tbswap.in 18.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10
#!/usr/bin/perl -w

#
# EMULAB-COPYRIGHT
# Copyright (c) 2000-2003 University of Utah and the Flux Group.
# All rights reserved.
#

use English;

11 12 13 14
# Returns 0 on success.
# Returns 1 on non-assign_wrapper failure.
# Returns (1 | assign_wrapper's errorcode) on assign_wrapper failure.

15 16 17 18 19
# 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
20 21 22
#
# XXX: for update, expt is swapped out on os_setup fail.
#      (we only recover if assign fails)
23 24 25

sub usage()
{
Chad Barb's avatar
Chad Barb committed
26
    print STDERR "Usage: $0 { in | out [-force] | update [-reboot] } pid eid\n";
27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51
    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.
#
52 53 54 55 56 57 58
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
59
sub UPDATE_RECOVER()  { return 0; }
60 61 62 63 64 65 66

#
# Turn off line buffering on output
#

$| = 1;

Chad Barb's avatar
Chad Barb committed
67
my $updateReboot = 0;
68 69
my $force  = 0;
my $errors = 0;
Chad Barb's avatar
Chad Barb committed
70
my $assignWrapperErrorCode = 0;
71

72 73 74 75 76 77
my $state;

my $os_setup_pid;
my $cleanvlans;

#
78
# First argument is either "in", "out", or "update";
79 80 81 82 83
# this value goes into $swapop.
#

my $swapop = shift;	

84 85 86 87
if (!$swapop || 
    (($swapop ne "in") && 
     ($swapop ne "out") &&
     ($swapop ne "update"))) {
88 89 90 91 92 93 94 95 96 97 98
    usage();
}

#
# Get other arguments.
#

while ($#ARGV > 1) {
    $arg = shift;
    if ($arg eq "-force") {
	$force = 1;
Chad Barb's avatar
Chad Barb committed
99 100
    } elsif ($arg eq "-reboot") {
	$updateReboot = 1;
101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142
    } else {
	usage();
    }
}
if ($#ARGV < 1) {
    usage();
}
my ($pid,$eid) = @ARGV;

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))) {
    print STDERR "*** No such experiment $pid/$eid\n";
    $errors = 1;
}

#
# Figure out which state we're looking for 
# and which state we're going to set according to the operation.
#

my $desiredState;
my $nextState;
my $suggestion;
my $suggestUseOfForce;

if ($swapop eq "in") {
    $desiredState = EXPTSTATE_SWAPPED;
    $nextState    = EXPTSTATE_ACTIVATING;
    $suggestion   = "Must be swapped out.";
    $suggestUseOfForce = 0;
143
} elsif ($swapop eq "out") {
144 145 146 147 148 149 150 151 152
    if (! $TESTMODE) {
	$desiredState = EXPTSTATE_ACTIVE;
	$suggestion   = "Must be running.";
    } else {
	$desiredState = EXPTSTATE_TESTING;
	$suggestion   = "Must be in testing state.";
    }
    $nextState    = EXPTSTATE_SWAPPING;
    $suggestUseOfForce = 1;
153 154 155 156 157
} elsif ($swapop eq "update") {
    $desiredState = EXPTSTATE_ACTIVE;
    $nextState    = EXPTSTATE_ACTIVATING; # XXX add an updating state?
    $suggestion   = "Must be running.";
    $suggestUseOfForce = 0;
158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199
}

#
# Check that experiment is in correct state.
# If it is not, print appropriate error message.
#

if (! $errors) {
    if ($state ne $desiredState) {
	if (! $force) {
	    if ($state eq EXPTSTATE_SWAPPED) {
		print STDERR "*** Experiment is already swapped out. $suggestion\n";
	    } elsif ($state eq EXPTSTATE_ACTIVE) {
		print STDERR "*** Experiment is already running. $suggestion\n";
	    } else {
		print STDERR "*** Experiment is in the wrong state: $state. $suggestion\n";
	    }
	    if ($suggestUseOfForce) {
		print STDERR "    Try using -force to ignore improper state.\n";
	    }
	    $errors = 1;
	} else {
	    print STDERR "*** WARNING: Ignoring improper state: $state.\n";
	}
    }
}

