
466

467

468
469
470
471
472
473
474
475
476
######################################################################
# Step 2 - Assign Loop
# 
# Here we loop up to maxrun times.  In each loop we snapshot the
# current testbed state into a ptop file.  We then run assign.  If
# assign succeeds we attempt to reserve the resources.  If that works
# we're done with step 2 otherwise we loop again.
#
#######################################################################
Leigh B. Stoller's avatar
Leigh B. Stoller committed
477

478
479
my $currentrun = 1;
my $canceled   = 0;
480
481

# XXX plab hack - only run assign once on plab topologies, since they're easy
482
# to map and the physical topology does not change frequently.
483
if ($plabcount && (keys(%virt_nodes) == $plabcount)) {
484
    $maxrun = 2;
485
}
486

487
488
TBDebugTimeStamp("assign_loop started");
while (1) {
489
    # Check cancel flag before continuing. 
490
    TBGetCancelFlag($pid, $eid, \$canceled);
491
492
    fatal($WRAPPER_FAILED|$WRAPPER_FAILED_CANRECOVER,
	  "Cancel flag set; aborting assign run!")
493
494
	if ($canceled);

495
    print "Assign Run $currentrun\n";
496

497
498
499
500
501
502
503
504
505
506
507
508
509
    #
    # RunAssign returns 0 if successful. Returns -1 if there was an
    # an error in RunAssign that is deemed fatal; do not try anymore.
    # The exitcode is returned in $exitcode. If RunAssign returns a
    # postive value, it means there was a problem, but we can keep
    # trying.
    #
    my $exitcode = 0;
    my $retval   = RunAssign(\$exitcode);

    # Success!
    last
	if ($retval == 0);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
510

511
512
513
514
    if ($currentrun >= $maxrun) {
	fatal($exitcode, "Reached run limit. Giving up.");
    }

515
516
517
518
    if ($retval < 0) {
	#
	# Failure in assign.
	#
519
	fatal($exitcode, "Unretriable error. Giving up.");
520
    }
521
    
522
523
524
525
526
527
528
529
530
531
532
533
534
535
    print "Waiting 5 seconds and trying again...\n";
    sleep(5);
    $currentrun++;
}
TBDebugTimeStamp("assign_loop finished");

#
# Run assign once.
# 
sub RunAssign ($)
{
    my ($retval) = @_;

    # Clear globals for each run.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
536
537
    undef %v2pmap;
    undef %p2vmap;
538
    undef %v2vmap;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
539
    undef %plinks;
540
    undef %virtnodes;
541
542
543

    my %toreserve = ();
    my %subnodes  = ();
Shashi Guruprasad's avatar
Shashi Guruprasad committed
544
    
545
    TBDebugTimeStamp("ptopgen started");
546
    # Snapshot physical resources.
Chad Barb's avatar
   
Chad Barb committed
547
548
549
550
    #
    # if updating (-u), include any resources that may already be
    # allocated to experiment in the PTOP results.
    #
551
552
    my $ptopargs = "-p $pid ";
    $ptopargs   .= "-e $eid "
553
554
555
	if ($updating);
    $ptopargs   .= "-m $multiplex_factor "
	if (defined($multiplex_factor));
556
557
    $ptopargs   .= "-v "
	if ($virtcount);
558
559
    $ptopargs   .= "-r "
	if ($remotecount);
560
561
    $ptopargs   .= "-S "
	if ($simcount);
562
    system("ptopgen $ptopargs > $ptopfile");
Chad Barb's avatar
   
Chad Barb committed
563

564
    TBDebugTimeStamp("ptopgen finished");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
565
566

    # Get number of nodes
567
    my $numnodes = TBAvailablePCs($pid);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
568
    
569
    if ($numnodes < ($minimum_nodes - $reserved_pcount)) {
570
571
	fatal($WRAPPER_FAILED|$WRAPPER_FAILED_CANRECOVER,
	      "Insufficient nodes available.");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
572
573
    }

574
    TBDebugTimeStamp("assign started");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
575
    # Run assign
Leigh B. Stoller's avatar
Leigh B. Stoller committed
576
    my $cmdargs = "$ptopfile $topfile";
577
    $cmdargs = "-uPod -c .75 $cmdargs"
578
	if ($virtcount);
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593

    my $cmd;

    # If doing an experiment with virtnodes, use the prepass wrapper for assign
    # Turned off for now, because it needs some work.
    #if ($virtcount) {
    #	$cmd = "assign_prepass";
    #	$cmdargs = "-m $multiplex_factor $cmdargs"
    #	    if ($multiplex_factor);
    #} else {
    #	$cmd = "assign";
    #}

