batch_daemon.in 15.3 KB
Newer Older
1 2 3 4 5 6 7 8 9
#!/usr/bin/perl -wT
use English;
use Getopt::Std;

#
# Create a batch experiment.
#
# usage: batch_daemon
#
Leigh B. Stoller's avatar
Leigh B. Stoller committed
10 11
# TODO: Use "logger" instead of writing a log file.
#
12 13
sub usage()
{
14 15
    print STDOUT "Usage: batch_daemon [-d]\n" .
	"Use the -d option to prevent daemonization\n";
16 17
    exit(-1);
}
18
my  $optlist = "d";
19 20 21 22 23 24 25

#
# Configure variables
#
my $TB       = "@prefix@";
my $DBNAME   = "@TBDBNAME@";
my $TBOPS    = "@TBOPSEMAIL@";
26
my $TBLOGS   = "@TBLOGSEMAIL@";
27

28 29 30 31 32 33 34
#
# Testbed Support libraries
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;

35 36 37 38
my $tbbindir = "$TB/bin/";
my $batchdir = "$TB/batch";
my $startexp = "$TB/bin/startexp";
my $endexp   = "$TB/bin/endexp";
39
my $savelogs = "$TB/bin/savelogs";
40
my $avail    = "$TB/sbin/avail";
41 42
my $batchlog = "$TB/log/batchlog";
my $projroot = "/proj";
43
my $debug    = 0;
44 45
my $dirname;

46 47 48 49 50
my $BSTATE_POSTED	= BATCHSTATE_POSTED;
my $BSTATE_ACTIVATING	= BATCHSTATE_ACTIVATING;
my $BSTATE_RUNNING	= BATCHSTATE_RUNNING;
my $BSTATE_TERMINATING	= BATCHSTATE_TERMINATING;

51 52 53 54 55 56
#
# 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
57
my $gid;
58
my $logname;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
59
my $nsfile;
60
my $user_name  = "Testbed Operations";
61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84
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($optlist, \%options)) {
    usage();
}
if (@ARGV != 0) {
    usage();
}
85 86 87
if (defined($options{"d"})) {
    $debug = $options{"d"};
}
88 89

# Go to ground.
90
if (! $debug) {
91 92 93
    if (TBBackGround($batchlog)) {
	exit(0);
    }
94
}
95 96 97 98 99

#
# Loop, looking for batch experiments that want to run.
# 
while (1) {
100
    my($count, $i, $query_result, $pending_result, $running_result);
101
    my(%row, %pending_row);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
102

103
    #
104
    # Need to lock the table here because of cancelation in endexp.
105 106 107 108 109 110
    # 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!
111
    # Thats would ne annoying to users.
112
    #
113
    # So, now you're wondering what my selection criteria is? Well, its
114 115
    # damn simplistic. I set the "started" datetime field each attempt,
    # and I pick the batch_experiment with the oldest time, thereby cycling
116 117 118 119
    # 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).
120 121
    #
    $query_result =
122 123
	DBQuery("lock tables experiments write, experiments as e1 write, ".
		"experiments as e2 write");
124 125
    if (! $query_result) {
	print "DB Error locking tables. Waiting a bit ...\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
126
	goto pause;
127 128 129
    }
    
    $pending_result =
130 131 132 133 134 135 136 137 138 139 140 141
	DBQueryWarn("SELECT e1.* FROM experiments as e1 ".
		    "left join experiments as e2 on ".
		    " e2.expt_head_uid=e1.expt_head_uid and ".
		    " e2.batchmode=1 and e2.batchstate='$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");
142

143
    $running_result =
144 145 146
	DBQuery("select * from experiments ".
		"where batchmode=1 and batchstate='$BSTATE_RUNNING' ".
		"ORDER BY expt_start");
147 148 149

    if (!$pending_result || !$running_result) {
	print "DB Error getting batch info. Waiting a bit ...\n";
150
	DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
151
	goto pause;
152 153 154
    }

    if (!$pending_result->numrows && !$running_result->numrows) {
155
	DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
156
	goto pause;
157 158 159
    }

    #
160
    # If we have a pending experiment to run, set its state to configuring
161
    # right away, while we have the tables locked. This prevents endexp
162
    # from seeing it as something it can cancel.
163
    #
164 165 166 167 168 169 170 171
    if ($pending_result->numrows) {
	%pending_row = $pending_result->fetchhash();

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

	$query_result = 
172 173 174
	    DBQuery("update experiments set expt_start=now(), ".
		    "batchstate='$BSTATE_ACTIVATING' ".
		    "where eid='$eid' and pid='$pid'");
175 176 177

	if (! $query_result) {
	    print "DB error setting batch $pid/$eid to configuring.\n";
178
	    DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
179
	    goto pause;
180 181
	}
    }
182
    DBQueryWarn("unlock tables");
183

184 185 186 187 188 189 190 191 192 193 194
    #
    # 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.
    #
195
    while (%row = $running_result->fetchhash()) {
196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212
	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
213 214
  pause:
    sleep(30);
215 216 217
}

