assign_wrapper.in 78.5 KB
Newer Older
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1
2
#!/usr/bin/perl -w

Leigh B. Stoller's avatar
Leigh B. Stoller committed
3
4
#
# EMULAB-COPYRIGHT
Leigh B. Stoller's avatar
Leigh B. Stoller committed
5
# Copyright (c) 2000-2003 University of Utah and the Flux Group.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
6
7
8
# All rights reserved.
#

Leigh B. Stoller's avatar
Leigh B. Stoller committed
9
10
11
12
13
14
15
16
17
# 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.

# Syntax: assign_wrapper <pid> <eid>

Chad Barb's avatar
Chad Barb committed
18
19
20
21
22
23
24
25
26
27
28
29
30
31
#
# Return code:
#
# 0  - success
# 1+ - error (Add other values:)
# 4  - bandwidth violation
# 8  - linkusers violation
# 16 - desires violation
# 32 - unassigned
# 64 - Set to indicate 'recoverability'
#      (E.g., no db or physical state was modified
#       by the time the error occurred.)
#

Leigh B. Stoller's avatar
Leigh B. Stoller committed
32
# Caveats:
Chad Barb's avatar
Chad Barb committed
33
#  The support for direct and interswitch links has not been tested much.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
34
35
36

# Settings
# delaythresh is the maximum delay in ms above which a delay node is needed.
37
#  (Note that the DB represents delays as floating point numbers)
Leigh B. Stoller's avatar
Leigh B. Stoller committed
38
# maxrun is maximum number of times we run assign.
39
40
# delaywithswitch if 1 will use the switch to delay when possible.  Currently
#  this only works for 10mbit links.
41
$delaythresh = 2;
42
$maxrun = 20;
43
$delaywithswitch=0;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
44

45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
#
# Some handy constants. Speed in Mbits/sec and Kbits/sec units.
# The conversion routine is to make explicit that we operate with two
# different sets of units. One is the topology, which is in Kbps now.
# The second is the physical description, which has been changed to
# Kbps in the DB (and in ptopgen).
#
# 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.
#
my $S10Mbs  = 10;
my $S100Mbs = 100;
my $S10Kbs  = 10000;
my $S100Kbs = 100000;
sub BWConvert($) {
    #
    # Use this if physical units are in Mbs (used to be).
    #
    #my $bw = $_[0] / 1000;
    
    my $bw = $_[0];
    return $bw;
}

70
$DELAYCAPACITY = @DELAYCAPACITY@;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
71
$TBROOT = "@prefix@";
72
$ENV{'PATH'} = "/usr/bin:$TBROOT/libexec:$TBROOT/sbin:$TBROOT/bin";
73

74
75
use lib '@prefix@/lib';
use libdb;
76
use libtestbed;
77
require exitonwarn;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
78

79
80
81
82
83
#
# Turn off line buffering on output
#
$| = 1;

84
85
use Getopt::Std;

Chad Barb's avatar
   
Chad Barb committed
86
getopts('vu',\%opt);
87
88
89
90

sub usage {
	print "Usage: $0 [-v] pid eid\n";
	print "		-v enables verbose output\n";
Chad Barb's avatar
   
Chad Barb committed
91
	print "         -u enables update functionality\n";
92
	exit(-1);
93
}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
94

Chad Barb's avatar
Chad Barb committed
95
96
97
98
99
100
101
102
sub fatal($$) {
    my $exitcode = shift;
    my $message  = shift;
    $message =~ s/\n$//;
    print STDERR "$message\n";
    exit($exitcode);
}

103
my $verbose = 0;
Chad Barb's avatar
   
Chad Barb committed
104
105
my $updating = 0;

106
if ($opt{v}) {
Chad Barb's avatar
   
Chad Barb committed
107
108
109
110
111
    $verbose = 1;
}

if ($opt{u}) {
    $updating = 1;
112
113
114
}

if (@ARGV != 2) {
Chad Barb's avatar
   
Chad Barb committed
115
    usage();
Leigh B. Stoller's avatar
Leigh B. Stoller committed
116
}
117

Leigh B. Stoller's avatar
Leigh B. Stoller committed
118
119
120
121
122
123
124
125
126
127
($pid,$eid) = @ARGV;

$ptopfile = "$pid-$eid-$$.ptop";

sub printdb {
    if ($verbose) {
	print $_[0];
    }
};

128
129
TBDebugTimeStamp("assign_wrapper started");

Leigh B. Stoller's avatar
Leigh B. Stoller committed
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
######################################################################
# Step 1 - Setup virtual topology
#
# 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
#
# Datastructures:
#  nodes is indexed by virtual node and contains the node type.
161
162
#  isremotenode is indexed by virtual node and says whether the node is
#    is remote. We let wanassign deal with those nodes.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
163
164
165
166
#  nodelans is indexed by virtual node and contains a list of
#    <port>:<lan> that it is connected to.
#  ips is indexed by nodeport and contains the IP addresses.
#  lannodes is indexed by physical name is the set of lan nodes.
167
#  portbw is indexed by virtual nodeport and contains the bandwidth
168
169
170
171
#    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.
172
173
#  fixed_nodes is indexed by virtual node name and points to physical node
#   name.
Christopher Alfeld's avatar
Christopher Alfeld committed
174
#  vtypes is indexed by vtype name and is a list of {weight types}.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
175
176
177
178
179
180
181
182
183
#
# Delay node names:
#  delay nodes are named tbdelayXX N > 2
#   and tbsdelayXX for N == 2.
# 
# Lan node nameS:
#  lan nodes are named lan/<virtual lan>
######################################################################

184
185
186
printdb "Generating TOP file.\n";
TBDebugTimeStamp("top started");

Leigh B. Stoller's avatar
Leigh B. Stoller committed
187
188
189
190
191
192
193
194
195
196
# Shark Hack
# For each LAN we replace all the sharks in the LAN with a single
# shark shelf node.  After this goes through assign we pull them
# all back out.
#
# sharkshelves is indexed by virtual shelf name and is a list of
# the virtual nodes in it.
# sharkshelfid is used to generate ids for shark shelves.
$sharkshelfid = 0;

197
# lans is indexed by virtual lan and is a list of nodeport members.
198
199
200
# emulated is 1 if the link/lan is emulated (virtual). This is
# controlled by the a variable in the virt_lans table. It also gets
# set when one of the nodes in the lan is a virtnode (jail).
201
202
203
my %lans     = ();
my %emulated = ();

204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
# delayinfo stores the virt_lans info, and is indexed by virtual
# lan:node:port and is a list of delay, bandwidth, lossrate, rdelay,
# rbandwidth, rlossrate.  Where r* indicates switch->node and the
# others indicate node->switch.
my %delayinfo = ();

