tbswap.in 20.6 KB
Newer Older
1 2 3 4 5 6 7 8 9
#!/usr/bin/perl -w

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

10 11 12 13
# Returns 0 on success.
# Returns 1 on non-assign_wrapper failure.
# Returns (1 | assign_wrapper's errorcode) on assign_wrapper failure.

14 15 16 17 18
# 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
19 20 21
#
# XXX: for update, expt is swapped out on os_setup fail.
#      (we only recover if assign fails)
22 23 24

sub usage()
{
Chad Barb's avatar
Chad Barb committed
25
    print STDERR "Usage: $0 { in | out [-force] | update [-reboot] } pid eid\n";
26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
    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.
#
51 52 53 54 55 56 57
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
58
sub UPDATE_RECOVER()  { return 0; }
59 60 61 62 63 64 65

#
# Turn off line buffering on output
#

$| = 1;

Chad Barb's avatar
Chad Barb committed
66
my $updateReboot = 0;
67 68
my $force  = 0;
my $errors = 0;
Chad Barb's avatar
Chad Barb committed
69
my $assignWrapperErrorCode = 0;
70

71 72 73 74 75 76
my $state;

my $os_setup_pid;
my $cleanvlans;

#
77
# First argument is either "in", "out", or "update";
78 79 80 81 82
# this value goes into $swapop.
#

my $swapop = shift;	

83 84 85 86
if (!$swapop || 
    (($swapop ne "in") && 
     ($swapop ne "out") &&
     ($swapop ne "update"))) {
87 88 89 90 91 92 93 94 95 96 97
    usage();
}

#
# Get other arguments.
#

while ($#ARGV > 1) {
    $arg = shift;
    if ($arg eq "-force") {
	$force = 1;
Chad Barb's avatar
Chad Barb committed
98 99
    } elsif ($arg eq "-reboot") {
	$updateReboot = 1;
100 101 102 103 104 105 106 107 108
    } else {
	usage();
    }
}
if ($#ARGV < 1) {
    usage();
}
my ($pid,$eid) = @ARGV;

109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124
#
# 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");
}

125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157
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.";
    $suggestUseOfForce = 0;
158
} elsif ($swapop eq "out") {
159 160 161 162 163 164 165 166 167
    if (! $TESTMODE) {
	$desiredState = EXPTSTATE_ACTIVE;
	$suggestion   = "Must be running.";
    } else {
	$desiredState = EXPTSTATE_TESTING;
	$suggestion   = "Must be in testing state.";
    }
    $nextState    = EXPTSTATE_SWAPPING;
    $suggestUseOfForce = 1;
168 169 170 171 172
} elsif ($swapop eq "update") {
    $desiredState = EXPTSTATE_ACTIVE;
    $nextState    = EXPTSTATE_ACTIVATING; # XXX add an updating state?
    $suggestion   = "Must be running.";
    $suggestUseOfForce = 0;
173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214
}

#
# 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
#

215
undef $nextState;
216 217

if (! $errors) {
218
    if ($swapop eq "out") {
219
	#
220
	# Swap out
221
	#
222 223 224 225 226 227 228
	$errors = doSwapout(REAL);
	if (! $errors) {
	    $nextState = EXPTSTATE_SWAPPED;
	} else {
	    # leave $nextState undefined
	    # experiment will stay in SWAPPING.
	}
229 230

	#
231 232 233 234 235
	# Update Accounting Information
	#
	TBSetExpSwapTime($pid, $eid);
    }
}
236

Chad Barb's avatar
Chad Barb committed
237 238 239
#
# Update, Phase One -- swap experiment partially out.
#
240 241
if (! $errors) {
    if ($swapop eq "update") {
Chad Barb's avatar
Chad Barb committed
242 243 244
	print STDERR "Backing up physical state...\n";
	TBExptBackupPhysicalState($pid,$eid,$$);

