tbswap.in 20.4 KB
Newer Older
Chad Barb's avatar
 
Chad Barb committed
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;

Chad Barb's avatar
 
Chad Barb committed
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.

Chad Barb's avatar
 
Chad Barb committed
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)
Chad Barb's avatar
 
Chad Barb committed
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";
Chad Barb's avatar
 
Chad Barb committed
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.
#
Chad Barb's avatar
 
Chad Barb committed
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; }
Chad Barb's avatar
 
Chad Barb committed
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;
Chad Barb's avatar
 
Chad Barb committed
67 68
my $force  = 0;
my $errors = 0;
Chad Barb's avatar
Chad Barb committed
69
my $assignWrapperErrorCode = 0;
Chad Barb's avatar
 
Chad Barb committed
70

Chad Barb's avatar
 
Chad Barb committed
71 72 73 74 75 76
my $state;

my $os_setup_pid;
my $cleanvlans;

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

my $swapop = shift;	

Chad Barb's avatar
 
Chad Barb committed
83 84 85 86
if (!$swapop || 
    (($swapop ne "in") && 
     ($swapop ne "out") &&
     ($swapop ne "update"))) {
Chad Barb's avatar
 
Chad Barb committed
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;
Chad Barb's avatar
 
Chad Barb committed
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");
}

Chad Barb's avatar
 
Chad Barb committed
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;
Chad Barb's avatar
 
Chad Barb committed
158
} elsif ($swapop eq "out") {
Chad Barb's avatar
 
Chad Barb committed
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;
Chad Barb's avatar
 
Chad Barb committed
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;
Chad Barb's avatar
 
Chad Barb committed
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
#

Chad Barb's avatar
 
Chad Barb committed
215
undef $nextState;
Chad Barb's avatar
 
Chad Barb committed
216 217

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

	#
Chad Barb's avatar
 
Chad Barb committed
231 232 233 234 235
	# Update Accounting Information
	#
	TBSetExpSwapTime($pid, $eid);
    }
}
Chad Barb's avatar
 
Chad Barb committed
236

Chad Barb's avatar
Chad Barb committed
237 238 239
#
# Update, Phase One -- swap experiment partially out.
#
Chad Barb's avatar
 
Chad Barb committed
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,$$);

Chad Barb's avatar
 
Chad Barb committed
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") {
Chad Barb's avatar
 
Chad Barb committed
268
	#
Chad Barb's avatar
Chad Barb committed
269
	# Update
Chad Barb's avatar
 
Chad Barb committed
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);
Chad Barb's avatar
 
Chad Barb committed
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;
	    }
Chad Barb's avatar
 
Chad Barb committed
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;
	    }
Chad Barb's avatar
 
Chad Barb committed
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);
Chad Barb's avatar
 
Chad Barb committed
337

Chad Barb's avatar
 
Chad Barb committed
338
	my $retries = 2;
Chad Barb's avatar
 
Chad Barb committed
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,
Chad Barb's avatar
 
Chad Barb committed
344 345
	#   c) we haven't tried too many times already.
	#
Chad Barb's avatar
 
Chad Barb committed
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 ) {
Chad Barb's avatar
 
Chad Barb committed
350 351 352
	    $retries--;

	    print STDERR "Cleaning up after errors; will try again.\n";
Chad Barb's avatar
 
Chad Barb committed
353
	    doSwapout(RETRY);
Chad Barb's avatar
 
Chad Barb committed
354 355

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

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

	    #
Chad Barb's avatar
Chad Barb committed
379
	    # Regardless of how well cleanup swapout worked,
Chad Barb's avatar
 
Chad Barb committed
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";
Chad Barb's avatar
 
Chad Barb committed
390
	$errors = 1;
Chad Barb's avatar
 
Chad Barb committed
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

Chad Barb's avatar
 
Chad Barb committed
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 );
Chad Barb's avatar
 
Chad Barb committed
406 407 408
} else {
    print "Successfully finished swap-$swapop for $pid/$eid. " . TBTimeStamp() . "\n";
    TBDebugTimeStamp("tbswap $swapop finished (succeeded)");
Chad Barb's avatar
 
Chad Barb committed
409
    exit( 0 );
Chad Barb's avatar
 
Chad Barb committed
410 411 412 413 414
}

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

##
Chad Barb's avatar
Chad Barb committed
415
#
Chad Barb's avatar
 
Chad Barb committed
416 417
# doSwapout - Swaps experiment out.
#
Chad Barb's avatar
Chad Barb committed
418
#             If in REAL or CLEANUP,
Chad Barb's avatar
 