    $cmd = "assign";
    print "$cmd $cmdargs\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
594

595
596
597
598
599
600
601
602
603
604
605
606
607
608
    #
    # Fork a child to run assign. Parent spins watching the cancel flag
    # and waiting for assign to finish.
    #
    if (my $childpid = fork()) {
	while (1) {
	    sleep(2);

	    if (waitpid($childpid, &WNOHANG) == $childpid) {
		$assignexitcode = $? >> 8;
		last;
	    }

	    # Check cancel flag.
609
	    TBGetCancelFlag($pid, $eid, \$canceled);
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
	    if ($canceled) {
		if ((my $pgrp = getpgrp($childpid)) > 0) {
		    kill('TERM', -$pgrp);
		    waitpid($childpid, 0);
	    
		    fatal($WRAPPER_FAILED|$WRAPPER_FAILED_CANRECOVER,
			  "Cancel flag set; aborting assign run!");
		}
		# Loop again to reap child above before exit.
	    }
	}
    }
    else {
	#
	# Change our session so the child can get a killpg without killing
	# the parent. 
	#
        POSIX::setsid();
628
	exec("nice $cmd $cmdargs > assign.log");
629
630
631
	die("*** $0:\n".
	    "    Could not start assign!\n");
    }
Chad Barb's avatar
Chad Barb committed
632

633
    # Check cancel flag before continuing. 
634
    TBGetCancelFlag($pid, $eid, \$canceled);
635
636
    fatal($WRAPPER_FAILED|$WRAPPER_FAILED_CANRECOVER, 
	  "Cancel flag set; aborting assign run!")
637
638
	if ($canceled);

639
640
641
    my $violations = 0;
    my $score = -1;

642
643
644
645
646

    # Saving up assign.log coz each swapin/modify is
    # different and it is nice to have every mapping
    # for debugging and archiving purposes
    system("/bin/cp assign.log assign-$pid-$eid-$$.log");
647
648
649
650
651
    open(ASSIGNFP, "assign.log") or
	fatal($WRAPPER_FAILED|$WRAPPER_FAILED_CANRECOVER,
	      "Could not open assign logfile!");

    if ($assignexitcode == 0) {
Chad Barb's avatar
Chad Barb committed
652
653
654
655
656
657
658
659
660
661
662
663
664
	# read output
	# Header
	printdb "Reading assign results.\n";
	while (<ASSIGNFP>) {
	    chop;
	    /No physical nodes of type (.+)$/ && do {
		$score=-2;
		print $_ . "\n";
	    };
	    /^With ([0-9]+) violations$/ && do {
		$violations = $1;
		last;
	    };
665
666
667
668
	    /^[ \t]+BEST SCORE: [ \t]+([0-9]+(\.[0-9]+)?)/ && do {
		$score=$1;
		print $_ . "\n";
	    };
Chad Barb's avatar
Chad Barb committed
669
670
671
	}
	if ($score == -2) {
	    # Type error
672
673
	    fatal($WRAPPER_FAILED|$WRAPPER_FAILED_CANRECOVER,
		  "Type error. Giving up!");
Chad Barb's avatar
Chad Barb committed
674
675
	}
	printdb "Found score $score, violations $violations.\n";
676
    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
677

678
679
680
681
682
683
684
685
686
    #
    # If we found violations above, or if the score is <0 then figure
    # out what happened, and return the result. I am not sure if this
    # stuff is set when the assign exitcode is 2 (no retry), but we end up
    # in here anyway. Should be harmless.
    #
    if ($violations || $score < 0 || $assignexitcode) {
	# Compute the exitcode and return it. Caller decides what to do.
	my $exitcode = $WRAPPER_FAILED|$WRAPPER_FAILED_CANRECOVER;
687

688
689
690
691
692
693
694
695
696
697
698
	# spit out up to nodes
	print "ASSIGN FAILED: \n";       
	while (<ASSIGNFP>) {
	    if (/link_users:\s*(\d+)$/) {
		$exitcode |= $WRAPPER_FAILED_LINKUSERS;
	    }
	    elsif (/bandwidth:\s*(\d+)$/) {
		$exitcode |= $WRAPPER_FAILED_BANDWIDTH;
	    }
	    elsif (/unassigned:\s*(\d+)$/) {
		$exitcode |= $WRAPPER_FAILED_UNASSIGNED;
699
	    }
700
701
702
703
704
705
706
707
	    elsif (/desires:\s*(\d+)$/) {
		$exitcode |= $WRAPPER_FAILED_DESIRES;
	    }
	    if (/^Nodes:/) {last;}
	    print "$_";
	}
	close(ASSIGNFP);
	$$retval = $exitcode;
Chad Barb's avatar
   