245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262
	$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
263 264 265 266 267
#
# Update, Phase Two -- swap experiment back in.
#
if (! $errors) { 
    if ($swapop eq "update") {
268
	#
Chad Barb's avatar
Chad Barb committed
269
	# Update
270
	#
Chad Barb's avatar
Chad Barb committed
271 272 273 274 275 276 277
	$errors = doSwapin(UPDATE);

	if (! $errors) {
	    #
	    # Update Accounting Information
	    #
	    TBSetExpSwapTime($pid, $eid);
278

Chad Barb's avatar
Chad Barb committed
279 280 281 282 283 284 285 286 287
	    #
	    # Swapin worked; exp is now ACTIVE.
	    # (or TESTING) 
	    #
	    if ($TESTMODE) {
		$nextState = EXPTSTATE_TESTING;
	    } else {
		$nextState = EXPTSTATE_ACTIVE;
	    }
288
	} else {
Chad Barb's avatar
Chad Barb committed
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
	    #
	    # 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;
	    }
327
	}
Chad Barb's avatar
Chad Barb committed
328 329 330 331 332 333 334 335 336
    }
}

if (! $errors) {
    if ($swapop eq "in") {
	#
	# Swap in
	#
	$errors = doSwapin(REAL);
337

338
	my $retries = 2;
339 340 341 342

	#
	# Attempt a retry if: 
	#   a) there were errors, 
Chad Barb's avatar
Chad Barb committed
343
	#   b) doswapin() indicated (via return code 3) a retry is appropriate,
344 345
	#   c) we haven't tried too many times already.
	#
346 347 348
	# NOTE that if $TESTMODE == 1, retries are impossible, since
        # $retry will never get set to 1.
	#
Chad Barb's avatar
Chad Barb committed
349
	while ($errors == 3 && $retries ) {
350 351 352
	    $retries--;

	    print STDERR "Cleaning up after errors; will try again.\n";
353
	    doSwapout(RETRY);
354 355

	    print STDERR "Trying again...\n";
356
	    $errors = doSwapin(RETRY);
357
	}
Chad Barb's avatar
Chad Barb committed
358

359 360 361 362 363 364 365 366
	if (! $errors) {
	    #
	    # Update Accounting Information
	    #
	    TBSetExpSwapTime($pid, $eid);

	    #
	    # Swapin worked; exp is now ACTIVE.
Chad Barb's avatar
Chad Barb committed
367
	    # (or TESTING)
368
	    #
369 370 371 372 373
	    if ($TESTMODE) {
		$nextState = EXPTSTATE_TESTING;
	    } else {
		$nextState = EXPTSTATE_ACTIVE;
	    }
Chad Barb's avatar
Chad Barb committed
374
	} else {
375
	    print STDERR "Cleaning up after errors.\n";
376
	    doSwapout(CLEANUP);
377 378

	    #
Chad Barb's avatar
Chad Barb committed
379
	    # Regardless of how well cleanup swapout worked,
380 381 382 383 384 385 386 387 388 389
	    # send exp to SWAPPED.
	    #
	    $nextState = EXPTSTATE_SWAPPED; 
	}
    }
}

if (defined $nextState) {
    if (! SetExpState($pid, $eid, $nextState)) {
	print STDERR "*** Failed to set experiment state.\n";
390
	$errors = 1;
391 392 393 394 395 396 397 398 399 400
    }
}

#
# 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
401

402 403 404 405
    # pass assign wrapper info along.
    # other codes in 'errors' (3 or 7) are meaningless and
    # should just be reported as 1's.
    exit( 1 | $assignWrapperErrorCode );
406 407 408
} else {
    print "Successfully finished swap-$swapop for $pid/$eid. " . TBTimeStamp() . "\n";
    TBDebugTimeStamp("tbswap $swapop finished (succeeded)");
409
    exit( 0 );
410 411 412 413 414
}

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

##
Chad Barb's avatar
Chad Barb committed
415
#
416 417
# doSwapout - Swaps experiment out.
#
Chad Barb's avatar
Chad Barb committed
418
#             If in REAL or CLEANUP,
419 420 421
#             this function will free all nodes for the 
#             experiment.
#
Chad Barb's avatar
Chad Barb committed
422
#             If in RETRY or UDPATE,
423 424 425 426 427 428
#             only nodes not in RES_READY will be freed.
#
#             Returns 0 on success, >0 on failure.
#
##

