swapexp.in 13.9 KB
Newer Older
1
#!/usr/bin/perl -wT
Leigh B. Stoller's avatar
Leigh B. Stoller committed
2 3 4

#
# EMULAB-COPYRIGHT
5
# Copyright (c) 2000-2003 University of Utah and the Flux Group.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
6 7 8
# All rights reserved.
#

9 10 11 12
use English;
use Getopt::Std;

#
Chad Barb's avatar
Chad Barb committed
13
# This gets invoked from the Web interface.
Chad Barb's avatar
 
Chad Barb committed
14
# Swap an experiment in, swap it out, restart or modify.
15
#
Chad Barb's avatar
Chad Barb committed
16

17 18
sub usage()
{
Chad Barb's avatar
Chad Barb committed
19
    print STDOUT "Usage: swapexp [-i] [-r] <-s in | out | restart | modify> " .
Chad Barb's avatar
 
Chad Barb committed
20
	         "<pid> <eid> [<nsfile>]\n";
21 22
    exit(-1);
}
Chad Barb's avatar
 
Chad Barb committed
23
my  $optlist = "s:ir";
24 25 26 27 28 29 30

#
# Configure variables
#
my $TB     = "@prefix@";
my $TBOPS  = "@TBOPSEMAIL@";
my $TBLOGS = "@TBLOGSEMAIL@";
31
my $TBINFO = "$TB/expinfo";
32 33 34 35 36 37 38 39 40

#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;

my $tbdir    = "$TB/bin/";
41
my $tbdata   = "tbdata";
42
my $batch    = 0;
43
my $idleswap = 0;
Chad Barb's avatar
Chad Barb committed
44
my $reboot   = 0;
45
my $errorstat= -1;
Chad Barb's avatar
 
Chad Barb committed
46

47 48 49 50 51
my $inout;
my $logname;
my $dbuid;
my $user_name;
my $user_email;
52
my @allnodes;
53
my @row;
54
my $action;
Chad Barb's avatar
 
Chad Barb committed
55

56 57 58 59 60 61 62 63 64 65 66
#
# Untaint the path
# 
$ENV{'PATH'} = '/bin:/usr/bin';
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

#
# Turn off line buffering on output
#
$| = 1;

67 68 69 70 71 72 73
#
# Set umask for start/swap. We want other members in the project to be
# able to swap/end experiments, so the log and intermediate files need
# to be 664 since some are opened for append.
#
umask(0002);

74 75 76 77 78 79 80 81
#
# Parse command arguments. Once we return from getopts, all that should
# left are the required arguments.
#
%options = ();
if (! getopts($optlist, \%options)) {
    usage();
}
82 83 84
if (defined($options{"i"})) {
    $idleswap = 1;
}
Chad Barb's avatar
 
Chad Barb committed
85 86 87
if (defined($options{"r"})) {
    $reboot = 1;
}
88 89 90
if (defined($options{"s"})) {
    $inout = $options{"s"};

Chad Barb's avatar
Chad Barb committed
91 92 93
    if ($inout ne "out"     &&
	$inout ne "in"      &&
	$inout ne "restart" &&
Chad Barb's avatar
 
Chad Barb committed
94
	$inout ne "modify") {
95 96 97 98 99 100 101
	usage();
    }
}
else {
    usage();
}

Chad Barb's avatar
 
Chad Barb committed
102 103 104 105 106 107 108 109
if (@ARGV != (($inout eq "modify") ? 3 : 2)) {
    usage();
}
my $pid   = $ARGV[0];
my $eid   = $ARGV[1];

my $nsfile;

Chad Barb's avatar
Chad Barb committed
110
if ($inout eq "modify") {
Chad Barb's avatar
 
Chad Barb committed
111 112 113 114 115 116 117 118 119 120 121 122
    $nsfile = $ARGV[2];

    #
    # Untaint nsfile argument; Allow slash.
    #
    if ($nsfile =~ /^([-\w.\/]+)$/) {
	$nsfile = $1;
    } else {
	die("Tainted nsfile name: $nsfile");
    }
}

123 124 125 126 127 128 129 130 131 132 133 134 135 136 137
#
# 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");
}
138
my $repfile = "$eid.report";
139 140
my $workdir = TBExptWorkDir($pid, $eid);
my $userdir = TBExptUserDir($pid, $eid);
141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158

#
# Verify user and get his DB uid.
#
if (! UNIX2DBUID($UID, \$dbuid)) {
    die("*** $0:\n".
	"    You do not exist in the Emulab Database.\n");
}

