tbswap.in 16.2 KB
Newer Older
Chad Barb's avatar
 
Chad Barb committed
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;

Chad Barb's avatar
 
Chad Barb committed
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

sub usage()
{
Chad Barb's avatar
 
Chad Barb committed
19
    print STDERR "Usage: $0 { in | out [-force] | update [-restart] } pid eid\n";
Chad Barb's avatar
 
Chad Barb committed
20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44
    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
45 46 47 48 49 50 51
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
52 53 54 55 56 57 58

#
# Turn off line buffering on output
#

$| = 1;

Chad Barb's avatar
 
Chad Barb committed
59
my $restart = 0;
Chad Barb's avatar
 
Chad Barb committed
60 61
my $force  = 0;
my $errors = 0;
Chad Barb's avatar
 
Chad Barb committed
62

Chad Barb's avatar
 
Chad Barb committed
63
my $retry  = 0;
Chad Barb's avatar
 
Chad Barb committed
64

Chad Barb's avatar
 
Chad Barb committed
65 66 67 68 69 70
my $state;

my $os_setup_pid;
my $cleanvlans;

#
Chad Barb's avatar
 
Chad Barb committed
71
# First argument is either "in", "out", or "update";
Chad Barb's avatar
 
Chad Barb committed
72 73 74 75 76
# this value goes into $swapop.
#

my $swapop = shift;	

Chad Barb's avatar
 
Chad Barb committed
77 78 79 80
if (!$swapop || 
    (($swapop ne "in") && 
     ($swapop ne "out") &&
     ($swapop ne "update"))) {
Chad Barb's avatar
 
Chad Barb committed
81 82 83 84 85 86 87 88 89 90 91
    usage();
}

#
# Get other arguments.
#

while ($#ARGV > 1) {
    $arg = shift;
    if ($arg eq "-force") {
	$force = 1;
Chad Barb's avatar
 
Chad Barb committed
92 93
    } elsif ($arg eq "-restart") {
	$restart = 1;
Chad Barb's avatar
 
Chad Barb committed
94 95 96 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
    } 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;
Chad Barb's avatar
 
Chad Barb committed
136
} elsif ($swapop eq "out") {
Chad Barb's avatar
 
Chad Barb committed
137 138 139 140 141 142 143 144 145
    if (! $TESTMODE) {
	$desiredState = EXPTSTATE_ACTIVE;
	$suggestion   = "Must be running.";
    } else {
	$desiredState = EXPTSTATE_TESTING;
	$suggestion   = "Must be in testing state.";
    }
    $nextState    = EXPTSTATE_SWAPPING;
    $suggestUseOfForce = 1;
Chad Barb's avatar
 
Chad Barb committed
146 147 148 149 150
} elsif ($swapop eq "update") {
    $desiredState = EXPTSTATE_ACTIVE;
    $nextState    = EXPTSTATE_ACTIVATING; # XXX add an updating state?
    $suggestion   = "Must be running.";
    $suggestUseOfForce = 0;
Chad Barb's avatar
 
Chad Barb committed
151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192
}

#
# 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
#

Chad Barb's avatar
 
Chad Barb committed
193
undef $nextState;
Chad Barb's avatar
 
Chad Barb committed
194 195

if (! $errors) {
Chad Barb's avatar
 
Chad Barb committed
196
    if ($swapop eq "out") {
Chad Barb's avatar
 
Chad Barb committed
197
	#
Chad Barb's avatar
 
Chad Barb committed
198
	# Swap out
Chad Barb's avatar
 
Chad Barb committed
199
	#
Chad Barb's avatar
 
Chad Barb committed
200 201 202 203 204 205 206
	$errors = doSwapout(REAL);
	if (! $errors) {
	    $nextState = EXPTSTATE_SWAPPED;
	} else {
	    # leave $nextState undefined
	    # experiment will stay in SWAPPING.
	}
Chad Barb's avatar
 
