tbswap.in 21.1 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
# Returns 0 on success.
# Returns 1 on non-assign_wrapper failure.
# Returns (1 | assign_wrapper's errorcode) on assign_wrapper failure.
13 14
# Returns |0x40 if update caused a swapout. Icky.
# Returns -1 on uncontrolled error (die called).
15

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

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

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

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

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

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

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

#
# Turn off line buffering on output
#

$| = 1;

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

#
80
# First argument is either "in", "out", or "update";
81 82 83 84 85
# this value goes into $swapop.
#

my $swapop = shift;	

86 87 88 89
if (!$swapop || 
    (($swapop ne "in") && 
     ($swapop ne "out") &&
     ($swapop ne "update"))) {
90 91 92 93 94 95 96 97 98 99 100
    usage();
}

#
# Get other arguments.
#

while ($#ARGV > 1) {
    $arg = shift;
    if ($arg eq "-force") {
	$force = 1;
Chad Barb's avatar
Chad Barb committed
101 102
    } elsif ($arg eq "-reboot") {
	$updateReboot = 1;
103 104 105 106 107 108 109 110 111
    } else {
	usage();
    }
}
if ($#ARGV < 1) {
    usage();
}
my ($pid,$eid) = @ARGV;

112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127
#
# 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");
}

128 129 130 131 132 133 134 135 136 137 138 139 140
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))) {
141 142
    die("*** $0:\n".
	"    No such experiment $pid/$eid\n");
143
}
144 145 146 147 148 149
# Sanity check the current state. 
if (!$force) {
    if ($swapop eq "in") {
	die("*** $0:\n".
	    "    Experiment should be ACTIVATING. Currently $state.\n")
	    if ($state ne EXPTSTATE_ACTIVATING);
150
    }
151 152 153 154
    elsif ($swapop eq "out") {
	die("*** $0:\n".
	    "    Experiment should be SWAPPING. Currently $state.\n")
	    if ($state ne EXPTSTATE_SWAPPING);
155
    }
156 157 158 159
    elsif ($swapop eq "update") {
	die("*** $0:\n".
	    "    Experiment should be MODIFY_RESWAP. Currently $state.\n")
	    if ($state ne EXPTSTATE_MODIFY_RESWAP);
160 161 162 163 164 165
    }
}

#
# Do actual swapping
#
166 167 168 169 170
if ($swapop eq "out") {
    #
    # Swap out
    #
    $errors = doSwapout(REAL);
171
}
172 173 174 175 176 177 178 179
elsif ($swapop eq "update") {
    #
    # Update.
    #
    # Phase One -- swap experiment partially out.
    #
    print STDERR "Backing up physical state...\n";
    TBExptBackupPhysicalState($pid,$eid);
180

181
    $errors = doSwapout(UPDATE);
Chad Barb's avatar
Chad Barb committed
182

183
    if ($errors) {
184
	#
185 186 187 188 189
	# Clean up the mess, leaving the experiment in the SWAPPED state,
	# 
	print STDERR "Cleaning up after errors.\n";
	doSwapout(CLEANUP);
	$updatehosed = 1;
190
    }
191
    else {
192
	#
193
	# Phase Two -- swap experiment back in.
194
	#
Chad Barb's avatar
Chad Barb committed
195 196
	$errors = doSwapin(UPDATE);

197
	if ($errors) {
Chad Barb's avatar
Chad Barb committed
198 199 200
	    #
	    # There were errors; see if we can recover.
	    #
201
	    my $NoRecover = 0;
Chad Barb's avatar
Chad Barb committed
202 203 204

	    if ($errors != 7) {
		print STDERR "Update failure occurred _after_ assign phase; ";
205
		$NoRecover = 1;
Chad Barb's avatar
Chad Barb committed
206 207
	    }