#
# Get email info for user.
#
if (! UserDBInfo($dbuid, \$user_name, \$user_email)) {
    die("*** $0:\n".
	"    Cannot determine your name and email address.\n");
}

#
Chad Barb's avatar
 
Chad Barb committed
159
# Verify that this person can muck with the experiment.
160 161 162 163 164
# Note that any script down the line has to do an admin check also. 
#
if ($UID && !TBAdmin($UID) &&
    !TBExptAccessCheck($dbuid, $pid, $eid, TB_EXPT_DESTROY)) {
    die("*** $0:\n".
Chad Barb's avatar
 
Chad Barb committed
165
	"    You do not have permission to swap or modify this experiment!\n");
166 167 168 169 170 171 172
}

#
# We have to protect against trying to end an experiment that is currently
# in the process of being terminated. We use a "wrapper" state (actually
# a timestamp so we can say when termination was requested) since
# terminating consists of a couple of different experiment states down inside
Chad Barb's avatar
Chad Barb committed
173
# the tb scripts.
174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190
#
DBQueryFatal("lock tables experiments write");

$query_result =
    DBQueryFatal("SELECT * FROM experiments WHERE eid='$eid' and pid='$pid'");

if (! $query_result->numrows) {
    die("*** $0:\n".
	"    No such experiment $pid/$eid exists!\n");
}
my %hashrow = $query_result->fetchhash();
my $expt_head_login = $hashrow{'expt_head_uid'};
my $estate          = $hashrow{'state'};
my $expt_path       = $hashrow{'path'};
my $isbatchexpt     = $hashrow{'batchmode'};
my $ebatchstate     = $hashrow{'batchstate'};

191
if (! chdir($workdir)) {
192
    die("*** $0:\n".
193
	"    Could not chdir to $workdir: $!\n");
194 195
}

196 197
if (defined($hashrow{'expt_locked'})) {
    $val = $hashrow{'expt_locked'};
Chad Barb's avatar
Chad Barb committed
198

199 200 201 202 203 204 205 206 207 208 209
    die("*** $0:\n".
	"    It appears that $pid/$eid went into transition at $val.\n".
	"    You will be notified via email when experiment transition ".
	"    is complete.\n");
}

#
# Disallow batch experiment swaps for now.
#
if ($isbatchexpt) {
    die("*** $0:\n".
Chad Barb's avatar
 
Chad Barb committed
210
	"    Batch experiments cannot be swapped or modified yet!");
211 212 213
}

#
Chad Barb's avatar
 
Chad Barb committed
214
# Okay, check state. We do not allow modification or swapping to start when the
215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234
# experiment is in transition. A future task would be to allow this,
# but for now the experiment must be in one of a few states to proceed
#
# Seems like too many states!
#
if ($estate ne EXPTSTATE_ACTIVE && $estate ne EXPTSTATE_SWAPPED) {
    die("*** $0:\n".
	"    It appears that experiment $pid/$eid is in transition.\n".
	"    The user that created the experiment will be notified via\n".
        "    email when the experiment is no longer in transition.\n");
}

if ($inout eq "in" && $estate eq EXPTSTATE_ACTIVE) {
    die("*** $0:\n".
	"    It appears that experiment $pid/$eid is already swapped in!");
}
if ($inout eq "out" && $estate eq EXPTSTATE_SWAPPED) {
    die("*** $0:\n".
	"    It appears that experiment $pid/$eid is already swapped out!");
}
235 236 237 238
if ($inout eq "restart" && $estate ne EXPTSTATE_ACTIVE) {
    die("*** $0:\n".
	"    It appears that experiment $pid/$eid is not active!");
}
239

Chad Barb's avatar
 
Chad Barb committed
240 241
# if $inout eq "modify", either EXPTSTATE_ACTIVE -or- EXPTSTATE_SWAPPED is ok.