Chad Barb committed
207 208

	#
Chad Barb's avatar
 
Chad Barb committed
209 210 211 212 213
	# Update Accounting Information
	#
	TBSetExpSwapTime($pid, $eid);
    }
}
Chad Barb's avatar
 
Chad Barb committed
214

Chad Barb's avatar
 
Chad Barb committed
215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245
if (! $errors) {
    if ($swapop eq "update") {
	$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);
    }
}

if (! $errors) {
    if ($swapop eq "in" || $swapop eq "update") {
	#
	# Swap in
	#

	if ($swapop eq "in") {
	    $errors = doSwapin(REAL);
	} else {
	    $errors = doSwapin(UPDATE);
	}
Chad Barb's avatar
 
Chad Barb committed
246

Chad Barb's avatar
 
Chad Barb committed
247
	my $retries = 2;
Chad Barb's avatar
 
Chad Barb committed
248 249 250 251 252 253

	#
	# Attempt a retry if: 
	#   a) there were errors, 
	#   b) doswapin() indicated (via $retry) a retry is appropriate,
	#   c) we haven't tried too many times already.
Chad Barb's avatar
 
Chad Barb committed
254
	#   d) (XXX temporary) We're not doing an update.
Chad Barb's avatar
 
Chad Barb committed
255
	#
Chad Barb's avatar
 
Chad Barb committed
256 257 258
	# NOTE that if $TESTMODE == 1, retries are impossible, since
        # $retry will never get set to 1.
	#
Chad Barb's avatar
 
Chad Barb committed
259
	while ($errors && $retry && $retries && $swapop eq "in") {
Chad Barb's avatar
 
Chad Barb committed
260 261 262
	    $retries--;

	    print STDERR "Cleaning up after errors; will try again.\n";
Chad Barb's avatar
 
Chad Barb committed
263
	    doSwapout(RETRY);
Chad Barb's avatar
 
Chad Barb committed
264 265

	    print STDERR "Trying again...\n";
Chad Barb's avatar
 
Chad Barb committed
266
	    $errors = doSwapin(RETRY);
Chad Barb's avatar
 
Chad Barb committed
267 268 269 270 271 272 273 274 275 276
	}
	
	if (! $errors) {
	    #
	    # Update Accounting Information
	    #
	    TBSetExpSwapTime($pid, $eid);

	    #
	    # Swapin worked; exp is now ACTIVE.
Chad Barb's avatar
 
Chad Barb committed
277
	    # (or TESTING) 
Chad Barb's avatar
 
Chad Barb committed
278
	    #
Chad Barb's avatar
 
Chad Barb committed
279 280 281 282 283
	    if ($TESTMODE) {
		$nextState = EXPTSTATE_TESTING;
	    } else {
		$nextState = EXPTSTATE_ACTIVE;
	    }
Chad Barb's avatar
 
Chad Barb committed
284 285
	} else {	    
	    print STDERR "Cleaning up after errors.\n";
Chad Barb's avatar
 
Chad Barb committed
286

Chad Barb's avatar
 
Chad Barb committed
287
	    doSwapout(CLEANUP);
Chad Barb's avatar
 
Chad Barb committed
288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338

	    #
	    # Regardless of how well cleanup swapout worked, 
	    # 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 $swapop finished (failed)");
} else {
    print "Successfully finished swap-$swapop for $pid/$eid. " . TBTimeStamp() . "\n";
    TBDebugTimeStamp("tbswap $swapop finished (succeeded)");
}

exit($errors);

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

##
# 
# doSwapout - Swaps experiment out.
#
#             If the global $retry flag is 0,
#             this function will free all nodes for the 
#             experiment.
#
#             If the global $retry flag is 1,
#             only nodes not in RES_READY will be freed.
#
#             Returns 0 on success, >0 on failure.
#
##

Chad Barb's avatar
 
Chad Barb committed
339 340
sub doSwapout($) {
    my $type = shift; # REAL==4, CLEANUP==3, RETRY==2, UPDATE==1.
Chad Barb's avatar
 
