tbswap.in 22.2 KB
Newer Older
Chad Barb's avatar
 
Chad Barb committed
1 2 3 4
#!/usr/bin/perl -w

#
# EMULAB-COPYRIGHT
5
# Copyright (c) 2000-2004 University of Utah and the Flux Group.
Chad Barb's avatar
 
Chad Barb committed
6 7 8 9
# All rights reserved.
#
use English;

Chad Barb's avatar
 
Chad Barb committed
10 11 12
# Returns 0 on success.
# Returns 1 on non-assign_wrapper failure.
# Returns (1 | assign_wrapper's errorcode) on assign_wrapper failure.
13 14
# Returns |0x40 if update caused a swapout. Icky.
# Returns -1 on uncontrolled error (die called).
Chad Barb's avatar
 
Chad Barb committed
15

Chad Barb's avatar
 
Chad Barb committed
16 17 18 19 20
# XXX: handle error cases for update? (backup the db?)
# XXX: Shouldn't do idempotent stuff twice for update.
# XXX: repush/calc routing for update??? (tbprerun)
# XXX: previz for update???              (tbprerun)
# XXX: make snmpit faster for update.
Chad Barb's avatar
Chad Barb committed
21 22 23
#
# XXX: for update, expt is swapped out on os_setup fail.
#      (we only recover if assign fails)
Chad Barb's avatar
 
Chad Barb committed
24 25 26

sub usage()
{
27
    print STDERR "Usage: $0 -force { in | out | update [-reboot] [-eventsys_restart] } pid eid\n";
Chad Barb's avatar
 
Chad Barb committed
28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52
    exit(-1);
}

#
# Configure variables
#
my $TBROOT         = "@prefix@";
my $TESTMODE       = @TESTMODE@;
my $DISABLE_EVENTS = "@DISABLE_EVENT_SCHED@";

# Untaint the path
$ENV{'PATH'} = "/usr/bin:$TBROOT/libexec:$TBROOT/libexec/ns2ir" . 
    ":$TBROOT/sbin:$TBROOT/bin";

#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;
require exitonwarn; # exitonwarn isn't really a module, so just require it

#
# Actual swap-in and swap-out functions, defined below.
#
Chad Barb's avatar
 
Chad Barb committed
53 54 55 56 57 58 59
sub doSwapout($);
sub doSwapin($);

sub REAL()    { return 4; }
sub CLEANUP() { return 3; }
sub RETRY()   { return 2; }
sub UPDATE()  { return 1; }
Chad Barb's avatar
Chad Barb committed
60
sub UPDATE_RECOVER()  { return 0; }
Chad Barb's avatar
 
Chad Barb committed
61 62 63 64 65 66 67

#
# Turn off line buffering on output
#

$| = 1;

68 69
my $updateReboot   = 0;
my $updateReconfig = 1;
70
my $updateEventsys_restart = 0;
Chad Barb's avatar
 
Chad Barb committed
71 72
my $force  = 0;
my $errors = 0;
73
my $updatehosed = 0;
Chad Barb's avatar
 
Chad Barb committed
74
my $state;
75
my $canceled;
Chad Barb's avatar
 
Chad Barb committed
76 77
my $os_setup_pid;
my $cleanvlans;
78
my $nextState;
Chad Barb's avatar
 
Chad Barb committed
79 80

#
Chad Barb's avatar
 
Chad Barb committed
81
# First argument is either "in", "out", or "update";
Chad Barb's avatar
 
Chad Barb committed
82 83 84 85 86
# this value goes into $swapop.
#

my $swapop = shift;	

Chad Barb's avatar
 
Chad Barb committed
87 88 89 90
if (!$swapop || 
    (($swapop ne "in") && 
     ($swapop ne "out") &&
     ($swapop ne "update"))) {
Chad Barb's avatar
 
Chad Barb committed
91 92 93 94 95 96 97 98 99 100 101
    usage();
}

#
# Get other arguments.
#

while ($#ARGV > 1) {
    $arg = shift;
    if ($arg eq "-force") {
	$force = 1;
Chad Barb's avatar
Chad Barb committed
102 103
    } elsif ($arg eq "-reboot") {
	$updateReboot = 1;
104
	$updateReconfig = 0;
105 106
    } elsif ($arg eq "-eventsys_restart" && $swapop eq "update") {
	$update_Eventsys_restart = 1;
Chad Barb's avatar
 
Chad Barb committed
107 108 109 110 111 112 113 114 115
    } else {
	usage();
    }
}
if ($#ARGV < 1) {
    usage();
}
my ($pid,$eid) = @ARGV;

116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131
#
# Untaint the arguments.
#
if ($pid =~ /^([-\@\w.]+)$/) {
    $pid = $1;
}
else {
    die("Tainted argument $pid!\n");
}
if ($eid =~ /^([-\@\w.]+)$/) {
    $eid = $1;
}
else {
    die("Tainted argument $eid!\n");
}