242 243 244 245 246 247 248 249 250 251 252 253 254
#
# Set the timestamp now, and unlock the experiments table.
#
DBQueryFatal("UPDATE experiments SET expt_locked=now() ".
	     "WHERE eid='$eid' and pid='$pid'");

DBQueryFatal("unlock tables");

#
# XXX - At this point a failure is going to leave things in an
# inconsistent state.
#

255 256 257 258 259 260 261 262 263
if ($inout eq "in") {
    $action = "swapped in";
}
if ($inout eq "out") {
    $action = "swapped out";
}
if ($inout eq "restart") {
    $action = "restarted";
}
Chad Barb's avatar
 
Chad Barb committed
264 265 266
if ($inout eq "modify") {
    $action = "modified";
}
267

268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286
#
# Get email address of the experiment head, which may be different than
# the person who is actually terminating the experiment, since its polite
# to let the original creator know whats going on. 
#
my $expt_head_name;
my $expt_head_email;

if (! UserDBInfo($expt_head_login, \$expt_head_name, \$expt_head_email)) {
    print STDERR "*** WARNING: ".
	         "Could not determine name/email for $expt_head_login.\n";
    $expt_head_name  = "TBOPS";
    $expt_head_email = $TBOPS;
}

#
# If not in batch mode, go into the background. Parent exits.
#
if (! $batch) {
287
    $logname = TBExptCreateLogFile($pid, $eid, "swapexp");
288
    TBExptSetLogFile($pid, $eid, $logname);
289
    TBExptOpenLogFile($pid, $eid);
Chad Barb's avatar
Chad Barb committed
290

291 292 293 294
    if (TBBackGround($logname)) {
	#
	# Parent exits normally
	#
295 296
	print "Experiment $pid/$eid is now being $action.\n".
	    "You will be notified via email when the this is done.\n";
297 298 299 300
	exit(0);
    }
}

301 302 303
#
# Remove old report file since its contents are going to be invalid.
#
304
if ($inout ne "restart" && -e $repfile) {
305 306 307
    unlink("$repfile");
}

308 309 310 311
#
# Sanity check states in case someone changes something.
#
if ($inout eq "out") {
Chad Barb's avatar
 
Chad Barb committed
312 313
    print STDOUT "Running 'tbswap out' with arguments: $pid $eid\n";
    if (system("$tbdir/tbswap out $pid $eid") != 0) {
314
	$errorstat = $? >> 8;
Chad Barb's avatar
 
Chad Barb committed
315
	fatal("tbswap out failed!\n");
316
    }
Chad Barb's avatar
Chad Barb committed
317

318 319 320 321 322
    $estate = ExpState($pid,$eid);
    if ($estate ne EXPTSTATE_SWAPPED) {
	fatal("Experiment is in the wrong state: $estate\n");
    }
}
323
elsif ($inout eq "in") {
Chad Barb's avatar
 
Chad Barb committed
324 325
    print STDOUT "Running 'tbswap in' with arguments: $pid $eid\n";
    if (system("$tbdir/tbswap in $pid $eid") != 0) {
326
	$errorstat = $? >> 8;
Chad Barb's avatar
 
Chad Barb committed
327
	fatal("tbswap in failed!\n");
328
    }
Chad Barb's avatar
Chad Barb committed
329

330 331 332 333
    $estate = ExpState($pid,$eid);
    if ($estate ne EXPTSTATE_ACTIVE) {
	fatal("Experiment is in the wrong state: $estate\n");
    }
334

335
    system("$tbdir/tbreport -b $pid $eid 2>&1 > $repfile");
Chad Barb's avatar
Chad Barb committed
336
}
Chad Barb's avatar
 
Chad Barb committed
337
elsif ($inout eq "modify") {
Chad Barb's avatar
Chad Barb committed
338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353
    my $modifyError = "";

    print "Backing up old experiment state ... " . TBTimeStamp() . "\n";
    if (TBExptBackupVirtualState($pid, $eid, $$)) {
	fatal("*** $0:\n".
	      "    Could not backup experiment state; cannot safely continue!\n");
    }
    TBExptRemoveVirtualState($pid, $eid);

    #
    # Rerun tbprerun if modifying.
    #
    if (system("$tbdir/tbprerun -m $pid $eid $nsfile") != 0) {
	$modifyError = "tbprerun failed!";
    }

Chad Barb's avatar
 
Chad Barb committed
354 355 356 357
    #
    # If experiment is currently swapped out, no need to do an update 
    # after modifying it.
    #
Chad Barb's avatar
Chad Barb committed
358
    if (! $modifyError && $estate eq EXPTSTATE_ACTIVE) {
Chad Barb's avatar
 
Chad Barb committed
359
	print STDOUT "Running 'tbswap update' with arguments: $pid $eid\n";
Chad Barb's avatar
Chad Barb committed
360 361 362 363

	my $rebootSwitch = "";
	if ($reboot) {
	    $rebootSwitch = "-reboot";
Chad Barb's avatar
 
