stated.in 58.5 KB
Newer Older
Robert Ricci's avatar
Robert Ricci committed
1
#!/usr/bin/perl -w
Leigh B. Stoller's avatar
Leigh B. Stoller committed
2 3
#
# EMULAB-COPYRIGHT
4
# Copyright (c) 2000-2011 University of Utah and the Flux Group.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
5 6 7
# All rights reserved.
#

Robert Ricci's avatar
Robert Ricci committed
8 9 10
#
# stated - A daemon to monitor the states of nodes in the testbed. Recives
# state change notification through the event system, and writes the new
Mac Newbold's avatar
Mac Newbold committed
11
# state into the database. Also watches for invalid transitions, timeouts,
12
# and performs other state-related control functions.
Robert Ricci's avatar
Robert Ricci committed
13 14 15 16
#
# Send it a HUP signal to get it to reload the timeout and transition
# information. Periodically reloads this information regardless, though.
#
17
# Will restart when sent SIGUSR1, by exec'ing its executable again.
Robert Ricci's avatar
Robert Ricci committed
18
#
19

Robert Ricci's avatar
Robert Ricci committed
20 21
# Configure variables
use lib '@prefix@/lib';
22
my $TB = "@prefix@";
23
my $REALTB = "/usr/testbed"; # So we know if we're the "real" stated or not
Robert Ricci's avatar
Robert Ricci committed
24
my $BOSSNODE = "@BOSSNODE@";
25
my $TBOPS = "@TBSTATEDEMAIL@";
26
my $REALTBOPS = "@TBOPSEMAIL@";
27
my $TBDBNAME = "@TBDBNAME@";
28
my $REALTBDBNAME = "tbdb"; # So we know if we're using the "real" db
29
my $pxeselect = "$TB/bin/pxe_select";
30
my $osselect = "$TB/bin/os_select";
31
my $nodereboot = "$TB/bin/node_reboot";
32
my $rebootlog  = "$TB/log/nodereboot.log";
33
my $power = "$TB/bin/power";
34
my $apod = "$TB/sbin/apod";
35
my $TBLOG = "@TBLOGFACIL@";
36
my $LOGFILE = "$TB/log/stated.log";
Robert Ricci's avatar
Robert Ricci committed
37 38 39 40 41

$| = 1;

use event;
use libdb;
42
use Node;
Robert Ricci's avatar
Robert Ricci committed
43
use libtestbed;
44
use TimeoutQueue;
Robert Ricci's avatar
Robert Ricci committed
45
use Getopt::Std;
46
#use strict;
Robert Ricci's avatar
Robert Ricci committed
47
use English;
Mac Newbold's avatar
Mac Newbold committed
48 49
use POSIX;			# for strftime, and sigprocmask and friends
use Fcntl;			# file constants for pidfile
50 51
use POSIX ":sys_wait_h";
use IO::Poll qw(POLLIN);
52
use POSIX qw(:errno_h);
Mac Newbold's avatar
Mac Newbold committed
53

54 55 56 57 58
# Set up some notification throttling
my $mailgap = 15;		# in seconds
my $lastmail = time() - $mailgap + 2; # Send a digest of startup msgs after 2s.
my %msgs = ();

Mac Newbold's avatar
Mac Newbold committed
59
# Number of iterations (roughly, seconds) after which we'll reload
Robert Ricci's avatar
Robert Ricci committed
60 61
# information from the database. This is so we don't end up with information
# that's _too_ out of sync.
62
my $reload_time = 600;
63
my $last_reload = time;
Robert Ricci's avatar
Robert Ricci committed
64

Mike Hibler's avatar
Mike Hibler committed
65 66 67
# For startup and reload: maximum time in the past for which we will schedule
# a timeout. Anything older is assumed to be ancient history and is ignored.
# Note it is a negative value.
68
my $maxpasttimeout = -(7 * 24 * 60 * 60);	# 1 week
Mike Hibler's avatar
Mike Hibler committed
69

70 71 72
# Handling of SECVIOLATIONS, eventually controlled by a sitevar
my $soft_secviolation = 1;

