tbswap.in 18.5 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 15
# 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
16 17 18
#
# XXX: for update, expt is swapped out on os_setup fail.
#      (we only recover if assign fails)
19 20 21

sub usage()
{
Chad Barb's avatar
Chad Barb committed
22
    print STDERR "Usage: $0 { in | out [-force] | update [-reboot] } pid eid\n";
23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47
    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.
#
48 49 50 51 52 53 54
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
55
sub UPDATE_RECOVER()  { return 0; }
56 57 58 59 60 61 62

#
# Turn off line buffering on output
#

$| = 1;

Chad Barb's avatar
Chad Barb committed
63
my $updateReboot = 0;
64 65
my $force  = 0;
my $errors = 0;
Chad Barb's avatar
Chad Barb committed
66
my $assignWrapperErrorCode = 0;
67

68 69 70 71 72 73
my $state;

my $os_setup_pid;
my $cleanvlans;

#
74
# First argument is either "in", "out", or "update";
75 76 77 78 79
# this value goes into $swapop.
#

my $swapop = shift;	

80 81 82 83
if (!$swapop || 
    (($swapop ne "in") && 
     ($swapop ne "out") &&
     ($swapop ne "update"))) {
84 85 86 87 88 89 90 91 92 93 94
    usage();
}

#
# Get other arguments.
#

while ($#ARGV > 1) {
    $arg = shift;
    if ($arg eq "-force") {
	$force = 1;
Chad Barb's avatar
Chad Barb committed
95 96
    } elsif ($arg eq "-reboot") {
	$updateReboot = 1;
97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138
    } else {
	usage();
    }
}
if ($#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.";
    $suggestUseOfForce = 0;
139
} elsif ($swapop eq "out") {
140 141 142 143 144 145 146 147 148
    if (! $TESTMODE) {
	$desiredState = EXPTSTATE_ACTIVE;
	$suggestion   = "Must be running.";
    } else {
	$desiredState = EXPTSTATE_TESTING;
	$suggestion   = "Must be in testing state.";
    }
    $nextState    = EXPTSTATE_SWAPPING;
    $suggestUseOfForce = 1;
149 150 151 152 153
} elsif ($swapop eq "update") {
    $desiredState = EXPTSTATE_ACTIVE;
    $nextState    = EXPTSTATE_ACTIVATING; # XXX add an updating state?
    $suggestion   = "Must be running.";
    $suggestUseOfForce = 0;
154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195
}

#
# 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
#

196
undef $nextState;
197 198

if (! $errors) {
199
    if ($swapop eq "out") {
200
	#
201
	# Swap out
202
	#
203 204 205 206 207 208 209
	$errors = doSwapout(REAL);
	if (! $errors) {
	    $nextState = EXPTSTATE_SWAPPED;
	} else {
	    # leave $nextState undefined
	    # experiment will stay in SWAPPING.
	}
210 211

	#
212 213 214 215 216
	# Update Accounting Information
	#
	TBSetExpSwapTime($pid, $eid);
    }
}
217

Chad Barb's avatar
Chad Barb committed
218 219 220
#
# Update, Phase One -- swap experiment partially out.
#
221 222
if (! $errors) {
    if ($swapop eq "update") {
Chad Barb's avatar
Chad Barb committed
223 224 225
	print STDERR "Backing up physical state...\n";
	TBExptBackupPhysicalState($pid,$eid,$$);

226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243
	$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
244 245 246 247 248
#
# Update, Phase Two -- swap experiment back in.
#
if (! $errors) { 
    if ($swapop eq "update") {
249
	#
Chad Barb's avatar
Chad Barb committed
250
	# Update
251
	#
Chad Barb's avatar
Chad Barb committed
252 253 254 255 256 257 258
	$errors = doSwapin(UPDATE);

	if (! $errors) {
	    #
	    # Update Accounting Information
	    #
	    TBSetExpSwapTime($pid, $eid);
259

Chad Barb's avatar
Chad Barb committed
260 261 262 263 264 265 266 267 268
	    #
	    # Swapin worked; exp is now ACTIVE.
	    # (or TESTING) 
	    #
	    if ($TESTMODE) {
		$nextState = EXPTSTATE_TESTING;
	    } else {
		$nextState = EXPTSTATE_ACTIVE;
	    }
269
	} else {
Chad Barb's avatar
Chad Barb committed
270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307
	    #
	    # 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;
	    }
308
	}
Chad Barb's avatar
Chad Barb committed
309 310 311 312 313 314 315 316 317
    }
}

if (! $errors) {
    if ($swapop eq "in") {
	#
	# Swap in
	#
	$errors = doSwapin(REAL);
318

