batch_daemon.in 19 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) 2000-2006 University of Utah and the Flux Group.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
6 7 8
# All rights reserved.
#

9 10
use English;
use Getopt::Std;
11 12
use Fcntl;
use IO::Handle;
13 14 15 16 17 18

#
# Create a batch experiment.
#
# usage: batch_daemon
#
Leigh B. Stoller's avatar
Leigh B. Stoller committed
19 20
# TODO: Use "logger" instead of writing a log file.
#
21 22
sub usage()
{
23 24
    print STDOUT "Usage: batch_daemon [-d]\n" .
	"Use the -d option to prevent daemonization\n";
25 26
    exit(-1);
}
27 28 29 30 31

sub dosomething($$);
sub isexpdone($);
sub TBBatchUnLockExp($$;$);

32
my  $optlist = "d";
33 34 35 36 37 38 39

#
# Configure variables
#
my $TB       = "@prefix@";
my $DBNAME   = "@TBDBNAME@";
my $TBOPS    = "@TBOPSEMAIL@";
40
my $TBLOGS   = "@TBLOGSEMAIL@";
41

42 43 44 45 46 47 48 49
#
# Only root can run this script.
#
if ($UID) {
    die("*** $0:\n".
	"    Only root can run this script!\n");
}

50 51 52 53 54 55 56
#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;

57 58
# Be careful not to exit on transient error; 0 means infinite retry.
$libdb::DBQUERY_MAXTRIES = 0;
59

60
my $tbbindir = "$TB/bin/";
61
my $swapexp  = "$TB/bin/swapexp";
62
my $endexp   = "$TB/bin/endexp";
63
my $savelogs = "$TB/bin/savelogs";
64
my $avail    = "$TB/sbin/avail";
65 66
my $batchlog = "$TB/log/batchlog";
my $projroot = "/proj";
67
my $debug    = 0;
68

69 70 71 72 73 74 75
my $BSTATE_POSTED	= EXPTSTATE_QUEUED;
my $BSTATE_ACTIVATING	= EXPTSTATE_ACTIVATING;
my $BSTATE_RUNNING	= EXPTSTATE_ACTIVE;
my $BSTATE_TERMINATING	= EXPTSTATE_TERMINATING;
my $BSTATE_PAUSED	= EXPTSTATE_SWAPPED;
my $BSTATE_LOCKED       = BATCHSTATE_LOCKED;
my $BSTATE_UNLOCKED     = BATCHSTATE_UNLOCKED;
76
my $TBOPSPID            = TBOPSPID;
77

78 79 80 81 82 83
#
# These are valid in the children, not the parent. I suppose I could use
# dynamically scoped variables, but hardly worth it.
#
my $eid;
my $pid;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
84
my $gid;
85
my $logname;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
86
my $nsfile;
87 88
my $userdir;
my $workdir;
89
my $user_name  = "Testbed Operations";
90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113
my $user_email = "$TBOPS";

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

#
# Untaint the path
# 
$ENV{'PATH'} = "/bin:/usr/bin:";
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

#
# Parse command arguments. Once we return from getopts, all that should be
# left are the required arguments.
#
%options = ();
if (! getopts($optlist, \%options)) {
    usage();
}
if (@ARGV != 0) {
    usage();
}
114
if (defined($options{"d"})) {
115
    $debug = 1;
116
}
117 118

# Go to ground.
119
if (! $debug) {
120 121 122
    if (TBBackGround($batchlog)) {
	exit(0);
    }
123
}
124 125 126 127 128

#
# Loop, looking for batch experiments that want to run.
# 
while (1) {
129
    my($count, $i, $query_result, $pending_result, $running_result);
130
    my(%pending_row);
131
    my $retry_wait = 60;
132
    my $disabled;
133

134 135 136 137 138 139 140 141 142
    if (! TBGetSiteVar("web/nologins", \$disabled)) {
	print "Error getting sitevar 'web/nologins'. Waiting a bit ...\n";
	goto pause;
    }
    if ($disabled) {
	print "Web interface is disabled; waiting a bit ...\n";
	sleep(60);
	goto pause;
    }
143 144 145 146
    if (! TBGetSiteVar("batch/retry_wait", \$retry_wait)) {
	print "Error getting sitevar 'batch/retry_wait'. Waiting a bit ...\n";
	goto pause;
    }
147 148 149
    # Do not allow zero!
    $retry_wait = 60
	if ($retry_wait == 0);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
150

