All new accounts created on Gitlab now require administrator approval. If you invite any collaborators, please let Flux staff know so they can approve the accounts.

batch_daemon.in 15.6 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 56 57 58
my $tbbindir = "$TB/bin/";
my $batchdir = "$TB/batch";
my $startexp = "$TB/bin/startexp";
my $endexp   = "$TB/bin/endexp";
59
my $savelogs = "$TB/bin/savelogs";
60
my $avail    = "$TB/sbin/avail";
61 62
my $batchlog = "$TB/log/batchlog";
my $projroot = "/proj";
63
my $debug    = 0;
64

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

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()) {
222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238
	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
239 240
  pause:
    sleep(30);
241 242 243
}

#
244
# Do something as the user. Either, start, end, or cancel an experiment.
245
#
246
sub dosomething($$)
247
{
248 249
    my($dowhat)   = shift;
    my(%exphash)  = @_;
250
    my($unix_uid, $unix_gid, $unix_gname, $row, $query_result);
251 252

    # Global vars
253 254 255
    $eid     = $exphash{'eid'};
    $pid     = $exphash{'pid'};
    $gid     = $exphash{'gid'};
256 257
    $userdir = $exphash{'path'};
    $workdir = TBExptWorkDir($pid, $eid);
258 259 260 261
    $nsfile  = "$eid.ns";
    
    # Locals
    my $creator = $exphash{'expt_head_uid'};
262

263 264 265 266 267 268 269
    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.
    #
270
    $logname = TBExptCreateLogFile($pid, $eid, "${dowhat}-batch");
271

272 273 274
    #
    # Start up a child to run the guts. The parent waits. If the
    # experiment configures okay, the parent can return to try something
275
    # else.
276 277 278
    #
    $childpid = fork();
    if ($childpid) {
279 280
	print "Child PID $childpid started to $dowhat $pid/$eid\n";

281
	waitpid($childpid, 0);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
282
	my $status = $? >> 8;
283

Leigh B. Stoller's avatar
Leigh B. Stoller committed
284
	print "Child PID $childpid exited with exit status $status\n";
285

286
        #
287
        # Close the log file.
288 289
	# The exp dir might be gone if the batch was killed/canceled.
        #
290
	if (-e $userdir) {
291 292
	    TBExptCloseLogFile($pid, $eid);
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
293
	return $status;
294
    }
295
    openlog($logname);
296 297
    TBExptSetLogFile($pid, $eid, $logname);
    TBExptOpenLogFile($pid, $eid);
298

299 300 301
    #
    # Get some user information. 
    #
302
    if (!UserDBInfo($creator, \$user_name, \$user_email)) {
303
	fatal("DB Error getting user information for uid $creator");
304
    }
305
    
306 307
    chdir("$workdir") or
	fatal("Could not cd into $workdir!");
308 309 310 311 312

    #
    # Figure out the unix uid/gid that the experiment configuration is
    # going to run as. 
    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
313
    (undef,undef,$unix_uid) = getpwnam($creator) or
314
	fatal("No such user $creator");
315 316
    TBGroupUnixInfo($pid, $gid, \$unix_gid, \$unix_gname) or
	fatal("No such group $pid/$gid");
317

318 319 320
    #
    # Change the ownership of the log file before we flip.
    #
321 322
    chown($unix_uid, $unix_gid, $logname) or
	fatal("Could not chown $logname to $unix_uid/$unix_gid!");
323 324

    # Flip to the user. We never flip back.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
325 326
    $EGID = $GID = $unix_gid;
    $EUID = $UID = $unix_uid;
327
    $ENV{'USER'} = $creator;
328
    
329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346
    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($)
{
    my(%exphash)  = @_;
347
    my($exit_status, $running, $query_result);
348 349 350

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

351
    #
352
    # Try to start the experiment. 
353
    #
354
    system("$startexp -b -g $gid $pid $eid $nsfile");
355 356
    $exit_status = $? >> 8;
    $running     = 1;
357 358 359 360 361
    if ($exit_status) {
	$running = 0;
    }
    
    #
362 363
    # Look for cancelation. If we get a DB error on this, just continue cause
    # we can pick up the cancelation later.
364 365
    #
    $query_result =
366
	DBQueryWarn("select canceled from experiments ".
367
		    "where eid='$eid' and pid='$pid'");
368

369 370
    if ($query_result) {
	@row = $query_result->fetchrow_array();
371

372 373 374 375 376 377 378 379
	if ($row[0]) {
	    cancelexp($running);
	    #
	    # Never returns, but just to be safe ...
	    #
	    exit(0);
	}
    }
380 381 382 383

    #
    # If the configuration failed for lack of nodes, then don't send
    # email unless the number of attempts starts to get big.
384
    #
385 386 387
    # 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. 
388 389
    #
    if (! $running) {
390 391 392 393
	#
	# XXX - What if this update fails?
	# 
	$query_result = 
394
	    DBQueryWarn("update experiments set attempts=attempts+1 ".
395
			"where eid='$eid' and pid='$pid'");
396