Chad Barb's avatar
 
Chad Barb committed
132 133 134 135 136 137 138 139 140 141 142 143 144
TBDebugTimeStampsOn();

#
# Print starting message.
#

print "Beginning swap-$swapop for $pid/$eid. " . TBTimeStamp() . "\n";
TBDebugTimeStamp("tbswap $swapop started");

#
# Get experiment state; verify that experiment exists.
#
if (! ($state = ExpState($pid, $eid))) {
145 146
    die("*** $0:\n".
	"    No such experiment $pid/$eid\n");
Chad Barb's avatar
 
Chad Barb committed
147
}
148 149 150 151 152 153
# Sanity check the current state. 
if (!$force) {
    if ($swapop eq "in") {
	die("*** $0:\n".
	    "    Experiment should be ACTIVATING. Currently $state.\n")
	    if ($state ne EXPTSTATE_ACTIVATING);
Chad Barb's avatar
 
Chad Barb committed
154
    }
155 156 157 158
    elsif ($swapop eq "out") {
	die("*** $0:\n".
	    "    Experiment should be SWAPPING. Currently $state.\n")
	    if ($state ne EXPTSTATE_SWAPPING);
Chad Barb's avatar
 
Chad Barb committed
159
    }
160 161 162 163
    elsif ($swapop eq "update") {
	die("*** $0:\n".
	    "    Experiment should be MODIFY_RESWAP. Currently $state.\n")
	    if ($state ne EXPTSTATE_MODIFY_RESWAP);
Chad Barb's avatar
 
Chad Barb committed
164 165 166 167 168 169
    }
}

#
# Do actual swapping
#
170 171 172 173 174
if ($swapop eq "out") {
    #
    # Swap out
    #
    $errors = doSwapout(REAL);
Chad Barb's avatar
 
Chad Barb committed
175
}
176 177 178 179 180 181 182 183
elsif ($swapop eq "update") {
    #
    # Update.
    #
    # Phase One -- swap experiment partially out.
    #
    print STDERR "Backing up physical state...\n";
    TBExptBackupPhysicalState($pid,$eid);
Chad Barb's avatar
 
Chad Barb committed
184

185
    $errors = doSwapout(UPDATE);
Chad Barb's avatar
Chad Barb committed
186

187
    if ($errors) {
Chad Barb's avatar
 
Chad Barb committed
188
	#
189 190 191 192 193
	# Clean up the mess, leaving the experiment in the SWAPPED state,
	# 
	print STDERR "Cleaning up after errors.\n";
	doSwapout(CLEANUP);
	$updatehosed = 1;
Chad Barb's avatar
 
Chad Barb committed
194
    }
195
    else {
Chad Barb's avatar
 
Chad Barb committed
196
	#
197
	# Phase Two -- swap experiment back in.
Chad Barb's avatar
 
Chad Barb committed
198
	#
Chad Barb's avatar
Chad Barb committed
199 200
	$errors = doSwapin(UPDATE);

201
	if ($errors) {
Chad Barb's avatar
Chad Barb committed
202 203 204
	    #
	    # There were errors; see if we can recover.
	    #
205
	    my $CanRecover = 1;
Chad Barb's avatar
Chad Barb committed
206 207 208

	    if ($errors != 7) {
		print STDERR "Update failure occurred _after_ assign phase; ";
209
		$CanRecover = 0;
Chad Barb's avatar
Chad Barb committed
210 211
	    }

212 213 214 215 216 217 218 219
	    if ($CanRecover) {
		print STDERR "Recovering virtual and physical state.\n";

		if (TBExptRemoveVirtualState($pid, $eid) ||
		    TBExptRestoreVirtualState($pid, $eid) ||
		    TBExptRestorePhysicalState($pid,$eid)) {
		    print STDERR "Could not restore backed-up state; ";
		    $CanRecover = 0;
Chad Barb's avatar
Chad Barb committed
220
		}
221 222 223 224 225 226 227
		else {
		    print STDERR "Doing a recovery swap-in of old state.\n";

		    if (doSwapin(UPDATE_RECOVER)) {
			print STDERR "Could not swap in old physical state; ";
			$CanRecover = 0;
		    }
Chad Barb's avatar
Chad Barb committed
228 229
		}
	    }
230 231 232 233 234 235 236

	    #
	    # Some part of the recovery failed; must swap it out. swapexp
	    # (caller) will then have to do more clean up, hence the special
	    # exit status indicated by $updatehosed.
	    # 
	    if (! $CanRecover) {
237
		print STDERR "Recovery aborted! Swapping experiment out.\n";
Chad Barb's avatar
Chad Barb committed
238
		doSwapout(CLEANUP);
239 240 241 242
		$updatehosed = 1;
	    }
	    else {
		print STDERR "Update recovery successful.\n";
Chad Barb's avatar
Chad Barb committed
243
	    }
Chad Barb's avatar
 