319
	my $retries = 2;
320 321 322 323

	#
	# Attempt a retry if: 
	#   a) there were errors, 
Chad Barb's avatar
Chad Barb committed
324
	#   b) doswapin() indicated (via return code 3) a retry is appropriate,
325 326
	#   c) we haven't tried too many times already.
	#
327 328 329
	# NOTE that if $TESTMODE == 1, retries are impossible, since
        # $retry will never get set to 1.
	#
Chad Barb's avatar
Chad Barb committed
330
	while ($errors == 3 && $retries ) {
331 332 333
	    $retries--;

	    print STDERR "Cleaning up after errors; will try again.\n";
334
	    doSwapout(RETRY);
335 336

	    print STDERR "Trying again...\n";
337
	    $errors = doSwapin(RETRY);
338
	}
Chad Barb's avatar
Chad Barb committed
339

340 341 342 343 344 345 346 347
	if (! $errors) {
	    #
	    # Update Accounting Information
	    #
	    TBSetExpSwapTime($pid, $eid);

	    #
	    # Swapin worked; exp is now ACTIVE.
Chad Barb's avatar
Chad Barb committed
348
	    # (or TESTING)
349
	    #
350 351 352 353 354
	    if ($TESTMODE) {
		$nextState = EXPTSTATE_TESTING;
	    } else {
		$nextState = EXPTSTATE_ACTIVE;
	    }
Chad Barb's avatar
Chad Barb committed
355
	} else {
356
	    print STDERR "Cleaning up after errors.\n";
357
	    doSwapout(CLEANUP);
358 359

	    #
Chad Barb's avatar
Chad Barb committed
360
	    # Regardless of how well cleanup swapout worked,
361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384
	    # send exp to SWAPPED.
	    #
	    $nextState = EXPTSTATE_SWAPPED; 
	}
    }
}

if (defined $nextState) {
    if (! SetExpState($pid, $eid, $nextState)) {
	print STDERR "*** Failed to set experiment state.\n";
	#
	# Don't obliterate a meaningful error code if this happens.
	#
	if (! $errors) { $errors = 1; }
    }
}

#
# 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
385 386 387

    # pass this info along.
    $errors |= $assignWrapperErrorCode;
388 389 390 391 392 393 394 395 396 397
} else {
    print "Successfully finished swap-$swapop for $pid/$eid. " . TBTimeStamp() . "\n";
    TBDebugTimeStamp("tbswap $swapop finished (succeeded)");
}

exit($errors);

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

##
Chad Barb's avatar
Chad Barb committed
398
#
399 400
# doSwapout - Swaps experiment out.
#
Chad Barb's avatar
Chad Barb committed
401
#             If in REAL or CLEANUP,
402 403 404
#             this function will free all nodes for the 
#             experiment.
#
Chad Barb's avatar
Chad Barb committed
405
#             If in RETRY or UDPATE,
406 407 408 409 410 411
#             only nodes not in RES_READY will be freed.
#
#             Returns 0 on success, >0 on failure.
#
##

412 413
sub doSwapout($) {
    my $type = shift; # REAL==4, CLEANUP==3, RETRY==2, UPDATE==1.
414 415 416 417 418 419 420 421 422 423 424
    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
425

426 427
    if (! $TESTMODE) { 
	if (! $DISABLE_EVENTS) {
428 429 430 431 432 433
	    if ($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;
		}
434 435
	    }
	}
Chad Barb's avatar
Chad Barb committed
436

437
	#
Chad Barb's avatar
Chad Barb committed
438
	# Clean up any VLANs in experiment.
439
	#
Chad Barb's avatar
Chad Barb committed
440 441 442 443 444 445 446
	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;
447
	}
Chad Barb's avatar
Chad Barb committed
448 449
	TBDebugTimeStamp("snmpit finished");

450 451 452
	#
	# This is a hack. We need a more general os_teardown, but for now
	# we just kill off the vnode stuff. 
453
	# (don't kill off vnodes when UPDATEing.)
454
	#
455 456 457 458 459 460 461 462
	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");
463 464 465
	}
    }

466
    if ($type >= CLEANUP) {
467 468 469
	#
	# We're not attempting a retry; 
	# remove all nodes from the experiment.
Chad Barb's avatar
Chad Barb committed
470
	# (nfree will send them to RES_FREE_DIRTY)
471 472 473 474 475 476 477 478
	#
	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");
479 480 481 482 483

	#
	# Since this is an actual swapout, 
	# reset our count of swap out nag emails sent.
	#
484 485
	DBQueryWarn("update experiments set swap_requests='' ".
		    "where eid='$eid' and pid='$pid'");
486 487
    } else {
	#
488
	# $type == RETRY or $type == UPDATE.
489 490
	# 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
491
	# (nfree will send deallocated nodes to RES_FREE_DIRTY)
492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508
	#

	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);
	    }
	}

509 510 511
	# XXX reboot nodes [e.g., put them in RES_DIRTY()] 
	# if doing an update -restart.

