batch_daemon.in 19.9 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) 2000-2007 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
#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;
56
use Template;
57
use Experiment;
58

59 60
# Be careful not to exit on transient error; 0 means infinite retry.
$libdb::DBQUERY_MAXTRIES = 0;
61

62
my $tbbindir = "$TB/bin/";
63
my $swapexp  = "$TB/bin/swapexp";
64
my $endexp   = "$TB/bin/endexp";
65
my $savelogs = "$TB/bin/savelogs";
66
my $avail    = "$TB/sbin/avail";
67
my $batchlog = "$TB/log/batchlog";
68
my $debug    = 0;
69

70 71 72
# New template stuff.
my $template_swapout  = "$TB/bin/template_swapout";

73 74 75 76 77 78 79
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;
80
my $TBOPSPID            = TBOPSPID;
81

82 83 84 85 86 87
#
# 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
88
my $gid;
89
my $logname;
90
my $nsfile;
91 92
my $userdir;
my $workdir;
93
my $user_name  = "Testbed Operations";
94
my $user_email = "$TBOPS";
95
my $template;
96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118

#
# 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($optlist, \%options)) {
    usage();
}
if (@ARGV != 0) {
    usage();
}
119
if (defined($options{"d"})) {
120
    $debug = 1;
121
}
122 123

# Go to ground.
124
if (! $debug) {
125 126 127
    if (TBBackGround($batchlog)) {
	exit(0);
    }
128
}
129 130 131 132 133

#
# Loop, looking for batch experiments that want to run.
# 
while (1) {
134
    my($count, $i, $query_result, $pending_result, $running_result);
135
    my(%pending_row);
136
    my $retry_wait = 60;
137
    my $disabled;
138

139 140 141 142 143 144 145 146 147
    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;
    }
148 149 150 151
    if (! TBGetSiteVar("batch/retry_wait", \$retry_wait)) {
	print "Error getting sitevar 'batch/retry_wait'. Waiting a bit ...\n";
	goto pause;
    }
152 153 154
    # Do not allow zero!
    $retry_wait = 60
	if ($retry_wait == 0);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
155

156
    #
157
    # Need to lock the table here because of cancelation in endexp.
158 159 160 161 162 163
    # 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!
164
    # Thats would ne annoying to users.
165
    #
166
    # So, now you're wondering what my selection criteria is? Well, its
167 168
    # damn simplistic. I set the "started" datetime field each attempt,
    # and I pick the batch_experiment with the oldest time, thereby cycling
169 170 171 172
    # 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).
173 174
    #
    $query_result =
175 176
	DBQuery("lock tables experiments write, experiments as e1 write, ".
		"experiments as e2 write");
177 178
    if (! $query_result) {
	print "DB Error locking tables. Waiting a bit ...\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
179
	goto pause;
180 181 182
    }
    
    $pending_result =
183 184 185
	DBQueryWarn("SELECT e1.* FROM experiments as e1 ".
		    "left join experiments as e2 on ".
		    " e2.expt_head_uid=e1.expt_head_uid and ".
186
		    " e2.batchmode=1 and e2.state='$BSTATE_RUNNING' and ".
187 188
		    " e1.pid=e2.pid and e1.eid!=e2.eid and ".
		    " e2.pid!='$TBOPSPID' ".
189 190
		    "WHERE e2.eid is null and ".
		    " e1.batchmode=1 and e1.canceled=0 and ".
191
		    " e1.expt_locked is null and ".
192
		    " e1.batchstate!='$BSTATE_LOCKED' and ".
193
		    " e1.state='$BSTATE_POSTED' and ".
194 195
		    "  (e1.attempts=0 or ".
		    "    ((UNIX_TIMESTAMP() - ".
196
		    "      UNIX_TIMESTAMP(e1.expt_start) > ($retry_wait)))) ".
197
		    "ORDER BY e1.expt_start LIMIT 1");
198