208 209 210
	    if (! $NoRecover) {
		print STDERR "Recovering physical state.\n";
		if (($NoRecover = TBExptRestorePhysicalState($pid,$eid))) {
Chad Barb's avatar
Chad Barb committed
211 212 213
		    print STDERR "Could not restore backed-up physical state; ";
		}
	    }
214 215 216
	    if (! $NoRecover) {
		print STDERR "Doing a recovery swap-in of old state.\n";
		if (($NoRecover = doSwapin(UPDATE_RECOVER))) {
Chad Barb's avatar
Chad Barb committed
217 218 219
		    print STDERR "Could not swap in old physical state; ";
		}
	    }
220 221
	    if ($NoRecover) {
		print STDERR "Recovery aborted! Swapping experiment out.\n";
Chad Barb's avatar
Chad Barb committed
222
		doSwapout(CLEANUP);
223 224 225 226
		$updatehosed = 1;
	    }
	    else {
		print STDERR "Update recovery successful.\n";
Chad Barb's avatar
Chad Barb committed
227
	    }
228
	}
Chad Barb's avatar
Chad Barb committed
229 230
    }
}
231 232 233 234 235 236 237
elsif ($swapop eq "in") {
    #
    # Swap in
    #
    my $retries = 2;
    
    $errors = doSwapin(REAL);
Chad Barb's avatar
Chad Barb committed
238

239 240 241 242 243 244 245 246 247 248
    #
    # Attempt a retry if: 
    #   a) there were errors, 
    #   b) doswapin() indicated (via return code 3) a retry is appropriate,
    #   c) we haven't tried too many times already.
    #   d) The cancelflag has not been set.
    #   e) $TESTMODE == 0.
    #
    while ($errors == 3 && $retries && !$canceled && !$TESTMODE) {
	$retries--;
249

250 251
	print STDERR "Cleaning up after errors; will try again.\n";
	doSwapout(RETRY);
252

253 254
	print STDERR "Trying again...\n";
	$errors = doSwapin(RETRY);
255
    }
256 257 258 259
    if ($errors || $canceled) {
	print STDERR "Cleaning up after " .
	    ($errors ? "errors" : "cancelation") . ".\n";
	doSwapout(CLEANUP);
260 261 262 263 264 265 266
    }
}

#
# Write appropriate message and exit.
#
if ($errors) {
267
    print "Failingly finished swap-$swapop for $pid/$eid. " .TBTimeStamp(). "\n";
268
    TBDebugTimeStamp("tbswap $swapop finished (failed)");
Chad Barb's avatar
Chad Barb committed
269

270 271 272
    # pass assign wrapper info along.
    # other codes in 'errors' (3 or 7) are meaningless and
    # should just be reported as 1's.
273
    exit(1 | $assignWrapperErrorCode | ($updatehosed ? 0x40 : 0));
274
}
275 276 277
print "Successfully finished swap-$swapop for $pid/$eid. " . TBTimeStamp() ."\n";
TBDebugTimeStamp("tbswap $swapop finished (succeeded)");
exit(0);
278 279 280 281

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

##
Chad Barb's avatar
Chad Barb committed
282
#
283 284
# doSwapout - Swaps experiment out.
#
Chad Barb's avatar
Chad Barb committed
285
#             If in REAL or CLEANUP,
286 287 288
#             this function will free all nodes for the 
#             experiment.
#
Chad Barb's avatar
Chad Barb committed
289
#             If in RETRY or UDPATE,
290 291 292 293 294 295
#             only nodes not in RES_READY will be freed.
#
#             Returns 0 on success, >0 on failure.
#
##

296 297
sub doSwapout($) {
    my $type = shift; # REAL==4, CLEANUP==3, RETRY==2, UPDATE==1.
298 299 300 301 302 303 304 305 306 307 308
    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
309

310 311
    if (! $TESTMODE) { 
	if (! $DISABLE_EVENTS) {
312 313 314 315 316 317
	    if ($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;
		}
318 319
	    }
	}
Chad Barb's avatar
Chad Barb committed
320

321
	#
Chad Barb's avatar
Chad Barb committed
322
	# Clean up any VLANs in experiment.
323
	#
Chad Barb's avatar
Chad Barb committed
324 325 326 327 328 329 330
	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;
331
	}
Chad Barb's avatar
Chad Barb committed
332
	TBDebugTimeStamp("snmpit finished");
333
    }
Chad Barb's avatar
Chad Barb committed
334