Chad Barb committed
708

709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
	# Signal retriable to caller, since assign thinks it is.
	return 1
	    if ($assignexitcode == 1);
	return -1;
    }

    #
    # Assign success; parse results.
    # 
    # read nodes section
    while (<ASSIGNFP> !~ /^Nodes:/) {}
    printdb "Nodes:\n";
    while (<ASSIGNFP>) {
	chop;
	/^End Nodes$/ && last;
	@info = split;
	my ($virtual,$physical) = @info[0,1];

	# We don't care about LAN nodes anymore.
	if (defined($lannodes{$virtual})) {
	    next;
	}

	if (physnodeallocated($physical)) {
	    #
	    # Mark node as being reused.
	    #
	    # Look at virtual node being mapped to node;
	    # if it wasn't in the previous map, mark node for reboot.
	    #
	    if (physnodereuse($physical) eq "reboot") {
		# No changes once it goes into reboot.
		;
	    }
	    elsif (virtnodeisvirt($virtual)) {
744
		#
745
746
747
748
749
750
751
752
753
754
755
756
757
758
		# A new virt virtual node on an existing physical node
		# does not force the physnode to be rebooted; we can
		# set up a new virtnode on it without a reboot. If its
		# an existing virtual on the same physnode, then mark
		# both as reused; no need to reboot either. If the 
		# virtnode has moved here from someplace else, no
		# reboot of the physnode either, but obviously the
		# vnode will be released and a new one allocated.  What
		# we cannot determine is if its just a renamed node
		# (which would require a reboot of the the virtual
		# node). 
		# 
		if (!exists($reserved_v2pmap{$virtual})) {
		    physnodesetreuse($physical, "reused");
Chad Barb's avatar
   
Chad Barb committed
759
		}
760
761
		elsif ($reserved_v2pmap{$virtual} eq $physical) {
		    my $reserved = $reserved_v2vmap{$virtual};
762

763
764
		    physnodesetreuse($reserved, "reused");
		    physnodesetreuse($physical, "reused");
765
766
		}
		else {
767
		    physnodesetreuse($physical, "reused");
768
769
770
		}
	    }
	    else {
Chad Barb's avatar
   
Chad Barb committed
771
		#
772
773
774
775
776
777
778
		# If a new virtual node mapped to this physnode (maybe
		# even the luser changed the name of the node), or if an
		# existing virtual node moved to this physnode, must
		# reboot the physnode. Else, the physnode is being
		# reused as is, and no need to mess with it. If the
		# user requested reboot, that will be handled outside
		# of this script.
779
		#
780
781
782
783
784
785
		if (!exists($reserved_v2pmap{$virtual}) ||
		    $reserved_v2pmap{$virtual} ne $physical) {
		    physnodesetreuse($physical, "reboot");
		}
		else {
		    physnodesetreuse($physical, "reused");
786
		}
787
	    }
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
	}
	else {
	    #
	    # This is a new node; we'll have to reserve it. Note that
	    # we do not reserve a widearea physnode when a virtual node
	    # is mapped to it; they are special.
	    #
	    $toreserve{$physical} = 1
		if (!virtnodeisremote($virtual));
	}
	
	if (virtnodeisvirt($virtual)) {
	    #
	    # If mapping a virtual node, then record that, since we need
	    # to allocate the virtnodes on that physnode, later.
	    #
	    if (!defined($virtnodes{$physical})) {
		$virtnodes{$physical} = [];
Shashi Guruprasad's avatar
Shashi Guruprasad committed
806
	    }
807
808
809
810
811
812
813
814
815
816
817
818
819
	    push(@{$virtnodes{$physical}}, $virtual);
	}
	elsif (virtnodeissubnode($virtual)) {
	    #
	    # Need to allocate the parent to. Should be optional?
	    # Save away and deal with once we have all the results.
	    #
	    $subnodes{$virtual} = $physical;
	}
	
	$v2pmap{$virtual} = $physical;
	if( ! defined($p2vmap{$physical}) ) {
	    $p2vmap{$physical} = [];
Leigh B. Stoller's avatar
Leigh B. Stoller committed
820
	}
821
822
823
	push(@{$p2vmap{$physical}}, $virtual);
	printdb "  $virtual $physical\n";
    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
824

825
826
827
828
829
830
831
832
833
    #
    # Process the subnodes. We have to allocate the parent at the same
    # time, lest it get sucked away for some other purpose by another
    # experiment. We might want to push this off into nalloc, but not
    # sure yet.
    #
    for my $virtual (keys(%subnodes)) {
	my $physical = $subnodes{$virtual};
	my $parent;
834