199
    $running_result =
200
	DBQuery("select * from experiments ".
201 202
		"where batchmode=1 and state='$BSTATE_RUNNING' and ".
		"      batchstate!='$BSTATE_LOCKED' ".
203
		"ORDER BY expt_start");
204 205 206

    if (!$pending_result || !$running_result) {
	print "DB Error getting batch info. Waiting a bit ...\n";
207
	DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
208
	goto pause;
209 210 211
    }

    if (!$pending_result->numrows && !$running_result->numrows) {
212
	DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
213
	goto pause;
214 215 216
    }

    #
217 218
    # If we have a pending experiment to run, the lock it right away,
    # while we have the tables locked. 
219
    #
220 221 222 223 224 225 226 227
    if ($pending_result->numrows) {
	%pending_row = $pending_result->fetchhash();

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

	$query_result = 
228
	    DBQuery("update experiments set ".
229
		    "       expt_locked=now(),expt_start=now(), ".
230
		    "       batchstate='$BSTATE_LOCKED' ".
231
		    "where eid='$eid' and pid='$pid'");
232 233 234

	if (! $query_result) {
	    print "DB error setting batch $pid/$eid to configuring.\n";
235
	    DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
236
	    goto pause;
237 238
	}
    }
239

240 241 242 243 244 245 246 247 248
    #
    # 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
249
    # we fire up again and look for them in the event that boss goes down.
250
    #
251 252 253 254
    if (!$running_result->numrows) {
	DBQueryWarn("unlock tables");
    }
    else {
255 256 257 258 259 260
	my %canterm = ();
	
	while (my %running_row = $running_result->fetchhash()) {
	    # Local vars!
	    my $eid         = $running_row{'eid'};
	    my $pid         = $running_row{'pid'};
261

262 263 264 265 266 267 268 269 270 271 272
	    #
	    # 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);
	}
273 274
	DBQueryWarn("unlock tables");

275 276 277 278 279 280 281 282 283 284 285 286 287 288 289
	#
	# 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);

290 291 292
	    if ($canceled) {
		# Look at the cancel flag.
		if ($canceled == EXPTCANCEL_TERM) {
293
		    dosomething("cancel", \%running_row);
294 295
		}
		elsif ($canceled == EXPTCANCEL_SWAP) {
296
		    dosomething("swap", \%running_row);
297 298 299 300
		}
		else {
		    print "Improper cancel flag: $canceled\n";
		}
301 302
	    }
	    else {
303
		if (isexpdone(\%running_row)) {
304 305 306
		    #
		    # Terminate the experiment. 
		    # 
307
		    dosomething("swap", \%running_row);
308 309 310 311 312 313 314
		}
		else {
		    #
		    # Unlock.
		    # 
		    TBBatchUnLockExp($pid, $eid);
		}
315
	    }
316 317 318 319 320 321
	}
    }
    #
    # Finally start an actual experiment!
    #
    if ($pending_result->numrows) {
322
	dosomething("start", \%pending_row);
323
    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
324
  pause:
325
    sleep(15);
326 327 328
}

#
329
# Do something as the user. Either, start, end, or cancel an experiment.
330
#
331
sub dosomething($$)
332
{
333
    my($dowhat,$exphash)   = @_;
334
    my($unix_uid, $unix_gid, $unix_gname, $row, $query_result);
335 336

    # Global vars
337 338 339 340
    $eid     = $exphash->{'eid'};
    $pid     = $exphash->{'pid'};
    $gid     = $exphash->{'gid'};
    $userdir = $exphash->{'path'};
341
    $workdir = TBExptWorkDir($pid, $eid);
342
    $nsfile  = "$eid.ns";
343
    $template= undef;
344 345
    
    # Locals
346 347 348 349 350
    my $experiment = Experiment->Lookup($pid, $eid);
    if (! defined($experiment)) {
	print "Could not find experiment object! Skipping ...\n";
	return -1;
    }
351 352 353 354 355 356 357
    my $creator = $experiment->GetCreator();
    if (! defined($creator)) {
	print "Could not find creator object! Skipping ...\n";
	return -1;
    }
    my $creator_uid = $creator->uid();
    