# queueinfo overlaps with delayinfo. It came later, and I added before
# I understood this stuff. It should be merged into delayinfo above.
# It holds the q_ stuff from virt_lans. 
my %queueinfo = ();

# delaylinks stores the actual link delay info, converted from the
# virt delay params above. It is indexed by link name and contains
# [delay,bw,loss] for each direction. 
my %delaylinks = ();

# 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 = ();

# 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. 
my $uselinkdelays   = 0;
my $forcelinkdelays = 0;

236
# And the per-lan controls, based on the above two variables, and
237
# per-lan variable in the virt_lans table. Indexed by lan name.
238
239
240
241
# The nobwshaping flag is used in conjunction with emulated links
# to turn off actual bw traffic shaping on an emulated link. This allows
# assign to match the specified bws, but not force them to be such
# with delay nodes (leaves it up to the user to moderate the bw). 
242
my %uselinkdelay    = ();
243
my %nobwshaping     = ();
244

245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
#
# 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,
		  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,
);

270
271
272
273
my $query_result =
    DBQueryFatal("SELECT uselinkdelays,forcelinkdelays from experiments ".
		 "where pid='$pid' and eid='$eid'");
($uselinkdelays,$forcelinkdelays) = $query_result->fetchrow_array();
Leigh B. Stoller's avatar
Leigh B. Stoller committed
274

275
276
277
# Figure out what kind of links we have. Indexed by bandwidth and is
# just a set.
my %okbandwidths = ();
Leigh B. Stoller's avatar
Leigh B. Stoller committed
278

279
280
$query_result =
    DBQueryFatal("SELECT type,max_speed from interface_types");
281

282
while (($type,$bandwidth) = $query_result->fetchrow_array()) {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
283
284
    $okbandwidths{$bandwidth} = 1;
}
285
286
287

# Load delay osids and default osids for types
my %delayosids   = ();
288
my %jailosids    = ();
289
290
291
292
293
294
my %defaultosids = ();

$query_result =
    DBQueryFatal("SELECT type,delay_osid,osid from node_types");

while (($type,$delayosid,$defosid) = $query_result->fetchrow_array()) {
295
296
    $delayosids{$type}   = $delayosid;
    $jailosids{$type}    = "emulab-ops-FBSD47-JAIL";
297
298
    $defaultosids{$type} = $defosid;
}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
299

Shashi Guruprasad's avatar
Shashi Guruprasad committed
300
301
302
303
# XXX NSE hack: List of simulated nodes. All these are gonna go
# into one pc850. Needs to change in distributed nse
my @simnodelist;
my %simnode_iplist = ();
304
305
my %iptonodemap    = ();
my $nsenode_id     = 0;
Shashi Guruprasad's avatar
Shashi Guruprasad committed
306

Leigh B. Stoller's avatar
Leigh B. Stoller committed
307
308
309
# XXX Remote/Virt node hacks. Turns out to be similar to NSE.
my %isremotenode = ();
my %isvirtnode   = ();
310
311
my $remotecount  = 0;
my $virtcount    = 0;
312
my $virtnode_id  = 0;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
313

Chad Barb's avatar
   
Chad Barb committed
314
315
my %alreadyAllocated = ();

Chad Barb's avatar
   
Chad Barb committed
316
317
318
319
320
321
322
323
if ($updating) {
    printdb "Fixing previously allocated nodes.\n";
	$result = 
	    DBQueryFatal("SELECT vname, node_id ".
			 "FROM reserved ".
			 "WHERE pid='$pid' AND eid='$eid'");
    
    while (($vname,$reserved) = $result->fetchrow_array) {
Chad Barb's avatar
   
Chad Barb committed
324
325
326
	$reserved_nodes{$vname} = $reserved;
	$fixed_nodes{$vname}    = $reserved;    
	$alreadyAllocated{$reserved} = "unused";
Chad Barb's avatar
   
Chad Barb committed
327
328
329
330
331
    }
    $result->finish;
}


Leigh B. Stoller's avatar
Leigh B. Stoller committed
332
printdb "Loading virt_nodes.\n";
333
$result =
Chad Barb's avatar
   
Chad Barb committed
334
    DBQueryFatal("select distinct vn.vname,vn.ips,vn.type,vn.fixed, ".
Chad Barb's avatar
   
Chad Barb committed
335
		 " nt.isremotenode,nt.isvirtnode ".
336
337
338
		 " from virt_nodes as vn ".
		 "left join node_types as nt on ".
		 " nt.type=vn.type or nt.class=vn.type ".
Chad Barb's avatar
   
Chad Barb committed
339
340
		 "where vn.pid='$pid' and vn.eid='$eid' ".
		 "order by vn.vname");
Chad Barb's avatar
   
Chad Barb committed
341

Chad Barb's avatar
   
Chad Barb committed
342
while (($vname,$ips,$type,$fixed,$isremote,$isvirt) = 
Chad Barb's avatar
   
Chad Barb committed
343
        $result->fetchrow_array){
Leigh B. Stoller's avatar
Leigh B. Stoller committed
344
345
346
347
    if (defined($fixed) && $fixed eq "") {
	undef($fixed);
    }
    
348
    # REMOTENODE HACK
349
    #
350
    # if its a vtype, no entry in node_types. vtypes break remote nodes.
351
352
353
    # Need to look inside the vtype and make sure no mixing of vnodes and
    # physnodes. Later ...
    #
354
    if (! defined($isremote)) {$isremote = 0;}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
355
356
357
358
    if (! defined($isvirt)) {$isvirt = 0;}
    
    printdb "  $vname $type $ips";
    printdb " " . (defined($fixed) ? $fixed : "") . " $isremote\n";
359
    
Leigh B. Stoller's avatar
Leigh B. Stoller committed
360
361
362
363
364
365
366
367
368
369
    # We need to check the names to make sure they won't clash with
    # our internal delay node names.
    if (($vname =~ /^tbdelay\d+/) ||
	($vname =~ /^tbsdelay\d+/)) {
	print STDERR "Warning: $vname is a reserved name.  Working around.\n";
	($num) = ($vname =~ /(\d+)/);
	$delayid = $num + 1;
    }
    $nodes{$vname} = $type;
    $nodelans{$vname} = [];
370

Shashi Guruprasad's avatar
Shashi Guruprasad committed
371
372
373
374
    if( $type eq "sim" ) {
      push( @simnodelist, $vname );
      $simnode_iplist{$vname} = [];
    }    
375
    # REMOTENODE HACK
376
    # 
377
    $isremotenode{$vname} = $isremote;
378
379
    $remotecount++
	if ($isremote);
380
381
382
383
384
385
386
387
388
389
390

