batch_daemon.in 15.9 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
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
my  $optlist = "d";
28 29 30 31 32 33 34

#
# Configure variables
#
my $TB       = "@prefix@";
my $DBNAME   = "@TBDBNAME@";
my $TBOPS    = "@TBOPSEMAIL@";
35
my $TBLOGS   = "@TBLOGSEMAIL@";
36

37 38 39 40 41 42 43 44
#
# Only root can run this script.
#
if ($UID) {
    die("*** $0:\n".
	"    Only root can run this script!\n");
}

45 46 47 48 49 50 51
#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;

52 53 54
# Be careful not to exit on transient error
$libdb::DBQUERY_MAXTRIES = 10;

55
my $tbbindir = "$TB/bin/";
56
my $swapexp  = "$TB/bin/swapexp";
57
my $endexp   = "$TB/bin/endexp";
58
my $savelogs = "$TB/bin/savelogs";
59
my $avail    = "$TB/sbin/avail";
60 61
my $batchlog = "$TB/log/batchlog";
my $projroot = "/proj";
62
my $debug    = 0;
63

64 65 66 67
my $BSTATE_POSTED	= BATCHSTATE_POSTED;
my $BSTATE_ACTIVATING	= BATCHSTATE_ACTIVATING;
my $BSTATE_RUNNING	= BATCHSTATE_RUNNING;
my $BSTATE_TERMINATING	= BATCHSTATE_TERMINATING;
68
my $BSTATE_PAUSED	= BATCHSTATE_PAUSED;
69

70 71 72 73 74 75
#
# 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
76
my $gid;
77
my $logname;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
78
my $nsfile;
79 80
my $userdir;
my $workdir;
81
my $user_name  = "Testbed Operations";
82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105
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($optlist, \%options)) {
    usage();
}
if (@ARGV != 0) {
    usage();
}
106 107 108
if (defined($options{"d"})) {
    $debug = $options{"d"};
}
109 110

# Go to ground.
111
if (! $debug) {
112 113 114
    if (TBBackGround($batchlog)) {
	exit(0);
    }
115
}
116 117 118 119 120

#
# Loop, looking for batch experiments that want to run.
# 
while (1) {
121
    my($count, $i, $query_result, $pending_result, $running_result);
122
    my(%row, %pending_row);
123 124 125 126 127
    my $retry_wait = TBGetSiteVar("batch/retry_wait");

    # Do not allow zero!
    $retry_wait = 60
	if ($retry_wait == 0);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
128

129
    #
130
    # Need to lock the table here because of cancelation in endexp.
131 132 133 134 135 136
    # 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!
137
    # Thats would ne annoying to users.
138
    #
139
    # So, now you're wondering what my selection criteria is? Well, its
140 141
    # damn simplistic. I set the "started" datetime field each attempt,
    # and I pick the batch_experiment with the oldest time, thereby cycling
142 143 144 145
    # 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).
146 147
    #
    $query_result =
148 149
	DBQuery("lock tables experiments write, experiments as e1 write, ".
		"experiments as e2 write");
150 151
    if (! $query_result) {
	print "DB Error locking tables. Waiting a bit ...\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
152
	goto pause;
153 154 155
    }
    
    $pending_result =
156 157 158 159 160 161 162 163 164 165
	DBQueryWarn("SELECT e1.* FROM experiments as e1 ".
		    "left join experiments as e2 on ".
		    " e2.expt_head_uid=e1.expt_head_uid and ".
		    " e2.batchmode=1 and e2.batchstate='$BSTATE_RUNNING' and ".
		    " e1.pid=e2.pid and e1.eid!=e2.eid ".
		    "WHERE e2.eid is null and ".
		    " e1.batchmode=1 and e1.canceled=0 and ".
		    " e1.batchstate='$BSTATE_POSTED' and ".
		    "  (e1.attempts=0 or ".
		    "    ((UNIX_TIMESTAMP() - ".
166
		    "      UNIX_TIMESTAMP(e1.expt_start) > ($retry_wait)))) ".
167
		    "ORDER BY e1.expt_start LIMIT 1");
168