358 359
    print "Doing a '$dowhat' to batch experiment $pid/$eid\n";

360 361 362
    # Clean before starting.
    $experiment->CleanLogFiles();

363 364 365 366 367
    #
    # 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.
    #
368
    $logname = TBExptCreateLogFile($pid, $eid, "${dowhat}-batch");
369

370 371 372 373 374
    #
    # If this is a template instantiation, then do not set the logfile.
    # We still generate the log file, but it is not available via the
    # web interface. Needs more thought.
    #
375 376 377 378
    my $exptidx  = $exphash->{'idx'};
    my $instance = Template::Instance->LookupByExptidx($exptidx);
    if (defined($instance)) {
	$template = $instance->template();
379 380
    }

381 382 383
    #
    # Start up a child to run the guts. The parent waits. If the
    # experiment configures okay, the parent can return to try something
384
    # else.
385 386 387
    #
    $childpid = fork();
    if ($childpid) {
388

389
	print TBDateTimeFSSafe() . "\n";
390 391
	print "Child PID $childpid started to $dowhat $pid/$eid\n";

392
	waitpid($childpid, 0);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
393
	my $status = $? >> 8;
394

395
	print TBDateTimeFSSafe() . "\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
396
	print "Child PID $childpid exited with exit status $status\n";
397

398
        #
399
        # Close the log file.
400 401
	# The exp dir might be gone if the batch was killed/canceled.
        #
402
	if (-e $userdir) {
403 404
	    TBExptCloseLogFile($pid, $eid);
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
405
	return $status;
406
    }
407
    openlog($logname);
408 409
    TBExptSetLogFile($pid, $eid, $logname);
    TBExptOpenLogFile($pid, $eid);
410

411 412 413
    #
    # Get some user information. 
    #
414 415
    $user_name  = $creator->name();
    $user_email = $creator->email();
416
    
417 418
    chdir("$workdir") or
	fatal("Could not cd into $workdir!");
419 420 421 422 423

    #
    # Figure out the unix uid/gid that the experiment configuration is
    # going to run as. 
    #
424 425
    (undef,undef,$unix_uid) = getpwnam($creator_uid) or
	fatal("No such user $creator_uid");
426 427
    TBGroupUnixInfo($pid, $gid, \$unix_gid, \$unix_gname) or
	fatal("No such group $pid/$gid");
428

429 430 431 432 433
    #
    # 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.
    #
434
    my $glist = `id -G $creator_uid`;
435 436 437 438
    if ($glist =~ /^([\d ]*)$/) {
	$glist = $1;
    }
    else {
439
	fatal("Unexpected results from 'id -G $creator_uid': $glist");
440 441 442 443 444 445 446 447 448
    }
    #
    # 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;

449 450 451
    #
    # Change the ownership of the log file before we flip.
    #
452 453
    chown($unix_uid, $unix_gid, $logname) or
	fatal("Could not chown $logname to $unix_uid/$unix_gid!");
454 455

    # Flip to the user. We never flip back.
456 457 458
    $GID            = $unix_gid;
    $EGID           = $glist;
    $EUID = $UID    = $unix_uid;
459 460
    $ENV{'USER'}    = $creator_uid;
    $ENV{'LOGNAME'} = $creator_uid;
461
    
462
    if ($dowhat eq "start") {
463
	startexp($exphash);
464
    }
465
    elsif ($dowhat eq "swap") {
466
	swapexp($exphash);
467 468
    }
    elsif ($dowhat eq "cancel") {
469
	cancelexp($exphash);
470 471 472 473 474 475 476 477 478
    }
    exit(0);
}

#
# Try to start an experiment. Never returns.
# 
sub startexp($)
{
479
    my($exphash)  = @_;
480
    my($exit_status, $running, $query_result);
481

482
    my $attempts  = $exphash->{'attempts'};
483

484
    #
485
    # Try to swap the experiment in. 
486
    #
487
    system("$swapexp -b -s in $pid $eid");
488 489
    $exit_status = $? >> 8;
    $running     = 1;
490 491 492 493 494
    if ($exit_status) {
	$running = 0;
    }
    