#
# Set intermediate experiment state.
#

if (! $errors) {
    if (! SetExpState($pid, $eid, $nextState)) {
	print STDERR "*** Failed to set intermediate experiment state.\n";
	$errors = 1;
    }
}

#
# Do actual swapping
#

200
undef $nextState;
201 202

if (! $errors) {
203
    if ($swapop eq "out") {
204
	#
205
	# Swap out
206
	#
207 208 209 210 211 212 213
	$errors = doSwapout(REAL);
	if (! $errors) {
	    $nextState = EXPTSTATE_SWAPPED;
	} else {
	    # leave $nextState undefined
	    # experiment will stay in SWAPPING.
	}
214 215

	#
216 217 218 219 220
	# Update Accounting Information
	#
	TBSetExpSwapTime($pid, $eid);
    }
}
221

Chad Barb's avatar
Chad Barb committed
222 223 224
#
# Update, Phase One -- swap experiment partially out.
#
225 226
if (! $errors) {
    if ($swapop eq "update") {
Chad Barb's avatar
Chad Barb committed
227 228 229
	print STDERR "Backing up physical state...\n";
	TBExptBackupPhysicalState($pid,$eid,$$);

230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247
	$errors = doSwapout(UPDATE);
	if (! $errors) {
	    # leave $nextState undefined. 
	    # (this will get defined below.)
	} else {
	    # Error; Clean up and put experiment in SWAPPED.
	    print STDERR "Cleaning up after errors.\n";
	    doSwapout(CLEANUP);
	    $nextState = EXPTSTATE_SWAPPED;
	}
	
	#
	# Update Accounting Information
	#
	TBSetExpSwapTime($pid, $eid);
    }
}

Chad Barb's avatar
Chad Barb committed
248 249 250 251 252
#
# Update, Phase Two -- swap experiment back in.
#
if (! $errors) { 
    if ($swapop eq "update") {
253
	#
Chad Barb's avatar
Chad Barb committed
254
	# Update
255
	#
Chad Barb's avatar
Chad Barb committed
256 257 258 259 260 261 262
	$errors = doSwapin(UPDATE);

	if (! $errors) {
	    #
	    # Update Accounting Information
	    #
	    TBSetExpSwapTime($pid, $eid);
263

Chad Barb's avatar
Chad Barb committed
264 265 266 267 268 269 270 271 272
	    #
	    # Swapin worked; exp is now ACTIVE.
	    # (or TESTING) 
	    #
	    if ($TESTMODE) {
		$nextState = EXPTSTATE_TESTING;
	    } else {
		$nextState = EXPTSTATE_ACTIVE;
	    }
273
	} else {
Chad Barb's avatar
Chad Barb committed
274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311
	    #
	    # There were errors; see if we can recover.
	    #
	    my $canStillRecover = 1;

	    if ($errors != 7) {
		print STDERR "Update failure occurred _after_ assign phase; ";
		$canStillRecover = 0;
	    }

	    if ($canStillRecover) {
		print STDERR "Recovering physical state...\n";
		my $recoverErrors = TBExptRestorePhysicalState($pid,$eid,$$);
		if ($recoverErrors) {
		    print STDERR "Could not restore backed-up physical state; ";
		    $canStillRecover = 0;
		}
	    }

	    if ($canStillRecover) {
		print STDERR "Doing a recovery swap-in of old state...\n";
		$recoverErrors = doSwapin(UPDATE_RECOVER);
		if ($recoverErrors) {
		    print STDERR "Could not swap in old physical state; ";
		    $canStillRecover = 0;
		}
	    }

	    if ($canStillRecover) {
		print STDERR "Update recovery successful.\n";
		$nextState = EXPTSTATE_ACTIVE;
	    } else {
		print STDERR "update recovery aborted!\n".
		             "Swapping experiment out...\n";
		doSwapout(CLEANUP);

		$nextState = EXPTSTATE_SWAPPED;
	    }
312
	}
Chad Barb's avatar
Chad Barb committed
313 314 315 316 317 318 319 320 321
    }
}

if (! $errors) {
    if ($swapop eq "in") {
	#
	# Swap in
	#
	$errors = doSwapin(REAL);
322