397 398 399 400 401 402 403 404 405
	#
	# 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 || $exit_status == -1) {
	    email_status("Experiment startup has failed with a fatal error!\n".
			 "Batch has been removed from the system.");
406
	    TBExptDestroy($pid, $eid);
407
	    exit($exit_status);
408
	}
409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434
	
	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);
	    
435
	exit($exit_status);
436 437 438 439 440
    }

    #
    # Well, it configured! Lets set it state to running.
    #
441 442
    TBSetBatchState($pid, $eid, $BSTATE_RUNNING);
    
443
    email_status("Batch Mode experiment $pid/$eid is now running!\n".
444
		 "Please consult the Web interface to see how it is doing.");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
445

446
    #
447
    # Done with this phase. Must exit.
448
    #
449 450
    exit(0);
}
451

452 453 454 455 456 457
#
# End an experiment. Never returns.
#
sub endexp($)
{
    my(%exphash)  = @_;
458

459 460 461
    #
    # Save tiplogs
    #
462 463
    system("$savelogs $pid $eid");

464 465 466 467 468
    #
    # Have to set the state to terminating or else endexp will not accept it.
    # 
    TBSetBatchState($pid, $eid, $BSTATE_TERMINATING);
    
469
    system("$endexp -b $pid $eid");
470
    if ($?) {
471 472 473 474 475
	#
	# TB admin is going to have to clean up. 
	# 
	fatal("Terminating Batch Mode experiment $pid/$eid");
    }
476
    email_status("Batch Mode experiment $pid/$eid has finished!");
477
   
478
    #
479
    # Child must exit!
480
    #
481 482 483 484 485 486 487 488 489 490 491
    exit(0);
}

#
# Cancel an experiment. Never returns.
#
sub cancelexp($$)
{
    my($running) = shift;
    my(%exphash) = @_;
    
492 493
    TBSetBatchState($pid, $eid, $BSTATE_TERMINATING);
    
494 495
    if ($running) {
	system("$endexp -b $pid $eid");
496 497 498 499 500 501 502 503 504
	if ($?) {
	    #
	    # TB admin is going to have to clean up. 
	    # 
	    fatal("Terminating Batch Mode experiment $pid/$eid");
	}
    }
    else {
	TBExptDestroy($pid, $eid);
505
    }
506
    donotify("Your Batch Mode experiment has been canceled!", "Canceled", 0);
507 508 509 510 511 512 513
   
    #
    # Child must exit!
    #
    exit(0);
}

514 515 516 517 518 519 520
#
# Check experiment status. Looks to see if all of the nodes in an
# experiment have reported in.
#
sub isexpdone($)
{
    my(%exphash)  = @_;
521
    my($query_result, @row);
522 523 524 525 526 527 528 529 530 531 532
    
    # 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.
    #
    $query_result =
533 534 535
	DBQueryWarn("SELECT startstatus,bootstatus FROM nodes ".
		    "LEFT JOIN reserved ON nodes.node_id=reserved.node_id ".
		    "WHERE reserved.eid='$eid' and reserved.pid='$pid'");
536 537 538 539 540

    if (! $query_result) {
	return 0;
    }

541 542 543 544 545 546 547 548 549 550 551 552 553
    #
    # 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;
554 555
	}
    }
556
    return 1;
557 558 559 560
}

#
# Start up a child, and set its descriptors talking to a log file.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
561
# The log file already exists, created with mktemp above.
562 563 564 565 566 567 568
# 
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
569 570
    # STDOUT away from the pipe. Otherwise the caller will continue to wait
    # even though the parent has exited. 
571 572
    #
    open(STDIN, "< /dev/null") or
573
	fatal("opening /dev/null for STDIN: $!");
574 575 576 577 578 579

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

580 581 582
    STDOUT->autoflush(1);
    STDERR->autoflush(1);

583 584 585
    return 0;
}

586 587 588 589 590
#
# 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.
# 
591
sub fatal($)
592 593 594
{
    my($mesg) = $_[0];

595
    donotify($mesg, "Failure", 1);
596 597 598 599

    exit(-1);
}

600 601 602
#
# Something the user cares about. 
# 
603
sub email_status($)
604
{
605
    my($mesg) = $_[0];
606

607
    donotify($mesg, "Status", 0);
608 609
}

610
sub donotify($$$)
611
{
612
    my($mesg, $subtext, $iserr) = @_;
613
    my($subject, $from, $to, $hdrs);
614
    my $MAIL;
615

616 617 618
    $mesg = "$mesg\n";

    print STDOUT "$mesg";
619

620
    $subject = "Batch Mode Experiment $subtext $pid/$eid";
621
    $from    = $TBOPS;
622 623 624 625 626 627
    $hdrs    = "Reply-To: $TBOPS";
    
    #
    # An error goes just to Testbed Operations. Normal status messages go
    # to the user and to the Testbed Logs address.
    # 
628
    if ($iserr) {
629
	$to = "$TBOPS";
630 631
    }
    else {
632 633 634
	$to   = "$user_name <$user_email>";
	$hdrs = "Bcc: $TBLOGS\n".
	        "$hdrs";
635 636
    }

637
    SENDMAIL($to, $subject, $mesg, $from, $hdrs,
638
	     ($logname, "assign.log", $nsfile));
639
}