Chad Barb committed
364 365
	}

Chad Barb's avatar
Chad Barb committed
366 367 368
	if (system("$tbdir/tbswap update $rebootSwitch $pid $eid") != 0) {
	    $errorstat = $? >> 8;
	    $modifyError = "tbswap update failed!";
Chad Barb's avatar
 
Chad Barb committed
369
	}
Chad Barb's avatar
 
Chad Barb committed
370

Chad Barb's avatar
Chad Barb committed
371 372 373 374
	if (! $modifyError) {
	    $estate = ExpState($pid, $eid);
	    if ($estate ne EXPTSTATE_ACTIVE) {
		fatal("Experiment is in the wrong state: $estate!");
Chad Barb's avatar
 
Chad Barb committed
375 376
	    }
	}
Chad Barb's avatar
Chad Barb committed
377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394

	if (! $modifyError) {
	    system("$tbdir/tbreport -b $pid $eid 2>&1 > $repfile");
	}
    }

    if ($modifyError) {
	print STDERR "*** $0:\n".
	             "    $modifyError\n";
	print STDOUT "Recovering experiment state...\n";
	TBExptRemoveVirtualState($pid, $eid );
	if (0 == TBExptRestoreVirtualState($pid, $eid, $$)) {
	    fatal("*** Update aborted; old state restored.\n");
	} else {
	    # Set state to NEW so experiment will get wiped.
	    SetExpState($pid, $eid, EXPTSTATE_NEW);
	    fatal("*** Experiment state could not be restored!\n");
	}
Chad Barb's avatar
 
Chad Barb committed
395
    }
396
}
Chad Barb's avatar
 
Chad Barb committed
397
else { # $inout eq "restart" assumed.
398 399 400 401 402
    print STDOUT "Running tbrestart with arguments: $pid $eid\n";
    if (system("$tbdir/tbrestart $pid $eid") != 0) {
	fatal("tbrestart failed!\n");
    }
}
403

404 405 406 407 408 409 410 411 412 413
#
# Try to copy off the files for testbed information gathering.
#
TBSaveExpLogFiles($pid, $eid);

#
# Make a copy of the work dir in the user visible space so the user
# can see the log files. This overwrites existing files of course,
# but thats okay.
#
414
system("cp -Rfp $workdir/ $userdir/tbdata/");
415

416 417
#
# Must unlock before exit.
418 419
#
TBUnLockExp($pid, $eid);
420

421 422 423 424
#
# Gather stats. 
#
if ($inout eq "in") {
425
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPIN, 0);
426 427
}
elsif ($inout eq "out") {
428 429
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPOUT, 0,
		    ($idleswap ? $TBDB_STATS_FLAGS_IDLESWAP : 0));
430 431
}
elsif ($inout eq "modify") {
432
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPMODIFY, 0);
433 434
}

435
print "Done!\n";
436 437 438 439 440 441 442 443

#
# In batch mode, just exit without sending email.
#
if ($batch) {
    exit(0);
}

Chad Barb's avatar
 
Chad Barb committed
444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460
#
# HACK! if successful, put new NS file in DB.
#

if ($inout eq "modify") {
    $nsdata_string = `cat $nsfile`;
    if (defined($nsdata_string)) {
	$nsdata_string = DBQuoteSpecial($nsdata_string);

	DBQueryWarn("delete from nsfiles WHERE eid='$eid' and pid='$pid'");
	DBQueryWarn("insert into nsfiles (pid, eid, nsfile) ".
		    "VALUES('$pid', '$eid', $nsdata_string)");
    } else {
	print "Warning!! Couldn't read nsfile '$nsfile'!\n";
    }
}

461 462 463 464 465 466 467
#
# Clear the log file so the web page stops spewing. 
#
if (defined($logname)) {
    TBExptCloseLogFile($pid, $eid);
}

468 469 470 471
#
# Send email notification to user.
#
my $message =
472 473 474 475 476 477 478 479 480 481 482 483
    "Experiment $eid in project $pid has been ";

if ($idleswap) {
    $message .= "forcibly swapped out by\n".
	"Testbed Operations because it was idle for too long.\n";
}
else {
    $message .= "$action.\n";
}

$message .=
    "\n".
484 485
    "Appended below is the output. If you have any questions or comments,\n" .
    "please include the output in your message to $TBOPS\n";
486 487