323
	my $retries = 2;
324 325 326 327

	#
	# Attempt a retry if: 
	#   a) there were errors, 
Chad Barb's avatar
Chad Barb committed
328
	#   b) doswapin() indicated (via return code 3) a retry is appropriate,
329 330
	#   c) we haven't tried too many times already.
	#
331 332 333
	# NOTE that if $TESTMODE == 1, retries are impossible, since
        # $retry will never get set to 1.
	#
Chad Barb's avatar
Chad Barb committed
334
	while ($errors == 3 && $retries ) {
335 336 337
	    $retries--;

	    print STDERR "Cleaning up after errors; will try again.\n";
338
	    doSwapout(RETRY);
339 340

	    print STDERR "Trying again...\n";
341
	    $errors = doSwapin(RETRY);
342
	}
Chad Barb's avatar
Chad Barb committed
343

344 345 346 347 348 349 350 351
	if (! $errors) {
	    #
	    # Update Accounting Information
	    #
	    TBSetExpSwapTime($pid, $eid);

	    #
	    # Swapin worked; exp is now ACTIVE.
Chad Barb's avatar
Chad Barb committed
352
	    # (or TESTING)
353
	    #
354 355 356 357 358
	    if ($TESTMODE) {
		$nextState = EXPTSTATE_TESTING;
	    } else {
		$nextState = EXPTSTATE_ACTIVE;
	    }
Chad Barb's avatar
Chad Barb committed
359
	} else {
360
	    print STDERR "Cleaning up after errors.\n";
361
	    doSwapout(CLEANUP);
362 363

	    #
Chad Barb's avatar
Chad Barb committed
364
	    # Regardless of how well cleanup swapout worked,
365 366 367 368 369 370 371 372 373 374
	    # send exp to SWAPPED.
	    #
	    $nextState = EXPTSTATE_SWAPPED; 
	}
    }
}

if (defined $nextState) {
    if (! SetExpState($pid, $eid, $nextState)) {
	print STDERR "*** Failed to set experiment state.\n";
375
	$errors = 1;
376 377 378 379 380 381 382 383 384 385
    }
}

#
# Write appropriate message and exit.
#

if ($errors) {
    print "Failingly finished swap-$swapop for $pid/$eid. " . TBTimeStamp() . "\n";
    TBDebugTimeStamp("tbswap $swapop finished (failed)");
Chad Barb's avatar
Chad Barb committed
386

387 388 389 390
    # pass assign wrapper info along.
    # other codes in 'errors' (3 or 7) are meaningless and
    # should just be reported as 1's.
    exit( 1 | $assignWrapperErrorCode );
391 392 393
} else {
    print "Successfully finished swap-$swapop for $pid/$eid. " . TBTimeStamp() . "\n";
    TBDebugTimeStamp("tbswap $swapop finished (succeeded)");
394
    exit( 0 );
395 396 397 398 399
}

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

##
Chad Barb's avatar
Chad Barb committed
400
#
401 402
# doSwapout - Swaps experiment out.
#
Chad Barb's avatar
Chad Barb committed
403
#             If in REAL or CLEANUP,
404 405 406
#             this function will free all nodes for the 
#             experiment.
#
Chad Barb's avatar
Chad Barb committed
407
#             If in RETRY or UDPATE,
408 409 410 411 412 413
#             only nodes not in RES_READY will be freed.
#
#             Returns 0 on success, >0 on failure.
#
##

414 415
sub doSwapout($) {
    my $type = shift; # REAL==4, CLEANUP==3, RETRY==2, UPDATE==1.
416 417 418 419 420 421 422 423 424 425 426
    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
427

428 429
    if (! $TESTMODE) { 
	if (! $DISABLE_EVENTS) {
430 431 432 433 434 435
	    if ($type >= RETRY) {
		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;
		}
436 437
	    }
	}
Chad Barb's avatar
Chad Barb committed
438

439
	#
Chad Barb's avatar
Chad Barb committed
440
	# Clean up any VLANs in experiment.
441
	#
Chad Barb's avatar
Chad Barb committed
442 443 444 445 446 447 448
	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;
449
	}
Chad Barb's avatar
Chad Barb committed
450 451
	TBDebugTimeStamp("snmpit finished");