151
    #
152
    # Need to lock the table here because of cancelation in endexp.
153 154 155 156 157 158
    # See the comments in there. We need to atomically grab the next
    # batch experiment we want to try, and then change its state from
    # new to configuring. We want to grab just one experiment, since
    # it takes a while to configure an experiment, and grabbing a bunch and
    # locking them up might result in having to wait a really long time
    # to cancel a batch experiment that hasn't really tried to start yet!
159
    # Thats would ne annoying to users.
160
    #
161
    # So, now you're wondering what my selection criteria is? Well, its
162 163
    # damn simplistic. I set the "started" datetime field each attempt,
    # and I pick the batch_experiment with the oldest time, thereby cycling
164 165 166 167
    # through in a "least recently attempted" manner. In addition, we want
    # to throttle the number simultaneous batches that one person can
    # can have running at a time (curently to just one batch), so check to
    # to see if the person has another batch active (thats e2 below).
168 169
    #
    $query_result =
170 171
	DBQuery("lock tables experiments write, experiments as e1 write, ".
		"experiments as e2 write");
172 173
    if (! $query_result) {
	print "DB Error locking tables. Waiting a bit ...\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
174
	goto pause;
175 176 177
    }
    
    $pending_result =
178 179 180
	DBQueryWarn("SELECT e1.* FROM experiments as e1 ".
		    "left join experiments as e2 on ".
		    " e2.expt_head_uid=e1.expt_head_uid and ".
181
		    " e2.batchmode=1 and e2.state='$BSTATE_RUNNING' and ".
182 183
		    " e1.pid=e2.pid and e1.eid!=e2.eid and ".
		    " e2.pid!='$TBOPSPID' ".
184 185
		    "WHERE e2.eid is null and ".
		    " e1.batchmode=1 and e1.canceled=0 and ".
186
		    " e1.expt_locked is null and ".
187
		    " e1.batchstate!='$BSTATE_LOCKED' and ".
188
		    " e1.state='$BSTATE_POSTED' and ".
189 190
		    "  (e1.attempts=0 or ".
		    "    ((UNIX_TIMESTAMP() - ".
191
		    "      UNIX_TIMESTAMP(e1.expt_start) > ($retry_wait)))) ".
192
		    "ORDER BY e1.expt_start LIMIT 1");
193

194
    $running_result =
195
	DBQuery("select * from experiments ".
196 197
		"where batchmode=1 and state='$BSTATE_RUNNING' and ".
		"      batchstate!='$BSTATE_LOCKED' ".
198
		"ORDER BY expt_start");
199 200 201

    if (!$pending_result || !$running_result) {
	print "DB Error getting batch info. Waiting a bit ...\n";
202
	DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
203
	goto pause;
204 205 206
    }

    if (!$pending_result->numrows && !$running_result->numrows) {
207
	DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
208
	goto pause;
209 210 211
    }

    #
212 213
    # If we have a pending experiment to run, the lock it right away,
    # while we have the tables locked. 
214
    #
215 216 217 218 219 220 221 222
    if ($pending_result->numrows) {
	%pending_row = $pending_result->fetchhash();

	# Local vars!
	my $eid = $pending_row{'eid'};
	my $pid = $pending_row{'pid'};

	$query_result = 
223
	    DBQuery("update experiments set ".
224
		    "       expt_locked=now(),expt_start=now(), ".
225
		    "       batchstate='$BSTATE_LOCKED' ".
226
		    "where eid='$eid' and pid='$pid'");
227 228 229

	if (! $query_result) {
	    print "DB error setting batch $pid/$eid to configuring.\n";
230
	    DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
231
	    goto pause;
232 233
	}
    }
234

235 236 237 238 239 240 241 242 243
    #
    # Okay, first we check the status of running batch mode experiments
    # since we want to end those before trying to start any new ones, cause
    # it would be nice to have as many nodes available as possible before
    # trying to add a new one. This can potentially delay startup, but thats
    # okay. Its a batch system.
    #
    # If you are wondering why I check for finished experiments in the main
    # loop instead of in the child that started the experiment, its so that
244
    # we fire up again and look for them in the event that boss goes down.
245
    #
246 247 248 249
    if (!$running_result->numrows) {
	DBQueryWarn("unlock tables");
    }
    else {
250 251 252 253 254 255
	my %canterm = ();
	
	while (my %running_row = $running_result->fetchhash()) {
	    # Local vars!
	    my $eid         = $running_row{'eid'};
	    my $pid         = $running_row{'pid'};
256