835
	TBPhysNodeID($physical, \$parent);
836

837
	printdb "  Subnode: $virtual $physical $parent\n";
838

839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
	#
	# See if we already have it. Swapmod, retry, or perhaps
	# the parent could be named separately? Or maybe there are
	# several subnodes on the physnode?
	#
	next
	    if (exists($p2vmap{$parent}));

	# Make up a name and add to the list.
	my $newvname = newvname($parent, "phost");

	$v2pmap{$newvname} = $parent;
	$p2vmap{$parent} = [ $newvname ];
	$toreserve{$parent} = 1;
	printdb "  Adding subnode host: $newvname $parent\n";
    }

    # read Edges
    # By convention, in plinks, the delay node is always the second
    # entry.
    while (<ASSIGNFP> !~ /^Edges:/) { }
    printdb "Edges:\n";
    EDGEWHILE: while (<ASSIGNFP>) {
	/^End Edges$/ && last EDGEWHILE;
	@info = split;
	$line = $_;
	$_ = $info[1]; # type
        SWITCH1: {
	    /^intraswitch$/ && do {
		($vlink,$rawA,$rawB) = @info[0,3,5];
		last SWITCH1;
	    };
	    /^interswitch$/ && do {
		($vlink,$rawA,$rawB) = @info[0,3,$#info];
		last SWITCH1;
	    };
	    /^direct$/ && do {
		($vlink,$rawA,$rawB) = @info[0,3,5];
		last SWITCH1;
	    };
	    /^trivial$/ && do {
		# we don't have plinks for trivial links
		$vlink = $info[0];
		$plinks{$vlink} = [];
		next EDGEWHILE;
	    };
	    print "Found garbage: $line\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
886
	}
887
888
889
890
891
892
	$nodeportA = &getnodeport($rawA);
	$nodeportB = &getnodeport($rawB);
	$nodeportA =~ s/\//:/;
	$nodeportB =~ s/\//:/;
	$plinks{$vlink} = [$nodeportA,$nodeportB];
	printdb "  $vlink " . join(" ",@{$plinks{$vlink}}) . "\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
893
894
    }
    close(ASSIGNFP);
895
    TBDebugTimeStamp("assign finished");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
896
897

    # Reserve resources
898
899
900
901
    if ($impotent) {
	print "Skipping physical reservation, as directed.\n";
	return 0;
    }
Chad Barb's avatar
Chad Barb committed
902

903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
    # From here, we can't recover anymore, coz we move
    # previously reserved pnodes/vnodes to the holding reservation
    # and back. By doing this, we will avoid any UNIQUE key issues
    # when a virt_node in the topology moves from one pnode to another
    # from previous to new mapping.
    # Another reason to do this just before nalloc of a new toreserve
    # nodes is that, we can get into name clashes
    # For example, lets say the user called his node pc2 and it was
    # actually mapped to pc99 in the initial swapin. If this was a
    # swapmod where the user asked for another node node0 which got 
    # mapped to pc2. nalloc of pc2 will result in UNIQUE key problems
    # since there exists a reserved vname pc2 (virtual name). By 
    # having this operation of moving the nodes into 
    # OLDRESERVED_PID/OLDRESERVED_EID and back before a new nalloc,
    # avoid this UNIQUE key problem. Also note that simply updating
    # the vname to be the same as the node_id field also won't
    # work all the time i.e. in the example discussed above
    my $oldreserved_pid = OLDRESERVED_PID;	
    my $oldreserved_eid = OLDRESERVED_EID;
    if (scalar(keys %oldreservednodes)) {
	# We can't recover after this coz we are making changes to
	# the DB
	$NoRecover = 1;
	TBDebugTimeStamp("Moving Old Reserved nodes to ".
	    		 "$oldreserved_pid/$oldreserved_eid ".
			 "and back started");
	system("nfree -o $pid $eid " . join(" ", keys(%oldreservednodes)) );
	system("nalloc $pid $eid " . join(" ", keys(%oldreservednodes)) );
	my $exitval  = $? >> 8;
	TBDebugTimeStamp("Moving Old Reserved nodes to ".
	    		 "$oldreserved_pid/$oldreserved_eid ".
			 "and back finished");
	#
	# If nalloc failed with a fatal error, lets give it up. No retry.
	# 
	if ($exitval != 0) {
	    fatal($WRAPPER_FAILED,
	    	  "Failed to move back Old Reserved nodes back to reserved\n");
	}
	# We need to move this back and forth the holding reservation only
	# once i.e. in the first call to RunAssign(). If it gets repeatedly
	# called coz only some pnode resources got nalloc'ed, we don't have
	# to do the above again.
	undef %oldreservednodes;
    }