73 74 75
# Command line opts.
my $dbtag = "";
my $debug = 0;
76 77
my $server = "localhost";
my $port   = @BOSSEVENTPORT@;
78 79
my $lockfile;
my $pidfile;
80
my $eventchild;
81

Robert Ricci's avatar
Robert Ricci committed
82 83 84
# Process command-line arguments

sub usage {
Mac Newbold's avatar
Mac Newbold committed
85
    print << "END";
86
Usage: $0 [-h] [-d] [-s server] [-p port] [-t dbtag]
87
-h              This message
88 89
-d              Turn on debugging output, and do not go into the background
-t tag          Use only those nodes with matching tag in nodes table
Robert Ricci's avatar
Robert Ricci committed
90 91
-s server       Use specified server, instead of this site's bossnode
-p port	        Use specified port
92
Send SIGHUP to reload database state, or SIGUSR1 to restart completely.
Robert Ricci's avatar
Robert Ricci committed
93
END
94
    exit(1);
Robert Ricci's avatar
Robert Ricci committed
95 96
}

Mac Newbold's avatar
Mac Newbold committed
97
# Only root should run this - it won't work when run as a user...
98
# (Or, let an admin run it if it isn't the real one in /usr/testbed/ )
99
if ($UID && ( $TB eq $REALTB || ! TBAdmin($UID) ) ) {
Mac Newbold's avatar
Mac Newbold committed
100 101 102
    die("Only root can run this script!\n");
}

103
my @args = @ARGV;    # save a copy for restart before we mess with them.
Robert Ricci's avatar
Robert Ricci committed
104
my %opt = ();
105
if (!getopts("ds:p:ht:l",\%opt)) { usage(); }
Robert Ricci's avatar
Robert Ricci committed
106

Mac Newbold's avatar
Mac Newbold committed
107 108 109 110 111 112
if ($opt{h}) {
    exit &usage;
}
if (@ARGV) {
    exit &usage;
}
Robert Ricci's avatar
Robert Ricci committed
113

Mac Newbold's avatar
Mac Newbold committed
114 115 116 117 118 119
if ($opt{s}) {
    $server = $opt{s};
}
if ($opt{p}) {
    $port = $opt{p};
}
120 121 122
if ($opt{t}) {
    $dbtag = $opt{t};
}
Mac Newbold's avatar
Mac Newbold committed
123 124 125
if ($opt{d}) {
    $debug = 1;
}
Robert Ricci's avatar
Robert Ricci committed
126

127
# Grab some constants into variables
128
my $TBANYMODE    = TBDB_NODEOPMODE_ANY;
129
my $TBPXERESET   = TBDB_TBCONTROL_PXERESET;
130 131
my $TBRESET      = TBDB_TBCONTROL_RESET;
my $TBTIMEOUT    = TBDB_TBCONTROL_TIMEOUT;
Mac Newbold's avatar
Mac Newbold committed
132 133 134
my $PXEBOOT      = TBDB_TBCONTROL_PXEBOOT;
my $BOOTING      = TBDB_TBCONTROL_BOOTING;
my $CHECKGENISUP = TBDB_TBCONTROL_CHECKGENISUP;
135 136 137
my $TBNOTIMEOUT  = TBDB_NO_STATE_TIMEOUT;
my $TBNODESTATE  = TBDB_TBEVENT_NODESTATE;
my $TBNODEOPMODE = TBDB_TBEVENT_NODEOPMODE;
138 139 140 141 142 143
my $TBCONTROL    = TBDB_TBEVENT_CONTROL;
my $TBCOMMAND    = TBDB_TBEVENT_COMMAND;
my $TBREBOOT     = TBDB_COMMAND_REBOOT;
my $TBPOWEROFF   = TBDB_COMMAND_POWEROFF;
my $TBPOWERON    = TBDB_COMMAND_POWERON;
my $TBPOWERCYCLE = TBDB_COMMAND_POWERCYCLE;
144
my $TBISUP       = TBDB_NODESTATE_ISUP;
145 146 147
my $PXEWAIT      = TBDB_NODESTATE_PXEWAIT;
my $PXEWAKEUP    = TBDB_NODESTATE_PXEWAKEUP;
my $PXEBOOTING   = TBDB_NODESTATE_PXEBOOTING;
148 149
my $TBRELOADDONEV1    = TBDB_TBCONTROL_RELOADDONE;
my $TBRELOADDONEV2    = TBDB_TBCONTROL_RELOADDONE_V2;
150 151 152 153
my $TBTIMEOUTREBOOT   = TBDB_STATED_TIMEOUT_REBOOT;
my $TBTIMEOUTNOTIFY   = TBDB_STATED_TIMEOUT_NOTIFY;
my $TBTIMEOUTCMDRETRY = TBDB_STATED_TIMEOUT_CMDRETRY;
my $TB_OSID_MBKERNEL  = TB_OSID_MBKERNEL;
154

