batch_daemon.in 18.9 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) 2000-2005 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

77 78 79 80 81 82
#
# 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
83
my $gid;
84
my $logname;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
85
my $nsfile;
86 87
my $userdir;
my $workdir;
88
my $user_name  = "Testbed Operations";
89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112
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($optlist, \%options)) {
    usage();
}
if (@ARGV != 0) {
    usage();
}
113
if (defined($options{"d"})) {
114
    $debug = 1;
115
}
116 117

# Go to ground.
118
if (! $debug) {
119 120 121
    if (TBBackGround($batchlog)) {
	exit(0);
    }
122
}
123 124 125 126 127

#
# Loop, looking for batch experiments that want to run.
# 
while (1) {
128
    my($count, $i, $query_result, $pending_result, $running_result);
129
    my(%pending_row);
130
    my $retry_wait = 60;
131
    my $disabled;
132

133 134 135 136 137 138 139 140 141
    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;
    }
142 143 144 145
    if (! TBGetSiteVar("batch/retry_wait", \$retry_wait)) {
	print "Error getting sitevar 'batch/retry_wait'. Waiting a bit ...\n";
	goto pause;
    }
146 147 148
    # Do not allow zero!
    $retry_wait = 60
	if ($retry_wait == 0);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
149

150
    #
151
    # Need to lock the table here because of cancelation in endexp.
152 153 154 155 156 157
    # 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!
158
    # Thats would ne annoying to users.
159
    #
160
    # So, now you're wondering what my selection criteria is? Well, its
161 162
    # damn simplistic. I set the "started" datetime field each attempt,
    # and I pick the batch_experiment with the oldest time, thereby cycling
163 164 165 166
    # 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).
167 168
    #
    $query_result =
169 170
	DBQuery("lock tables experiments write, experiments as e1 write, ".
		"experiments as e2 write");
171 172
    if (! $query_result) {
	print "DB Error locking tables. Waiting a bit ...\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
173
	goto pause;
174 175 176
    }
    
    $pending_result =
177 178 179
	DBQueryWarn("SELECT e1.* FROM experiments as e1 ".
		    "left join experiments as e2 on ".
		    " e2.expt_head_uid=e1.expt_head_uid and ".
180
		    " e2.batchmode=1 and e2.state='$BSTATE_RUNNING' and ".
181 182 183
		    " e1.pid=e2.pid and e1.eid!=e2.eid ".
		    "WHERE e2.eid is null and ".
		    " e1.batchmode=1 and e1.canceled=0 and ".
184
		    " e1.expt_locked is null and ".
185
		    " e1.state='$BSTATE_POSTED' and ".
186 187
		    "  (e1.attempts=0 or ".
		    "    ((UNIX_TIMESTAMP() - ".
188
		    "      UNIX_TIMESTAMP(e1.expt_start) > ($retry_wait)))) ".
189
		    "ORDER BY e1.expt_start LIMIT 1");
190

191
    $running_result =
192
	DBQuery("select * from experiments ".
193
		"where batchmode=1 and state='$BSTATE_RUNNING' ".
194
		"ORDER BY expt_start");
195 196 197

    if (!$pending_result || !$running_result) {
	print "DB Error getting batch info. Waiting a bit ...\n";
198
	DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
199
	goto pause;
200 201 202
    }

    if (!$pending_result->numrows && !$running_result->numrows) {
203
	DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
204
	goto pause;
205 206 207
    }

    #
208 209
    # If we have a pending experiment to run, the lock it right away,
    # while we have the tables locked. 
210
    #
211 212 213 214 215 216 217 218
    if ($pending_result->numrows) {
	%pending_row = $pending_result->fetchhash();

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

	$query_result = 
219
	    DBQuery("update experiments set ".
220
		    "       expt_locked=now(),expt_start=now(), ".
221
		    "       batchstate='$BSTATE_LOCKED' ".
222
		    "where eid='$eid' and pid='$pid'");
223 224 225

	if (! $query_result) {
	    print "DB error setting batch $pid/$eid to configuring.\n";
226
	    DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
227
	    goto pause;
228 229
	}
    }
230