Chad Barb committed
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,
Chad Barb's avatar
 
Chad Barb committed
423 424 425 426 427 428
#             only nodes not in RES_READY will be freed.
#
#             Returns 0 on success, >0 on failure.
#
##

Chad Barb's avatar
 
Chad Barb committed
429 430
sub doSwapout($) {
    my $type = shift; # REAL==4, CLEANUP==3, RETRY==2, UPDATE==1.
Chad Barb's avatar
 
Chad Barb committed
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

Chad Barb's avatar
 
Chad Barb committed
443 444
    if (! $TESTMODE) { 
	if (! $DISABLE_EVENTS) {
Chad Barb's avatar
 
Chad Barb committed
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;
		}
Chad Barb's avatar
 
Chad Barb committed
451 452
	    }
	}
Chad Barb's avatar
Chad Barb committed
453

Chad Barb's avatar
 
Chad Barb committed
454
	#
Chad Barb's avatar
Chad Barb committed
455
	# Clean up any VLANs in experiment.
Chad Barb's avatar
 
Chad Barb committed
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;
Chad Barb's avatar
 
Chad Barb committed
464
	}
Chad Barb's avatar
Chad Barb committed
465 466
	TBDebugTimeStamp("snmpit finished");

Chad Barb's avatar
 
Chad Barb committed
467 468 469
	#
	# 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
470
	# (don't kill off vnodes when UPDATEing.)
Chad Barb's avatar
 
Chad Barb committed
471
	#
Chad Barb's avatar
 
Chad Barb committed
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");
Chad Barb's avatar
 
Chad Barb committed
480 481 482
	}
    }

Chad Barb's avatar
 
Chad Barb committed
483
    if ($type >= CLEANUP) {
Chad Barb's avatar
 
Chad Barb committed
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)
Chad Barb's avatar
 
Chad Barb committed
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");
Chad Barb's avatar
 
Chad Barb committed
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'");
Chad Barb's avatar
 
Chad Barb committed
503 504
    } else {
	#
Chad Barb's avatar
 
Chad Barb committed
505
	# $type == RETRY or $type == UPDATE.
Chad Barb's avatar
 
Chad Barb committed
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)
Chad Barb's avatar
 
Chad Barb committed
509 510 511 512 513 514 515 516 517 518 519
	#

	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'");

520
	while (my ($node,$allocstate) = $db_result->fetchrow_array) {
Chad Barb's avatar
 
Chad Barb committed
521 522 523 524 525
	    if ($allocstate ne TBDB_ALLOCSTATE_RES_READY()) {
		push(@failedNodes, $node);
	    }
	}

Chad Barb's avatar
 
Chad Barb committed
526 527 528
	# XXX reboot nodes [e.g., put them in RES_DIRTY()] 
	# if doing an update -restart.

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

Chad Barb's avatar
 
Chad Barb committed
563 564 565 566
	#
	# Resetting named maps and email lists is fast and idempotent,
	# so whatever.
	#
Chad Barb's avatar
 
Chad Barb committed
567 568 569 570 571 572
	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
573

Chad Barb's avatar
 
Chad Barb committed
574 575 576 577 578 579 580 581
	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
582 583 584
    #
    # Wipe the DB clean.
    #
Chad Barb's avatar
Chad Barb committed
585

Chad Barb's avatar
 
Chad Barb committed
586
    print STDERR "Resetting DB.\n";
Chad Barb's avatar
 
Chad Barb committed
587
    TBExptRemovePhysicalState( $pid, $eid );
Chad Barb's avatar
 
Chad Barb committed
588 589 590 591 592

    return $swapout_errors;
}

##
Chad Barb's avatar
Chad Barb committed
593
#
Chad Barb's avatar
 
Chad Barb committed
594 595
# doSwapin - Swaps experiment in.
#
Chad Barb's avatar
Chad Barb committed
596 597 598 599 600
#            Returns:
#              0 - successful swapin
#              1 - failed swapin; cleanup required.
#              3 - failed swapin; cleanup required; can retry.
#              7 - failed swapin; assign failed; no cleanup.
Chad Barb's avatar
 
Chad Barb committed
601 602 603 604 605
#
#            Will set $retry = 1 if os_setup failed.
#
##

Chad Barb's avatar
 
Chad Barb committed
606
sub doSwapin($) {
Chad Barb's avatar
Chad Barb committed
607
    my $type = shift; # REAL==4, RETRY==2, UPDATE==1, UPDATE_RECOVER=0. 
Chad Barb's avatar
 
