batch_daemon.in 15.4 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);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
123

124
    #
125
    # Need to lock the table here because of cancelation in endexp.
126
127
128
129
130
131
    # 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!
132
    # Thats would ne annoying to users.
133
    #
134
    # So, now you're wondering what my selection criteria is? Well, its
135
136
    # damn simplistic. I set the "started" datetime field each attempt,
    # and I pick the batch_experiment with the oldest time, thereby cycling
137
138
139
140
    # 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).
141
142
    #
    $query_result =
143
144
	DBQuery("lock tables experiments write, experiments as e1 write, ".
		"experiments as e2 write");
145
146
    if (! $query_result) {
	print "DB Error locking tables. Waiting a bit ...\n";
Leigh B. Stoller's avatar
Leigh B. Stoller committed
147
	goto pause;
148
149
150
    }
    
    $pending_result =
151
152
153
154
155
156
157
158
159
160
161
162
	DBQueryWarn("SELECT e1.* FROM experiments as e1 ".
		    "left join experiments as e2 on ".
		    " e2.expt_head_uid=e1.expt_head_uid and ".
		    " e2.batchmode=1 and e2.batchstate='$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");
163

164
    $running_result =
165
166
167
	DBQuery("select * from experiments ".
		"where batchmode=1 and batchstate='$BSTATE_RUNNING' ".
		"ORDER BY expt_start");
168
169
170

    if (!$pending_result || !$running_result) {
	print "DB Error getting batch info. Waiting a bit ...\n";
171
	DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
172
	goto pause;
173
174
175
    }

    if (!$pending_result->numrows && !$running_result->numrows) {
176
	DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
177
	goto pause;
178
179
180
    }

    #
181
    # If we have a pending experiment to run, set its state to configuring
182
    # right away, while we have the tables locked. This prevents endexp
183
    # from seeing it as something it can cancel.
184
    #
185
186
187
188
189
190
191
192
    if ($pending_result->numrows) {
	%pending_row = $pending_result->fetchhash();

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

	$query_result = 
193
194
195
	    DBQuery("update experiments set expt_start=now(), ".
		    "batchstate='$BSTATE_ACTIVATING' ".
		    "where eid='$eid' and pid='$pid'");
196
197
198

	if (! $query_result) {
	    print "DB error setting batch $pid/$eid to configuring.\n";
199
	    DBQuery("unlock tables");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
200
	    goto pause;
201
202
	}
    }
203
    DBQueryWarn("unlock tables");
204

205
206
207
208
209
210
211
212
213
214
215
    #
    # 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.
    #
216
    while (%row = $running_result->fetchhash()) {
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
	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
234
235
  pause:
    sleep(30);
236
237
238
}

#
239
# Do something as the user. Either, start, end, or cancel an experiment.
240
#
241
sub dosomething($$)
242
{
243
244
    my($dowhat)   = shift;
    my(%exphash)  = @_;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
245
    my($unix_uid, $unix_gid, $row, $query_result);
246
247

    # Global vars
248
249
250
    $eid     = $exphash{'eid'};
    $pid     = $exphash{'pid'};
    $gid     = $exphash{'gid'};
251
252
    $userdir = $exphash{'path'};
    $workdir = TBExptWorkDir($pid, $eid);
253
254
255
256
    $nsfile  = "$eid.ns";
    
    # Locals
    my $creator = $exphash{'expt_head_uid'};
257

258
259
260
261
262
263
264
    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.
    #
265
    $logname = TBExptCreateLogFile($pid, $eid, "${dowhat}-batch");
266

267
268
269
    #
    # Start up a child to run the guts. The parent waits. If the
    # experiment configures okay, the parent can return to try something
270
    # else.
271
272
273
    #
    $childpid = fork();
    if ($childpid) {
274
275
	print "Child PID $childpid started to $dowhat $pid/$eid\n";

276
	waitpid($childpid, 0);
Leigh B. Stoller's avatar
Leigh B. Stoller committed
277
	my $status = $? >> 8;
278