Chad Barb committed
244
	}
Chad Barb's avatar
Chad Barb committed
245 246
    }
}
247 248 249 250 251 252 253
elsif ($swapop eq "in") {
    #
    # Swap in
    #
    my $retries = 2;
    
    $errors = doSwapin(REAL);
Chad Barb's avatar
Chad Barb committed
254

255 256 257 258 259 260 261 262 263 264
    #
    # Attempt a retry if: 
    #   a) there were errors, 
    #   b) doswapin() indicated (via return code 3) a retry is appropriate,
    #   c) we haven't tried too many times already.
    #   d) The cancelflag has not been set.
    #   e) $TESTMODE == 0.
    #
    while ($errors == 3 && $retries && !$canceled && !$TESTMODE) {
	$retries--;
Chad Barb's avatar
 
Chad Barb committed
265

266 267
	print STDERR "Cleaning up after errors; will try again.\n";
	doSwapout(RETRY);
Chad Barb's avatar
 
Chad Barb committed
268

269 270
	print STDERR "Trying again...\n";
	$errors = doSwapin(RETRY);
Chad Barb's avatar
 
Chad Barb committed
271
    }
272 273
    if ($errors || $canceled) {
	print STDERR "Cleaning up after " .
274
	    ($canceled ? "cancelation" : "errors") . ".\n";
275
	doSwapout(CLEANUP);
Chad Barb's avatar
 
Chad Barb committed
276 277 278 279 280 281 282
    }
}

#
# Write appropriate message and exit.
#
if ($errors) {
283
    print "Failingly finished swap-$swapop for $pid/$eid. ".TBTimeStamp()."\n";
Chad Barb's avatar
 
Chad Barb committed
284
    TBDebugTimeStamp("tbswap $swapop finished (failed)");
Chad Barb's avatar
Chad Barb committed
285

286 287
    # Pass out magic value to indicate that update failed!
    exit(1 | ($updatehosed ? 0x40 : 0));
Chad Barb's avatar
 
Chad Barb committed
288
}
289
print "Successfully finished swap-$swapop for $pid/$eid. " .TBTimeStamp()."\n";
290 291
TBDebugTimeStamp("tbswap $swapop finished (succeeded)");
exit(0);
Chad Barb's avatar
 
Chad Barb committed
292 293 294 295

#################################

##
Chad Barb's avatar
Chad Barb committed
296
#
Chad Barb's avatar
 
Chad Barb committed
297 298
# doSwapout - Swaps experiment out.
#
Chad Barb's avatar
Chad Barb committed
299
#             If in REAL or CLEANUP,
Chad Barb's avatar
 
Chad Barb committed
300 301 302
#             this function will free all nodes for the 
#             experiment.
#
Chad Barb's avatar
Chad Barb committed
303
#             If in RETRY or UDPATE,
Chad Barb's avatar
 
Chad Barb committed
304 305 306 307 308 309
#             only nodes not in RES_READY will be freed.
#
#             Returns 0 on success, >0 on failure.
#
##

Chad Barb's avatar
 
Chad Barb committed
310 311
sub doSwapout($) {
    my $type = shift; # REAL==4, CLEANUP==3, RETRY==2, UPDATE==1.
Chad Barb's avatar
 
Chad Barb committed
312 313 314 315 316 317 318 319 320 321 322
    my $swapout_errors = 0;

    #
    # wait for os_setup;
    # this only applies if called after a failed doswapin.
    #
    if ($os_setup_pid) {
	print "Waiting for os_setup to finish\n";
	waitpid($os_setup_pid, 0);
	undef $os_setup_pid;
    }
Chad Barb's avatar
Chad Barb committed
323

Chad Barb's avatar
 
Chad Barb committed
324 325
    if (! $TESTMODE) { 
	if (! $DISABLE_EVENTS) {
326
	    if ($type >= RETRY || ($update_Eventsys_restart && $type == UPDATE) ) {
Chad Barb's avatar
 
Chad Barb committed
327 328 329 330 331
		print "Stopping the event system\n";
		if (system("eventsys_control stop $pid $eid")) {
		    print STDERR "*** Failed to stop the event system.\n";
		    $swapout_errors = 1;
		}
Chad Barb's avatar
 
Chad Barb committed
332 333
	    }
	}
Chad Barb's avatar
Chad Barb committed
334

Chad Barb's avatar
 
Chad Barb committed
335
	#
Chad Barb's avatar
Chad Barb committed
336
	# Clean up any VLANs in experiment.
Chad Barb's avatar
 
Chad Barb committed
337
	#
Chad Barb's avatar
Chad Barb committed
338 339 340 341 342 343 344
	TBDebugTimeStamp("snmpit started");
	print STDERR "Removing VLANs.\n";
	if (system("snmpit -r $pid $eid")) {
	    print STDERR "*** Failed to reset VLANs\n";
	    $swapout_errors = 1;
	} else {
	    $cleanvlans = 0;
Chad Barb's avatar
 
