swapexp.in 13.1 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 14
# This gets invoked from the Web interface. 
# Swap an experiment in, swap it out, restart or modify.
15 16 17
#
sub usage()
{
Chad Barb's avatar
 
Chad Barb committed
18
    print STDOUT "Usage: swapexp [-i] [-r] <-s in | out | restart | modify> ". 
Chad Barb's avatar
 
Chad Barb committed
19
	         "<pid> <eid> [<nsfile>]\n";
20 21
    exit(-1);
}
Chad Barb's avatar
 
Chad Barb committed
22
my  $optlist = "s:ir";
23 24 25 26 27 28 29

#
# Configure variables
#
my $TB     = "@prefix@";
my $TBOPS  = "@TBOPSEMAIL@";
my $TBLOGS = "@TBLOGSEMAIL@";
30
my $TBINFO = "$TB/expinfo";
31 32 33 34 35 36 37 38 39

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

my $tbdir    = "$TB/bin/";
40
my $tbdata   = "tbdata";
41
my $batch    = 0;
42
my $idleswap = 0;
Chad Barb's avatar
 
Chad Barb committed
43
my $reboot   = 0;    
44
my $errorstat= -1;
Chad Barb's avatar
 
Chad Barb committed
45

46 47 48 49 50
my $inout;
my $logname;
my $dbuid;
my $user_name;
my $user_email;
51
my @allnodes;
52
my @row;
53
my $action;
Chad Barb's avatar
 
Chad Barb committed
54

55 56 57 58 59 60 61 62 63 64 65
#
# Untaint the path
# 
$ENV{'PATH'} = '/bin:/usr/bin';
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

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

66 67 68 69 70 71 72
#
# 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);

73 74 75 76 77 78 79 80
#
# Parse command arguments. Once we return from getopts, all that should
# left are the required arguments.
#
%options = ();
if (! getopts($optlist, \%options)) {
    usage();
}
81 82 83
if (defined($options{"i"})) {
    $idleswap = 1;
}
Chad Barb's avatar
 
Chad Barb committed
84 85 86
if (defined($options{"r"})) {
    $reboot = 1;
}
87 88 89
if (defined($options{"s"})) {
    $inout = $options{"s"};

Chad Barb's avatar
 
Chad Barb committed
90 91 92 93
    if ($inout ne "out" && 
	$inout ne "in" && 
	$inout ne "restart" && 
	$inout ne "modify") {
94 95 96 97 98 99 100
	usage();
    }
}
else {
    usage();
}

Chad Barb's avatar
 
Chad Barb committed
101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121
if (@ARGV != (($inout eq "modify") ? 3 : 2)) {
    usage();
}
my $pid   = $ARGV[0];
my $eid   = $ARGV[1];

my $nsfile;

if ($inout eq "modify") { 
    $nsfile = $ARGV[2];

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

122 123 124 125 126 127 128 129 130 131 132 133 134 135 136
#
# 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");
}
137
my $repfile = "$eid.report";
138 139
my $workdir = TBExptWorkDir($pid, $eid);
my $userdir = TBExptUserDir($pid, $eid);
140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157

#
# 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
158
# Verify that this person can muck with the experiment.
159 160 161 162 163
# 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
164
	"    You do not have permission to swap or modify this experiment!\n");
165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189
}

#
# 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
# the tb scripts. 
#
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'};

190
if (! chdir($workdir)) {
191
    die("*** $0:\n".
192
	"    Could not chdir to $workdir: $!\n");
193 194
}

195 196 197 198 199 200 201 202 203 204 205 206 207 208
if (defined($hashrow{'expt_locked'})) {
    $val = $hashrow{'expt_locked'};
    
    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
209
	"    Batch experiments cannot be swapped or modified yet!");
210 211 212
}

#
Chad Barb's avatar
 
Chad Barb committed
213
# Okay, check state. We do not allow modification or swapping to start when the
214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233
# 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!");
}
234 235 236 237
if ($inout eq "restart" && $estate ne EXPTSTATE_ACTIVE) {
    die("*** $0:\n".
	"    It appears that experiment $pid/$eid is not active!");
}
238

Chad Barb's avatar
 
Chad Barb committed
239 240
# if $inout eq "modify", either EXPTSTATE_ACTIVE -or- EXPTSTATE_SWAPPED is ok.