949
950
951
952
    TBDebugTimeStamp("reserving started");
    system("nalloc -p $pid $eid " . join(" ", keys(%toreserve)));
    TBDebugTimeStamp("reserving finished");
    my $exitval  = $? >> 8;
Chad Barb's avatar
   
Chad Barb committed
953

954
955
956
957
958
959
960
961
    #
    # If nalloc failed with a fatal error, lets give it up. No retry.
    # 
    if ($exitval < 0) {
	print "Failed to reserve any nodes.\n";
	$$retval = $WRAPPER_FAILED|$WRAPPER_FAILED_CANRECOVER;
	return -1; 
    }
Chad Barb's avatar
   
Chad Barb committed
962

963
964
965
966
967
968
969
    #
    # Okay, if nalloc got anything, we have to set the norecover bit,
    # since tbswap operates on the principle that any change in the DB
    # means no recover is possible. This can probably me dealt with by
    # deallocating any nodes we allocated in the wrapper before exiting.
    #
    $NoRecover = 1;
Chad Barb's avatar
   
Chad Barb committed
970

971
972
973
974
975
976
977
978
979
980
981
982
983
984
    #
    # Otherwise, all newly allocated nodes MUST go to the INIT_DIRTY
    # allocstate since the user now has control of them. If we eventually
    # fail, nodes not in RES_READY are deallocated (retry/modify). 
    #
    if ($exitval > 0) {
	my @reserved = ExpNodes($pid, $eid);

	# We got only some. Need to figure out which.
	print "Reserved some nodes ($exitval) we needed, but not all.\n";
	
	foreach my $node (@reserved) {
	    if (exists($toreserve{$node})) {
		TBSetNodeAllocState($node, TBDB_ALLOCSTATE_RES_INIT_DIRTY());
Chad Barb's avatar
   
Chad Barb committed
985
	    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
986
987
	}

988
989
990
991
992
993
994
995
996
997
998
999
	#
	# We check to see if were were able to reserve all the fixed
	# nodes we needed. If we couldn't get the fixed list, then
	# this experiment is unlikely to map in the "near" future, so
	# give up now (no retry).
	#
	foreach my $node (values(%fixed_nodes)) {
	    if (! grep {$_ eq $node} @reserved) {
		printdb "  Could not allocate fixed node $node!\n";
		$$retval = $WRAPPER_FAILED;
		return -1;
	    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1000
	}
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
	#
	# Okay, must extend the fixed list with newly allocated nodes
	# so that we can recreate the top file, and try again with a
	# new set.
	#
	foreach my $node (@reserved) {
	    if (exists($toreserve{$node})) {
		foreach my $vname (@{$p2vmap{$node}}) {
		    $fixed_nodes{$vname} = $node;
		}
	    }
1012
	}
1013
1014
1015
1016
	CreateTopFile();
	$$retval = 0;
	return 1;
    }
Chad Barb's avatar
Chad Barb committed
1017

1018
1019
1020
1021
1022
1023
1024
1025
1026
    #
    # We got all the nodes we wanted. All newly allocated nodes MUST
    # go to the INIT_DIRTY allocstate since the user now has control
    # of them.
    #
    print "Successfully reserved all physical nodes we needed.\n";
	
    foreach my $node (keys(%toreserve)) {
	TBSetNodeAllocState($node, TBDB_ALLOCSTATE_RES_INIT_DIRTY());
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1027
    }
Chad Barb's avatar
Chad Barb committed
1028

1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
    #
    # Release phys and virt nodes no longer needed. They are marked
    # for teardown. They need to be freed by SOMEONE, currently the
    # wrapper (tbswap), since this only happens when in update mode
    # (swapmod).
    #
    foreach my $pnode (keys(%phys_nodes)) {
	my $reuse = physnodereuse($pnode);
		
	if ($reuse eq "unused") {
	    #
	    # Node was used in previous incarnation, but not any more.
	    #
	    TBSetNodeAllocState($pnode, TBDB_ALLOCSTATE_RES_TEARDOWN());
	}
	elsif ($reuse eq "reboot") {
	    #
	    # Node is being reused, but for a different purpose, so
	    # it should be rebooted.
	    #
	    TBSetNodeAllocState($pnode, TBDB_ALLOCSTATE_RES_INIT_DIRTY());
	}
    }
    return 0;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1053
}
1054