Leigh B. Stoller's avatar
Leigh B. Stoller committed
279
	print "Child PID $childpid exited with exit status $status\n";
280

281
        #
282
        # Close the log file.
283
284
	# The exp dir might be gone if the batch was killed/canceled.
        #
285
	if (-e $userdir) {
286
287
	    TBExptCloseLogFile($pid, $eid);
	}
Leigh B. Stoller's avatar
Leigh B. Stoller committed
288
	return $status;
289
    }
290
    openlog($logname);
291
292
    TBExptSetLogFile($pid, $eid, $logname);
    TBExptOpenLogFile($pid, $eid);
293

294
295
296
    #
    # Get some user information. 
    #
297
    if (!UserDBInfo($creator, \$user_name, \$user_email)) {
298
	fatal("DB Error getting user information for uid $creator");
299
    }
300
    
301
302
    chdir("$workdir") or
	fatal("Could not cd into $workdir!");
303
304
305
306
307

    #
    # Figure out the unix uid/gid that the experiment configuration is
    # going to run as. 
    #
Leigh B. Stoller's avatar
Leigh B. Stoller committed
308
    (undef,undef,$unix_uid) = getpwnam($creator) or
309
	fatal("No such user $creator");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
310
311
    (undef,undef,$unix_gid) = getgrnam($gid) or
	fatal("No such group $gid");
312

313
314
315
    #
    # Change the ownership of the log file before we flip.
    #
316
317
    chown($unix_uid, $unix_gid, $logname) or
	fatal("Could not chown $logname to $unix_uid/$unix_gid!");
318
319

    # Flip to the user. We never flip back.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
320
321
    $EGID = $GID = $unix_gid;
    $EUID = $UID = $unix_uid;
322
    $ENV{'USER'} = $creator;
323
    
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
    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($)
{
    my(%exphash)  = @_;
342
    my($exit_status, $running, $query_result);
343
344
345

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

346
    #
347
    # Try to start the experiment. 
348
    #
349
    system("$startexp -b -g $gid $pid $eid $nsfile");
350
351
    $exit_status = $? >> 8;
    $running     = 1;
352
353
354
355
356
    if ($exit_status) {
	$running = 0;
    }
    
    #
357
358
    # Look for cancelation. If we get a DB error on this, just continue cause
    # we can pick up the cancelation later.
359
360
    #
    $query_result =
361
	DBQueryWarn("select canceled from experiments ".
362
		    "where eid='$eid' and pid='$pid'");
363

364
365
    if ($query_result) {
	@row = $query_result->fetchrow_array();
366

367
368
369
370
371
372
373
374
	if ($row[0]) {
	    cancelexp($running);
	    #
	    # Never returns, but just to be safe ...
	    #
	    exit(0);
	}
    }
375
376
377
378

    #
    # If the configuration failed for lack of nodes, then don't send
    # email unless the number of attempts starts to get big.
379
    #
380
381
382
    # 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. 
383
384
    #
    if (! $running) {
385
386
387
388
	#
	# XXX - What if this update fails?
	# 
	$query_result = 
389
	    DBQueryWarn("update experiments set attempts=attempts+1 ".
390
			"where eid='$eid' and pid='$pid'");
391

392
393
394
395
396
397
398
399
400
	#
	# 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 || $exit_status == -1) {
	    email_status("Experiment startup has failed with a fatal error!\n".
			 "Batch has been removed from the system.");
401
	    TBExptDestroy($pid, $eid);
402
	    exit($exit_status);
403
	}
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
	
	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);
	    
430
	exit($exit_status);
431
432
433
434
435
    }

    #
    # Well, it configured! Lets set it state to running.
    #
436
437
    TBSetBatchState($pid, $eid, $BSTATE_RUNNING);
    
438
    email_status("Batch Mode experiment $pid/$eid is now running!\n".
439
		 "Please consult the Web interface to see how it is doing.");
Leigh B. Stoller's avatar
Leigh B. Stoller committed
440

441
    #
442
    # Done with this phase. Must exit.
443
    #
444
445
    exit(0);
}
446