    # stats
    my @iplist  = split(" ", $ips);
    my $ipcount = scalar(@iplist);

    $expt_stats{"maxlinks"} = $ipcount
	if ($ipcount > $expt_stats{"maxlinks"});
    $expt_stats{"minlinks"} = $ipcount
	if ($ipcount < $expt_stats{"minlinks"});

    foreach $ipinfo (@iplist) {
391
392
	($port,$ip) = split(":",$ipinfo);
	$ips{"$vname:$port"} = $ip;
Shashi Guruprasad's avatar
Shashi Guruprasad committed
393
394
395
396
	if( $type eq "sim" ) {
	  push(@{$simnode_iplist{$vname}}, $ip);
	}
	$iptonodemap{$ip} = $vname;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
397
    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
398
    $isvirtnode{$vname} = $isvirt;
399
400
    $virtcount++
	if ($isvirt);
Chad Barb's avatar
   
Chad Barb committed
401

Leigh B. Stoller's avatar
Leigh B. Stoller committed
402
    if (defined($fixed)) {
403
404
	$fixed_nodes{$vname} = $fixed;
    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
405
}
406
$result->finish;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
407

408
409
410
411
412
# Stats
$expt_stats{"vnodes"}   = $virtcount;
$expt_stats{"wanodes"}  = $remotecount;
$expt_stats{"simnodes"} = scalar(@simnodelist);

Leigh B. Stoller's avatar
Leigh B. Stoller committed
413
printdb "Loading virt_lans.\n";
414
415
$result =
    DBQueryFatal("select vname,member,delay,bandwidth,lossrate," .
416
		 "rdelay,rbandwidth,rlossrate,widearea, ".
417
		 "emulated,uselinkdelay,nobwshaping " .
Chad Barb's avatar
   
Chad Barb committed
418
419
		 "from virt_lans where pid='$pid' and eid='$eid' ".
                 "order by vname,member");
420
421

#
422
# REMOTENODE HACK: Remote nodes are special.
423
#
424
# A list of all lans that have remote nodes as members,
425
my %rnodelans = ();
426
427
428
# A list of all the tunnels we need to build. Each list member is a list
# of the nodes in the tunnel.
my %tunnels   = ();
429
430

#
431
# Process the virt_lans table.
432
# 
433
434
while (my ($vname,$member,$delay,$bandwidth,$lossrate,
	   $rdelay,$rbandwidth,$rlossrate,$widearea,
435
	   $isemulated,$uselinkdelay,$nobwshaping) = $result->fetchrow_array) {
436
    my ($node,$port) = split(":",$member);
437
438

    #
439
    # REMOTENODE HACK: 
440
    #
441
    # If its a duplex link involving a remotenode, skip it. 
442
    # We do not want to have assign deal with these. The nodes are
443
444
    # allocated as unconnected by another program, and we deal with
    # it later by creating tunnels.
445
    # 
446
    if ($widearea) {
447
448
449
450
	if (! defined($tunnels{$vname})) {
	    $tunnels{$vname} = [];
	}
	push(@{$tunnels{$vname}},$member);
451
	$rnodelans{$vname} = 1;
452
	$expt_stats{"walinks"} += 1;
453
454
	printdb "    Added $member to tunnels of $vname\n";
	next;
455
456
    }
    
Leigh B. Stoller's avatar
Leigh B. Stoller committed
457
458
459
460
    if (! defined($lans{$vname})) {
	$lans{$vname} = [];
    }
    push(@{$lans{$vname}},$member);
Christopher Alfeld's avatar
Christopher Alfeld committed
461
462
    $delayinfo{"$vname:$member"} = [$delay,$bandwidth,$lossrate,
				    $rdelay,$rbandwidth,$rlossrate];
463
464
465
466
467
468
469
470
471

    #
    # Grab the Q stuff from virt_lans. I'm keeping this separate for
    # now until I understand whats going on. There are no "r" params
    # either. I have no idea how do to this stuff for lans, and for
    # duplex links the "r" params are not necessary. Each virt_lans
    # entry gives the params towards the switch, which equal the
    # reverse params for the other member. 
    #
472
    $query_result =
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
	DBQueryFatal("select q_limit,q_maxthresh,q_minthresh,q_weight, ".
		     "       q_linterm,q_qinbytes,q_bytes,q_meanpsize, ".
		     "       q_wait,q_setbit,q_droptail,q_red,q_gentle ".
		     "from virt_lans ".
		     "where pid='$pid' and eid='$eid' and ".
		     "      vname='$vname' and member='$member'");
    
    my ($q_limit,$q_maxthresh,$q_minthresh,$q_weight,$q_linterm,
	$q_qinbytes,$q_bytes,$q_meanpsize,$q_wait,$q_setbit,
	$q_droptail,$q_red,$q_gentle) = $query_result->fetchrow_array;
    
    $queueinfo{"$vname:$member"} =
	[$q_limit,$q_maxthresh,$q_minthresh,$q_weight,$q_linterm,
	 $q_qinbytes,$q_bytes,$q_meanpsize,$q_wait,$q_setbit,
	 $q_droptail,$q_red,$q_gentle];

    if ($q_red) {
	$mustdelay{$vname} = 1;
    }
492
493
    $emulated{$vname} = $isemulated;
    $uselinkdelay{$vname} = $uselinkdelay;
494
    $nobwshaping{$vname} = $nobwshaping;
495
    
496
497
498
499
500
501
502
503
504
505
506
507
    #
    # XXX - Whenever a delay node is inserted, port speeds are set to
    #       100Mbs, even if they requested exactly 10Mbs. This is a
    #       simplification. At some point we might want to force all the
    #       ports along the way to 10Mbs, and have the delay node worry
    #       about delay only, and not bandwidth. That will be harder to
    #       to do in this mess. See companion XXX below where the delays
    #       table is initialized. Initially, we set the speed to 10Mbs,
    #       if a delay node is insterted below, it resets this to 100Mbs.
    # 
    if ($bandwidth == $S10Kbs && $delaywithswitch) {
	$portbw{$member} = $S10Mbs;
508
    } else {
509
	$portbw{$member} = $S100Mbs;
510
    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
511
    push(@{$nodelans{$node}},"$port:$vname");
512
    printdb "  portbw of $member = $portbw{$member}\n";
Christopher Alfeld's avatar
Christopher Alfeld committed
513
    printdb "  $vname $member - $delay $bandwidth $lossrate $rdelay $rbandwidth $rlossrate\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
514
515
    printdb "    Added $port:$vname to nodelans of $node\n";
}
516
$result->finish;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
517