1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
###########################################################################
# Step 2A
#
# We run the wanassigner to allocate remote nodes. We do this after cause
# it takes so long. We run it just once.
#
# wanassign does its own nalloc.
#
###########################################################################

1065
#
1066
# VIRTNODES HACK: Allocate the remote virtual nodes.
1067
#
1068
if ($needwanassign) {
1069
    my $success  = 0;
1070
    my %wanmap   = ();
1071

1072
1073
    print "Running 'wanassign -d $pid $eid'\n";
    open(WANFP,"wanassign -d $pid $eid 2>&1 | tee wanassign.log |") or
1074
1075
	fatal($WRAPPER_FAILED|$WRAPPER_FAILED_CANRECOVER,
	      "Failed to start wanassign: $!");
1076
1077
1078
1079
1080

    printdb "Reading wanassign results.\n";
    while (<WANFP>) {
	chop;
	if ($_ =~ /(\S+) mapsto (\S+)/) {
1081
1082
	    $wanmap{$1} = $2;
	    printdb "  $1 mapsto $2\n";
1083
	}
1084
1085
1086
1087
1088
1089
	if ($_ =~ /^Success/) {
	    $success = 1;
	}
	# Skip other output. Usually its debugging output.
    }
    close(WANFP) or
1090
1091
1092
1093
	fatal($WRAPPER_FAILED|$WRAPPER_FAILED_CANRECOVER,
	      "wanassign: " .
	      ($? ? "exited with status: $?." :
	            "error closing pipe: $!"));
1094

1095
    if (!$success) {
1096
1097
	fatal($WRAPPER_FAILED|$WRAPPER_FAILED_CANRECOVER,
	      "wanassign could not find a solution!");
1098
    }
1099
1100
    foreach my $virtual (keys(%wanmap)) {
	my $physical = $wanmap{$virtual};
1101

1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
	fatal($WRAPPER_FAILED|$WRAPPER_FAILED_CANRECOVER,
	      "Improper mapping from wanassign: $virtual/$physical")
	    if (!virtnodeisvirt($virtual));
	
	#
	# If mapping a virtual node, then record that, since we need
	# to allocate the virtnodes on that physnode, later.
	#
	if (!defined($virtnodes{$physical})) {
	    $virtnodes{$physical} = [];
1112
	}
1113
1114
1115
1116
1117
	push(@{$virtnodes{$physical}}, $virtual);
	
	$v2pmap{$virtual} = $physical;
	if( ! defined($p2vmap{$physical}) ) {
	    $p2vmap{$physical} = [];
1118
	}
1119
	push(@{$p2vmap{$physical}}, $virtual);
1120
    }
1121
    TBDebugTimeStamp("wanassign finished");
1122
1123
}

Chad Barb's avatar
Chad Barb committed
1124
1125
#
# Recoverability ends.
1126
# All fatal() calls from this point do not have the recoverable '64' bit set.
Chad Barb's avatar
Chad Barb committed
1127
#
1128
$NoRecover = 1;
1129

1130
1131
1132
# VIRTNODES HACK: Local virtnodes have to be mapped now. This is a little
# hokey in that the virtnodes just need to be allocated from the pool that
# is on the real node. We know they are free, but we should go through
1133
# nalloc anyway. If anything fails, no point in retry.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1134
#
1135
1136
foreach my $pnode (keys(%virtnodes)) {
    my @vlist = @{$virtnodes{$pnode}};
1137
    my $numvs = @vlist;
1138
    my @plist = ();
1139
1140
    my @oplist = ();
    my @ovlist = ();
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1141

1142
1143
    printdb "On pnode $pnode: vnodes @vlist\n";

1144
    # Check cancel flag before continuing. 
1145
    TBGetCancelFlag($pid, $eid, \$canceled);
1146
1147
    fatal($WRAPPER_FAILED,
	  "Cancel flag set; aborting assign run!")
1148
1149
	if ($canceled);

Leigh B. Stoller's avatar
Leigh B. Stoller committed
1150
    #
1151
1152
1153
    # If updating, need to watch for nodes that are already reserved.
    # We save that info in oplist/ovlist, and build a new vlist for
    # avail, of just the nodes we need in this run. 
1154
    #
1155
    if ($updating) {
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
	my @oldvlist = ();
	if ( defined($reserved_p2vmap{$pnode}) ) {
	    @oldvlist = @{$reserved_p2vmap{$pnode}};
	} 
        printdb "On pnode $pnode: oldvlist: @oldvlist\n";