Chad Barb committed
345
	}
Chad Barb's avatar
Chad Barb committed
346
	TBDebugTimeStamp("snmpit finished");
347
    }
Chad Barb's avatar
Chad Barb committed
348

349 350 351
    if ($type >= CLEANUP) {
	#
	# We're not attempting a retry;
Chad Barb's avatar
 
Chad Barb committed
352
	#
353
	# Stop all of the vnodes.
Chad Barb's avatar
 
Chad Barb committed
354
	#
355
	if (! $TESTMODE) { 	
Chad Barb's avatar
 
Chad Barb committed
356 357 358 359 360 361 362
	    print "Tearing down virtual nodes.\n";
	    TBDebugTimeStamp("vnode_setup -k started");
	    if (system("vnode_setup -d -k $pid $eid")) {
		print STDERR "*** Failed to tear down vnodes.\n";
		$swapout_errors = 1;
	    }
	    TBDebugTimeStamp("vnode_setup finished");
Chad Barb's avatar
 
Chad Barb committed
363 364 365 366
	}

	#
	# remove all nodes from the experiment.
Chad Barb's avatar
 
Chad Barb committed
367
	# (nfree will send them to RES_FREE_DIRTY)
Chad Barb's avatar
 
Chad Barb committed
368 369 370 371 372 373 374 375
	#
	print STDERR "Freeing nodes.\n";
	TBDebugTimeStamp("nfree started");
	if (system("nfree $pid $eid")) {
	    print STDERR "*** Could not free nodes.\n";
	    $swapout_errors = 1;
	}
	TBDebugTimeStamp("nfree finished");
Chad Barb's avatar
 
Chad Barb committed
376 377 378 379 380

	#
	# Since this is an actual swapout, 
	# reset our count of swap out nag emails sent.
	#
381
	DBQueryWarn("update experiments set swap_requests='',sim_reswap_count='0' ".
382
		    "where eid='$eid' and pid='$pid'");
Chad Barb's avatar
 
Chad Barb committed
383 384
    } else {
	#
Chad Barb's avatar
 
Chad Barb committed
385
	# $type == RETRY or $type == UPDATE.
Chad Barb's avatar
 
Chad Barb committed
386 387
	# Therefore, don't deallocate nodes which have been successfully
	# incorporated into the experiment (i.e., are RES_READY).
Chad Barb's avatar
 
Chad Barb committed
388
	# (nfree will send deallocated nodes to RES_FREE_DIRTY)
Chad Barb's avatar
 
Chad Barb committed
389
	#
390 391 392 393 394 395 396 397
	my @failedpnodes = ();
	my @failedvnodes = ();
	
	my $db_result =
	    DBQueryFatal("select rv.node_id,n.allocstate,nt.isvirtnode ".
                         "  from reserved as rv ".
			 "left join nodes as n on n.node_id = rv.node_id ".
			 "left join node_types as nt on nt.type=n.type ".
Chad Barb's avatar
 
Chad Barb committed
398 399
			 "where rv.pid='$pid' and rv.eid='$eid'");

400
	while (my ($node,$allocstate,$isvirt) = $db_result->fetchrow_array) {
Chad Barb's avatar
 
Chad Barb committed
401
	    if ($allocstate ne TBDB_ALLOCSTATE_RES_READY()) {
402 403 404 405 406 407
		if ($isvirt) {
		    push(@failedvnodes, $node);
		}
		else {
		    push(@failedpnodes, $node);
		}
Chad Barb's avatar
 
Chad Barb committed
408 409 410
	    }
	}

411 412 413 414 415 416 417 418 419 420 421 422
	#
	# Tear down failed vnodes. Perhaps not needed?
	# 
	if (!$TESTMODE && @failedvnodes > 0) {
	    print "Tearing down failed virtual nodes.\n";
	    TBDebugTimeStamp("vnode_setup -k started");
	    if (system("vnode_setup -d -k $pid $eid @failedvnodes")) {
		print STDERR "*** Failed to tear down vnodes.\n";
		$swapout_errors = 1;
	    }
	    TBDebugTimeStamp("vnode_setup -k finished");
	}
Chad Barb's avatar
 
Chad Barb committed
423

424 425 426 427
	#
	# Release all failed nodes.
	# 
	if (@failedpnodes > 0 || @failedvnodes > 0) {
428 429
	    print STDERR "Freeing failed nodes.\n";
	    
Chad Barb's avatar
 
Chad Barb committed
430
	    TBDebugTimeStamp("nfree started");
Chad Barb's avatar
 
Chad Barb committed
431 432 433 434 435
	    #
	    # Specify -x switch so when a physical node gets freed,
	    # any virtual nodes (owned by this experiment)
	    # sitting on top of it are freed as well.
	    #
436 437
	    if (system("nfree -x $pid $eid " .
		       join(" ", (@failedpnodes, @failedvnodes)))) {
Chad Barb's avatar
 