#
218
# Do something as the user. Either, start, end, or cancel an experiment.
219
#
220
sub dosomething($$)
221
{
222 223
    my($dowhat)   = shift;
    my(%exphash)  = @_;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
224
    my($unix_uid, $unix_gid, $row, $query_result);
225 226

    # Global vars
227 228 229 230 231 232 233 234
    $eid     = $exphash{'eid'};
    $pid     = $exphash{'pid'};
    $gid     = $exphash{'gid'};
    $dirname = $exphash{'path'};
    $nsfile  = "$eid.ns";
    
    # Locals
    my $creator = $exphash{'expt_head_uid'};
235

236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251
    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.
    #
    $logname = `mktemp /tmp/$dowhat-batch-$pid-$eid.XXXXXX`;

    # Note different taint check (allow /).
    if ($logname =~ /^([-\@\w.\/]+)$/) {
	$logname = $1;
    } else {
	die "Bad data in $logname";
    }

252 253 254
    #
    # Start up a child to run the guts. The parent waits. If the
    # experiment configures okay, the parent can return to try something
255
    # else.
256 257 258
    #
    $childpid = fork();
    if ($childpid) {
259 260
	print "Child PID $childpid started to $dowhat $pid/$eid\n";

261
	waitpid($childpid, 0);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
262
	my $status = $? >> 8;
263

Leigh B. Stoller's avatar
Leigh B. Stoller committed
264
	print "Child PID $childpid exited with exit status $status\n";
265

Leigh B. Stoller's avatar
Leigh B. Stoller committed
266
	sleep(5);
267

268
	unlink($logname);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
269
	return $status;
270
    }
271 272
    openlog($logname);

273 274 275
    #
    # Get some user information. 
    #
276
    if (!UserDBInfo($creator, \$user_name, \$user_email)) {
277
	fatal("DB Error getting user information for uid $creator");
278
    }
279 280 281
    
    chdir("$dirname/tbdata") or
	fatal("Could not cd into $dirname/tbdata!");
282 283 284 285 286

    #
    # Figure out the unix uid/gid that the experiment configuration is
    # going to run as. 
    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
287
    (undef,undef,$unix_uid) = getpwnam($creator) or
288
	fatal("No such user $creator");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
289 290
    (undef,undef,$unix_gid) = getgrnam($gid) or
	fatal("No such group $gid");
291

292 293 294
    #
    # Change the ownership of the log file before we flip.
    #
295 296
    chown($unix_uid, $unix_gid, $logname) or
	fatal("Could not chown $logname to $unix_uid/$unix_gid!");
297 298

    # Flip to the user. We never flip back.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
299 300
    $EGID = $GID = $unix_gid;
    $EUID = $UID = $unix_uid;
301
    $ENV{'USER'} = $creator;
302
    
303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320
    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($)
{
    my(%exphash)  = @_;
321
    my($exit_status, $running, $query_result);
322 323 324

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

325
    #
326
    # Try to start the experiment. 
327
    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
328
    system("$startexp -b $logname -g $gid $pid $eid $nsfile");
329 330
    $exit_status = $? >> 8;
    $running     = 1;
331 332 333 334 335
    if ($exit_status) {
	$running = 0;
    }
    
    #
336 337
    # Look for cancelation. If we get a DB error on this, just continue cause
    # we can pick up the cancelation later.
338 339
    #
    $query_result =
340
	DBQueryWarn("select canceled from experiments ".
341
		    "where eid='$eid' and pid='$pid'");
342

343 344
    if ($query_result) {
	@row = $query_result->fetchrow_array();
345

346 347 348 349 350 351 352 353
	if ($row[0]) {
	    cancelexp($running);
	    #
	    # Never returns, but just to be safe ...
	    #
	    exit(0);
	}
    }
354 355 356 357

    #
    # If the configuration failed for lack of nodes, then don't send
    # email unless the number of attempts starts to get big.
358
    #
359 360 361
    # 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. 
362 363
    #
    if (! $running) {
364 365 366 367
	#
	# XXX - What if this update fails?
	# 
	$query_result = 
368
	    DBQueryWarn("update experiments set attempts=attempts+1 ".
369
			"where eid='$eid' and pid='$pid'");
370

371 372 373 374 375 376 377 378 379 380
	#
	# 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 || $exit_status == -1) {
	    email_status("Experiment startup has failed with a fatal error!\n".
			 "Batch has been removed from the system.");
	    ExptCleanup();
381
	    exit($exit_status);
382
	}
383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408
	
	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);
	    
409
	exit($exit_status);
410 411 412 413 414
    }

    #
    # Well, it configured! Lets set it state to running.
    #
415 416
    TBSetBatchState($pid, $eid, $BSTATE_RUNNING);
    
417
    email_status("Batch Mode experiment $pid/$eid is now running!\n".
418
		 "Please consult the Web interface to see how it is doing.");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
419