169
    $running_result =
170 171 172
	DBQuery("select * from experiments ".
		"where batchmode=1 and batchstate='$BSTATE_RUNNING' ".
		"ORDER BY expt_start");
173 174 175

    if (!$pending_result || !$running_result) {
	print "DB Error getting batch info. Waiting a bit ...\n";
176
	DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
177
	goto pause;
178 179 180
    }

    if (!$pending_result->numrows && !$running_result->numrows) {
181
	DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
182
	goto pause;
183 184 185
    }

    #
186
    # If we have a pending experiment to run, set its state to configuring
187
    # right away, while we have the tables locked. This prevents endexp
188
    # from seeing it as something it can cancel.
189
    #
190 191 192 193 194 195 196 197
    if ($pending_result->numrows) {
	%pending_row = $pending_result->fetchhash();

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

	$query_result = 
198 199 200
	    DBQuery("update experiments set expt_start=now(), ".
		    "batchstate='$BSTATE_ACTIVATING' ".
		    "where eid='$eid' and pid='$pid'");
201 202 203

	if (! $query_result) {
	    print "DB error setting batch $pid/$eid to configuring.\n";
204
	    DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
205
	    goto pause;
206 207
	}
    }
208
    DBQueryWarn("unlock tables");
209

210 211 212 213 214 215 216 217 218 219 220
    #
    # 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
    # we fire up again and look for them in the event that paper goes down.
    #
221
    while (%row = $running_result->fetchhash()) {
222 223
	my $canceled = $row{'canceled'};
	if ($canceled) {
224 225 226 227 228 229 230
	    # Look at the cancel flag.
	    if ($canceled == BATCHMODE_CANCELTERM) {
		dosomething("cancel", %row);
	    }
	    elsif ($canceled == BATCHMODE_CANCELSWAP) {
		dosomething("swap", %row);
	    }
231 232 233
	    next;
	}
	if (isexpdone(%row)) {
234
	    dosomething("swap", %row);
235 236 237 238 239 240 241 242 243 244
	    next;
	}
    }

    #
    # Finally start an actual experiment!
    #
    if ($pending_result->numrows) {
	dosomething("start", %pending_row);
    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
245 246
  pause:
    sleep(30);
247 248 249
}

#
250
# Do something as the user. Either, start, end, or cancel an experiment.
251
#
252
sub dosomething($$)
253
{
254 255
    my($dowhat)   = shift;
    my(%exphash)  = @_;
256
    my($unix_uid, $unix_gid, $unix_gname, $row, $query_result);
257 258

    # Global vars
259 260 261
    $eid     = $exphash{'eid'};
    $pid     = $exphash{'pid'};
    $gid     = $exphash{'gid'};
262 263
    $userdir = $exphash{'path'};
    $workdir = TBExptWorkDir($pid, $eid);
264 265 266 267
    $nsfile  = "$eid.ns";
    
    # Locals
    my $creator = $exphash{'expt_head_uid'};
268

269 270 271 272 273 274 275
    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.
    #
276
    $logname = TBExptCreateLogFile($pid, $eid, "${dowhat}-batch");
277

278 279 280
    #
    # Start up a child to run the guts. The parent waits. If the
    # experiment configures okay, the parent can return to try something
281
    # else.
282 283 284
    #
    $childpid = fork();
    if ($childpid) {
285 286
	print "Child PID $childpid started to $dowhat $pid/$eid\n";

287
	waitpid($childpid, 0);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
288
	my $status = $? >> 8;
289

Leigh B. Stoller's avatar
Leigh B. Stoller committed
290
	print "Child PID $childpid exited with exit status $status\n";
291

292
        #
293
        # Close the log file.
294 295
	# The exp dir might be gone if the batch was killed/canceled.
        #
296
	if (-e $userdir) {
297 298
	    TBExptCloseLogFile($pid, $eid);
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
299
	return $status;
300
    }
301
    openlog($logname);
302 303
    TBExptSetLogFile($pid, $eid, $logname);
    TBExptOpenLogFile($pid, $eid);
304

305 306 307
    #
    # Get some user information. 
    #
308
    if (!UserDBInfo($creator, \$user_name, \$user_email)) {
309
	fatal("DB Error getting user information for uid $creator");
310
    }
311
    
312 313
    chdir("$workdir") or
	fatal("Could not cd into $workdir!");
314 315 316 317 318