231 232 233 234 235 236 237 238 239
    #
    # 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
240
    # we fire up again and look for them in the event that boss goes down.
241
    #
242 243 244 245
    if (!$running_result->numrows) {
	DBQueryWarn("unlock tables");
    }
    else {
246 247 248 249 250 251
	my %canterm = ();
	
	while (my %running_row = $running_result->fetchhash()) {
	    # Local vars!
	    my $eid         = $running_row{'eid'};
	    my $pid         = $running_row{'pid'};
252

253 254 255 256 257 258 259 260 261 262 263
	    #
	    # 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);
	}
264 265
	DBQueryWarn("unlock tables");

266 267 268 269 270 271 272 273 274 275 276 277 278 279 280
	#
	# 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);

281 282 283 284 285 286 287 288 289 290 291
	    if ($canceled) {
		# Look at the cancel flag.
		if ($canceled == EXPTCANCEL_TERM) {
		    dosomething("cancel", %running_row);
		}
		elsif ($canceled == EXPTCANCEL_SWAP) {
		    dosomething("swap", %running_row);
		}
		else {
		    print "Improper cancel flag: $canceled\n";
		}
292 293
	    }
	    else {
294 295 296 297 298 299 300 301 302 303 304 305
		if (isexpdone(%running_row)) {
		    #
		    # Terminate the experiment. 
		    # 
		    dosomething("swap", %running_row);
		}
		else {
		    #
		    # Unlock.
		    # 
		    TBBatchUnLockExp($pid, $eid);
		}
306
	    }
307 308 309 310 311 312 313 314
	}
    }
    #
    # Finally start an actual experiment!
    #
    if ($pending_result->numrows) {
	dosomething("start", %pending_row);
    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
315
  pause:
316
    sleep(15);
317 318 319
}

#
320
# Do something as the user. Either, start, end, or cancel an experiment.
321
#
322
sub dosomething($$)
323
{
324 325
    my($dowhat)   = shift;
    my(%exphash)  = @_;
326
    my($unix_uid, $unix_gid, $unix_gname, $row, $query_result);
327 328

    # Global vars
329 330 331
    $eid     = $exphash{'eid'};
    $pid     = $exphash{'pid'};
    $gid     = $exphash{'gid'};
332 333
    $userdir = $exphash{'path'};
    $workdir = TBExptWorkDir($pid, $eid);
334 335 336 337
    $nsfile  = "$eid.ns";
    
    # Locals
    my $creator = $exphash{'expt_head_uid'};
338

339 340 341 342 343 344 345
    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.
    #
346
    $logname = TBExptCreateLogFile($pid, $eid, "${dowhat}-batch");
347

348 349 350
    #
    # Start up a child to run the guts. The parent waits. If the
    # experiment configures okay, the parent can return to try something
351
    # else.
352 353 354
    #
    $childpid = fork();
    if ($childpid) {
355

356
	print TBDateTimeFSSafe() . "\n";
357 358
	print "Child PID $childpid started to $dowhat $pid/$eid\n";

359
	waitpid($childpid, 0);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
360
	my $status = $? >> 8;
361

362
	print TBDateTimeFSSafe() . "\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
363
	print "Child PID $childpid exited with exit status $status\n";
364

365
        #
366
        # Close the log file.
367 368
	# The exp dir might be gone if the batch was killed/canceled.
        #
369
	if (-e $userdir) {
370 371
	    TBExptCloseLogFile($pid, $eid);
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
372
	return $status;
373
    }
374
    openlog($logname);
375 376
    TBExptSetLogFile($pid, $eid, $logname);
    TBExptOpenLogFile($pid, $eid);
377

378 379 380
    #
    # Get some user information. 
    #
381
    if (!UserDBInfo($creator, \$user_name, \$user_email)) {
382
	fatal("DB Error getting user information for uid $creator");
383
    }
384
    
385 386
    chdir("$workdir") or
	fatal("Could not cd into $workdir!");
387 388 389 390 391

    #
    # Figure out the unix uid/gid that the experiment configuration is
    # going to run as. 
    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
392
    (undef,undef,$unix_uid) = getpwnam($creator) or
393
	fatal("No such user $creator");
394 395
    TBGroupUnixInfo($pid, $gid, \$unix_gid, \$unix_gname) or
	fatal("No such group $pid/$gid");
396

397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416
    #
    # 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;

417 418 419
    #
    # Change the ownership of the log file before we flip.
    #
420 421
    chown($unix_uid, $unix_gid, $logname) or
	fatal("Could not chown $logname to $unix_uid/$unix_gid!");
422 423