241 242 243 244 245 246 247 248 249 250 251 252 253
#
# 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.
#

254 255 256 257 258 259 260 261 262
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
263 264 265
if ($inout eq "modify") {
    $action = "modified";
}
266

267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285
#
# 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) {
286
    $logname = TBExptCreateLogFile($pid, $eid, "swapexp");
287
    TBExptSetLogFile($pid, $eid, $logname);
288
    TBExptOpenLogFile($pid, $eid);
289 290 291 292 293
    
    if (TBBackGround($logname)) {
	#
	# Parent exits normally
	#
294 295
	print "Experiment $pid/$eid is now being $action.\n".
	    "You will be notified via email when the this is done.\n";
296 297 298 299
	exit(0);
    }
}

300 301 302
#
# Remove old report file since its contents are going to be invalid.
#
303
if ($inout ne "restart" && -e $repfile) {
304 305 306
    unlink("$repfile");
}

Chad Barb's avatar
 
Chad Barb committed
307 308 309 310 311 312 313 314 315
#
# Rerun tbprerun if modifying.
#
if ($inout eq "modify") {
    if (system("$tbdir/tbprerun -m $pid $eid $nsfile") != 0) {
	fatal("tbprerun failed!\n"); 
    }
}

316 317 318 319
#
# Sanity check states in case someone changes something.
#
if ($inout eq "out") {
Chad Barb's avatar
 
Chad Barb committed
320 321
    print STDOUT "Running 'tbswap out' with arguments: $pid $eid\n";
    if (system("$tbdir/tbswap out $pid $eid") != 0) {
322
	$errorstat = $? >> 8;
Chad Barb's avatar
 
Chad Barb committed
323
	fatal("tbswap out failed!\n");
324 325 326 327 328 329 330
    }
    
    $estate = ExpState($pid,$eid);
    if ($estate ne EXPTSTATE_SWAPPED) {
	fatal("Experiment is in the wrong state: $estate\n");
    }
}
331
elsif ($inout eq "in") {
Chad Barb's avatar
 
Chad Barb committed
332 333
    print STDOUT "Running 'tbswap in' with arguments: $pid $eid\n";
    if (system("$tbdir/tbswap in $pid $eid") != 0) {
334
	$errorstat = $? >> 8;
Chad Barb's avatar
 
Chad Barb committed
335
	fatal("tbswap in failed!\n");
336 337 338 339 340 341
    }
    
    $estate = ExpState($pid,$eid);
    if ($estate ne EXPTSTATE_ACTIVE) {
	fatal("Experiment is in the wrong state: $estate\n");
    }
342

343
    system("$tbdir/tbreport -b $pid $eid 2>&1 > $repfile");
Chad Barb's avatar
 
Chad Barb committed
344 345 346 347 348 349 350 351 352
} 
elsif ($inout eq "modify") {
    #
    # If experiment is currently swapped out, no need to do an update 
    # after modifying it.
    #
    if ($estate eq EXPTSTATE_ACTIVE) {
	print STDOUT "Running 'tbswap update' with arguments: $pid $eid\n";
	if (system("$tbdir/tbswap update $pid $eid") != 0) {
353
	    $errorstat = $? >> 8;
Chad Barb's avatar
 
Chad Barb committed
354 355 356 357 358 359 360 361 362
	    fatal("tbswap update failed!\n");
	}

	$estate = ExpState($pid, $eid);
	if ($estate ne EXPTSTATE_ACTIVE) {
	    fatal("Experiment is in the wrong state: $estate\n");
	}
	
	system("$tbdir/tbreport -b $pid $eid 2>&1 > $repfile");
Chad Barb's avatar
 
Chad Barb committed
363 364 365 366 367 368

	if ($reboot) {
	    if (system("$tbdir/node_reboot -e $pid,$eid") != 0) {
		fatal("node reboot failed!\n");
	    }
	}
Chad Barb's avatar
 
