batch_daemon.in 15.2 KB
Newer Older
1
#!/usr/bin/perl -wT
Leigh B. Stoller's avatar
Leigh B. Stoller committed
2 3 4 5 6 7 8

#
# EMULAB-COPYRIGHT
# Copyright (c) 2000-2002 University of Utah and the Flux Group.
# 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
#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;

44 45 46 47
my $tbbindir = "$TB/bin/";
my $batchdir = "$TB/batch";
my $startexp = "$TB/bin/startexp";
my $endexp   = "$TB/bin/endexp";
48
my $savelogs = "$TB/bin/savelogs";
49
my $avail    = "$TB/sbin/avail";
50 51
my $batchlog = "$TB/log/batchlog";
my $projroot = "/proj";
52
my $debug    = 0;
53

54 55 56 57 58
my $BSTATE_POSTED	= BATCHSTATE_POSTED;
my $BSTATE_ACTIVATING	= BATCHSTATE_ACTIVATING;
my $BSTATE_RUNNING	= BATCHSTATE_RUNNING;
my $BSTATE_TERMINATING	= BATCHSTATE_TERMINATING;

59 60 61 62 63 64
#
# 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
65
my $gid;
66
my $logname;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
67
my $nsfile;
68 69
my $userdir;
my $workdir;
70
my $user_name  = "Testbed Operations";
71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94
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($optlist, \%options)) {
    usage();
}
if (@ARGV != 0) {
    usage();
}
95 96 97
if (defined($options{"d"})) {
    $debug = $options{"d"};
}
98 99

# Go to ground.
100
if (! $debug) {
101 102 103
    if (TBBackGround($batchlog)) {
	exit(0);
    }
104
}
105 106 107 108 109

#
# Loop, looking for batch experiments that want to run.
# 
while (1) {
110
    my($count, $i, $query_result, $pending_result, $running_result);
111
    my(%row, %pending_row);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
112

113
    #
114
    # Need to lock the table here because of cancelation in endexp.
115 116 117 118 119 120
    # 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!
121
    # Thats would ne annoying to users.
122
    #
123
    # So, now you're wondering what my selection criteria is? Well, its
124 125
    # damn simplistic. I set the "started" datetime field each attempt,
    # and I pick the batch_experiment with the oldest time, thereby cycling
126 127 128 129
    # 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).
130 131
    #
    $query_result =
132 133
	DBQuery("lock tables experiments write, experiments as e1 write, ".
		"experiments as e2 write");
134 135
    if (! $query_result) {
	print "DB Error locking tables. Waiting a bit ...\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
136
	goto pause;
137 138 139
    }
    
    $pending_result =
140 141 142 143 144 145 146 147 148 149 150 151
	DBQueryWarn("SELECT e1.* FROM experiments as e1 ".
		    "left join experiments as e2 on ".
		    " e2.expt_head_uid=e1.expt_head_uid and ".
		    " e2.batchmode=1 and e2.batchstate='$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() - ".
		    "      UNIX_TIMESTAMP(e1.expt_start) > (60 * 15)))) ".
		    "ORDER BY e1.expt_start LIMIT 1");
152

153
    $running_result =
154 155 156
	DBQuery("select * from experiments ".
		"where batchmode=1 and batchstate='$BSTATE_RUNNING' ".
		"ORDER BY expt_start");
157 158 159

    if (!$pending_result || !$running_result) {
	print "DB Error getting batch info. Waiting a bit ...\n";
160
	DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
161
	goto pause;
162 163 164
    }

    if (!$pending_result->numrows && !$running_result->numrows) {
165
	DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
166
	goto pause;
167 168 169
    }

    #
170
    # If we have a pending experiment to run, set its state to configuring
171
    # right away, while we have the tables locked. This prevents endexp
172
    # from seeing it as something it can cancel.
173
    #
174 175 176 177 178 179 180 181
    if ($pending_result->numrows) {
	%pending_row = $pending_result->fetchhash();

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

	$query_result = 
182 183 184
	    DBQuery("update experiments set expt_start=now(), ".
		    "batchstate='$BSTATE_ACTIVATING' ".
		    "where eid='$eid' and pid='$pid'");
185 186 187

	if (! $query_result) {
	    print "DB error setting batch $pid/$eid to configuring.\n";
188
	    DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
189
	    goto pause;
190 191
	}
    }
192
    DBQueryWarn("unlock tables");
193

194 195 196 197 198 199 200 201 202 203 204
    #
    # 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.
    #
205
    while (%row = $running_result->fetchhash()) {
206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222
	my $canceled = $row{'canceled'};
	if ($canceled) {
	    dosomething("cancel", %row);
	    next;
	}
	if (isexpdone(%row)) {
	    dosomething("end", %row);
	    next;
	}
    }

    #
    # Finally start an actual experiment!
    #
    if ($pending_result->numrows) {
	dosomething("start", %pending_row);
    }
Leigh B. Stoller's avatar
Leigh B. Stoller committed
223 224
  pause:
    sleep(30);
225 226 227
}