429 430
sub doSwapout($) {
    my $type = shift; # REAL==4, CLEANUP==3, RETRY==2, UPDATE==1.
431 432 433 434 435 436 437 438 439 440 441
    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
442

443 444
    if (! $TESTMODE) { 
	if (! $DISABLE_EVENTS) {
445 446 447 448 449 450
	    if ($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;
		}
451 452
	    }
	}
Chad Barb's avatar
Chad Barb committed
453

454
	#
Chad Barb's avatar
Chad Barb committed
455
	# Clean up any VLANs in experiment.
456
	#
Chad Barb's avatar
Chad Barb committed
457 458 459 460 461 462 463
	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;
464
	}
Chad Barb's avatar
Chad Barb committed
465 466
	TBDebugTimeStamp("snmpit finished");

467 468 469
	#
	# This is a hack. We need a more general os_teardown, but for now
	# we just kill off the vnode stuff. 
470
	# (don't kill off vnodes when UPDATEing.)
471
	#
472 473 474 475 476 477 478 479
	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");
480 481 482
	}
    }

483
    if ($type >= CLEANUP) {
484 485 486
	#
	# We're not attempting a retry; 
	# remove all nodes from the experiment.
Chad Barb's avatar
Chad Barb committed
487
	# (nfree will send them to RES_FREE_DIRTY)
488 489 490 491 492 493 494 495
	#
	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");
496 497 498 499 500

	#
	# Since this is an actual swapout, 
	# reset our count of swap out nag emails sent.
	#
501 502
	DBQueryWarn("update experiments set swap_requests='' ".
		    "where eid='$eid' and pid='$pid'");
503 504
    } else {
	#
505
	# $type == RETRY or $type == UPDATE.
506 507
	# 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
508
	# (nfree will send deallocated nodes to RES_FREE_DIRTY)
509 510 511 512 513 514 515 516 517 518
	#

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

519
	while (my ($node,$allocstate) = $db_result->fetchrow_array) {
520 521 522 523 524
	    if ($allocstate ne TBDB_ALLOCSTATE_RES_READY()) {
		push(@failedNodes, $node);
	    }
	}

525 526 527
	# XXX reboot nodes [e.g., put them in RES_DIRTY()] 
	# if doing an update -restart.

528
	if (@failedNodes > 0) {
529 530
	    print STDERR "Freeing failed nodes.\n";
	    
531
	    TBDebugTimeStamp("nfree started");
Chad Barb's avatar
Chad Barb committed
532 533 534 535 536 537
	    #
	    # 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))) {
538 539 540 541 542 543 544 545 546 547 548 549 550 551 552
		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
553
	# not a failed swapin(cleanup), update, or retry.
554
	#
555
	if ($type == REAL) {
556 557 558 559 560 561 562
	    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
563

564 565 566 567
	#
	# Resetting named maps and email lists is fast and idempotent,
	# so whatever.
	#
568 569 570 571 572 573
	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
574

575 576 577 578 579 580 581 582
	print "Resetting email lists.\n";
	TBDebugTimeStamp("genelists started");
	if (system("genelists")) {
	    print "*** WARNING: Failed to reset email lists.\n";
	}
	TBDebugTimeStamp("genelists finished");
    }

583 584 585
    #
    # Wipe the DB clean.
    #
Chad Barb's avatar
Chad Barb committed
586

587
    print STDERR "Resetting DB.\n";
588
    TBExptRemovePhysicalState( $pid, $eid );
589 590 591 592 593

    return $swapout_errors;
}

##
Chad Barb's avatar
Chad Barb committed
594
#
595 596
# doSwapin - Swaps experiment in.
#
Chad Barb's avatar
Chad Barb committed
597 598 599 600 601
#            Returns:
#              0 - successful swapin
#              1 - failed swapin; cleanup required.
#              3 - failed swapin; cleanup required; can retry.
#              7 - failed swapin; assign failed; no cleanup.
602 603 604 605 606
#
#            Will set $retry = 1 if os_setup failed.
#
##