518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
#
# Check event list. Anytime we find an event to control a link, we need
# to drop a delay node in. start/stop especially, since thats the easiest
# way to do that, even if the link has no other traffic shaping in it. 
# 
printdb "Checking events for LINK commands.\n";
$result =
    DBQueryFatal("select distinct vname from eventlist as ex ".
		 "left join event_eventtypes as et on ex.eventtype=et.idx ".
		 "left join event_objecttypes as ot on ex.objecttype=ot.idx ".
		 "where ot.type='LINK' and ex.pid='$pid' and ex.eid='$eid'");
while (($vname) = $result->fetchrow_array) {
    $mustdelay{$vname} = 1;
}

Leigh B. Stoller's avatar
Leigh B. Stoller committed
533
534
535
536
537
538
539
# Shark hack
foreach $lan (keys(%lans)) {
    $realmembers = [];
    $sharks = [];
    $hassharks = 0;
    foreach $member (@{$lans{$lan}}) {
	($node) = (split(":",$member))[0];
540
	if (($nodes{$node} eq "shark") || ($nodes{$node} eq "dnard")) {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
541
542
543
544
545
546
547
548
549
550
551
552
	    push(@$sharks,$member);
	    $hassharks = 1;
	} else {
	    push(@$realmembers,$member);
	}
    }
    if ($hassharks) {
	$shelfid = "sharkshelf$sharkshelfid";
	printdb "  Creating shark shelf: $shelfid (" . 
	    join(" ",@$sharks) . ")\n";
	$sharkshelfid++;
	$sharkshelves{$shelfid} = $sharks;
553
	$delayinfo{"$lan:$shelfid:uplink"} = [0,$S100Kbs,0.0,
554
					      0,$S100Kbs,0.0];
Leigh B. Stoller's avatar
Leigh B. Stoller committed
555
556
557
558
559
560
561
	push(@$realmembers,"$shelfid:uplink");
	$nodes{$shelfid} = "shark-shelf";
    }
    $lans{$lan} = $realmembers;
}
# End shark hack

Christopher Alfeld's avatar
Christopher Alfeld committed
562
563
564
565
566
567
568
569
570
# Load virt types
printdb "Loading virt_vtypes.\n";
$result = DBQueryFatal("SELECT name,weight,members from virt_vtypes" .
		       " where pid=\"$pid\" and eid=\"$eid\"");
while (($name,$weight,$types) = $result->fetchrow_array) {
    printdb "  $name $weight $types\n";
    $vtypes{$name} = "$weight $types";
}

Leigh B. Stoller's avatar
Leigh B. Stoller committed
571
# Open the TOP file
572
$topfile = "$eid.top";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
573
open(TOPFILE,"> $topfile") || do {
Chad Barb's avatar
Chad Barb committed
574
575
    fatal(65,"*** $0:\n".
	    "    Could not open $topfile.\n");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
576
577
};

Christopher Alfeld's avatar
Christopher Alfeld committed
578
579
580
581
# Print out vtypes
foreach $vtype (keys(%vtypes)) {
    print TOPFILE "make-vclass $vtype $vtypes{$vtype}\n";
}
582
583
584

$nodes=0;

Leigh B. Stoller's avatar
Leigh B. Stoller committed
585
586
foreach $node (keys(%nodes)) {
    # Shark hack
587
    if (($nodes{$node} ne "shark") &&
588
	($nodes{$node} ne "dnard") && !$isremotenode{$node}) {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
589
	print TOPFILE "node $node $nodes{$node}\n";
590
591
592
	if ($nodes{$node} ne "shark-shelf") {
	    $nodes++;
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
593
594
595
596
    }
    # End Shark hack
}

Shashi Guruprasad's avatar
Shashi Guruprasad committed
597
598
599
600
#
# lans that have simulated nodes
my %simnodelans = ();

Leigh B. Stoller's avatar
Leigh B. Stoller committed
601
602
foreach $lan (keys(%lans)) {
    @members = @{$lans{$lan}};
603
    printdb "$lan - " . join(" ",@members) . "\n";
604
605
606
    # Shark hack for rvr
    $sharks = 0;
    $nonsharks = 0;
Shashi Guruprasad's avatar
Shashi Guruprasad committed
607
608
    $simnodes = 0;
    $realnodes = 0;
609
610
611
    my $emulated = $emulated{$lan};
    my $uselinkdelay = $uselinkdelay{$lan};
    my $mustdelay = $mustdelay{$lan};
612
    my $isdelayed = 0;
613
    my $nobwshaping = $nobwshaping{$lan};
614
    $trivial_ok = 0;
615
616
617
618
    foreach $member (@members) {
	($node) = (split(":",$member))[0];
	if ($nodes{$node} eq "shark-shelf") {
	    $sharks++;
619
	} elsif ($nodes{$node} eq "sim") {
Shashi Guruprasad's avatar
Shashi Guruprasad committed
620
621
	    $simnodes++;
	    $simnodelans{$lan} = 1;
622
	} else {
623
	    # We always use an emulated link when its a virtnode.
624
625
	    if ($isvirtnode{$node}) {
		$emulated++;
626
		$emulated{$lan} = 1;
627
	    }
628
	    $nonsharks++;
Shashi Guruprasad's avatar
Shashi Guruprasad committed
629
	    $realnodes++;
630
631
	}
    }
632
633
634
635
    #
    # Hmm, no emulated lans for now. 
    #
    if ($emulated && @members > 2) {
Chad Barb's avatar
Chad Barb committed
636
637
	fatal(65,"*** $0:\n".
	        "    Emulated lans ($lan) not allowed yet! Only links.\n");
638
    }
639
640
641
    if ($simnodes > 0 && $realnodes == 0 && $sharks == 0) {
	$trivial_ok = 1;
    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
642
    if ($#members == 1) {
643
	$expt_stats{"links"} += 1;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
644
645
646
	($nodeport0,$nodeport1) = @members;
	$node0 = (split(":",$nodeport0))[0];
	$node1 = (split(":",$nodeport1))[0];
Christopher Alfeld's avatar
Christopher Alfeld committed
647
648
649
650
651
652
653
654
655
656
657
658
659
	($delay0,$bw0,$loss0,
	 $rdelay0,$rbw0,$rloss0) = @{$delayinfo{"$lan:$nodeport0"}};
	($delay1,$bw1,$loss1,
	 $rdelay1,$rbw1,$rloss1) = @{$delayinfo{"$lan:$nodeport1"}};
	# Here the r's aregoing to be 1->0 and the others 0->1
	$delay = $delay0+$rdelay1;
	$loss = 1-(1-$loss0)*(1-$rloss1);
	$bw = &min($bw0,$rbw1);
	$rdelay = $rdelay0+$delay1;
	$rloss = 1-(1-$rloss0)*(1-$loss1);
	$rbw = &min($rbw0,$bw1);
	$bandwidth = &getbandwidth(&min($bw0,$rbw1));
	$rbandwidth = &getbandwidth(&min($rbw0,$bw1));
Shashi Guruprasad's avatar
Shashi Guruprasad committed
660
	if (((($delay >= $delaythresh) ||
661
662
663
664
665
	      (!$nobwshaping &&
	       ((($bw != $S100Kbs) && ($bw != $S10Kbs)) ||
		(($delaywithswitch == 0) && 
		 (($bw != $S100Kbs) &&
		  (($sharks == 0) || ($nonsharks > 1)))))) || 
666
667
	      ($loss != 0)) ||
	     (($rdelay >= $delaythresh) ||
668
669
670
671
672
	      (!$nobwshaping &&
	       ((($rbw != $S100Kbs) && ($rbw != $S10Kbs)) ||
		(($delaywithswitch == 0) && !$nobwshaping &&
		 (($rbw != $S100Kbs) &&
		  (($sharks == 0) || ($nonsharks > 1)))))) || 
673
674
675
676
677
	      ($rloss != 0)) ||
	     # Link must be shaped for other reasons (q_red).
	     $mustdelay ||
	     # Global force, or per-link force. 
	     $forcelinkdelays || $uselinkdelay) &&
678
679
680
681
	    # XXX simulated nodes hack. We don't want to put delay nodes
	    # between simulated nodes. If there is a link between a
	    # simulated and a real node, we might need to put in delay
	    # nodes
682
	    ($realnodes != 0)) {
683
	    $isdelayed = 1;
684
685
686
687
688
689
690
691
	    #
	    # We use a linkdelay if the link is emulated, globally forced,
	    # globally preferred if the link is shaped, or if the per-link
	    # flag was set (which means to put in a link delay regardless
	    # of whether the link is shaped).
	    #
	    if ($emulated ||
		$forcelinkdelays || $uselinkdelays || $uselinkdelay) {
692
693
694
695
696
697
		my $plink = "linksimple/$lan/$nodeport0,$nodeport1";
		