	# newvlist = elements in vlist but not in oldvlist
	#            i.e. newly mapped to this pnode
	my @newvlist = array_diff( \@vlist, \@oldvlist );
        printdb "On pnode $pnode: newvlist: @newvlist\n";

	# curvlist = elements in both vlist and oldvlist
	#            i.e. vnodes mapped the same way from previous
	#            to current
	# This is the same as @vlist intersection @oldvlist
	# since the lists have no duplicates
	my @curvlist = array_diff( \@vlist, \@newvlist );
        printdb "On pnode $pnode: curvlist: @curvlist\n";

	foreach my $vnode (@curvlist) {
	    if( defined($reserved_v2vmap{$vnode}) ) {
		push( @oplist, $reserved_v2vmap{$vnode} );
		push( @ovlist, $vnode );
1179
	    }
1180
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1181

1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
	# delvlist = elements in oldvlist not in vlist
	#            i.e. vnodes that moved to another pnode or
	#            went away
	my @delvlist = array_diff( \@oldvlist, \@vlist );
        printdb "On pnode $pnode: delvlist: @delvlist\n";

	while( scalar(@newvlist) && scalar(@delvlist) ) {
	    my $del_vnode = pop(@delvlist);
	    if ( defined($reserved_v2vmap{$del_vnode}) ) {
		my $reserved_node = $reserved_v2vmap{$del_vnode};
		push( @oplist, $reserved_node );
		my $new_vnode = pop(@newvlist);
		push( @ovlist, $new_vnode );
	    }
1196
	}
1197
	
1198
1199
1200
	# These are the new nodes we need to allocate
	@vlist = @newvlist;
	$numvs = scalar(@vlist);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1201

1202
1203
1204
1205
	if (@oplist) {
	    print "Reusing vnodes @oplist\n"; 
	}
    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1206

1207
1208
1209
1210
1211
1212
1213
1214
    #
    # Still need to allocate some virtnodes?
    #
    if ($numvs) {
	#
	# Run avail to get the list of virtnodes on the phys node. We
	# already know there are enough, since assign knows that.
	#
1215
	printdb "Asking avail for $numvs for vnodes: @vlist on $pnode\n";
1216
1217
    
	open(AVAIL,"$TBROOT/sbin/avail virtonly=$pnode rand limit=$numvs |")
1218
	    or fatal($WRAPPER_FAILED, "avail failed");
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229

	while (<AVAIL>) {
	    next
		if (! /^\|/);
	    next
		if (/node_id/);

	    if ($_ =~ /^\|([-a-zA-Z0-9]+)\s*\|(\w+)\s*\|(\w+)\s*\|$/) {
		push(@plist, $1);
	    }
	    else {
1230
1231
		fatal($WRAPPER_FAILED,
		      "Bad line from avail: $_");
1232
	    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1233
	}
1234
1235
1236
1237
	close(AVAIL);

	# Sanity check.
	if (scalar(@vlist) != scalar(@plist)) {
1238
	    printdb "avail gave " . scalar(@plist) . " vnodes: @plist on $pnode\n";
1239
1240
	    fatal($WRAPPER_FAILED,
		  "Could not map some virtual nodes on $pnode");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1241
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1242

1243
1244
1245
1246
	#
	# Try to allocate. Note, if this fails we are done for. Okay for now
	# since it is never the case that it should fail!
	#
1247
	if ($impotent) {
1248
	    print "Selected for $pnode: @plist\n";
1249
1250
1251
	    print "Skipping physical reservation, as directed.\n";
	}
	else {
1252
	    print "Reserving on $pnode: @plist ...\n";
1253
	    if (system("nalloc $pid $eid @plist")) {
1254
1255
		fatal($WRAPPER_FAILED,
		      "Failed to reserve @plist (on $pnode)");
1256
	    }
1257
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1258
1259
    }

1260
1261
1262
1263
1264
1265
1266
    if ($updating) {
	#
	# Append the lists we created above, so that we get all of them
	# in the loop below.
	#
	@plist = (@plist, @oplist);
	@vlist = (@vlist, @ovlist);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1267
    }
1268
    
1269
1270
1271
    while (@plist) {
	my $physical = pop(@plist);
	my $virtual  = pop(@vlist);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1272

1273
1274
	$v2vmap{$virtual}  = $physical;
	printdb "  Mapping $virtual to $physical on $pnode\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1275