    #
495 496
    # Look for cancelation. If we get a DB error on this, just continue cause
    # we can pick up the cancelation later.
497 498
    #
    $query_result =
499
	DBQueryWarn("select canceled from experiments ".
500
		    "where eid='$eid' and pid='$pid'");
501

502
    if ($query_result) {
503
	my ($canceled) = $query_result->fetchrow_array();
504
	$exphash->{'canceled'} = $canceled;
505

506
	# Yuck: This is strictly for the benefit of swapexp() below.
507
	$exphash->{'state'} = EXPTSTATE_ACTIVE
508 509 510 511
	    if ($running);
	
	if ($canceled) {
	    # Look at the cancel flag.
512
	    if ($canceled == EXPTCANCEL_TERM) {
513
		cancelexp($exphash);
514
	    }
515
	    elsif ($canceled == EXPTCANCEL_SWAP) {
516
		swapexp($exphash);
517 518 519 520
	    }
	    else {
		print "Improper cancel flag: $canceled\n";
	    }
521 522 523
	    #
	    # Never returns, but just to be safe ...
	    #
524
	    exit(-1);
525 526
	}
    }
527 528 529 530

    #
    # If the configuration failed for lack of nodes, then don't send
    # email unless the number of attempts starts to get big.
531
    #
532 533 534
    # 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. 
535 536
    #
    if (! $running) {
537 538 539 540
	#
	# XXX - What if this update fails?
	# 
	$query_result = 
541
	    DBQueryWarn("update experiments set attempts=attempts+1 ".
542
			"where eid='$eid' and pid='$pid'");
543

544
	#
545 546 547 548
	# 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.
549 550
	# We leave it up the user to kill the batch if it looks like its
	# never going to work.
551
	#
552
	if ($exit_status == 255) {
553
	    TBBatchUnLockExp($pid, $eid, EXPTSTATE_SWAPPED());
554
	    
555
	    email_status("Experiment startup has failed with a fatal error!\n".
556 557
			 "Batch has been dequeued so that you may check it.");

558
	    exit($exit_status);
559
	}
560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577
	
	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($msg);
	}
578
	TBBatchUnLockExp($pid, $eid, EXPTSTATE_QUEUED());
579
	exit($exit_status);
580 581 582
    }

    #
583
    # Well, it configured! We can now unlock it.
584
    #
585
    TBBatchUnLockExp($pid, $eid);
586
    
587
    email_status("Batch Mode experiment $pid/$eid is now running!\n".
588
		 "Please consult the Web interface to see how it is doing.");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
589

590
    #
591
    # Done with this phase. Must exit.
592
    #
593 594
    exit(0);
}
595

596
#
597
# A batch has completed. Swap it out. 
598
#
599
sub swapexp($;$)
600
{
601 602 603
    my($exphash)  = @_;
    my $canceled  = $exphash->{'canceled'};
    my $running   = ($exphash->{'state'} eq EXPTSTATE_ACTIVE);
604

605
    if ($running) {
606 607 608 609 610
	if (defined($template)) {
	    my $guid = $template->guid();
	    my $vers = $template->vers();
	    
	    system("$template_swapout -b -e $eid $guid/$vers");
611 612 613 614
	}
	else {
	    system("$swapexp -b -s out $pid $eid");
	}
615 616 617 618 619 620
	if ($?) {
	    #
	    # TB admin is going to have to clean up. 
	    # 
	    fatal("Swapping out Batch Mode experiment $pid/$eid");
	}
621 622 623 624 625
    }
    #
    # Set the state to paused to ensure that it is not run again until
    # the user wants to.
    # 
626 627
    TBSetCancelFlag($pid, $eid, EXPTCANCEL_CLEAR);
    TBBatchUnLockExp($pid, $eid);
628
    
629 630 631 632 633
    if ($canceled) {
	email_status("Batch Mode experiment $pid/$eid has been stopped!");
    }
    else {
	email_status("Batch Mode experiment $pid/$eid has finished!");
634
    }
635
   