		print(TOPFILE "link $plink $node0 $node1 ".
		      max($bw,$rbw) . " 0 0" .
		      ($emulated ? " emulated\n" : "\n"));

698
699
700
701
702
703
704
705
706
707
708
709
710
		#
		# We allow users to oversubscribe by letting them turn
		# off the bandwidth shaping. If however, the link was
		# shaped for some other reason (like a delay), then
		# turn off just the bw shaping part by setting them to 0.
		# This is special; means no limits in ipfw.
		#
		if ($nobwshaping) {
		    $bw = $rbw = 0;
		}
		$delaylinks{$plink} = [$nodeport0,$delay,$bw,$loss,
				       $nodeport1,$rdelay,$rbw,$rloss];
	    
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
		printdb "Delay link $plink = " . 
		    join(" ",@{$delaylinks{$plink}}) . "\n";
	    }
	    else {
		my $delayname = "tbsdelay" . $delayid++;
		my $plink     = "linksdelaysrc/$lan/$nodeport0,$nodeport1";
		
		$delaylinks{$plink} = [$nodeport0,$delay,$bw,$loss,
				       $nodeport1,$rdelay,$rbw,$rloss];

		print TOPFILE "node $delayname delay\n";
		print TOPFILE "link linksdelaysrc/$lan/$nodeport0,$nodeport1 ".
		    "$node0 $delayname $bandwidth 0 0\n";
		print TOPFILE "link linksdelaydst/$lan/$nodeport1,$nodeport0 ".
		    "$node1 $delayname $bandwidth 0 0\n";

		$delaynodes{$delayname} = $delayname;
		
		printdb "Delay node $plink ($delayname) = " . 
		    join(" ",@{$delaylinks{$plink}}) . "\n";
	    }
732
733
734
735
736
737
738
739
	    #
	    # Ports are set to 100Mbs when a link gets a delay node.
	    # This can override initialization above cause we could not
	    # tell earlier if the link was going to get a real delay node
	    # or just a delaywithswitch.
	    #
	    $portbw{$nodeport0} = $S100Mbs;	    
	    $portbw{$nodeport1} = $S100Mbs;	    
Leigh B. Stoller's avatar
Leigh B. Stoller committed
740
	} else {
741
	    print TOPFILE "link linksimple/$lan/$nodeport0,$nodeport1 ".
742
		"$node0 $node1";
743
	    if ($emulated) {
744
		print TOPFILE " " . max($bw,$rbw) . " 0 0 emulated";
745
746
747
	    }
	    else {
		print TOPFILE " $bandwidth 0 0";
748
	    }
749
750
751
	    if ($trivial_ok) {
		print TOPFILE " trivial_ok";
	    }
752
	    print TOPFILE "\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
753
	}
754
	$expt_stats{"shapedlinks"} += $isdelayed;
755
    } elsif ($#members != 0) {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
756
	print TOPFILE "node lan/$lan lan\n";
757
	$expt_stats{"lans"} += 1;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
758
759
	$lannodes{"lan/$lan"} = 1;
	foreach $member (@members) {
Christopher Alfeld's avatar
Christopher Alfeld committed
760
761
	    ($delay,$bw,$loss,
	     $rdelay,$rbw,$rloss) = @{$delayinfo{"$lan:$member"}};
Leigh B. Stoller's avatar
Leigh B. Stoller committed
762
	    $bandwidth = &getbandwidth($bw);
Christopher Alfeld's avatar
Christopher Alfeld committed
763
	    $rbandwidth = &getbandwidth($rbw);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
764
	    ($node) = (split(":",$member))[0];
765
766
# XXX The expression below should be modified for better bandwidth support.
# Probably needs to happen post assign somehow.
Shashi Guruprasad's avatar
Shashi Guruprasad committed
767
	    if (((($delay >= $delaythresh) ||
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
		  (($bw != $S100Kbs) && ($bw != $S10Kbs))  ||
		  (($delaywithswitch == 0) && 
		   (($bw != $S100Kbs) && (($sharks == 0) ||
					  ($nonsharks > 1)))) ||
		  ($loss != 0)) || 
		 (($rdelay >= $delaythresh) ||
		  (($rbw != $S100Kbs) && ($rbw != $S10Kbs))  ||
		  (($delaywithswitch == 0) && 
		   (($rbw != $S100Kbs) && (($sharks == 0) ||
					   ($nonsharks > 1)))) ||
		  ($rloss != 0)) ||
		 # Link must be shaped for other reasons (q_red).
		 $mustdelay ||
		 # Global force, or per-lan force. 
		 $forcelinkdelays || $uselinkdelay) &&
783
784
		# if we have 1 real node in the LAN, we may need to create
		# a lan
785
		($realnodes != 0)) {
786
		$isdelayed = 1;
787
788
789
790
791
792
793
794
795
796
		#
		# We use a linkdelay if the link is emulated, globally forced,
		# globally preferred if the link is shaped, or if the per-link
		# flag was set (which means to put in a link delay regardless
		# of whether the link is shaped).
		#
		# NB: Emulated lans not supported at this time. Caught above.
		#
		if ($emulated ||
		    $forcelinkdelays || $uselinkdelays || $uselinkdelay) {
797
798
799
800
801
802
803
		    my $plink = "linklan/$lan/$member";
		    