Chad Barb committed
438 439 440 441 442 443 444
		print STDERR "*** Could not free nodes.\n";
		$swapout_errors = 1;
	    }
	    TBDebugTimeStamp("nfree finished");
	}
    }

445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468
    if (! $TESTMODE) {
	#
	# If the experiment has no Plab dslice nodes left, but still has
	# a Plab slice, destroy the slice
	#

	# Does the slice exist?
	$db_result =
	    DBQueryFatal("select slicename from plab_slices ".
			 "where pid='$pid' and eid='$eid'");

	if ($db_result->numrows) {
	    # Are there any dslice nodes left?
	    $db_result =
		DBQueryFatal("select n.node_id from nodes as n ".
			     "left join node_types as nt on n.type = nt.type ".
			     "left join reserved as r ".
			     " on r.node_id = n.node_id ".
			     "where r.pid='$pid' and r.eid='$eid' ".
			     " and nt.isplabdslice = 1");

	    if (!$db_result->numrows) {
		print "Destroying Planetlab slice.\n";
		TBDebugTimeStamp("plabslice destroy started");
469
		if (system("plabslice destroy $pid $eid")) {
470 471 472 473 474 475 476 477
		    print STDERR "*** Failed to destroy Plab dslice\n";
		    $swapout_errors = 1;
		}
		TBDebugTimeStamp("plabslice destroy finished");
	    }
	}
    }

Chad Barb's avatar
 
Chad Barb committed
478 479 480 481 482 483 484 485
    if (! $TESTMODE) {
	#
	# All of these errors are non-fatal on swapout. We find out about them
	# via email sent from the individual scripts.
	#

	#
	# Only reset mountpoints if this is an actual swapout, and
Chad Barb's avatar
 
Chad Barb committed
486
	# not a failed swapin(cleanup), update, or retry.
Chad Barb's avatar
 
Chad Barb committed
487
	#
Chad Barb's avatar
 
Chad Barb committed
488
	if ($type == REAL) {
Chad Barb's avatar
 
Chad Barb committed
489 490 491 492 493 494 495
	    print "Resetting mountpoints.\n";
	    TBDebugTimeStamp("exports started");
	    if (system("exports_setup")) {
		print STDERR "*** Failed to reset mountpoints.\n";
	    }
	    TBDebugTimeStamp("exports finished");
	}
Chad Barb's avatar
Chad Barb committed
496

Chad Barb's avatar
 
Chad Barb committed
497 498 499 500
	#
	# Resetting named maps and email lists is fast and idempotent,
	# so whatever.
	#
Chad Barb's avatar
 
Chad Barb committed
501 502 503 504 505 506
	print "Resetting named maps.\n";
	TBDebugTimeStamp("named started");
	if (system("named_setup")) {
	    print "*** WARNING: Failed to reset named map.\n";
	}
	TBDebugTimeStamp("named finished");
Chad Barb's avatar
Chad Barb committed
507

Chad Barb's avatar
 
Chad Barb committed
508 509
	print "Resetting email lists.\n";
	TBDebugTimeStamp("genelists started");
510
	if (system("genelists -t")) {
Chad Barb's avatar
 
Chad Barb committed
511 512 513 514 515
	    print "*** WARNING: Failed to reset email lists.\n";
	}
	TBDebugTimeStamp("genelists finished");
    }

Chad Barb's avatar
 
Chad Barb committed
516
    #
517 518 519
    # Wipe the DB clean except during UPDATE or RETRY. In those
    #    cases, assign_wrapper will reset the DB after reading
    #    the info.
Chad Barb's avatar
 
Chad Barb committed
520
    #
521 522 523 524
    if ( $type >= CLEANUP ) {
	print STDERR "Resetting DB.\n";
	TBExptRemovePhysicalState( $pid, $eid );
    }
Chad Barb's avatar
 
Chad Barb committed
525 526 527 528 529

    return $swapout_errors;
}

##
Chad Barb's avatar
Chad Barb committed
530
#
Chad Barb's avatar
 
Chad Barb committed
531 532
# doSwapin - Swaps experiment in.
#
Chad Barb's avatar
Chad Barb committed
533 534 535 536 537
#            Returns:
#              0 - successful swapin
#              1 - failed swapin; cleanup required.
#              3 - failed swapin; cleanup required; can retry.
#              7 - failed swapin; assign failed; no cleanup.
Chad Barb's avatar
 
Chad Barb committed
538 539
##

Chad Barb's avatar
 
Chad Barb committed
540
sub doSwapin($) {
Chad Barb's avatar
Chad Barb committed
541
    my $type = shift; # REAL==4, RETRY==2, UPDATE==1, UPDATE_RECOVER=0. 
Chad Barb's avatar
 
Chad Barb committed
542

Chad Barb's avatar
 
Chad Barb committed
543 544 545 546 547
    #
    # assign_wrapper does all the virtual to physical mapping 
    # and updating the DB state.
    #