636
    #
637
    # Child must exit!
638
    #
639 640 641 642 643 644
    exit(0);
}

#
# Cancel an experiment. Never returns.
#
645
sub cancelexp($)
646
{
647
    my($exphash) = @_;
648
    
649 650 651 652
    #
    # It does not matter if the experiment is running; endexp does the
    # right thing.
    # 
653 654 655 656 657
    if (defined($template)) {
	my $guid = $template->guid();
	my $vers = $template->vers();
	
	system("$template_swapout -b -e $eid $guid/$vers");
658 659
    }
    else {
660
	system("$endexp -b $pid,$eid");
661
    }
662 663 664 665 666
    if ($?) {
	#
	# TB admin is going to have to clean up. 
	# 
	fatal("Terminating Batch Mode experiment $pid/$eid");
667
    }
668
    donotify("Your Batch Mode experiment has been canceled!", "Canceled", 0);
669 670 671 672 673 674 675
   
    #
    # Child must exit!
    #
    exit(0);
}

676 677 678 679 680 681
#
# Check experiment status. Looks to see if all of the nodes in an
# experiment have reported in.
#
sub isexpdone($)
{
682
    my($exphash)  = @_;
683
    my($query_result, @row);
684 685
    
    # Global vars
686 687
    $eid = $exphash->{'eid'};
    $pid = $exphash->{'pid'};
688

689
    print TBDateTimeFSSafe() . "\n";
690 691 692 693 694 695
    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.
    #
    $query_result =
696 697 698
	DBQueryWarn("SELECT startstatus,bootstatus FROM nodes ".
		    "LEFT JOIN reserved ON nodes.node_id=reserved.node_id ".
		    "WHERE reserved.eid='$eid' and reserved.pid='$pid'");
699 700 701 702 703

    if (! $query_result) {
	return 0;
    }

704 705 706 707 708 709 710 711 712 713 714 715 716
    #
    # 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;
717 718
	}
    }
719
    return 1;
720 721 722 723
}

#
# Start up a child, and set its descriptors talking to a log file.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
724
# The log file already exists, created with mktemp above.
725 726 727 728 729 730 731
# 
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
732 733
    # STDOUT away from the pipe. Otherwise the caller will continue to wait
    # even though the parent has exited. 
734 735
    #
    open(STDIN, "< /dev/null") or
736
	fatal("opening /dev/null for STDIN: $!");
737 738 739 740 741 742

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

743 744 745
    STDOUT->autoflush(1);
    STDERR->autoflush(1);

746 747 748
    return 0;
}

749 750 751 752 753
#
# 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.
# 
754
sub fatal($)
755 756 757
{
    my($mesg) = $_[0];

758
    donotify($mesg, "Failure", 1);
759 760 761 762

    exit(-1);
}

763 764 765
#
# Something the user cares about. 
# 
766
sub email_status($)
767
{
768
    my($mesg) = $_[0];
769

770
    donotify($mesg, "Status", 0);
771 772
}

773
sub donotify($$$)
774
{
775
    my($mesg, $subtext, $iserr) = @_;
776
    my($subject, $from, $to, $hdrs);
777
    my $MAIL;
778

779 780 781
    $mesg = "$mesg\n";

    print STDOUT "$mesg";
782

783
    $subject = "Batch Mode Experiment $subtext $pid/$eid";
784
    $from    = $TBOPS;
785 786 787 788 789 790
    $hdrs    = "Reply-To: $TBOPS";
    
    #
    # An error goes just to Testbed Operations. Normal status messages go
    # to the user and to the Testbed Logs address.
    # 
791
    if ($iserr) {
792
	$to = "$TBOPS";
793 794
    }
    else {
795 796 797
	$to   = "$user_name <$user_email>";
	$hdrs = "Bcc: $TBLOGS\n".
	        "$hdrs";
798 799
    }

800
    SENDMAIL($to, $subject, $mesg, $from, $hdrs,
801
	     ($logname, "assign.log", $nsfile));
802
}
803