    # Flip to the user. We never flip back.
424 425 426 427 428
    $GID            = $unix_gid;
    $EGID           = $glist;
    $EUID = $UID    = $unix_uid;
    $ENV{'USER'}    = $creator;
    $ENV{'LOGNAME'} = $creator;
429
    
430 431 432
    if ($dowhat eq "start") {
	startexp(%exphash);
    }
433 434
    elsif ($dowhat eq "swap") {
	swapexp(%exphash);
435 436
    }
    elsif ($dowhat eq "cancel") {
437
	cancelexp(%exphash);
438 439 440 441 442 443 444 445 446 447
    }
    exit(0);
}

#
# Try to start an experiment. Never returns.
# 
sub startexp($)
{
    my(%exphash)  = @_;
448
    my($exit_status, $running, $query_result);
449 450 451

    my $attempts  = $exphash{'attempts'};

452
    #
453
    # Try to swap the experiment in. 
454
    #
455
    system("$swapexp -b -s in $pid $eid");
456 457
    $exit_status = $? >> 8;
    $running     = 1;
458 459 460 461 462
    if ($exit_status) {
	$running = 0;
    }
    
    #
463 464
    # Look for cancelation. If we get a DB error on this, just continue cause
    # we can pick up the cancelation later.
465 466
    #
    $query_result =
467
	DBQueryWarn("select canceled from experiments ".
468
		    "where eid='$eid' and pid='$pid'");
469

470
    if ($query_result) {
471 472
	my ($canceled) = $query_result->fetchrow_array();
	$exphash{'canceled'} = $canceled;
473

474
	# Yuck: This is strictly for the benefit of swapexp() below.
475
	$exphash{'state'} = EXPTSTATE_ACTIVE
476 477 478 479
	    if ($running);
	
	if ($canceled) {
	    # Look at the cancel flag.
480
	    if ($canceled == EXPTCANCEL_TERM) {
481 482
		cancelexp(%exphash);
	    }
483
	    elsif ($canceled == EXPTCANCEL_SWAP) {
484 485 486 487 488
		swapexp(%exphash);
	    }
	    else {
		print "Improper cancel flag: $canceled\n";
	    }
489 490 491
	    #
	    # Never returns, but just to be safe ...
	    #
492
	    exit(-1);
493 494
	}
    }
495 496 497 498

    #
    # If the configuration failed for lack of nodes, then don't send
    # email unless the number of attempts starts to get big.
499
    #
500 501 502
    # 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. 
503 504
    #
    if (! $running) {
505 506 507 508
	#
	# XXX - What if this update fails?
	# 
	$query_result = 
509
	    DBQueryWarn("update experiments set attempts=attempts+1 ".
510
			"where eid='$eid' and pid='$pid'");
511

512
	#
513 514 515 516
	# 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.
517 518
	# We leave it up the user to kill the batch if it looks like its
	# never going to work.
519
	#
520
	if ($exit_status == 255) {
521
	    TBBatchUnLockExp($pid, $eid, EXPTSTATE_SWAPPED());
522
	    
523
	    email_status("Experiment startup has failed with a fatal error!\n".
524 525
			 "Batch has been dequeued so that you may check it.");

526
	    exit($exit_status);
527
	}
528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545
	
	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($msg);
	}
546
	TBBatchUnLockExp($pid, $eid, EXPTSTATE_QUEUED());
547
	exit($exit_status);
548 549 550
    }

    #
551
    # Well, it configured! We can now unlock it.
552
    #
553
    TBBatchUnLockExp($pid, $eid);
554
    
555
    email_status("Batch Mode experiment $pid/$eid is now running!\n".
556
		 "Please consult the Web interface to see how it is doing.");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
557

558
    #
559
    # Done with this phase. Must exit.
560
    #
561 562
    exit(0);
}
563

564
#
565
# A batch has completed. Swap it out. 
566
#
567
sub swapexp($;$)
568 569
{
    my(%exphash)  = @_;
570
    my $canceled  = $exphash{'canceled'};
571
    my $running   = ($exphash{'state'} eq EXPTSTATE_ACTIVE);
572