257 258 259 260 261 262 263 264 265 266 267
	    #
	    # Lock so user cannot mess with it.
	    # 
	    $query_result = 
		DBQuery("update experiments set ".
			"       expt_locked=now(), ".
			"       batchstate='$BSTATE_LOCKED' ".
			"where eid='$eid' and pid='$pid'");

	    $canterm{"$pid:$eid"} = ($query_result ? 1 : 0);
	}
268 269
	DBQueryWarn("unlock tables");

270 271 272 273 274 275 276 277 278 279 280 281 282 283 284
	#
	# Reset and go through again.
	#
	$running_result->dataseek(0);
	
	while (my %running_row = $running_result->fetchhash()) {
	    my $canceled    = $running_row{'canceled'};
	    # Local vars!
	    my $eid         = $running_row{'eid'};
	    my $pid         = $running_row{'pid'};
	    my $canterm     = $canterm{"$pid:$eid"};

	    next
		if (!$canterm);

285 286 287
	    if ($canceled) {
		# Look at the cancel flag.
		if ($canceled == EXPTCANCEL_TERM) {
288
		    dosomething("cancel", \%running_row);
289 290
		}
		elsif ($canceled == EXPTCANCEL_SWAP) {
291
		    dosomething("swap", \%running_row);
292 293 294 295
		}
		else {
		    print "Improper cancel flag: $canceled\n";
		}
296 297
	    }
	    else {
298
		if (isexpdone(\%running_row)) {
299 300 301
		    #
		    # Terminate the experiment. 
		    # 
302
		    dosomething("swap", \%running_row);
303 304 305 306 307 308 309
		}
		else {
		    #
		    # Unlock.
		    # 
		    TBBatchUnLockExp($pid, $eid);
		}
310
	    }
311 312 313 314 315 316
	}
    }
    #
    # Finally start an actual experiment!
    #
    if ($pending_result->numrows) {
317
	dosomething("start", \%pending_row);
318
    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
319
  pause:
320
    sleep(15);
321 322 323
}

#
324
# Do something as the user. Either, start, end, or cancel an experiment.
325
#
326
sub dosomething($$)
327
{
328
    my($dowhat,$exphash)   = @_;
329
    my($unix_uid, $unix_gid, $unix_gname, $row, $query_result);
330 331

    # Global vars
332 333 334 335
    $eid     = $exphash->{'eid'};
    $pid     = $exphash->{'pid'};
    $gid     = $exphash->{'gid'};
    $userdir = $exphash->{'path'};
336
    $workdir = TBExptWorkDir($pid, $eid);
337 338 339
    $nsfile  = "$eid.ns";
    
    # Locals
340
    my $creator = $exphash->{'expt_head_uid'};
341

342 343 344 345 346 347 348
    print "Doing a '$dowhat' to batch experiment $pid/$eid\n";

    #
    # Create a temporary name for a log file. We do this in the parent so
    # we can remove it when the child ends. The child could remove it, but
    # since it is open in the child, it has the tendency to stick around.
    #
349
    $logname = TBExptCreateLogFile($pid, $eid, "${dowhat}-batch");
350

351 352 353
    #
    # Start up a child to run the guts. The parent waits. If the
    # experiment configures okay, the parent can return to try something
354
    # else.
355 356 357
    #
    $childpid = fork();
    if ($childpid) {
358

359
	print TBDateTimeFSSafe() . "\n";
360 361
	print "Child PID $childpid started to $dowhat $pid/$eid\n";

362
	waitpid($childpid, 0);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
363
	my $status = $? >> 8;
364

365
	print TBDateTimeFSSafe() . "\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
366
	print "Child PID $childpid exited with exit status $status\n";
367

368
        #
369
        # Close the log file.
370 371
	# The exp dir might be gone if the batch was killed/canceled.
        #
372
	if (-e $userdir) {
373 374
	    TBExptCloseLogFile($pid, $eid);
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
375
	return $status;
376
    }
377
    openlog($logname);
378 379
    TBExptSetLogFile($pid, $eid, $logname);
    TBExptOpenLogFile($pid, $eid);
380