Chad Barb's avatar
Chad Barb committed
548 549 550 551 552
    if ($type > UPDATE_RECOVER) {
	print "Mapping to physical reality ...\n";
	TBDebugTimeStamp("assign_wrapper started");

	#
553 554 555 556
	# Pass the -u (update) switch into assign_wrapper, which turns on
	# update mode. When doing a retry, must also fix the current nodes
	# to avoid stuff jumping around when simply trying to replace a node
	# that did not boot.
Chad Barb's avatar
Chad Barb committed
557 558
	#
	my $exitcode;
559 560 561
	my $wrapper = "assign_wrapper -u";
	$wrapper .= " -f"
	    if ($type == RETRY);
562
	
563
	if (system("$wrapper $pid $eid")) {
564 565 566 567
	    $exitcode = $? >> 8;

	    print STDERR "*** Failed ($exitcode) to map to reality.\n";

568 569
	    # Wrapper sets this bit when recovery is possible.
	    if ($exitcode & 64) {
570
		# We can recover. 
Chad Barb's avatar
Chad Barb committed
571
		return 7;
572 573
	    }
	    else {
574
		# No recovery, no retry.
Chad Barb's avatar
Chad Barb committed
575 576 577 578
		return 1;
	    }
	}
	TBDebugTimeStamp("assign_wrapper finished");
Chad Barb's avatar
 
Chad Barb committed
579

Chad Barb's avatar
Chad Barb committed
580 581
	print "Mapped to physical reality!\n";
    }
Chad Barb's avatar
 
Chad Barb committed
582

583
    # Check cancel flag before continuing. No retry, 
584
    TBGetCancelFlag($pid, $eid, \$canceled);
585 586 587
    return 1
	if ($canceled);

588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625
    #
    # Look for any nodes in RES_TEARDOWN. These need to be released,
    # and if a virtnode, they need to be torn down. We cannot wait for
    # the virtnodes to go down with the physnode they are hosted on,
    # so teardown and release the virtnodes first, and then do the
    # physnodes.
    #
    # Errors are fatal; no recovery or retry.
    #
    if ($type == UPDATE) {
	my $allocstate = TBDB_ALLOCSTATE_RES_TEARDOWN();
	
	$db_result =
	    DBQueryFatal("select r.node_id,nt.isvirtnode,nt.isremotenode ".
			 "  from reserved as r ".
			 "left join nodes as n on n.node_id=r.node_id ".
			 "left join node_types as nt on nt.type=n.type ".
			 "where r.pid='$pid' and r.eid='$eid' and ".
			 "      n.allocstate='$allocstate'");

	if ($db_result->numrows) {
	    my @virtnodes = ();
	    my @physnodes = ();
	    
	    print "Tearing down and releasing unused nodes\n";

	    # First teardown/release virtnodes. 
	    while (my ($node,$isvirt,$isrem) = $db_result->fetchrow_array()) {
		if ($isvirt) {
		    push(@virtnodes, $node);
		}
		elsif (!$isrem) {
		    push(@physnodes, $node);
		}
	    }
	    if (@virtnodes) {
		TBDebugTimeStamp("vnode_setup started");
		
626
		if (system("vnode_setup -k $pid $eid @virtnodes")) {
627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645
		    print "Failed to tear down unused virtnodes!\n";
		    return 1;
		}
		TBDebugTimeStamp("vnode_setup finished");
		
		if (system("nfree $pid $eid @virtnodes")) {
		    print "Failed to nfree unused virtnodes!\n";
		    return 1;
		}
	    }
	    if (@physnodes) {
		if (system("nfree $pid $eid @physnodes")) {
		    print "Failed to nfree unused physnodes!\n";
		    return 1;
		}
	    }
	}
    }

Chad Barb's avatar
 
Chad Barb committed
646 647 648 649 650 651
    # Exit here if we are testing.
    if ($TESTMODE) {
	print "Testing run - Stopping here.\n";
	return 0;
    }

652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667
    #
    # Handle tarballs - we might need to fetch some from URLs if the user
    # asked for that.
    #
    print "Fetching tarballs and RPMs (if any) ...\n";
    TBDebugTimeStamp("tarfiles_setup started");

    if (system("tarfiles_setup $pid $eid")) {
	#
	# No recovery for now - what would we do?
	#
	print STDERR "*** Failed to set up tarballs.\n";
	return 1;
    }
    TBDebugTimeStamp("tarfiles_setup finished");