SENDMAIL("$user_name <$user_email>",
488
	 "Experiment $pid/$eid $action",
489
	 $message,
490
	 ($idleswap ? $TBOPS : "$user_name <$user_email>"),
491 492
	 "Cc:  $expt_head_name <$expt_head_email>\n".
	 "Bcc: $TBLOGS",
493
	 (($inout eq "restart") ? ($logname) : ($repfile, $logname)));
494 495 496 497 498 499

exit 0;

sub fatal($)
{
    my($mesg) = $_[0];
Chad Barb's avatar
Chad Barb committed
500

501 502
    print STDOUT $mesg;

503 504 505 506 507 508 509 510 511 512 513 514 515
    #
    # Gather stats. 
    #
    if ($inout eq "in") {
	GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPIN, $errorstat);
    }
    elsif ($inout eq "out") {
	GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPOUT, $errorstat);
    }
    elsif ($inout eq "modify") {
	GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPMODIFY, $errorstat);
    }

Chad Barb's avatar
 
Chad Barb committed
516 517 518 519 520 521 522 523 524 525 526 527
    #
    # if $hosed == 1, we entirely terminate the experiment.
    #
    my $hosed = 0;

    # If we're doing a modify,
    # and tbprerun sent the experiment to "NEW",
    # we're hosed.
    if ($inout eq "modify" && ExpState($pid,$eid) eq EXPTSTATE_NEW) {
	$hosed = 1;
    }

Chad Barb's avatar
Chad Barb committed
528
    if ($hosed) {
Chad Barb's avatar
 
Chad Barb committed
529 530 531 532 533 534 535 536 537 538
	#
	# Note: $estate is still set to the state which the experiment was in
	# when we began.
	#
	if ($estate eq EXPTSTATE_ACTIVE) {
	    print "Running 'tbswap out' with arguments: $pid $eid\n";
	    if (system("$tbdir/tbswap out -force $pid $eid") != 0) {
		print "tbswap out failed!\n";
	    }
	}
Chad Barb's avatar
Chad Barb committed
539

Chad Barb's avatar
 
Chad Barb committed
540 541 542 543 544 545
	print "Running tbend with arguments: -force $pid $eid\n";
	if (system("$tbdir/tbend -force $pid $eid") != 0) {
	    print "tbend failed!\n";
	}
    }

546 547 548 549 550 551
    #
    # Kill this for convenience later.
    # 
    DBQueryWarn("update experiments set expt_locked=NULL ".
		"WHERE eid='$eid' and pid='$pid'");

552 553 554
    # Copy over the log files so the user can see them.
    system("/bin/cp -Rfp $workdir/ $userdir/tbdata");

555
    #
Chad Barb's avatar
Chad Barb committed
556
    # In batch mode, exit without sending the email.
557 558
    #
    if ($batch) {
559
	TBUnLockExp($pid, $eid);
560 561 562
	exit(-1);
    }

563
    #
Chad Barb's avatar
Chad Barb committed
564
    # Clear the log file so the web page stops spewing.
565 566 567 568 569
    #
    if (defined($logname)) {
	TBExptCloseLogFile($pid, $eid);
    }

570 571 572 573
    #
    # Send a message to the testbed list. Append the logfile.
    #
    SENDMAIL("$user_name <$user_email>",
574
	     "Swap ${inout} Failure: $pid/$eid",
575
	     $mesg,
576
	     ($idleswap ? $TBOPS : "$user_name <$user_email>"),
577
	     "Cc:  $expt_head_name <$expt_head_email>\n".
Leigh B. Stoller's avatar
Leigh B. Stoller committed
578
	     "Cc:  $TBOPS",
579 580
	     ($logname));

Chad Barb's avatar
 
Chad Barb committed
581 582 583 584 585 586 587 588 589 590
    if ($hosed) {
	#
	# Copy off the workdir to the user directory, Then back up both of
	# them for post-mortem debugging.
	#
	system("/bin/cp -Rfp $workdir/ $userdir/tbdata");
	system("/bin/rm -rf  ${workdir}-failed");
	system("/bin/mv -f   $workdir ${workdir}-failed");
	system("/bin/rm -rf  ${userdir}-failed");
	system("/bin/mv -f   $userdir ${userdir}-failed");
Chad Barb's avatar
Chad Barb committed
591
	TBExptDestroy($pid, $eid);
Chad Barb's avatar
 
Chad Barb committed
592 593
    }

594 595 596
    exit(-1);
}

Chad Barb's avatar
Chad Barb committed
597 598