452 453 454
	#
	# This is a hack. We need a more general os_teardown, but for now
	# we just kill off the vnode stuff. 
455
	# (don't kill off vnodes when UPDATEing.)
456
	#
457 458 459 460 461 462 463 464
	if ($type >= RETRY) {
	    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");
465 466 467
	}
    }

468
    if ($type >= CLEANUP) {
469 470 471
	#
	# We're not attempting a retry; 
	# remove all nodes from the experiment.
Chad Barb's avatar
Chad Barb committed
472
	# (nfree will send them to RES_FREE_DIRTY)
473 474 475 476 477 478 479 480
	#
	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");
481 482 483 484 485

	#
	# Since this is an actual swapout, 
	# reset our count of swap out nag emails sent.
	#
486 487
	DBQueryWarn("update experiments set swap_requests='' ".
		    "where eid='$eid' and pid='$pid'");
488 489
    } else {
	#
490
	# $type == RETRY or $type == UPDATE.
491 492
	# 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
493
	# (nfree will send deallocated nodes to RES_FREE_DIRTY)
494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510
	#

	my @failedNodes = ();
	print STDERR "Freeing failed nodes.\n";
	$db_result =
	    DBQueryFatal("select rv.node_id, n.allocstate ".
                         "from reserved as rv ".
			 "left join nodes as n on ".
			 "n.node_id = rv.node_id ".
			 "where rv.pid='$pid' and rv.eid='$eid'");

	while (($node,$allocstate) = $db_result->fetchrow_array) {
	    if ($allocstate ne TBDB_ALLOCSTATE_RES_READY()) {
		push(@failedNodes, $node);
	    }
	}

511 512 513
	# XXX reboot nodes [e.g., put them in RES_DIRTY()] 
	# if doing an update -restart.

514 515
	if (@failedNodes > 0) {
	    TBDebugTimeStamp("nfree started");
Chad Barb's avatar
Chad Barb committed
516 517 518 519 520 521
	    #
	    # 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.
	    #
	    if (system("nfree -x $pid $eid " . join(" ", @failedNodes))) {
522 523 524 525 526 527 528 529 530 531 532 533 534 535 536
		print STDERR "*** Could not free nodes.\n";
		$swapout_errors = 1;
	    }
	    TBDebugTimeStamp("nfree finished");
	}
    }

    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
537
	# not a failed swapin(cleanup), update, or retry.
538
	#
539
	if ($type == REAL) {
540 541 542 543 544 545 546
	    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
547

548 549 550 551
	#
	# Resetting named maps and email lists is fast and idempotent,
	# so whatever.
	#
552 553 554 555 556 557
	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
558

559 560 561 562 563 564 565 566
	print "Resetting email lists.\n";
	TBDebugTimeStamp("genelists started");
	if (system("genelists")) {
	    print "*** WARNING: Failed to reset email lists.\n";
	}
	TBDebugTimeStamp("genelists finished");
    }

567 568 569
    #
    # Wipe the DB clean.
    #
Chad Barb's avatar
Chad Barb committed
570

571
    print STDERR "Resetting DB.\n";
572
    TBExptRemovePhysicalState( $pid, $eid );
573 574 575 576 577

    return $swapout_errors;
}

##
Chad Barb's avatar
Chad Barb committed
578
#
579 580
# doSwapin - Swaps experiment in.
#
Chad Barb's avatar
Chad Barb committed
581 582 583 584 585
#            Returns:
#              0 - successful swapin
#              1 - failed swapin; cleanup required.
#              3 - failed swapin; cleanup required; can retry.
#              7 - failed swapin; assign failed; no cleanup.
586 587 588 589 590
#
#            Will set $retry = 1 if os_setup failed.
#
##

591
sub doSwapin($) {
Chad Barb's avatar
Chad Barb committed
592
    my $type = shift; # REAL==4, RETRY==2, UPDATE==1, UPDATE_RECOVER=0. 
593

594 595 596 597 598
    #
    # assign_wrapper does all the virtual to physical mapping 
    # and updating the DB state.
    #