Chad Barb committed
608

Chad Barb's avatar
 
Chad Barb committed
609 610 611 612 613
    #
    # assign_wrapper does all the virtual to physical mapping 
    # and updating the DB state.
    #

Chad Barb's avatar
Chad Barb committed
614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631
    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) {
Chad Barb's avatar
 
Chad Barb committed
632 633
		# so batchexp doesn't choke.
		$assignWrapperErrorCode -= 64;
Chad Barb's avatar
Chad Barb committed
634 635 636 637 638 639
		return 7;
	    } else {
		return 1;
	    }
	}
	TBDebugTimeStamp("assign_wrapper finished");
Chad Barb's avatar
 
Chad Barb committed
640

Chad Barb's avatar
Chad Barb committed
641 642
	print "Mapped to physical reality!\n";
    }
Chad Barb's avatar
 
Chad Barb committed
643

644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701
    #
    # 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");
		
		if (system("vnode_setup -f -k $pid $eid @virtnodes")) {
		    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;
		}
	    }
	}
    }

Chad Barb's avatar
 
Chad Barb committed
702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720
    # 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
721

Chad Barb's avatar
 
Chad Barb committed
722 723 724 725 726 727
    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
728
	#
Chad Barb's avatar
 
Chad Barb committed
729 730
    }
    TBDebugTimeStamp("named finished");
Chad Barb's avatar
Chad Barb committed
731 732 733 734


    #
    # If user specified -reboot to update,
Chad Barb's avatar
 
Chad Barb committed
735
    # and we are successfully performing the update,
Chad Barb's avatar
Chad Barb committed
736 737 738
    # then mark all nodes in experiment
    # so os_setup will reboot them.
    #
Chad Barb's avatar
 
Chad Barb committed
739
    if (($type == UPDATE) && $updateReboot) {
Chad Barb's avatar
Chad Barb committed
740 741 742 743 744 745 746 747
	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'");

748
	while (my ($node) = $db_result->fetchrow_array) {
Chad Barb's avatar
Chad Barb committed
749 750 751 752
	    TBSetNodeAllocState( $node, TBDB_ALLOCSTATE_RES_INIT_DIRTY() );
	}
    }

Chad Barb's avatar
 
Chad Barb committed
753 754 755 756 757 758 759 760 761 762 763 764
    #
    # 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
765

Chad Barb's avatar
 
Chad Barb committed
766
    #
767 768 769 770 771
    # 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.)
Chad Barb's avatar
 
Chad Barb committed
772 773 774 775 776 777 778 779 780
    #

    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
781

Chad Barb's avatar
 
Chad Barb committed
782 783 784 785
    #
    # An error now means that the VLANS need to be cleaned up.
    #
    $cleanvlans = 1;
Chad Barb's avatar
Chad Barb committed
786

Chad Barb's avatar
 
Chad Barb committed
787 788 789 790 791 792 793 794 795
    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
796

Chad Barb's avatar
 
Chad Barb committed
797 798 799 800 801 802 803 804 805 806 807 808 809 810
    #
    # 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
811
    }
Chad Barb's avatar
Chad Barb committed
812

Chad Barb's avatar
 
Chad Barb committed
813 814 815 816 817 818 819 820 821
    #
    # 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
822 823
	    # Use returncode from os_setup process to
	    # set global $retry flag, indicating to caller
Chad Barb's avatar
 
Chad Barb committed
824 825 826
	    # that it may be beneficial to attempt
	    # a doSwapin() again.
	    #
Chad Barb's avatar
 
Chad Barb committed
827
            if (($CHILD_ERROR >> 8) == 1) {
Chad Barb's avatar
Chad Barb committed
828
		return 3;
Chad Barb's avatar
 
Chad Barb committed
829 830
	    } else {
		print STDERR "Not retrying due to error type.\n";
Chad Barb's avatar
Chad Barb committed
831
		return 1;
Chad Barb's avatar
 
Chad Barb committed
832
	    }
Chad Barb's avatar
 
Chad Barb committed
833 834 835 836 837 838 839
	}
    } 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
840

Chad Barb's avatar
 
Chad Barb committed
841 842 843 844
    #
    # 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
845 846
    # 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
847 848
    # 
    if (! $DISABLE_EVENTS) {
Chad Barb's avatar
Chad Barb committed
849
	if ($type != UPDATE && $type != UPDATE_RECOVER) {
Chad Barb's avatar
 
Chad Barb committed
850 851 852 853 854 855 856
	    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
857 858
	}
    }
Chad Barb's avatar
Chad Barb committed
859

Chad Barb's avatar
 
Chad Barb committed
860 861
    return 0;
}