512 513
	if (@failedNodes > 0) {
	    TBDebugTimeStamp("nfree started");
Chad Barb's avatar
Chad Barb committed
514 515 516 517 518 519
	    #
	    # 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))) {
520 521 522 523 524 525 526 527 528 529 530 531 532 533 534
		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
535
	# not a failed swapin(cleanup), update, or retry.
536
	#
537
	if ($type == REAL) {
538 539 540 541 542 543 544
	    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
545

546 547 548 549
	#
	# Resetting named maps and email lists is fast and idempotent,
	# so whatever.
	#
550 551 552 553 554 555
	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
556

557 558 559 560 561 562 563 564
	print "Resetting email lists.\n";
	TBDebugTimeStamp("genelists started");
	if (system("genelists")) {
	    print "*** WARNING: Failed to reset email lists.\n";
	}
	TBDebugTimeStamp("genelists finished");
    }

565 566 567
    #
    # Wipe the DB clean.
    #
Chad Barb's avatar
Chad Barb committed
568

569
    print STDERR "Resetting DB.\n";
570
    TBExptRemovePhysicalState( $pid, $eid );
571 572 573 574 575

    return $swapout_errors;
}

##
Chad Barb's avatar
Chad Barb committed
576
#
577 578
# doSwapin - Swaps experiment in.
#
Chad Barb's avatar
Chad Barb committed
579 580 581 582 583
#            Returns:
#              0 - successful swapin
#              1 - failed swapin; cleanup required.
#              3 - failed swapin; cleanup required; can retry.
#              7 - failed swapin; assign failed; no cleanup.
584 585 586 587 588
#
#            Will set $retry = 1 if os_setup failed.
#
##

589
sub doSwapin($) {
Chad Barb's avatar
Chad Barb committed
590
    my $type = shift; # REAL==4, RETRY==2, UPDATE==1, UPDATE_RECOVER=0. 
591

592 593 594 595 596
    #
    # assign_wrapper does all the virtual to physical mapping 
    # and updating the DB state.
    #

Chad Barb's avatar
Chad Barb committed
597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614
    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) {
615 616
		# so batchexp doesn't choke.
		$assignWrapperErrorCode -= 64;
Chad Barb's avatar
Chad Barb committed
617 618 619 620 621 622
		return 7;
	    } else {
		return 1;
	    }
	}
	TBDebugTimeStamp("assign_wrapper finished");
623

Chad Barb's avatar
Chad Barb committed
624 625
	print "Mapped to physical reality!\n";
    }
626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645

    # 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
646

647 648 649 650 651 652
    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
653
	#
654 655
    }
    TBDebugTimeStamp("named finished");
Chad Barb's avatar
Chad Barb committed
656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677


    #
    # If user specified -reboot to update,
    # then mark all nodes in experiment
    # so os_setup will reboot them.
    #
    if (($type == UPDATE ||
         $type == UPDATE_RECOVER) && $updateReboot) {
	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() );
	}
    }

678 679 680 681 682 683 684 685 686 687 688 689
    #
    # 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
690

691 692 693 694 695 696 697 698 699 700 701 702 703 704
    #
    # 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
705

706 707 708 709
    #
    # An error now means that the VLANS need to be cleaned up.
    #
    $cleanvlans = 1;
Chad Barb's avatar
Chad Barb committed
710

711 712 713 714 715 716 717 718 719
    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
720

721 722 723 724 725 726 727 728 729 730 731 732 733 734
    #
    # 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");
735
    }
Chad Barb's avatar
Chad Barb committed
736

737 738 739 740 741 742 743 744 745
    #
    # 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";
	    #
746 747
	    # Use returncode from os_setup process to
	    # set global $retry flag, indicating to caller
748 749 750
	    # that it may be beneficial to attempt
	    # a doSwapin() again.
	    #
751
            if (($CHILD_ERROR >> 8) == 1) {
Chad Barb's avatar
Chad Barb committed
752
		return 3;
753 754
	    } else {
		print STDERR "Not retrying due to error type.\n";
Chad Barb's avatar
Chad Barb committed
755
		return 1;
756
	    }
757 758 759 760 761 762 763
	}
    } 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
764

765 766 767 768
    #
    # 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.
769 770
    # Don't do this during an update, since we didn't kill the 
    # event system previously, so starting it again will fail!
771 772
    # 
    if (! $DISABLE_EVENTS) {
Chad Barb's avatar
Chad Barb committed
773
	if ($type != UPDATE && $type != UPDATE_RECOVER) {
774 775 776 777 778 779 780
	    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");
781 782
	}
    }
Chad Barb's avatar
Chad Barb committed
783

784 785
    return 0;
}