Chad Barb's avatar
Chad Barb committed
599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616
    if ($type > UPDATE_RECOVER) {
	print "Mapping to physical reality ...\n";
	TBDebugTimeStamp("assign_wrapper started");

	#
	# Pass the -u (update) switch into assign_wrapper,
	# So any nodes already in the experiment are
	# fixed.
	#
	my $exitcode;
	if ($exitcode = system("assign_wrapper -u $pid $eid")) {
	    print STDERR "*** Failed to map to reality.\n";

	    #
	    # save this off so it will get passed back later.
	    #
	    $assignWrapperErrorCode = $exitcode;
	    if ((($exitcode >> 8) & 64) != 0) {
617 618
		# so batchexp doesn't choke.
		$assignWrapperErrorCode -= 64;
Chad Barb's avatar
Chad Barb committed
619 620 621 622 623 624
		return 7;
	    } else {
		return 1;
	    }
	}
	TBDebugTimeStamp("assign_wrapper finished");
625

Chad Barb's avatar
Chad Barb committed
626 627
	print "Mapped to physical reality!\n";
    }
628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647

    # Exit here if we are testing.
    if ($TESTMODE) {
	print "Testing run - Stopping here.\n";
	return 0;
    }

    #
    # 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
648

649 650 651 652 653 654
    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
655
	#
656 657
    }
    TBDebugTimeStamp("named finished");
Chad Barb's avatar
Chad Barb committed
658 659 660 661


    #
    # If user specified -reboot to update,
662
    # and we are successfully performing the update,
Chad Barb's avatar
Chad Barb committed
663 664 665
    # then mark all nodes in experiment
    # so os_setup will reboot them.
    #
666
    if (($type == UPDATE) && $updateReboot) {
Chad Barb's avatar
Chad Barb committed
667 668 669 670 671 672 673 674 675 676 677 678 679
	print STDERR "Marking nodes for reboot.\n";
	$db_result =
	    DBQueryFatal("select rv.node_id ".
                         "from reserved as rv ".
			 "left join nodes as n on ".
			 "n.node_id = rv.node_id ".
			 "where rv.pid='$pid' and rv.eid='$eid'");

	while (($node) = $db_result->fetchrow_array) {
	    TBSetNodeAllocState( $node, TBDB_ALLOCSTATE_RES_INIT_DIRTY() );
	}
    }

680 681 682 683 684 685 686 687 688 689 690 691
    #
    # 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";
    TBDebugTimeStamp("os_setup started");
    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
692

693 694 695 696 697 698 699 700 701 702 703 704 705 706
    #
    # 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.)
    #

    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
707

708 709 710 711
    #
    # An error now means that the VLANS need to be cleaned up.
    #
    $cleanvlans = 1;
Chad Barb's avatar
Chad Barb committed
712

713 714 715 716 717 718 719 720 721
    print "Setting up email lists.\n";
    TBDebugTimeStamp("genelists started");
    if (system("genelists")) {
	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
722

723 724 725 726 727 728 729 730 731 732 733 734 735 736
    #
    # 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");
737
    }
Chad Barb's avatar
Chad Barb committed
738

739 740 741 742 743 744 745 746 747
    #
    # 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";
	    #
748 749
	    # Use returncode from os_setup process to
	    # set global $retry flag, indicating to caller
750 751 752
	    # that it may be beneficial to attempt
	    # a doSwapin() again.
	    #
753
            if (($CHILD_ERROR >> 8) == 1) {
Chad Barb's avatar
Chad Barb committed
754
		return 3;
755 756
	    } else {
		print STDERR "Not retrying due to error type.\n";
Chad Barb's avatar
Chad Barb committed
757
		return 1;
758
	    }
759 760 761 762 763 764 765
	}
    } 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
766

767 768 769 770
    #
    # 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.
771 772
    # Don't do this during an update, since we didn't kill the 
    # event system previously, so starting it again will fail!
773 774
    # 
    if (! $DISABLE_EVENTS) {
Chad Barb's avatar
Chad Barb committed
775
	if ($type != UPDATE && $type != UPDATE_RECOVER) {
776 777 778 779 780 781 782
	    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");
783 784
	}
    }
Chad Barb's avatar
Chad Barb committed
785

786 787
    return 0;
}