#
228
# Do something as the user. Either, start, end, or cancel an experiment.
229
#
230
sub dosomething($$)
231
{
232 233
    my($dowhat)   = shift;
    my(%exphash)  = @_;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
234
    my($unix_uid, $unix_gid, $row, $query_result);
235 236

    # Global vars
237 238 239
    $eid     = $exphash{'eid'};
    $pid     = $exphash{'pid'};
    $gid     = $exphash{'gid'};
240 241
    $userdir = $exphash{'path'};
    $workdir = TBExptWorkDir($pid, $eid);
242 243 244 245
    $nsfile  = "$eid.ns";
    
    # Locals
    my $creator = $exphash{'expt_head_uid'};
246

247 248 249 250 251 252 253
    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.
    #
254
    $logname = TBExptCreateLogFile($pid, $eid, "${dowhat}-batch");
255

256 257 258
    #
    # Start up a child to run the guts. The parent waits. If the
    # experiment configures okay, the parent can return to try something
259
    # else.
260 261 262
    #
    $childpid = fork();
    if ($childpid) {
263 264
	print "Child PID $childpid started to $dowhat $pid/$eid\n";

265
	waitpid($childpid, 0);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
266
	my $status = $? >> 8;
267

Leigh B. Stoller's avatar
Leigh B. Stoller committed
268
	print "Child PID $childpid exited with exit status $status\n";
269

270
        #
271
        # Close the log file.
272 273
	# The exp dir might be gone if the batch was killed/canceled.
        #
274
	if (-e $userdir) {
275 276
	    TBExptCloseLogFile($pid, $eid);
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
277
	return $status;
278
    }
279
    openlog($logname);
280 281
    TBExptSetLogFile($pid, $eid, $logname);
    TBExptOpenLogFile($pid, $eid);
282

283 284 285
    #
    # Get some user information. 
    #
286
    if (!UserDBInfo($creator, \$user_name, \$user_email)) {
287
	fatal("DB Error getting user information for uid $creator");
288
    }
289
    
290 291
    chdir("$workdir") or
	fatal("Could not cd into $workdir!");
292 293 294 295 296

    #
    # Figure out the unix uid/gid that the experiment configuration is
    # going to run as. 
    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
297
    (undef,undef,$unix_uid) = getpwnam($creator) or
298
	fatal("No such user $creator");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
299 300
    (undef,undef,$unix_gid) = getgrnam($gid) or
	fatal("No such group $gid");
301

302 303 304
    #
    # Change the ownership of the log file before we flip.
    #
305 306
    chown($unix_uid, $unix_gid, $logname) or
	fatal("Could not chown $logname to $unix_uid/$unix_gid!");
307 308

    # Flip to the user. We never flip back.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
309 310
    $EGID = $GID = $unix_gid;
    $EUID = $UID = $unix_uid;
311
    $ENV{'USER'} = $creator;
312
    
313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330
    if ($dowhat eq "start") {
	startexp(%exphash);
    }
    elsif ($dowhat eq "end") {
	endexp(%exphash);
    }
    elsif ($dowhat eq "cancel") {
	cancelexp(1, %exphash);
    }
    exit(0);
}

#
# Try to start an experiment. Never returns.
# 
sub startexp($)
{
    my(%exphash)  = @_;
331
    my($exit_status, $running, $query_result);
332 333 334

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

335
    #
336
    # Try to start the experiment. 
337
    #
338
    system("$startexp -b -g $gid $pid $eid $nsfile");
339 340
    $exit_status = $? >> 8;
    $running     = 1;
341 342 343 344 345
    if ($exit_status) {
	$running = 0;
    }
    
    #
346 347
    # Look for cancelation. If we get a DB error on this, just continue cause
    # we can pick up the cancelation later.
348 349
    #
    $query_result =
350
	DBQueryWarn("select canceled from experiments ".
351
		    "where eid='$eid' and pid='$pid'");
352

353 354
    if ($query_result) {
	@row = $query_result->fetchrow_array();
355

356 357 358 359 360 361 362 363
	if ($row[0]) {
	    cancelexp($running);
	    #
	    # Never returns, but just to be safe ...
	    #
	    exit(0);
	}
    }
364 365 366 367

    #
    # If the configuration failed for lack of nodes, then don't send
    # email unless the number of attempts starts to get big.
368
    #
369 370 371
    # 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. 
372 373
    #
    if (! $running) {
374 375 376 377
	#
	# XXX - What if this update fails?
	# 
	$query_result = 
378
	    DBQueryWarn("update experiments set attempts=attempts+1 ".
379
			"where eid='$eid' and pid='$pid'");
380

381 382 383 384 385 386 387 388 389
	#
	# 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 || $exit_status == -1) {
	    email_status("Experiment startup has failed with a fatal error!\n".
			 "Batch has been removed from the system.");
390
	    TBExptDestroy($pid, $eid);
391
	    exit($exit_status);
392
	}
393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418
	
	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.
	# 
	SetExpState($pid, $eid, EXPTSTATE_NEW);
	TBSetBatchState($pid, $eid, $BSTATE_POSTED);
	    