1276
1277
1278
1279
1280
	#
	# New virtual nodes are always clean. Old ones stay in whatever
	# state they were in so that os_setup/vnode_setup know they
	# need to reboot them.
	#
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
	if (!$impotent) {
	    if (!defined($reserved_v2vmap{$virtual})) {
		TBSetNodeAllocState($physical,
				    TBDB_ALLOCSTATE_RES_INIT_CLEAN());
	    }
	    elsif ($reserved_v2vmap{$virtual} ne $physical) {
		# Node has moved! Nuts!
		TBSetNodeAllocState($physical,
				    TBDB_ALLOCSTATE_RES_INIT_DIRTY());
	    }
1291
	}
1292
    }
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303

    # Since we have some extra physical vnode reserved on this pnode,
    # we will let tbswap tear them down
    while (@delvlist) {
	my $del_vnode = pop(@delvlist);
	if ( defined($reserved_v2vmap{$del_vnode}) ) {
	    my $reserved_node = $reserved_v2vmap{$del_vnode};
	    TBSetNodeAllocState($reserved_node,
				TBDB_ALLOCSTATE_RES_TEARDOWN());
	}		
    }
1304
}
1305

1306
# Check cancel flag before continuing. 
1307
TBGetCancelFlag($pid, $eid, \$canceled);
1308
1309
fatal($WRAPPER_FAILED,
      "Cancel flag set; aborting assign run!")
1310
1311
    if ($canceled);

1312
# Set port range (see below for how we deal with update).
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1313
TBExptSetPortRange();
1314

1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
# Load the current physical resources. This avoids lots of repeated
# queries to the DB.
LoadPhysResources();

#
# For update, wipe old interfaces in DB (normally done by nfree.)
# These will get rebuilt soon. 
#
if ($updating && !$impotent) {
    foreach my $pnode (keys(%phys_nodes)) {
	#
	# Do not need to do this for phys nodes that are to be
	# released, or for virtnodes since they do not have interfaces
	# associated with them directly. This is probably a bad assumption
	# though, and perhaps this entire function should be moved to the
	# library. 
	#
	next
	    if (physnodeisvirtnode($pnode) ||
		physnodereuse($pnode) eq "unused");

1336
1337
	DBQueryFatal("update interfaces set IP='',IPaliases=NULL,mask=NULL,".
		     "       rtabid='0',vnode_id=NULL " .
1338
1339
		     "where node_id='$pnode' and ".
		     "  role='" . TBDB_IFACEROLE_EXPERIMENT() . "'");
1340
1341
1342
1343
1344
1345

	# Clean the veth_interfaces table for this node too.
	DBQueryFatal("delete from veth_interfaces where node_id='$pnode'");
    }
}

Leigh B. Stoller's avatar
Leigh B. Stoller committed
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
######################################################################
# Step 3 - Convert to vlans, delays, and portmap
# 
# Here we convert the plinks into vlans, delays, and portmap.  We
# convert them first into internal datastructure.  After Step 4
# when we do some port swapping we'll upload the modified versions
# of these structures into the database.
#
# delays is indexed by an internal ID and contains:
#  [pnode, int0, int1, vname, delay, bandwidth, lossrate]
# portmap is indexed by <virtual node>:<virtual port> and contains
#  the physical port.
#
# vlan ids
#  vlan ids are increasing integers in the case of node<->delay connections.
#  In the case of actual LANs either of real node or of delay nodes
#  they are indexed by virtual lan name.
# delay ids
#  delay ids are increasing integers.  We could have used a list of
# delays just as well.  Having it as an array may prove useful for
# future changes however.
######################################################################

$delayid = 0;

printdb "Interpreting results.\n";
1372
TBDebugTimeStamp("interpreting started");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1373
foreach $plink (keys(%plinks)) {
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
    # trivial links do not have physical links, so no delay nodes. But,
    # we *do* use trivial links for intranode links, and thus there could
    # be link delays (ie: two jailed nodes on a link/lan assigned to the
    # same phys node).
    my $trivial = 0;

    if (scalar(@{$plinks{$plink}})) {
	($nodeportA,$nodeportB) = @{$plinks{$plink}};
	($nodeA,$portA) = split(":", $nodeportA);
	($nodeB,$portB) = split(":", $nodeportB);
	printdb "plink $plink - $nodeportA $nodeportB\n";
    }
    else {
	$trivial = 1;
	printdb "plink $plink - trivial\n";
Shashi Guruprasad's avatar
Shashi Guruprasad committed
1389
    }
1390

1391
    if (($lan,$virtA,$virtC) =
1392
	   ($plink =~ m|^linksdelaysrc/(.+)/(.+),(.+)$|)) {
1393
1394
1395
1396
	# trivial links do not have physical links, so no delay nodes.
	if ($trivial) {
	    next;
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1397
1398
1399
	# Node has a single entry in lan.
	# Node is nodeportA
	# Delay node is nodeportB