668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695
    #
    # If there are any Plab dslice nodes in the experiment, create the
    # dslice now
    #
    if ($type > UPDATE_RECOVER) {
	# Are there any Plab nodes?
	$db_result =
	    DBQueryFatal("select n.node_id from nodes as n ".
			 "left join node_types as nt on n.type = nt.type ".
			 "left join reserved as r on r.node_id = n.node_id ".
			 "where r.pid='$pid' and r.eid='$eid' ".
			 " and nt.isplabdslice = 1");

	if ($db_result->numrows) {
	    # Does slice already exist?
	    $db_result =
		DBQueryFatal("select slicename from plab_slices ".
			     "where pid='$pid' and eid='$eid'");

	    if (! $db_result->numrows) {
		my @plabnodes = ();
		
		while (my ($node) = $db_result->fetchrow_array()) {
		    push(@plabnodes, $node);
		}
		
		print "Creating Planetlab slice.\n";
		TBDebugTimeStamp("plabslice create started");
696
		if (system("plabslice create $pid $eid")) {
697 698 699 700 701 702 703 704
		    print STDERR "*** Failed to create Plab dslice\n";
		    return 3;
		}
		TBDebugTimeStamp("plabslice alloc finished");
	    }
	}
    }

705
    # Check cancel flag before continuing. No retry, 
706
    TBGetCancelFlag($pid, $eid, \$canceled);
707 708 709
    return 1
	if ($canceled);

Chad Barb's avatar
 
Chad Barb committed
710 711 712 713 714 715 716 717 718 719 720 721 722
    #
    # These things need to get started before the nodes come up, so we'll
    # do them before the os_setup. Everything else can done in parallel with
    # os_setup. (Actually, these probably can too, since they should finish
    # long before the nodes reboot, but better safe than sorry)
    #
    print "Setting up mountpoints.\n";
    TBDebugTimeStamp("mountpoints started");
    if (system("exports_setup")) {
	print STDERR "*** Failed to setup mountpoints.\n";
	return 1;
    }
    TBDebugTimeStamp("mountpoints finished");
Chad Barb's avatar
Chad Barb committed
723

Chad Barb's avatar
 
Chad Barb committed
724 725 726 727 728 729
    TBDebugTimeStamp("named started");
    print "Setting up named maps.\n";
    if (system("named_setup")) {
	print STDERR "*** WARNING: Failed to add node names to named map.\n";
	#
	# This is a non-fatal error.
Chad Barb's avatar
Chad Barb committed
730
	#
Chad Barb's avatar
 
Chad Barb committed
731 732
    }
    TBDebugTimeStamp("named finished");
Chad Barb's avatar
Chad Barb committed
733

734
    # Check cancel flag before continuing. No retry, 
735
    TBGetCancelFlag($pid, $eid, \$canceled);
736 737
    return 1
	if ($canceled);
Chad Barb's avatar
Chad Barb committed
738 739 740

    #
    # If user specified -reboot to update,
Chad Barb's avatar
 
Chad Barb committed
741
    # and we are successfully performing the update,
742
    # then mark all nodes in experiment so os_setup will reboot them.
Chad Barb's avatar
Chad Barb committed
743
    #
744 745 746
    if (($type == UPDATE) &&
	($updateReboot || $updateReconfig)) {
	print STDERR "Marking nodes for reboot/reconfig.\n";
Chad Barb's avatar
Chad Barb committed
747
	$db_result =
748 749 750
	    DBQueryFatal("select r.node_id,n.allocstate from reserved as r ".
			 "left join nodes as n on n.node_id=r.node_id ".
			 "where r.pid='$pid' and r.eid='$eid'");
Chad Barb's avatar
Chad Barb committed
751

752 753 754 755 756 757 758 759
	while (my ($node,$allocstate) = $db_result->fetchrow_array) {
	    #
	    # If the node is INIT_CLEAN, leave it alone. It will still get
	    # rebooted, but will not falsely be tagged as dirty. This is
	    # important for vnodes too, where INIT_CLEAN indicated the vnode
	    # does not even exist yet (plab nodes).
	    #
	    if ($allocstate ne TBDB_ALLOCSTATE_RES_INIT_CLEAN()) {
760 761 762 763
		TBSetNodeAllocState($node,
				    ($updateReboot ?
				     TBDB_ALLOCSTATE_RES_INIT_DIRTY() :
				     TBDB_ALLOCSTATE_RES_RECONFIG()));
764
	    }
Chad Barb's avatar
Chad Barb committed
765 766 767
	}
    }

Chad Barb's avatar
 
Chad Barb committed
768 769 770 771 772
    #
    # Since it'll take a while for the nodes to reboot, we'll start now, and
    # wait for the os_setup to finish, down below
    #
    print "Resetting OS and rebooting.\n";
773
    TBDebugTimeStamp("launching os_setup");
Chad Barb's avatar
 
Chad Barb committed
774 775 776 777 778 779
    if (!($os_setup_pid = fork())) { 
	exec("os_setup $pid $eid") or return 1;
    } elsif ($os_setup_pid == -1) {
	print STDERR "*** Fork failed.\n";
	return 1;
    }
Chad Barb's avatar
Chad Barb committed
780