		    $delaylinks{$plink} =
			[$member,$delay,$bw,$loss,
			 $member,$rdelay,$rbw,$rloss];

		    print(TOPFILE "link $plink $node lan/$lan " .
804
			  max($bw,$rbw) . " 0 0" .
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
			  ($emulated ? " emulated\n" : "\n"));

		    printdb "Delay link $plink = " .
			    join(" ", @{$delaylinks{$plink}}) . "\n";
		}
		else {
		    my $delayname = "tbdelay" . $delayid++;
		    my $plink     = "linkdelaysrc/$lan/$member";

		    $delaylinks{$plink} =
			[$member,$delay,$bw,$loss,
			 $member,$rdelay,$rbw,$rloss];
		    
		    print TOPFILE "node $delayname delay\n";
		    print TOPFILE "link linkdelaysrc/$lan/$member" .
			" $node $delayname $bandwidth 0 0\n";
		    print TOPFILE "link linkdelaydst/$lan/$member" .
			" lan/$lan $delayname $bandwidth 0 0\n";
		    
		    $delaynodes{$delayname} = $delayname;

		    printdb "Delay node $plink ($delayname) = " .
			    join(" ", @{$delaylinks{$plink}}) . "\n";
		}
829
830
831
832
833
834
835
		#
		# Port is set to 100Mbs when the link gets a delay node.
		# This can override initialization above cause we could not
		# tell earlier if the link was going to get a real delay node
		# or just a delaywithswitch.
		#
		$portbw{$member} = $S100Mbs;	    
Leigh B. Stoller's avatar
Leigh B. Stoller committed
836
837
	    } else {
		print TOPFILE "link linklan/$lan/$member $node lan/$lan" .
838
839
840
841
		    " $bandwidth 0 0";
		if ($emulated) {
		    print TOPFILE " emulated";
		}
842
843
844
		if ($trivial_ok) {
		    print TOPFILE " trivial_ok";
		}
845
		print TOPFILE "\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
846
847
	    }
	}
848
	$expt_stats{"shapedlans"} += $isdelayed;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
849
    }
850
    # If a LAN has only one member we don't do anything.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
851
852
}

Shashi Guruprasad's avatar
Shashi Guruprasad committed
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
# XXX NSE hack
# Will find a free randomly chosen pc850 and fix sim nodes to it
# Assign's scoring needs to be fixed so that the solution has
# as many sim nodes on a phy node as possible while maxing out
# its interfaces. Currently having sim:N where N is a large
# number in the ptop file does not guarantee that all these
# get mapped to a single phy node even when that mapping has no
# violations and a low score. distributed nse will fix this
#$result =
#  DBQueryFatal("select a.node_id,a.type from nodes as a ".
#	   "left join reserved as b on a.node_id=b.node_id ".
#	   "left join reserved as m on a.phys_nodeid=m.node_id ".
#	   "where b.node_id is null and (a.role='testnode' and ".
#	   "      a.type='pc850' and ".
#	   "       (m.node_id is null or ".
#	   "        m.pid!='$DEADPID' or m.eid!='$DEADEID'))");
if( scalar(@simnodelist) > 0 ) {
  open(AVAIL,"$TBROOT/sbin/avail type=pc rand |")
Chad Barb's avatar
Chad Barb committed
871
872
    or fatal(65, "*** $0:\n".
                "    avail failed\n");
Shashi Guruprasad's avatar
Shashi Guruprasad committed
873
874
875
876
877
878
879
880
881
882
883
884

  my $num = 0;
  while (<AVAIL>) {
    if (! /^\|/) {next};
    if (/node_id/) {next;}
    ($fixednode,$type) = /^\|([-a-zA-Z0-9]+)\s*\|(\w+)\s*\|(\w+)\s*\|$/;    
    $num++;
    last;
  }
  close(AVAIL);
  
  if( $num == 0 ) {
Chad Barb's avatar
Chad Barb committed
885
    fatal(65, "$0: *** Insufficient PCs available.\n");
Shashi Guruprasad's avatar
Shashi Guruprasad committed
886
887
888
889
890
891
892
  }

  foreach $simnode (@simnodelist) {
    print TOPFILE "fix-node $simnode $fixednode\n";
  }
}

893
# Print out fixed nodes
Chad Barb's avatar
   
Chad Barb committed
894
895
896
897
# But _not_ nodes which aren't in the experiment any more.
# CRB you can only fix nodes and delaynodes, right?
$reused_count = 0;

898
foreach $fixed (keys(%fixed_nodes)) {
Chad Barb's avatar
   
Chad Barb committed
899
900
    if (!$isremotenode{$fixed} && 
	(exists $nodes{$fixed} || exists $delaynodes{$fixed}) ) {
901
	print TOPFILE "fix-node $fixed $fixed_nodes{$fixed}\n";
Chad Barb's avatar
   
Chad Barb committed
902
	if ($reserved_nodes{$fixed}) { $reused_count++; }
903
    }
904
905
}

Leigh B. Stoller's avatar
Leigh B. Stoller committed
906
907
close TOPFILE;

908
# Set estimations
Chad Barb's avatar
   
Chad Barb committed
909
910
$minimum_nodes = $nodes + keys(%delaynodes)/$DELAYCAPACITY - $reused_count;
$maximum_nodes = $nodes + keys(%delaynodes) - $reused_count;
911
DBQueryFatal("UPDATE experiments set maximum_nodes=$maximum_nodes, " .
912
	 "minimum_nodes=$minimum_nodes where pid=\"$pid\" and eid=\"$eid\"");
913
print "Minimum nodes = $minimum_nodes\n";
914
915
print "Maximum nodes = $maximum_nodes\n";

916
917
TBDebugTimeStamp("top finished");