420
    #
421
    # Done with this phase. Must exit.
422
    #
423 424
    exit(0);
}
425

426 427 428 429 430 431
#
# End an experiment. Never returns.
#
sub endexp($)
{
    my(%exphash)  = @_;
432

433 434 435
    #
    # Save tiplogs
    #
436 437
    system("$savelogs $pid $eid");

438 439 440 441 442
    #
    # Have to set the state to terminating or else endexp will not accept it.
    # 
    TBSetBatchState($pid, $eid, $BSTATE_TERMINATING);
    
443
    system("$endexp -b $pid $eid");
444 445 446 447 448 449 450 451
    my $exit_status = $? >> 8;

    if ($exit_status) {
	#
	# TB admin is going to have to clean up. 
	# 
	fatal("Terminating Batch Mode experiment $pid/$eid");
    }
452 453

    ExptCleanup();
454
    email_status("Batch Mode experiment $pid/$eid has finished!");
455
   
456
    #
457
    # Child must exit!
458
    #
459 460 461 462 463 464 465 466 467 468 469
    exit(0);
}

#
# Cancel an experiment. Never returns.
#
sub cancelexp($$)
{
    my($running) = shift;
    my(%exphash) = @_;
    
470 471
    TBSetBatchState($pid, $eid, $BSTATE_TERMINATING);
    
472 473 474 475
    if ($running) {
	system("$endexp -b $pid $eid");
    }

476 477
    ExptCleanup();
    donotify("Your Batch Mode experiment has been canceled!", "Canceled", 0);
478 479 480 481 482 483 484
   
    #
    # Child must exit!
    #
    exit(0);
}

485 486 487 488 489 490 491
#
# Check experiment status. Looks to see if all of the nodes in an
# experiment have reported in.
#
sub isexpdone($)
{
    my(%exphash)  = @_;
492
    my($query_result, @row);
493 494 495 496 497 498 499 500 501 502 503
    
    # 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.
    #
    $query_result =
504 505 506
	DBQueryWarn("SELECT startstatus,bootstatus FROM nodes ".
		    "LEFT JOIN reserved ON nodes.node_id=reserved.node_id ".
		    "WHERE reserved.eid='$eid' and reserved.pid='$pid'");
507 508 509 510 511

    if (! $query_result) {
	return 0;
    }

512 513 514 515 516 517 518 519 520 521 522 523 524
    #
    # 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;
525 526
	}
    }
527
    return 1;
528 529
}

530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552
#
# Remove all trace.
# 
sub ExptCleanup()
{
    if (system("rm -rf $dirname")) {
	print "*** WARNING: Not able to remove experiment directory.\n";
	print "             Someone will need to do this by hand.\n";
    }

    #
    # Remove all trace from the DB.
    # 
    DBQueryWarn("DELETE from nsfiles ".
		"WHERE eid='$eid' and pid='$pid'");

    DBQueryWarn("DELETE from exppid_access ".
		"WHERE exp_eid='$eid' and exp_pid='$pid'");

    DBQueryWarn("DELETE from experiments ".
		"WHERE eid='$eid' and pid='$pid'");
}

553 554
#
# Start up a child, and set its descriptors talking to a log file.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
555
# The log file already exists, created with mktemp above.
556 557 558 559 560 561 562
# 
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
563 564
    # STDOUT away from the pipe. Otherwise the caller will continue to wait
    # even though the parent has exited. 
565 566
    #
    open(STDIN, "< /dev/null") or
567
	fatal("opening /dev/null for STDIN: $!");
568 569 570 571 572 573 574 575 576

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

    return 0;
}

577 578 579 580 581
#
# 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.
# 
582
sub fatal($)
583 584 585
{
    my($mesg) = $_[0];

586
    donotify($mesg, "Failure", 1);
587 588 589 590

    exit(-1);
}

591 592 593
#
# Something the user cares about. 
# 
594
sub email_status($)
595
{
596
    my($mesg) = $_[0];
597

598
    donotify($mesg, "Status", 0);
599 600
}

601
sub donotify($$$)
602
{
603
    my($mesg, $subtext, $iserr) = @_;
604
    my($subject, $from, $to, $hdrs);
605
    my $MAIL;
606

607 608 609
    $mesg = "$mesg\n";

    print STDOUT "$mesg";
610

611
    $subject = "Batch Mode Experiment $subtext $pid/$eid";
612
    $from    = $TBOPS;
613 614 615 616 617 618
    $hdrs    = "Reply-To: $TBOPS";
    
    #
    # An error goes just to Testbed Operations. Normal status messages go
    # to the user and to the Testbed Logs address.
    # 
619
    if ($iserr) {
620
	$to = "$TBOPS";
621 622
    }
    else {
623 624 625
	$to   = "$user_name <$user_email>";
	$hdrs = "Bcc: $TBLOGS\n".
	        "$hdrs";
626 627
    }

628
    SENDMAIL($to, $subject, $mesg, $from, $hdrs,
629
	     ($logname, "assign.log", $nsfile));
630
}