Chad Barb committed
369
    }
370
}
Chad Barb's avatar
 
Chad Barb committed
371
else { # $inout eq "restart" assumed.
372 373 374 375 376
    print STDOUT "Running tbrestart with arguments: $pid $eid\n";
    if (system("$tbdir/tbrestart $pid $eid") != 0) {
	fatal("tbrestart failed!\n");
    }
}
377

378 379 380 381 382 383 384 385 386 387
#
# 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.
#
388
system("cp -Rfp $workdir/ $userdir/tbdata/");
389

390 391
#
# Must unlock before exit.
392 393
#
TBUnLockExp($pid, $eid);
394

395 396 397 398
#
# Gather stats. 
#
if ($inout eq "in") {
399
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPIN, 0);
400 401
}
elsif ($inout eq "out") {
402 403
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPOUT, 0,
		    ($idleswap ? $TBDB_STATS_FLAGS_IDLESWAP : 0));
404 405
}
elsif ($inout eq "modify") {
406
    GatherSwapStats($pid, $eid, $dbuid, TBDB_STATS_SWAPMODIFY, 0);
407 408
}

409
print "Done!\n";
410 411 412 413 414 415 416 417

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

Chad Barb's avatar
 
Chad Barb committed
418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434
#
# 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";
    }
}

435 436 437 438 439 440 441
#
# Clear the log file so the web page stops spewing. 
#
if (defined($logname)) {
    TBExptCloseLogFile($pid, $eid);
}

442 443 444 445
#
# Send email notification to user.
#
my $message =
446 447 448 449 450 451 452 453 454 455 456 457
    "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".
458 459
    "Appended below is the output. If you have any questions or comments,\n" .
    "please include the output in your message to $TBOPS\n";
460 461

SENDMAIL("$user_name <$user_email>",
462
	 "Experiment $pid/$eid $action",
463
	 $message,
464
	 ($idleswap ? $TBOPS : "$user_name <$user_email>"),
465 466
	 "Cc:  $expt_head_name <$expt_head_email>\n".
	 "Bcc: $TBLOGS",
467
	 (($inout eq "restart") ? ($logname) : ($repfile, $logname)));
468 469 470 471 472 473 474 475 476

exit 0;

sub fatal($)
{
    my($mesg) = $_[0];
    
    print STDOUT $mesg;

477 478 479 480 481 482 483 484 485 486 487 488 489
    #
    # 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
490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519
    #
    # 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;
    }

    if ($hosed) { 
	#
	# 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";
	    }
	}
	
	print "Running tbend with arguments: -force $pid $eid\n";
	if (system("$tbdir/tbend -force $pid $eid") != 0) {
	    print "tbend failed!\n";
	}
    }

520 521 522 523 524 525
    #
    # Kill this for convenience later.
    # 
    DBQueryWarn("update experiments set expt_locked=NULL ".
		"WHERE eid='$eid' and pid='$pid'");

526 527 528
    # Copy over the log files so the user can see them.
    system("/bin/cp -Rfp $workdir/ $userdir/tbdata");

529 530 531 532
    #
    # In batch mode, exit without sending the email. 
    #
    if ($batch) {
533
	TBUnLockExp($pid, $eid);
534 535 536
	exit(-1);
    }

537 538 539 540 541 542 543
    #
    # Clear the log file so the web page stops spewing. 
    #
    if (defined($logname)) {
	TBExptCloseLogFile($pid, $eid);
    }

544 545 546 547
    #
    # Send a message to the testbed list. Append the logfile.
    #
    SENDMAIL("$user_name <$user_email>",
548
	     "Swap ${inout} Failure: $pid/$eid",
549
	     $mesg,
550
	     ($idleswap ? $TBOPS : "$user_name <$user_email>"),
551
	     "Cc:  $expt_head_name <$expt_head_email>\n".
Leigh B. Stoller's avatar
Leigh B. Stoller committed
552
	     "Cc:  $TBOPS",
553 554
	     ($logname));

Chad Barb's avatar
 
Chad Barb committed
555 556 557 558 559 560 561 562 563 564 565 566 567
    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 ${userdir}-failed");
	TBExptDestroy($pid, $eid);        
    }

568 569 570
    exit(-1);
}