Leigh B. Stoller's avatar
Leigh B. Stoller committed
918
919
920
921
922
923
924
925
926
######################################################################
# 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.
#
# v2pmap is indexed by virtual and contains the physical node.
Shashi Guruprasad's avatar
Shashi Guruprasad committed
927
# p2vmap is indexed by physical and contains one or more virtual nodes
Leigh B. Stoller's avatar
Leigh B. Stoller committed
928
929
930
931
932
# plinks is indexed by virtual name and contains 
#  (pnodeportA,pnodeportB) .  If one is a delay node it is always
#  the second.
#######################################################################

933
TBDebugTimeStamp("assign_loop started");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
934
935
936
937
938
$currentrun = 1;
while (1) {
    print "Assign Run $currentrun\n";

    # Violation counts
939
    $unassigned = -1;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
940
941
942
943
944
945
946
947
    $linkusers = -1;
    $bandwidth = -1;
    $desires = -1;

    # Clear v2pmap, p2vmap, and plinks
    undef %v2pmap;
    undef %p2vmap;
    undef %plinks;
948
949
    undef %toreserve;
    undef %virtnodes;
Shashi Guruprasad's avatar
Shashi Guruprasad committed
950
    
951
    TBDebugTimeStamp("ptopgen started");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
952
    # Snapshot
Chad Barb's avatar
   
Chad Barb committed
953
954
955
956
957
958
959
960
961
962
    #
    # if updating (-u), include any resources that may already be
    # allocated to experiment in the PTOP results.
    #
    if ($updating) {
	system("ptopgen -e $pid/$eid > $ptopfile");
    } else {
	system("ptopgen > $ptopfile");
    }

963
    TBDebugTimeStamp("ptopgen finished");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
964
965

    # Get number of nodes
966
967
    my $numnodes_result = 
	DBQueryFatal("select a.node_id,a.type from" .
Leigh B. Stoller's avatar
Leigh B. Stoller committed
968
969
970
		     " nodes as a left join reserved as b" .
		     " on a.node_id=b.node_id" .
		     " where b.node_id is null" .
971
		     " and a.role='testnode' and a.type!='dnard'");
972
    $numnodes = $numnodes_result->numrows;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
973
974
    
    if ($numnodes < $minimum_nodes) {
Chad Barb's avatar
Chad Barb committed
975
	fatal(65, "$0: *** Insufficient nodes available.\n");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
976
977
    }

978
    TBDebugTimeStamp("assign started");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
979
980
    # Run assign
    $fail = 0;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
981
    my $cmdargs = "$ptopfile $topfile";
982
    $cmdargs = "-p $cmdargs"
983
	if ($virtcount);
984
    print "assign $cmdargs\n";
Chad Barb's avatar
Chad Barb committed
985
986
987
988
    if (-1 == system "assign $cmdargs > assign.log") {
	fatal(65, "*** Couldn't run assign!");
    }
    
Leigh B. Stoller's avatar
Leigh B. Stoller committed
989
990
    $violations = 0;
    $score = -1;
Chad Barb's avatar
Chad Barb committed
991
    $assignexitcode = $? >> 8;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
992

Chad Barb's avatar
Chad Barb committed
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
    open(ASSIGNFP, "assign.log")
	or fatal(65, "*** Couldn't open assign logfile!");

    if ($assignexitcode == 0)
    {
	# read output
	# Header
	printdb "Reading assign results.\n";
	while (<ASSIGNFP>) {
	    chop;
	    /No physical nodes of type (.+)$/ && do {
		$score=-2;
		print $_ . "\n";
	    };
	    /^With ([0-9]+) violations$/ && do {
		$violations = $1;
		last;
	    };
	    /^[ \t]+BEST SCORE: [ \t]+([0-9]+(\.[0-9]+)?)/ && ($score=$1);
	}
	if ($score == -2) {
	    # Type error
	    fatal(65, "Giving up.\n" );
	}
	printdb "Found score $score, violations $violations.\n";
1018
    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028

    # We don't bother reading anything else if violations occured.
    if (($violations == 0) && ($score != -1)) {
	# read nodes section
	while (<ASSIGNFP> !~ /^Nodes:/) {}
	printdb "Nodes:\n";
	while (<ASSIGNFP>) {
	    chop;
	    /^End Nodes$/ && last;
	    @info = split;
1029
1030
1031
1032
1033
1034
	    ($virtual,$physical) = @info[0,1];

	    # We don't care about LAN nodes anymore.
	    if (defined($lannodes{$virtual})) {
		next;
	    }
Chad Barb's avatar
   
Chad Barb committed
1035

Chad Barb's avatar
   
Chad Barb committed
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
	    if ($alreadyAllocated{$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 (! exists $reserved_nodes{$virtual} ||
		    $reserved_nodes{$virtual} ne $physical ||
		    $alreadyAllocated{$physical} eq "reboot") {
		    $alreadyAllocated{$physical} = "reboot";
		} else {
		    $alreadyAllocated{$physical} = "reused";
		}
	    } else {
		#
		# This is a new node; we'll have to reserve it.
		#
Chad Barb's avatar
   
Chad Barb committed
1054
1055
		$toreserve{$physical} = 1;	    
	    }
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
	    
	    if ($isvirtnode{$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} = [];
		}
		push(@{$virtnodes{$physical}}, $virtual);
	    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1067
	    $v2pmap{$virtual} = $physical;
Shashi Guruprasad's avatar
Shashi Guruprasad committed
1068
	    if( ! defined($p2vmap{$physical}) ) {
1069
		$p2vmap{$physical} = [];
Shashi Guruprasad's avatar
Shashi Guruprasad committed
1070
1071
	    }
	    push(@{$p2vmap{$physical}}, $virtual);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1072
1073
1074
1075
1076
1077
	    printdb "  $virtual $physical\n";
	}