381 382 383
    #
    # Get some user information. 
    #
384
    if (!UserDBInfo($creator, \$user_name, \$user_email)) {
385
	fatal("DB Error getting user information for uid $creator");
386
    }
387
    
388 389
    chdir("$workdir") or
	fatal("Could not cd into $workdir!");
390 391 392 393 394

    #
    # Figure out the unix uid/gid that the experiment configuration is
    # going to run as. 
    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
395
    (undef,undef,$unix_uid) = getpwnam($creator) or
396
	fatal("No such user $creator");
397 398
    TBGroupUnixInfo($pid, $gid, \$unix_gid, \$unix_gname) or
	fatal("No such group $pid/$gid");
399

400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419
    #
    # Need the entire group list for the user, cause of subgroups, and cause
    # thats the correct thing to do. Too bad perl does not have a getgrouplist
    # function like the C library.
    #
    my $glist = `id -G $creator`;
    if ($glist =~ /^([\d ]*)$/) {
	$glist = $1;
    }
    else {
	fatal("Unexpected results from 'id -G $creator': $glist");
    }
    #
    # Remove current group from glist, then add gid twice at the front
    # of the list. Order matters here, or we won't pick up all the
    # groups we need.
    #
    $glist =~ s/ ?\b$unix_gid\b ?//;
    $glist = $unix_gid . " " . $unix_gid . " " . $glist;

420 421 422
    #
    # Change the ownership of the log file before we flip.
    #
423 424
    chown($unix_uid, $unix_gid, $logname) or
	fatal("Could not chown $logname to $unix_uid/$unix_gid!");
425 426

    # Flip to the user. We never flip back.
427 428 429 430 431
    $GID            = $unix_gid;
    $EGID           = $glist;
    $EUID = $UID    = $unix_uid;
    $ENV{'USER'}    = $creator;
    $ENV{'LOGNAME'} = $creator;
432
    
433
    if ($dowhat eq "start") {
434
	startexp($exphash);
435
    }
436
    elsif ($dowhat eq "swap") {
437
	swapexp($exphash);
438 439
    }
    elsif ($dowhat eq "cancel") {
440
	cancelexp($exphash);
441 442 443 444 445 446 447 448 449
    }
    exit(0);
}

#
# Try to start an experiment. Never returns.
# 
sub startexp($)
{
450
    my($exphash)  = @_;
451
    my($exit_status, $running, $query_result);
452

453
    my $attempts  = $exphash->{'attempts'};
454

455
    #
456
    # Try to swap the experiment in. 
457
    #
458
    system("$swapexp -b -s in $pid $eid");
459 460
    $exit_status = $? >> 8;
    $running     = 1;
461 462 463 464 465
    if ($exit_status) {
	$running = 0;
    }
    
    #
466 467
    # Look for cancelation. If we get a DB error on this, just continue cause
    # we can pick up the cancelation later.
468 469
    #
    $query_result =
470
	DBQueryWarn("select canceled from experiments ".
471
		    "where eid='$eid' and pid='$pid'");
472

473
    if ($query_result) {
474
	my ($canceled) = $query_result->fetchrow_array();
475
	$exphash->{'canceled'} = $canceled;
476

477
	# Yuck: This is strictly for the benefit of swapexp() below.
478
	$exphash->{'state'} = EXPTSTATE_ACTIVE
479 480 481 482
	    if ($running);
	
	if ($canceled) {
	    # Look at the cancel flag.
483
	    if ($canceled == EXPTCANCEL_TERM) {
484
		cancelexp($exphash);
485
	    }
486
	    elsif ($canceled == EXPTCANCEL_SWAP) {
487
		swapexp($exphash);
488 489 490 491
	    }
	    else {
		print "Improper cancel flag: $canceled\n";
	    }
492 493 494
	    #
	    # Never returns, but just to be safe ...
	    #
495
	    exit(-1);
496 497
	}
    }
498 499 500 501