Chad Barb committed
341 342 343 344 345 346 347 348 349 350 351 352 353 354
    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;
    }
    
    if (! $TESTMODE) { 
	if (! $DISABLE_EVENTS) {
Chad Barb's avatar
 
Chad Barb committed
355 356 357 358 359 360
	    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;
		}
Chad Barb's avatar
 
Chad Barb committed
361 362 363 364
	    }
	}
	
	#
Chad Barb's avatar
 
Chad Barb committed
365
	# If this is an actual swapout, or an update/retry, always clean up VLANs
Chad Barb's avatar
 
Chad Barb committed
366 367
	# Otherwise, only do it if swapin previously messed them up.
	#
Chad Barb's avatar
 
Chad Barb committed
368 369
	# XXX Is doing this all the time cromulent?
	if ($type != CLEANUP || $cleanvlans) {
Chad Barb's avatar
 
Chad Barb committed
370 371 372 373 374 375 376 377 378 379 380 381 382 383
	    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;
	    }
	    TBDebugTimeStamp("snmpit finished");
	}
	
	#
	# This is a hack. We need a more general os_teardown, but for now
	# we just kill off the vnode stuff. 
Chad Barb's avatar
 
Chad Barb committed
384
	# (don't kill off vnodes when UPDATEing.)
Chad Barb's avatar
 
Chad Barb committed
385
	#
Chad Barb's avatar
 
Chad Barb committed
386 387 388 389 390 391 392 393
	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");
Chad Barb's avatar
 
Chad Barb committed
394 395 396
	}
    }

Chad Barb's avatar
 
Chad Barb committed
397
    if ($type >= CLEANUP) {
Chad Barb's avatar
 
Chad Barb committed
398 399 400
	#
	# We're not attempting a retry; 
	# remove all nodes from the experiment.
Chad Barb's avatar
 
Chad Barb committed
401
	# (nfree will send them to RES_FREE_DIRTY)
Chad Barb's avatar
 
Chad Barb committed
402 403 404 405 406 407 408 409
	#
	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
410 411 412 413 414

	#
	# Since this is an actual swapout, 
	# reset our count of swap out nag emails sent.
	#
415 416
	DBQueryWarn("update experiments set swap_requests='' ".
		    "where eid='$eid' and pid='$pid'");
Chad Barb's avatar
 
Chad Barb committed
417 418
    } else {
	#
Chad Barb's avatar
 
Chad Barb committed
419
	# $type == RETRY or $type == UPDATE.
Chad Barb's avatar
 
Chad Barb committed
420 421
	# 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
422
	# (nfree will send deallocated nodes to RES_FREE_DIRTY)
Chad Barb's avatar
 
Chad Barb committed
423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439
	#

	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);
	    }
	}

Chad Barb's avatar
 
Chad Barb committed
440 441 442
	# XXX reboot nodes [e.g., put them in RES_DIRTY()] 
	# if doing an update -restart.

Chad Barb's avatar
 
Chad Barb committed
443 444
	if (@failedNodes > 0) {
	    TBDebugTimeStamp("nfree started");
Chad Barb's avatar
 
Chad Barb committed
445 446 447 448 449 450
	    #
	    # 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))) {
Chad Barb's avatar
 
Chad Barb committed
451 452 453 454 455 456 457 458 459 460 461 462 463 464 465
		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
Chad Barb's avatar
 
Chad Barb committed
466
	# not a failed swapin(cleanup), update, or retry.
Chad Barb's avatar
 
Chad Barb committed
467
	#
Chad Barb's avatar
 
Chad Barb committed
468
	if ($type == REAL) {
Chad Barb's avatar
 
Chad Barb committed
469 470 471 472 473 474 475 476
	    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
477 478 479 480
	#
	# Resetting named maps and email lists is fast and idempotent,
	# so whatever.
	#
Chad Barb's avatar
 
Chad Barb committed
481 482 483 484 485 486 487 488 489 490 491 492 493 494 495
	print "Resetting named maps.\n";
	TBDebugTimeStamp("named started");
	if (system("named_setup")) {
	    print "*** WARNING: Failed to reset named map.\n";
	}
	TBDebugTimeStamp("named finished");
	
	print "Resetting email lists.\n";
	TBDebugTimeStamp("genelists started");
	if (system("genelists")) {
	    print "*** WARNING: Failed to reset email lists.\n";
	}
	TBDebugTimeStamp("genelists finished");
    }