    #
    # Figure out the unix uid/gid that the experiment configuration is
    # going to run as. 
    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
319
    (undef,undef,$unix_uid) = getpwnam($creator) or
320
	fatal("No such user $creator");
321 322
    TBGroupUnixInfo($pid, $gid, \$unix_gid, \$unix_gname) or
	fatal("No such group $pid/$gid");
323

324 325 326
    #
    # Change the ownership of the log file before we flip.
    #
327 328
    chown($unix_uid, $unix_gid, $logname) or
	fatal("Could not chown $logname to $unix_uid/$unix_gid!");
329 330

    # Flip to the user. We never flip back.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
331 332
    $EGID = $GID = $unix_gid;
    $EUID = $UID = $unix_uid;
333
    $ENV{'USER'} = $creator;
334
    
335 336 337
    if ($dowhat eq "start") {
	startexp(%exphash);
    }
338 339
    elsif ($dowhat eq "swap") {
	swapexp(%exphash);
340 341 342 343 344 345 346 347 348 349 350 351 352
    }
    elsif ($dowhat eq "cancel") {
	cancelexp(1, %exphash);
    }
    exit(0);
}

#
# Try to start an experiment. Never returns.
# 
sub startexp($)
{
    my(%exphash)  = @_;
353
    my($exit_status, $running, $query_result);
354 355 356

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

357
    #
358
    # Try to swap the experiment in. 
359
    #
360
    system("$swapexp -b -s in $pid $eid");
361 362
    $exit_status = $? >> 8;
    $running     = 1;
363 364 365 366 367
    if ($exit_status) {
	$running = 0;
    }
    
    #
368 369
    # Look for cancelation. If we get a DB error on this, just continue cause
    # we can pick up the cancelation later.
370 371
    #
    $query_result =
372
	DBQueryWarn("select canceled from experiments ".
373
		    "where eid='$eid' and pid='$pid'");
374

375 376
    if ($query_result) {
	@row = $query_result->fetchrow_array();
377

378 379 380 381 382 383 384 385
	if ($row[0]) {
	    cancelexp($running);
	    #
	    # Never returns, but just to be safe ...
	    #
	    exit(0);
	}
    }
386 387 388 389

    #
    # If the configuration failed for lack of nodes, then don't send
    # email unless the number of attempts starts to get big.
390
    #
391 392 393
    # 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. 
394 395
    #
    if (! $running) {
396 397 398 399
	#
	# XXX - What if this update fails?
	# 
	$query_result = 
400
	    DBQueryWarn("update experiments set attempts=attempts+1 ".
401
			"where eid='$eid' and pid='$pid'");
402

403 404 405 406 407 408 409 410 411
	#
	# The exit value is important. If its -1 or 1, thats bad. Kill the
	# batch off. Anything else implies an assign violation that is
	# (hopefully) transient. We leave it up the user to kill cancel the
	# batch if it looks like its never going to work.
	#
	if ($exit_status == 1 || $exit_status == -1) {
	    email_status("Experiment startup has failed with a fatal error!\n".
			 "Batch has been removed from the system.");
412
	    TBExptDestroy($pid, $eid);
413
	    exit($exit_status);
414
	}
415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437
	
	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);
	}

	#
	# There is some state that needs to be reset so that another
	# attempt can be made.
	# 
438
	SetExpState($pid, $eid, EXPTSTATE_SWAPPED);
439 440
	TBSetBatchState($pid, $eid, $BSTATE_POSTED);
	    
441
	exit($exit_status);
442 443 444 445 446
    }

    #
    # Well, it configured! Lets set it state to running.
    #
447 448
    TBSetBatchState($pid, $eid, $BSTATE_RUNNING);
    
449
    email_status("Batch Mode experiment $pid/$eid is now running!\n".
450
		 "Please consult the Web interface to see how it is doing.");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
451

452
    #
453
    # Done with this phase. Must exit.
454
    #
455 456
    exit(0);
}
457

458
#
459
# A batch has completed. Swap it out. 
460
#
461
sub swapexp($)
462 463
{
    my(%exphash)  = @_;
464
    my $canceled  = $exphash{'canceled'};
465

466
    #
467
    # Have to set the state to terminating or else swapexp will not accept it.
468 469 470
    # 
    TBSetBatchState($pid, $eid, $BSTATE_TERMINATING);
    