335 336 337
    if ($type >= CLEANUP) {
	#
	# We're not attempting a retry;
338
	#
339
	# Stop all of the vnodes.
340
	#
341
	if (! $TESTMODE) { 	
342 343 344 345 346 347 348
	    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");
349 350 351 352
	}

	#
	# remove all nodes from the experiment.
Chad Barb's avatar
Chad Barb committed
353
	# (nfree will send them to RES_FREE_DIRTY)
354 355 356 357 358 359 360 361
	#
	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");
362 363 364 365 366

	#
	# Since this is an actual swapout, 
	# reset our count of swap out nag emails sent.
	#
367 368
	DBQueryWarn("update experiments set swap_requests='' ".
		    "where eid='$eid' and pid='$pid'");
369 370
    } else {
	#
371
	# $type == RETRY or $type == UPDATE.
372 373
	# 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
374
	# (nfree will send deallocated nodes to RES_FREE_DIRTY)
375
	#
376 377 378 379 380 381 382 383
	my @failedpnodes = ();
	my @failedvnodes = ();
	
	my $db_result =
	    DBQueryFatal("select rv.node_id,n.allocstate,nt.isvirtnode ".
                         "  from reserved as rv ".
			 "left join nodes as n on n.node_id = rv.node_id ".
			 "left join node_types as nt on nt.type=n.type ".
384 385
			 "where rv.pid='$pid' and rv.eid='$eid'");

386
	while (my ($node,$allocstate,$isvirt) = $db_result->fetchrow_array) {
387
	    if ($allocstate ne TBDB_ALLOCSTATE_RES_READY()) {
388 389 390 391 392 393
		if ($isvirt) {
		    push(@failedvnodes, $node);
		}
		else {
		    push(@failedpnodes, $node);
		}
394 395 396
	    }
	}

397 398 399 400 401 402 403 404 405 406 407 408
	#
	# Tear down failed vnodes. Perhaps not needed?
	# 
	if (!$TESTMODE && @failedvnodes > 0) {
	    print "Tearing down failed virtual nodes.\n";
	    TBDebugTimeStamp("vnode_setup -k started");
	    if (system("vnode_setup -d -k $pid $eid @failedvnodes")) {
		print STDERR "*** Failed to tear down vnodes.\n";
		$swapout_errors = 1;
	    }
	    TBDebugTimeStamp("vnode_setup -k finished");
	}
409

410 411 412 413
	#
	# Release all failed nodes.
	# 
	if (@failedpnodes > 0 || @failedvnodes > 0) {
414 415
	    print STDERR "Freeing failed nodes.\n";
	    
416
	    TBDebugTimeStamp("nfree started");
Chad Barb's avatar
Chad Barb committed
417 418 419 420 421
	    #
	    # 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.
	    #
422 423
	    if (system("nfree -x $pid $eid " .
		       join(" ", (@failedpnodes, @failedvnodes)))) {
424 425 426 427 428 429 430
		print STDERR "*** Could not free nodes.\n";
		$swapout_errors = 1;
	    }
	    TBDebugTimeStamp("nfree finished");
	}
    }

431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454
    if (! $TESTMODE) {
	#
	# If the experiment has no Plab dslice nodes left, but still has
	# a Plab slice, destroy the slice
	#

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

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

	    if (!$db_result->numrows) {
		print "Destroying Planetlab slice.\n";
		TBDebugTimeStamp("plabslice destroy started");
455
		if (system("plabslice destroy $pid $eid")) {
456 457 458 459 460 461 462 463
		    print STDERR "*** Failed to destroy Plab dslice\n";
		    $swapout_errors = 1;
		}
		TBDebugTimeStamp("plabslice destroy finished");
	    }
	}
    }

464 465 466 467 468 469 470 471
    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
472
	# not a failed swapin(cleanup), update, or retry.
473
	#
474
	if ($type == REAL) {
475 476 477 478 479 480 481
	    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
482