    #
    # If the configuration failed for lack of nodes, then don't send
    # email unless the number of attempts starts to get big.
502
    #
503 504 505
    # If the configuration failed for some other reason, then send email.
    # We have to reset the state to "new" so that it will be retried again
    # later. 
506 507
    #
    if (! $running) {
508 509 510 511
	#
	# XXX - What if this update fails?
	# 
	$query_result = 
512
	    DBQueryWarn("update experiments set attempts=attempts+1 ".
513
			"where eid='$eid' and pid='$pid'");
514

515
	#
516 517 518 519
	# The exit status does not tell us if the experiment can ever be
	# mapped. In fact, that is really hard to know at this level;
	# it depends on what resources the testbed actually has. So,
	# unless status is -1 (really really fatal) just keep going.
520 521
	# We leave it up the user to kill the batch if it looks like its
	# never going to work.
522
	#
523
	if ($exit_status == 255) {
524
	    TBBatchUnLockExp($pid, $eid, EXPTSTATE_SWAPPED());
525
	    
526
	    email_status("Experiment startup has failed with a fatal error!\n".
527 528
			 "Batch has been dequeued so that you may check it.");

529
	    exit($exit_status);
530
	}
531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548
	
	if (($attempts % 30) == 0) {
	    $attempts++;

	    my $msg =
		"Could not configure Batch Mode experiment $pid/$eid.\n".
		"\n".
		"There was an assignment violation (please check the log)\n".
		"that prevented it from being scheduled. The violation\n".
		"might result from not enough nodes or not enough link\n".
		"bandwidth. If you feel that the violation is in error,\n".
		"please cancel the batch and notify $TBOPS\n".
		"Otherwise, another attempt will be made in a little while.\n".
		"\n". 
		"There have been $attempts attempts to start this batch.";

	    email_status($msg);
	}
549
	TBBatchUnLockExp($pid, $eid, EXPTSTATE_QUEUED());
550
	exit($exit_status);
551 552 553
    }

    #
554
    # Well, it configured! We can now unlock it.
555
    #
556
    TBBatchUnLockExp($pid, $eid);
557
    
558
    email_status("Batch Mode experiment $pid/$eid is now running!\n".
559
		 "Please consult the Web interface to see how it is doing.");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
560

561
    #
562
    # Done with this phase. Must exit.
563
    #
564 565
    exit(0);
}
566

567
#
568
# A batch has completed. Swap it out. 
569
#
570
sub swapexp($;$)
571
{
572 573 574
    my($exphash)  = @_;
    my $canceled  = $exphash->{'canceled'};
    my $running   = ($exphash->{'state'} eq EXPTSTATE_ACTIVE);
575

576 577 578 579 580 581 582 583
    if ($running) {
	system("$swapexp -b -s out $pid $eid");
	if ($?) {
	    #
	    # TB admin is going to have to clean up. 
	    # 
	    fatal("Swapping out Batch Mode experiment $pid/$eid");
	}
584 585 586 587 588
    }
    #
    # Set the state to paused to ensure that it is not run again until
    # the user wants to.
    # 
589 590
    TBSetCancelFlag($pid, $eid, EXPTCANCEL_CLEAR);
    TBBatchUnLockExp($pid, $eid);
591
    
592 593 594 595 596
    if ($canceled) {
	email_status("Batch Mode experiment $pid/$eid has been stopped!");
    }
    else {
	email_status("Batch Mode experiment $pid/$eid has finished!");
597
    }
598
   
599
    #
600
    # Child must exit!
601
    #
602 603 604 605 606 607
    exit(0);
}

#
# Cancel an experiment. Never returns.
#
608
sub cancelexp($)
609
{
610
    my($exphash) = @_;
611
    
612 613 614 615
    #
    # It does not matter if the experiment is running; endexp does the
    # right thing.
    # 
616 617 618 619 620 621
    system("$endexp -b $pid $eid");
    if ($?) {
	#
	# TB admin is going to have to clean up. 
	# 
	fatal("Terminating Batch Mode experiment $pid/$eid");
622
    }
623
    donotify("Your Batch Mode experiment has been canceled!", "Canceled", 0);
624 625 626 627 628 629 630
   
    #
    # Child must exit!
    #
    exit(0);
}

631 632 633 634 635 636
#
# Check experiment status. Looks to see if all of the nodes in an
# experiment have reported in.
#
sub isexpdone($)
{
637
    my($exphash)  = @_;
638
    my($query_result, @row);
639 640
    