447
448
449
450
451
452
#
# End an experiment. Never returns.
#
sub endexp($)
{
    my(%exphash)  = @_;
453

454
455
456
    #
    # Save tiplogs
    #
457
458
    system("$savelogs $pid $eid");

459
460
461
462
463
    #
    # Have to set the state to terminating or else endexp will not accept it.
    # 
    TBSetBatchState($pid, $eid, $BSTATE_TERMINATING);
    
464
    system("$endexp -b $pid $eid");
465
    if ($?) {
466
467
468
469
470
	#
	# TB admin is going to have to clean up. 
	# 
	fatal("Terminating Batch Mode experiment $pid/$eid");
    }
471
    email_status("Batch Mode experiment $pid/$eid has finished!");
472
   
473
    #
474
    # Child must exit!
475
    #
476
477
478
479
480
481
482
483
484
485
486
    exit(0);
}

#
# Cancel an experiment. Never returns.
#
sub cancelexp($$)
{
    my($running) = shift;
    my(%exphash) = @_;
    
487
488
    TBSetBatchState($pid, $eid, $BSTATE_TERMINATING);
    
489
490
    if ($running) {
	system("$endexp -b $pid $eid");
491
492
493
494
495
496
497
498
499
	if ($?) {
	    #
	    # TB admin is going to have to clean up. 
	    # 
	    fatal("Terminating Batch Mode experiment $pid/$eid");
	}
    }
    else {
	TBExptDestroy($pid, $eid);
500
    }
501
    donotify("Your Batch Mode experiment has been canceled!", "Canceled", 0);
502
503
504
505
506
507
508
   
    #
    # Child must exit!
    #
    exit(0);
}

509
510
511
512
513
514
515
#
# Check experiment status. Looks to see if all of the nodes in an
# experiment have reported in.
#
sub isexpdone($)
{
    my(%exphash)  = @_;
516
    my($query_result, @row);
517
518
519
520
521
522
523
524
525
526
527
    
    # 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.
    #
    $query_result =
528
529
530
	DBQueryWarn("SELECT startstatus,bootstatus FROM nodes ".
		    "LEFT JOIN reserved ON nodes.node_id=reserved.node_id ".
		    "WHERE reserved.eid='$eid' and reserved.pid='$pid'");
531
532
533
534
535

    if (! $query_result) {
	return 0;
    }

536
537
538
539
540
541
542
543
544
545
546
547
548
    #
    # 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;
549
550
	}
    }
551
    return 1;
552
553
554
555
}

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

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

575
576
577
    STDOUT->autoflush(1);
    STDERR->autoflush(1);

578
579
580
    return 0;
}

581
582
583
584
585
#
# 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.
# 
586
sub fatal($)
587
588
589
{
    my($mesg) = $_[0];

590
    donotify($mesg, "Failure", 1);
591
592
593
594

    exit(-1);
}

595
596
597
#
# Something the user cares about. 
# 
598
sub email_status($)
599
{
600
    my($mesg) = $_[0];
601

602
    donotify($mesg, "Status", 0);
603
604
}

605
sub donotify($$$)
606
{
607
    my($mesg, $subtext, $iserr) = @_;
608
    my($subject, $from, $to, $hdrs);
609
    my $MAIL;
610

611
612
613
    $mesg = "$mesg\n";

    print STDOUT "$mesg";
614

615
    $subject = "Batch Mode Experiment $subtext $pid/$eid";
616
    $from    = $TBOPS;
617
618
619
620
621
622
    $hdrs    = "Reply-To: $TBOPS";
    
    #
    # An error goes just to Testbed Operations. Normal status messages go
    # to the user and to the Testbed Logs address.
    # 
623
    if ($iserr) {
624
	$to = "$TBOPS";
625
626
    }
    else {
627
628
629
	$to   = "$user_name <$user_email>";
	$hdrs = "Bcc: $TBLOGS\n".
	        "$hdrs";
630
631
    }

632
    SENDMAIL($to, $subject, $mesg, $from, $hdrs,
633
	     ($logname, "assign.log", $nsfile));
634
}