Chad Barb's avatar
 
Chad Barb committed
496 497 498 499
    #
    # Wipe the DB clean.
    #
    
Chad Barb's avatar
 
Chad Barb committed
500
    print STDERR "Resetting DB.\n";
Chad Barb's avatar
 
Chad Barb committed
501
    TBExptRemovePhysicalState( $pid, $eid );
Chad Barb's avatar
 
Chad Barb committed
502 503 504 505 506 507 508 509 510 511 512 513 514 515 516

    return $swapout_errors;
}

##
# 
# doSwapin - Swaps experiment in.
#
#            Returns 0 on success, 1 on failure,
#            assign_wrapper's exitcode if that failed.
#
#            Will set $retry = 1 if os_setup failed.
#
##

Chad Barb's avatar
 
Chad Barb committed
517 518 519 520 521 522 523
sub doSwapin($) {
    my $type = shift; # REAL==4, RETRY==2, UPDATE==1. (CLEANUP is nonsense.)
    #
    # By default, errors are not tagged as 'retry-suitable.'
    #
    $retry = 0;

Chad Barb's avatar
 
Chad Barb committed
524 525 526 527 528 529 530
    #
    # assign_wrapper does all the virtual to physical mapping 
    # and updating the DB state.
    #
    print "Mapping to physical reality ...\n";
    TBDebugTimeStamp("assign_wrapper started");

Chad Barb's avatar
 
Chad Barb committed
531 532 533 534 535
    #
    # Pass the -u (update) switch into assign_wrapper,
    # So any nodes already in the experiment are
    # fixed.
    #
Chad Barb's avatar
 
Chad Barb committed
536
    my $exitcode;
Chad Barb's avatar
 
Chad Barb committed
537
    if ($exitcode = system("assign_wrapper -u $pid $eid")) {
Chad Barb's avatar
 
Chad Barb committed
538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618
	print STDERR "*** Failed to map to reality.\n";
	# Pass exit code through
	return $exitcode >> 8;
    }
    TBDebugTimeStamp("assign_wrapper finished");

    print "Mapped to physical reality!\n";

    # 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");
    
    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.
	# 
    }
    TBDebugTimeStamp("named finished");
    
    #
    # 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;
    }
    
    #
    # 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");
    
    #
    # An error now means that the VLANS need to be cleaned up.
    #
    $cleanvlans = 1;
    
    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
619 620 621 622 623 624 625 626 627 628 629 630 631 632
    #
    # 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
633 634 635 636 637 638 639 640 641 642 643
    }
    
    #
    # 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
644 645
	    # Use returncode from os_setup process to
	    # set global $retry flag, indicating to caller
Chad Barb's avatar
 
Chad Barb committed
646 647 648
	    # that it may be beneficial to attempt
	    # a doSwapin() again.
	    #
Chad Barb's avatar
 
Chad Barb committed
649 650 651 652 653 654
            if (($CHILD_ERROR >> 8) == 1) {
		$retry = 1;
	    } else {
		print STDERR "Not retrying due to error type.\n";
		# leave $retry == 0.
	    }
Chad Barb's avatar
 
Chad Barb committed
655 656 657 658 659 660 661 662 663 664 665 666 667 668

	    return 1;
	}
    } else {
	undef $os_setup_pid;
	print STDERR "*** Error waiting for os_setup to finish.\n";
	return 1;
    }
    TBDebugTimeStamp("os_setup finished");
    
    #
    # 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
669 670
    # 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
671 672
    # 
    if (! $DISABLE_EVENTS) {
Chad Barb's avatar
 
Chad Barb committed
673 674 675 676 677 678 679 680
	if ($type != UPDATE) {
	    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
681 682 683 684 685 686
	}
    }
    
    return 0;
}