    # Global vars
641 642
    $eid = $exphash->{'eid'};
    $pid = $exphash->{'pid'};
643

644
    print TBDateTimeFSSafe() . "\n";
645 646 647 648 649 650
    print "Checking to see if $pid/$eid has finished up yet\n";

    #
    # Look to see if any nodes yet to report status. If so, spin again.
    #
    $query_result =
651 652 653
	DBQueryWarn("SELECT startstatus,bootstatus FROM nodes ".
		    "LEFT JOIN reserved ON nodes.node_id=reserved.node_id ".
		    "WHERE reserved.eid='$eid' and reserved.pid='$pid'");
654 655 656 657 658

    if (! $query_result) {
	return 0;
    }

659 660 661 662 663 664 665 666 667 668 669 670 671
    #
    # Well, right now a node is considered finished up only if its
    # boot did not fail, and it has reported start command status.
    # The idea being that if the boot failed, then its status will
    # never be reported anyway, and we might as well consider the node
    # done (else the experiment would never end).
    # 
    while (@row = $query_result->fetchrow_array()) {
	if ($row[1] eq NODEBOOTSTATUS_FAILED) {
	    next;
	}
	if ($row[0] eq NODESTARTSTATUS_NOSTATUS) {
	    return 0;
672 673
	}
    }
674
    return 1;
675 676 677 678
}

#
# Start up a child, and set its descriptors talking to a log file.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
679
# The log file already exists, created with mktemp above.
680 681 682 683 684 685 686
# 
sub openlog($)
{
    my($logname) = $_[0];
	
    #
    # We have to disconnect from the caller by redirecting both STDIN and
Leigh B. Stoller's avatar
Leigh B. Stoller committed
687 688
    # STDOUT away from the pipe. Otherwise the caller will continue to wait
    # even though the parent has exited. 
689 690
    #
    open(STDIN, "< /dev/null") or
691
	fatal("opening /dev/null for STDIN: $!");
692 693 694 695 696 697

    open(STDERR, ">> $logname") or
	fatal("opening $logname for STDERR: $!");
    open(STDOUT, ">> $logname") or
	fatal("opening $logname for STDOUT: $!");

698 699 700
    STDOUT->autoflush(1);
    STDERR->autoflush(1);

701 702 703
    return 0;
}

704 705 706 707 708
#
# A fatal error is something that the user does not need to know about.
# Caused by a breakdown in the TB system. Generally speaking, once the
# experiment is running, this should not be used.
# 
709
sub fatal($)
710 711 712
{
    my($mesg) = $_[0];

713
    donotify($mesg, "Failure", 1);
714 715 716 717

    exit(-1);
}

718 719 720
#
# Something the user cares about. 
# 
721
sub email_status($)
722
{
723
    my($mesg) = $_[0];
724

725
    donotify($mesg, "Status", 0);
726 727
}

728
sub donotify($$$)
729
{
730
    my($mesg, $subtext, $iserr) = @_;
731
    my($subject, $from, $to, $hdrs);
732
    my $MAIL;
733

734 735 736
    $mesg = "$mesg\n";

    print STDOUT "$mesg";
737

738
    $subject = "Batch Mode Experiment $subtext $pid/$eid";
739
    $from    = $TBOPS;
740 741 742 743 744 745
    $hdrs    = "Reply-To: $TBOPS";
    
    #
    # An error goes just to Testbed Operations. Normal status messages go
    # to the user and to the Testbed Logs address.
    # 
746
    if ($iserr) {
747
	$to = "$TBOPS";
748 749
    }
    else {
750 751 752
	$to   = "$user_name <$user_email>";
	$hdrs = "Bcc: $TBLOGS\n".
	        "$hdrs";
753 754
    }

755
    SENDMAIL($to, $subject, $mesg, $from, $hdrs,
756
	     ($logname, "assign.log", $nsfile));
757
}
758 759 760 761 762 763 764 765 766 767 768

sub TBBatchUnLockExp($$;$)
{
    my($pid, $eid, $newstate) = @_;

    DBQueryWarn("update experiments set expt_locked=NULL, ".
		"       batchstate='$BSTATE_UNLOCKED' ".
		(defined($newstate) ? ",state='$newstate' " : "") .
		"where eid='$eid' and pid='$pid'");
    return 1;
}