471
    system("$swapexp -b -s out $pid $eid");
472
    if ($?) {
473 474 475
	#
	# TB admin is going to have to clean up. 
	# 
476 477 478 479 480 481 482 483 484 485 486 487 488
	fatal("Swapping out Batch Mode experiment $pid/$eid");
    }
    #
    # Set the state to paused to ensure that it is not run again until
    # the user wants to.
    # 
    TBSetBatchState($pid, $eid, $BSTATE_PAUSED);

    if ($canceled) {
	email_status("Batch Mode experiment $pid/$eid has been stopped!");
    }
    else {
	email_status("Batch Mode experiment $pid/$eid has finished!");
489
    }
490
   
491
    #
492
    # Child must exit!
493
    #
494 495 496 497 498 499 500 501 502 503 504
    exit(0);
}

#
# Cancel an experiment. Never returns.
#
sub cancelexp($$)
{
    my($running) = shift;
    my(%exphash) = @_;
    
505 506
    TBSetBatchState($pid, $eid, $BSTATE_TERMINATING);
    
507 508 509 510 511 512
    system("$endexp -b $pid $eid");
    if ($?) {
	#
	# TB admin is going to have to clean up. 
	# 
	fatal("Terminating Batch Mode experiment $pid/$eid");
513
    }
514
    donotify("Your Batch Mode experiment has been canceled!", "Canceled", 0);
515 516 517 518 519 520 521
   
    #
    # Child must exit!
    #
    exit(0);
}

522 523 524 525 526 527 528
#
# Check experiment status. Looks to see if all of the nodes in an
# experiment have reported in.
#
sub isexpdone($)
{
    my(%exphash)  = @_;
529
    my($query_result, @row);
530 531 532 533 534 535 536 537 538 539 540
    
    # Global vars
    $eid = $exphash{'eid'};
    $pid = $exphash{'pid'};

    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.
    #
    $query_result =
541 542 543
	DBQueryWarn("SELECT startstatus,bootstatus FROM nodes ".
		    "LEFT JOIN reserved ON nodes.node_id=reserved.node_id ".
		    "WHERE reserved.eid='$eid' and reserved.pid='$pid'");
544 545 546 547 548

    if (! $query_result) {
	return 0;
    }

549 550 551 552 553 554 555 556 557 558 559 560 561
    #
    # 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;
562 563
	}
    }
564
    return 1;
565 566 567 568
}

#
# Start up a child, and set its descriptors talking to a log file.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
569
# The log file already exists, created with mktemp above.
570 571 572 573 574 575 576
# 
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
577 578
    # STDOUT away from the pipe. Otherwise the caller will continue to wait
    # even though the parent has exited. 
579 580
    #
    open(STDIN, "< /dev/null") or
581
	fatal("opening /dev/null for STDIN: $!");
582 583 584 585 586 587

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

588 589 590
    STDOUT->autoflush(1);
    STDERR->autoflush(1);

591 592 593
    return 0;
}

594 595 596 597 598
#
# 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.
# 
599
sub fatal($)
600 601 602
{
    my($mesg) = $_[0];

603
    donotify($mesg, "Failure", 1);
604 605 606 607

    exit(-1);
}

608 609 610
#
# Something the user cares about. 
# 
611
sub email_status($)
612
{
613
    my($mesg) = $_[0];
614

615
    donotify($mesg, "Status", 0);
616 617
}

618
sub donotify($$$)
619
{
620
    my($mesg, $subtext, $iserr) = @_;
621
    my($subject, $from, $to, $hdrs);
622
    my $MAIL;
623

624 625 626
    $mesg = "$mesg\n";

    print STDOUT "$mesg";
627

628
    $subject = "Batch Mode Experiment $subtext $pid/$eid";
629
    $from    = $TBOPS;
630 631 632 633 634 635
    $hdrs    = "Reply-To: $TBOPS";
    
    #
    # An error goes just to Testbed Operations. Normal status messages go
    # to the user and to the Testbed Logs address.
    # 
636
    if ($iserr) {
637
	$to = "$TBOPS";
638 639
    }
    else {
640 641 642
	$to   = "$user_name <$user_email>";
	$hdrs = "Bcc: $TBLOGS\n".
	        "$hdrs";
643 644
    }

645
    SENDMAIL($to, $subject, $mesg, $from, $hdrs,
646
	     ($logname, "assign.log", $nsfile));
647
}