	# read Edges
	# By convention, in plinks, the delay node is always the second
	# entry.
Chad Barb's avatar
Chad Barb committed
1078
	while (<ASSIGNFP> !~ /^Edges:/) { }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1079
	printdb "Edges:\n";
Shashi Guruprasad's avatar
Shashi Guruprasad committed
1080
1081
	EDGEWHILE: while (<ASSIGNFP>) {
	    /^End Edges$/ && last EDGEWHILE;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1082
1083
1084
1085
1086
1087
1088
1089
1090
	    @info = split;
	    $line = $_;
	    $_ = $info[1]; # type
	  SWITCH1: {
	      /^intraswitch$/ && do {
		  ($vlink,$rawA,$rawB) = @info[0,3,5];
		  last SWITCH1;
	      };
	      /^interswitch$/ && do {
1091
1092
		  ($vlink,$rawA,$rawB) = @info[0,3,$#info];
		  last SWITCH1;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1093
1094
	      };
	      /^direct$/ && do {
Chad Barb's avatar
Chad Barb committed
1095
1096
		  fatal(65, "*** $0:\n".
			   "    Unsupported link type: direct.\n");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1097
	      };
Shashi Guruprasad's avatar
Shashi Guruprasad committed
1098
1099
1100
1101
1102
1103
	      /^trivial$/ && do {
		  # we don't have plinks for trivial links
		  $vlink = $info[0];
		  $plinks{$vlink} = [];
		  next EDGEWHILE;
	      };
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1104
1105
	      print "Found garbage: $line\n";
	  }
1106
1107
1108
1109
	    $nodeportA = &getnodeport($rawA);
	    $nodeportB = &getnodeport($rawB);
	    $nodeportA =~ s/\//:/;
	    $nodeportB =~ s/\//:/;
1110
	    $plinks{$vlink} = [$nodeportA,$nodeportB];
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1111
1112
1113
1114
	    printdb "  $vlink " . join(" ",@{$plinks{$vlink}}) . "\n";
	}
    } else {
	# spit out up to nodes
Chad Barb's avatar
Chad Barb committed
1115
	print "ASSIGN FAILED: \n";       
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1116
	while (<ASSIGNFP>) {
1117
	    if (/link_users:\s*(\d+)$/) {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1118
		$linkusers = $1;
1119
	    } elsif (/bandwidth:\s*(\d+)$/) {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1120
		$bandwidth = $1;
1121
	    } elsif (/unassigned:\s*(\d+)$/) {
1122
		$unassigned = $1;
1123
	    } elsif (/desires:\s*(\d+)$/) {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1124
1125
1126
1127
1128
1129
1130
1131
1132
		$desires = $1;
	    }
	    if (/^Nodes:/) {last;}
	    print "$_";
	}
	$fail = 1;
    }
    while (<ASSIGNFP>) { } # Read anything left in the pipe before closing
    close(ASSIGNFP);
1133
    TBDebugTimeStamp("assign finished");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1134
1135

    # Reserve resources
1136
    if (!$fail) {
Chad Barb's avatar
Chad Barb committed
1137
1138
	TBDebugTimeStamp("reserving started");

1139
	if (system("nalloc $pid $eid " . join(" ", keys(%toreserve)))) {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1140
	    print "Failed to reserve nodes. Trying again.\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1141
	} else {
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1142
	    print "Successfully reserved physical nodes\n";
Chad Barb's avatar
   
Chad Barb committed
1143
1144
1145
1146
1147
1148

	    foreach $node (keys(%toreserve)) {
		# in future, this will be a fully enforced state machine.
		TBSetNodeAllocState( $node, TBDB_ALLOCSTATE_RES_INIT_DIRTY() );
	    }

1149
	    TBDebugTimeStamp("reserving finished");
Chad Barb's avatar
   
Chad Barb committed
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172

	    my %tolose = ();
	    foreach $node (keys(%alreadyAllocated)) {
		if ($alreadyAllocated{$node} eq "unused") {
		    #
		    # Node was used in previous incarnation, but not any more.
		    #
		    $tolose{$node} = 1;
		} elsif ($alreadyAllocated{$node} eq "reboot") {
		    #
		    # Node is being reused, but for a different purpose, so
		    # it should be rebooted.
		    #
		    TBSetNodeAllocState( $node, TBDB_ALLOCSTATE_RES_INIT_DIRTY() );
		}
	    }
	    
	    if ((keys %tolose) > 0) {
		if (system("nfree $pid $eid " . join(" ", keys(%tolose)))) {
		    print "Failed to free no-longer-needed nodes!";
		}
	    }

Leigh B. Stoller's avatar
Leigh B. Stoller committed
1173
1174
1175
1176
1177
	    last;
	}
    }

    # Check for exit
Chad Barb's avatar
Chad Barb committed
1178
1179
    if ($assignexitcode == 2 || $currentrun >= $maxrun) {
	$exitcode = 65;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1180
1181
	if ($bandwidth > 0) {
	    $exitcode += 4;
Chad Barb's avatar
Chad Barb committed
1182
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1183
1184
1185
1186
1187
1188
	if ($linkusers > 0) {
	    $exitcode += 8;
	}
	if ($desires > 0) {
	    $exitcode += 16;
	}
1189
1190
1191
	if ($unassigned > 0) {
	    $exitcode += 32;
	}
Chad Barb's avatar
Chad Barb committed
1192
1193
1194
1195
1196
1197
1198
1199

	if ($assignexitcode == 2) {
	    fatal($exitcode, "*** $0:\n".
		     "    Unretriable error. Giving up.\n");
	} else {
	    fatal($exitcode, "*** $0:\n".
		     "    Reached run limit. Giving up.\n");
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
1200
    }
Chad Barb's avatar
Chad Barb committed
1201
1202
1203
    print "Waiting 5 seconds and trying again...\n";
    sleep(5);

Leigh B. Stoller's avatar
Leigh B. Stoller committed
1204
1205
1206
    $currentrun++;
}

1207
1208
TBDebugTimeStamp("assign_loop finished");

1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
###########################################################################
# 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.
#
###########################################################################

1219
1220
1221
1222
1223
1224
1225
1226
#
# VIRTNODES HACK: physical nodes (from the nodes table) might really
# be a virtual node :-) Must record the underlying phys_nodeid and an
# indicator for jail.
#
my %jailed = ();

# Now allocate the remote nodes. 
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
if (scalar(keys(%isremotenode))) {
    my $success  = 0;
    my $doremote = 0;
    
    foreach my $value (values(%isremotenode)) {
	if ($value) {
	    $doremote = 1;
	}
    }
    
    if ($doremote) {
	print "Running 'wanassign -d $pid $eid'\n";
	open(WANFP,"wanassign -d $pid $eid 2>&1 | tee wanassign.log |") or
Chad Barb's avatar
Chad Barb committed
1240
1241
	    fatal(65, "*** $0:\n".
		     "    Failed to start wanassign: $!\n");
1242
1243
1244
1245
1246

	printdb "Reading wanassign results.\n";
	while (<WANFP>) {
	    chop;
	    if ($_ =~ /(\S+) mapsto (\S+)/) {
1247
		$v2vmap{$1} = $2;
1248
1249
1250
1251
1252
1253
1254
1255
		printdb "  $1 $2\n";
	    }
	    if ($_ =~ /^Success/) {
		$success = 1;
	    }
	    # Skip other output. Usually its debugging output.
	}
	close(WANFP) or
Chad Barb's avatar
Chad Barb committed
1256
1257
	    fatal(65, "*** $0:\n".
		     "    wanassign: " . $? ? "exited with status: $?.\n" :
1258
1259
1260
		                         "error closing pipe: $!\n");

	if (!$success) {