483 484 485 486
	#
	# Resetting named maps and email lists is fast and idempotent,
	# so whatever.
	#
487 488 489 490 491 492
	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
493

494 495 496 497 498 499 500 501
	print "Resetting email lists.\n";
	TBDebugTimeStamp("genelists started");
	if (system("genelists")) {
	    print "*** WARNING: Failed to reset email lists.\n";
	}
	TBDebugTimeStamp("genelists finished");
    }

502 503 504
    #
    # Wipe the DB clean.
    #
Chad Barb's avatar
Chad Barb committed
505

506
    print STDERR "Resetting DB.\n";
507
    TBExptRemovePhysicalState( $pid, $eid );
508 509 510 511 512

    return $swapout_errors;
}

##
Chad Barb's avatar
Chad Barb committed
513
#
514 515
# doSwapin - Swaps experiment in.
#
Chad Barb's avatar
Chad Barb committed
516 517 518 519 520
#            Returns:
#              0 - successful swapin
#              1 - failed swapin; cleanup required.
#              3 - failed swapin; cleanup required; can retry.
#              7 - failed swapin; assign failed; no cleanup.
521 522
##

523
sub doSwapin($) {
Chad Barb's avatar
Chad Barb committed
524
    my $type = shift; # REAL==4, RETRY==2, UPDATE==1, UPDATE_RECOVER=0. 
525

526 527 528 529 530
    #
    # assign_wrapper does all the virtual to physical mapping 
    # and updating the DB state.
    #

Chad Barb's avatar
Chad Barb committed
531 532 533 534 535 536 537 538 539 540
    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;
541
	my $wrapper = "assign_wrapper";
542
	
543
	if (system("$wrapper -u $pid $eid")) {
Chad Barb's avatar
Chad Barb committed
544 545
	    #
	    # save this off so it will get passed back later.
546
	    # Note that -1 is an uncontrolled error. No recovery.
Chad Barb's avatar
Chad Barb committed
547
	    #
548 549 550 551 552 553
	    $assignWrapperErrorCode = $?;
	    $exitcode = $? >> 8;

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

	    if (($exitcode & 64) && ($exitcode != 255)) {
554
		# so batchexp doesn't choke.
555 556 557
		$assignWrapperErrorCode -= 64
		    if ($exitcode != 255);
		# We can recover. 
Chad Barb's avatar
Chad Barb committed
558
		return 7;
559 560 561
	    }
	    else {
		# No recovery.
Chad Barb's avatar
Chad Barb committed
562 563 564 565
		return 1;
	    }
	}
	TBDebugTimeStamp("assign_wrapper finished");
566

Chad Barb's avatar
Chad Barb committed
567 568
	print "Mapped to physical reality!\n";
    }
569

570
    # Check cancel flag before continuing. No retry, 
571
    TBGetCancelFlag($pid, $eid, \$canceled);
572 573 574
    return 1
	if ($canceled);

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
    #
    # 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");
		
613
		if (system("vnode_setup -k $pid $eid @virtnodes")) {
614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632
		    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;
		}
	    }
	}
    }

633 634 635 636 637 638
    # Exit here if we are testing.
    if ($TESTMODE) {
	print "Testing run - Stopping here.\n";
	return 0;
    }

639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666
    #
    # If there are any Plab dslice nodes in the experiment, create the
    # dslice now
    #
    if ($type > UPDATE_RECOVER) {
	# Are there any Plab nodes?
	$db_result =
	    DBQueryFatal("select n.node_id from nodes as n ".
			 "left join node_types as nt on n.type = nt.type ".
			 "left join reserved as r on r.node_id = n.node_id ".
			 "where r.pid='$pid' and r.eid='$eid' ".
			 " and nt.isplabdslice = 1");

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

	    if (! $db_result->numrows) {
		my @plabnodes = ();
		
		while (my ($node) = $db_result->fetchrow_array()) {
		    push(@plabnodes, $node);
		}
		
		print "Creating Planetlab slice.\n";
		TBDebugTimeStamp("plabslice create started");
667
		if (system("plabslice create $pid $eid")) {
668 669 670 671 672 673 674 675
		    print STDERR "*** Failed to create Plab dslice\n";
		    return 3;
		}
		TBDebugTimeStamp("plabslice alloc finished");
	    }
	}
    }