573 574 575 576 577 578 579 580
    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");
	}
581 582 583 584 585
    }
    #
    # Set the state to paused to ensure that it is not run again until
    # the user wants to.
    # 
586 587
    TBSetCancelFlag($pid, $eid, EXPTCANCEL_CLEAR);
    TBBatchUnLockExp($pid, $eid);
588
    
589 590 591 592 593
    if ($canceled) {
	email_status("Batch Mode experiment $pid/$eid has been stopped!");
    }
    else {
	email_status("Batch Mode experiment $pid/$eid has finished!");
594
    }
595
   
596
    #
597
    # Child must exit!
598
    #
599 600 601 602 603 604
    exit(0);
}

#
# Cancel an experiment. Never returns.
#
605
sub cancelexp($)
606 607 608
{
    my(%exphash) = @_;
    
609 610 611 612
    #
    # It does not matter if the experiment is running; endexp does the
    # right thing.
    # 
613 614 615 616 617 618
    system("$endexp -b $pid $eid");
    if ($?) {
	#
	# TB admin is going to have to clean up. 
	# 
	fatal("Terminating Batch Mode experiment $pid/$eid");
619
    }
620
    donotify("Your Batch Mode experiment has been canceled!", "Canceled", 0);
621 622 623 624 625 626 627
   
    #
    # Child must exit!
    #
    exit(0);
}

628 629 630 631 632 633 634
#
# Check experiment status. Looks to see if all of the nodes in an
# experiment have reported in.
#
sub isexpdone($)
{
    my(%exphash)  = @_;
635
    my($query_result, @row);
636 637 638 639 640
    
    # Global vars
    $eid = $exphash{'eid'};
    $pid = $exphash{'pid'};

641
    print TBDateTimeFSSafe() . "\n";
642 643 644 645 646 647
    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.
    #
    $query_result =
648 649 650
	DBQueryWarn("SELECT startstatus,bootstatus FROM nodes ".
		    "LEFT JOIN reserved ON nodes.node_id=reserved.node_id ".
		    "WHERE reserved.eid='$eid' and reserved.pid='$pid'");
651 652 653 654 655

    if (! $query_result) {
	return 0;
    }

656 657 658 659 660 661 662 663 664 665 666 667 668
    #
    # 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;
669 670
	}
    }
671
    return 1;
672 673 674 675
}

#
# Start up a child, and set its descriptors talking to a log file.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
676
# The log file already exists, created with mktemp above.
677 678 679 680 681 682 683
# 
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
684 685
    # STDOUT away from the pipe. Otherwise the caller will continue to wait
    # even though the parent has exited. 
686 687
    #
    open(STDIN, "< /dev/null") or
688
	fatal("opening /dev/null for STDIN: $!");
689 690 691 692 693 694

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

695 696 697
    STDOUT->autoflush(1);
    STDERR->autoflush(1);

698 699 700
    return 0;
}

701 702 703 704 705
#
# 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.
# 
706
sub fatal($)
707 708 709
{
    my($mesg) = $_[0];

710
    donotify($mesg, "Failure", 1);
711 712 713 714

    exit(-1);
}

715 716 717
#
# Something the user cares about. 
# 
718
sub email_status($)
719
{
720
    my($mesg) = $_[0];
721

722
    donotify($mesg, "Status", 0);
723 724
}

725
sub donotify($$$)
726
{
727
    my($mesg, $subtext, $iserr) = @_;
728
    my($subject, $from, $to, $hdrs);
729
    my $MAIL;
730

731 732 733
    $mesg = "$mesg\n";

    print STDOUT "$mesg";
734

735
    $subject = "Batch Mode Experiment $subtext $pid/$eid";
736
    $from    = $TBOPS;
737 738 739 740 741 742
    $hdrs    = "Reply-To: $TBOPS";
    
    #
    # An error goes just to Testbed Operations. Normal status messages go
    # to the user and to the Testbed Logs address.
    # 
743
    if ($iserr) {
744
	$to = "$TBOPS";
745 746
    }
    else {
747 748 749
	$to   = "$user_name <$user_email>";
	$hdrs = "Bcc: $TBLOGS\n".
	        "$hdrs";
750 751
    }

752
    SENDMAIL($to, $subject, $mesg, $from, $hdrs,
753
	     ($logname, "assign.log", $nsfile));
754
}
755 756 757 758 759 760 761 762 763 764 765

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;
}