Chad Barb's avatar
 
Chad Barb committed
781
    #
782 783 784 785 786
    # XXX
    # Don't add any steps between here and the waitpid() call below
    # without verifying that 1) It's OK for nodes to come up before
    # the step has completed and 2) It's OK for the command to run in
    # parallel with os_setup (no DB dependencies, etc.)
Chad Barb's avatar
 
Chad Barb committed
787 788 789 790 791 792 793 794 795
    #

    print "Setting up VLANs.\n";
    TBDebugTimeStamp("snmpit started");
    if (system("snmpit -t $pid $eid")) {
	print STDERR "*** Failed to set up VLANs.\n";
	return 1;
    }
    TBDebugTimeStamp("snmpit finished");
Chad Barb's avatar
Chad Barb committed
796

Chad Barb's avatar
 
Chad Barb committed
797 798 799 800
    #
    # An error now means that the VLANS need to be cleaned up.
    #
    $cleanvlans = 1;
Chad Barb's avatar
Chad Barb committed
801

Chad Barb's avatar
 
Chad Barb committed
802 803
    print "Setting up email lists.\n";
    TBDebugTimeStamp("genelists started");
804
    if (system("genelists -t")) {
Chad Barb's avatar
 
Chad Barb committed
805 806 807 808 809 810
	print STDERR "*** WARNING: Failed to update email lists.\n";
	#
	# This is a non-fatal error.
	# 
    }
    TBDebugTimeStamp("genelists finished");
Chad Barb's avatar
Chad Barb committed
811

Chad Barb's avatar
 
Chad Barb committed
812 813 814 815 816 817 818 819 820 821 822 823 824 825
    #
    # Don't clear port counters on UPDATE.
    # (XXX should clear new nodes' port counters.)

    if ($type >= RETRY) {
	print "Clearing port counters.\n";
	TBDebugTimeStamp("portstats started");
	if (system("portstats -z -a -q $pid $eid")) {
	    print STDERR "*** WARNING: Failed to clear port counters.\n";
	    #
	    # This is a non-fatal error.
	    # 
	}
	TBDebugTimeStamp("portstats finished");
Chad Barb's avatar
 
Chad Barb committed
826
    }
Chad Barb's avatar
Chad Barb committed
827

Chad Barb's avatar
 
Chad Barb committed
828 829 830 831 832 833 834 835 836
    #
    # OK, let's see how that os_setup did
    #
    $kid = waitpid($os_setup_pid,0);
    if ($kid == $os_setup_pid) {
	undef $os_setup_pid; # Make sure doswapout() doesn't wait for it.
	if ($CHILD_ERROR) {
	    print STDERR "*** Failed to reset OS and reboot nodes.\n";
	    #
Chad Barb's avatar
 
Chad Barb committed
837 838
	    # Use returncode from os_setup process to
	    # set global $retry flag, indicating to caller
Chad Barb's avatar
 
Chad Barb committed
839 840 841
	    # that it may be beneficial to attempt
	    # a doSwapin() again.
	    #
Chad Barb's avatar
 
Chad Barb committed
842
            if (($CHILD_ERROR >> 8) == 1) {
Chad Barb's avatar
Chad Barb committed
843
		return 3;
Chad Barb's avatar
 
Chad Barb committed
844 845
	    } else {
		print STDERR "Not retrying due to error type.\n";
Chad Barb's avatar
Chad Barb committed
846
		return 1;
Chad Barb's avatar
 
Chad Barb committed
847
	    }
Chad Barb's avatar
 
Chad Barb committed
848 849 850 851 852 853 854
	}
    } else {
	undef $os_setup_pid;
	print STDERR "*** Error waiting for os_setup to finish.\n";
	return 1;
    }
    TBDebugTimeStamp("os_setup finished");
Chad Barb's avatar
Chad Barb committed
855

Chad Barb's avatar
 
Chad Barb committed
856 857 858 859
    #
    # Okay, start the event system now that we know all the nodes have
    # rebooted (os_setup is done). This only takes a moment (puts itself
    # in the background), so its not enough of a delay to worry about.
Chad Barb's avatar
 
Chad Barb committed
860 861
    # Don't do this during an update, since we didn't kill the 
    # event system previously, so starting it again will fail!
Chad Barb's avatar
 
Chad Barb committed
862 863
    # 
    if (! $DISABLE_EVENTS) {
864 865
	if ( $update_Eventsys_restart || 
	    ($type != UPDATE && $type != UPDATE_RECOVER) ) {
Chad Barb's avatar
 
Chad Barb committed
866 867 868 869 870 871 872
	    print "Starting the event system.\n";
	    TBDebugTimeStamp("eventsys_control started");
	    if (system("eventsys_control start $pid $eid")) {
		print STDERR "*** Failed to start the event system.\n";
		return 1;
	    }
	    TBDebugTimeStamp("eventsys_control finished");
Chad Barb's avatar
 
Chad Barb committed
873 874
	}
    }
Chad Barb's avatar
Chad Barb committed
875

Chad Barb's avatar
 
Chad Barb committed
876 877
    return 0;
}