676
    # Check cancel flag before continuing. No retry, 
677
    TBGetCancelFlag($pid, $eid, \$canceled);
678 679 680
    return 1
	if ($canceled);

681 682 683 684 685 686 687 688 689 690 691 692 693
    #
    # 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
694

695 696 697 698 699 700
    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
701
	#
702 703
    }
    TBDebugTimeStamp("named finished");
Chad Barb's avatar
Chad Barb committed
704

705
    # Check cancel flag before continuing. No retry, 
706
    TBGetCancelFlag($pid, $eid, \$canceled);
707 708
    return 1
	if ($canceled);
Chad Barb's avatar
Chad Barb committed
709 710 711

    #
    # If user specified -reboot to update,
712
    # and we are successfully performing the update,
713
    # then mark all nodes in experiment so os_setup will reboot them.
Chad Barb's avatar
Chad Barb committed
714
    #
715
    if (($type == UPDATE) && $updateReboot) {
Chad Barb's avatar
Chad Barb committed
716 717
	print STDERR "Marking nodes for reboot.\n";
	$db_result =
718 719 720
	    DBQueryFatal("select r.node_id,n.allocstate from reserved as r ".
			 "left join nodes as n on n.node_id=r.node_id ".
			 "where r.pid='$pid' and r.eid='$eid'");
Chad Barb's avatar
Chad Barb committed
721

722 723 724 725 726 727 728 729 730 731
	while (my ($node,$allocstate) = $db_result->fetchrow_array) {
	    #
	    # If the node is INIT_CLEAN, leave it alone. It will still get
	    # rebooted, but will not falsely be tagged as dirty. This is
	    # important for vnodes too, where INIT_CLEAN indicated the vnode
	    # does not even exist yet (plab nodes).
	    #
	    if ($allocstate ne TBDB_ALLOCSTATE_RES_INIT_CLEAN()) {
		TBSetNodeAllocState($node, TBDB_ALLOCSTATE_RES_INIT_DIRTY());
	    }
Chad Barb's avatar
Chad Barb committed
732 733 734
	}
    }

735 736 737 738 739 740 741 742 743 744 745 746
    #
    # 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
747

748
    #
749 750 751 752 753
    # 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.)
754 755 756 757 758 759 760 761 762
    #

    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
763

764 765 766 767
    #
    # An error now means that the VLANS need to be cleaned up.
    #
    $cleanvlans = 1;
Chad Barb's avatar
Chad Barb committed
768

769 770 771 772 773 774 775 776 777
    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
778

779 780 781 782 783 784 785 786 787 788 789 790 791 792
    #
    # 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");
793
    }
Chad Barb's avatar
Chad Barb committed
794

795 796 797 798 799 800 801 802 803
    #
    # 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";
	    #
804 805
	    # Use returncode from os_setup process to
	    # set global $retry flag, indicating to caller
806 807 808
	    # that it may be beneficial to attempt
	    # a doSwapin() again.
	    #
809
            if (($CHILD_ERROR >> 8) == 1) {
Chad Barb's avatar
Chad Barb committed
810
		return 3;
811 812
	    } else {
		print STDERR "Not retrying due to error type.\n";
Chad Barb's avatar
Chad Barb committed
813
		return 1;
814
	    }
815 816 817 818 819 820 821
	}
    } 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
822

823 824 825 826
    #
    # 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.
827 828
    # Don't do this during an update, since we didn't kill the 
    # event system previously, so starting it again will fail!
829 830
    # 
    if (! $DISABLE_EVENTS) {
Chad Barb's avatar
Chad Barb committed
831
	if ($type != UPDATE && $type != UPDATE_RECOVER) {
832 833 834 835 836 837 838
	    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");
839 840
	}
    }
Chad Barb's avatar
Chad Barb committed
841

842 843
    return 0;
}