419
	exit($exit_status);
420 421 422 423 424
    }

    #
    # Well, it configured! Lets set it state to running.
    #
425 426
    TBSetBatchState($pid, $eid, $BSTATE_RUNNING);
    
427
    email_status("Batch Mode experiment $pid/$eid is now running!\n".
428
		 "Please consult the Web interface to see how it is doing.");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
429

430
    #
431
    # Done with this phase. Must exit.
432
    #
433 434
    exit(0);
}
435

436 437 438 439 440 441
#
# End an experiment. Never returns.
#
sub endexp($)
{
    my(%exphash)  = @_;
442

443 444 445
    #
    # Save tiplogs
    #
446 447
    system("$savelogs $pid $eid");

448 449 450 451 452
    #
    # Have to set the state to terminating or else endexp will not accept it.
    # 
    TBSetBatchState($pid, $eid, $BSTATE_TERMINATING);
    
453
    system("$endexp -b $pid $eid");
454
    if ($?) {
455 456 457 458 459
	#
	# TB admin is going to have to clean up. 
	# 
	fatal("Terminating Batch Mode experiment $pid/$eid");
    }
460
    email_status("Batch Mode experiment $pid/$eid has finished!");
461
   
462
    #
463
    # Child must exit!
464
    #
465 466 467 468 469 470 471 472 473 474 475
    exit(0);
}

#
# Cancel an experiment. Never returns.
#
sub cancelexp($$)
{
    my($running) = shift;
    my(%exphash) = @_;
    
476 477
    TBSetBatchState($pid, $eid, $BSTATE_TERMINATING);
    
478 479
    if ($running) {
	system("$endexp -b $pid $eid");
480 481 482 483 484 485 486 487 488
	if ($?) {
	    #
	    # TB admin is going to have to clean up. 
	    # 
	    fatal("Terminating Batch Mode experiment $pid/$eid");
	}
    }
    else {
	TBExptDestroy($pid, $eid);
489
    }
490
    donotify("Your Batch Mode experiment has been canceled!", "Canceled", 0);
491 492 493 494 495 496 497
   
    #
    # Child must exit!
    #
    exit(0);
}

498 499 500 501 502 503 504
#
# Check experiment status. Looks to see if all of the nodes in an
# experiment have reported in.
#
sub isexpdone($)
{
    my(%exphash)  = @_;
505
    my($query_result, @row);
506 507 508 509 510 511 512 513 514 515 516
    
    # 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.
    #
    $query_result =
517 518 519
	DBQueryWarn("SELECT startstatus,bootstatus FROM nodes ".
		    "LEFT JOIN reserved ON nodes.node_id=reserved.node_id ".
		    "WHERE reserved.eid='$eid' and reserved.pid='$pid'");
520 521 522 523 524

    if (! $query_result) {
	return 0;
    }

525 526 527 528 529 530 531 532 533 534 535 536 537
    #
    # 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;
538 539
	}
    }
540
    return 1;
541 542 543 544
}

#
# Start up a child, and set its descriptors talking to a log file.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
545
# The log file already exists, created with mktemp above.
546 547 548 549 550 551 552
# 
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
553 554
    # STDOUT away from the pipe. Otherwise the caller will continue to wait
    # even though the parent has exited. 
555 556
    #
    open(STDIN, "< /dev/null") or
557
	fatal("opening /dev/null for STDIN: $!");
558 559 560 561 562 563

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

564 565 566
    STDOUT->autoflush(1);
    STDERR->autoflush(1);

567 568 569
    return 0;
}

570 571 572 573 574
#
# 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.
# 
575
sub fatal($)
576 577 578
{
    my($mesg) = $_[0];

579
    donotify($mesg, "Failure", 1);
580 581 582 583

    exit(-1);
}

584 585 586
#
# Something the user cares about. 
# 
587
sub email_status($)
588
{
589
    my($mesg) = $_[0];
590

591
    donotify($mesg, "Status", 0);
592 593
}

594
sub donotify($$$)
595
{
596
    my($mesg, $subtext, $iserr) = @_;
597
    my($subject, $from, $to, $hdrs);
598
    my $MAIL;
599

600 601 602
    $mesg = "$mesg\n";

    print STDOUT "$mesg";
603

604
    $subject = "Batch Mode Experiment $subtext $pid/$eid";
605
    $from    = $TBOPS;
606 607 608 609 610 611
    $hdrs    = "Reply-To: $TBOPS";
    
    #
    # An error goes just to Testbed Operations. Normal status messages go
    # to the user and to the Testbed Logs address.
    # 
612
    if ($iserr) {
613
	$to = "$TBOPS";
614 615
    }
    else {
616 617 618
	$to   = "$user_name <$user_email>";
	$hdrs = "Bcc: $TBLOGS\n".
	        "$hdrs";
619 620
    }

621
    SENDMAIL($to, $subject, $mesg, $from, $hdrs,
622
	     ($logname, "assign.log", $nsfile));
623
}