607
sub doSwapin($) {
Chad Barb's avatar
Chad Barb committed
608
    my $type = shift; # REAL==4, RETRY==2, UPDATE==1, UPDATE_RECOVER=0. 
609

610 611 612 613 614
    #
    # assign_wrapper does all the virtual to physical mapping 
    # and updating the DB state.
    #

Chad Barb's avatar
Chad Barb committed
615 616 617 618 619 620 621 622 623 624
    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;
625
	my $wrapper = (($pid eq "testbed" || $pid eq "tbmeasure") ?
626 627 628
		       "assign_wrapper-new" : "assign_wrapper");
	
	if ($exitcode = system("$wrapper -u $pid $eid")) {
Chad Barb's avatar
Chad Barb committed
629 630 631 632 633 634 635
	    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) {
636 637
		# so batchexp doesn't choke.
		$assignWrapperErrorCode -= 64;
Chad Barb's avatar
Chad Barb committed
638 639 640 641 642 643
		return 7;
	    } else {
		return 1;
	    }
	}
	TBDebugTimeStamp("assign_wrapper finished");
644

Chad Barb's avatar
Chad Barb committed
645 646
	print "Mapped to physical reality!\n";
    }
647

648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685
    #
    # 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");
		
686
		if (system("vnode_setup -k $pid $eid @virtnodes")) {
687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705
		    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;
		}
	    }
	}
    }

706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724
    # 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
725

726 727 728 729 730 731
    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
732
	#
733 734
    }
    TBDebugTimeStamp("named finished");
Chad Barb's avatar
Chad Barb committed
735 736 737 738


    #
    # If user specified -reboot to update,
739
    # and we are successfully performing the update,
Chad Barb's avatar
Chad Barb committed
740 741 742
    # then mark all nodes in experiment
    # so os_setup will reboot them.
    #
743
    if (($type == UPDATE) && $updateReboot) {
Chad Barb's avatar
Chad Barb committed
744 745 746 747 748 749 750 751
	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'");

752
	while (my ($node) = $db_result->fetchrow_array) {
Chad Barb's avatar
Chad Barb committed
753 754 755 756
	    TBSetNodeAllocState( $node, TBDB_ALLOCSTATE_RES_INIT_DIRTY() );
	}
    }

757 758 759 760 761 762 763 764 765 766 767 768
    #
    # 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
769

770
    #
771 772 773 774 775
    # 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.)
776 777 778 779 780 781 782 783 784
    #

    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
785

786 787 788 789
    #
    # An error now means that the VLANS need to be cleaned up.
    #
    $cleanvlans = 1;
Chad Barb's avatar
Chad Barb committed
790

791 792 793 794 795 796 797 798 799
    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
800

801 802 803 804 805 806 807 808 809 810 811 812 813 814
    #
    # 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");
815
    }
Chad Barb's avatar
Chad Barb committed
816

817 818 819 820 821 822 823 824 825
    #
    # 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";
	    #
826 827
	    # Use returncode from os_setup process to
	    # set global $retry flag, indicating to caller
828 829 830
	    # that it may be beneficial to attempt
	    # a doSwapin() again.
	    #
831
            if (($CHILD_ERROR >> 8) == 1) {
Chad Barb's avatar
Chad Barb committed
832
		return 3;
833 834
	    } else {
		print STDERR "Not retrying due to error type.\n";
Chad Barb's avatar
Chad Barb committed
835
		return 1;
836
	    }
837 838 839 840 841 842 843
	}
    } 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
844

845 846 847 848
    #
    # 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.
849 850
    # Don't do this during an update, since we didn't kill the 
    # event system previously, so starting it again will fail!
851 852
    # 
    if (! $DISABLE_EVENTS) {
Chad Barb's avatar
Chad Barb committed
853
	if ($type != UPDATE && $type != UPDATE_RECOVER) {
854 855 856 857 858 859 860
	    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");
861 862
	}
    }
Chad Barb's avatar
Chad Barb committed
863

864 865
    return 0;
}