155 156
# Special PXEBOOT state machine that all local nodes use.
my $PXEKERNEL	 = "PXEKERNEL";
157

158 159
# Even special-er SECUREBOOT state machine that local nodes may use.
my $SECUREBOOT	 = "SECUREBOOT";
160

161 162 163 164
# Protos.
sub debug(@);
sub fatal($);
sub notify($;$);
165
sub info($);
166 167 168 169 170 171 172
sub getTimeouts();
sub getValid();
sub getModeTrans();
sub getTriggers();
sub readStates(;@);
sub handleCtrlEvent($$);
sub reload();
173 174
sub StartEvents();
sub PollEvents($$);
175
    
176 177 178
if (!$debug) {
    if ( $TB eq $REALTB ) {
	$pidfile = "/var/run/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
179
    } else {
180
	$pidfile = "$TB/locks/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
181
    }
182
    debug("Using pidfile $pidfile\n");
Mac Newbold's avatar
Mac Newbold committed
183

184 185 186 187 188 189 190 191 192 193 194 195 196 197 198
    if (-e $pidfile) {
	my $otherpid = `cat $pidfile`;
	my $running = `ps -auxww | grep $otherpid | grep -v grep`;
	if ($running ne "") {
	    fatal("Lockfile $pidfile exists, and process $otherpid appears ".
		  "to be running.\n");
	} else {
	    notify("Lockfile exists, but process $otherpid appears to be dead".
		   "\n".
		   "Removing lock file...\n");
	}
	system("rm $pidfile") &&
	    fatal("Couldn't remove $pidfile: $? $!\n");
    }
    # Background
199
    if (TBBackGround($LOGFILE)) {
Mac Newbold's avatar
Mac Newbold committed
200 201
	exit(0);
    }
202
    TBdbfork();
Robert Ricci's avatar
Robert Ricci committed
203
}
204 205 206 207 208 209 210 211 212

if (defined($pidfile)) {
    sysopen(PIDFILE, $pidfile, O_WRONLY | O_EXCL | O_CREAT) ||
	fatal("Couldn't create '$pidfile': $? $!\n");
    print PIDFILE "$$";
    close PIDFILE;
    # If I make it to here, I'll need to clean up the lock file
    $lockfile = $pidfile;
}
Robert Ricci's avatar
Robert Ricci committed
213

214 215 216
# Change my $0 so that it is easier to see in a ps/top
$0 = "$0";

217 218 219 220 221
#
# Start up the event system interface.
#
if (StartEvents() != 0) {
    fatal("Error starting events");
Robert Ricci's avatar
Robert Ricci committed
222
}
223 224
# We want to exit on any warning. 
$SIG{__WARN__} = sub { print STDERR $_[0]; exit(-1); };
Robert Ricci's avatar
Robert Ricci committed
225 226 227

# Read in the pre-existing node states, and timeout and valid transition
# information from the database
228 229 230 231
my %timeouts  = getTimeouts();
my %valid     = getValid();
my %modeTrans = getModeTrans();
my %triggers  = getTriggers();
232
my %nodes     = readStates();
233
my %timeouttag= ();
234
if ($debug) { qshow(); }
Robert Ricci's avatar
Robert Ricci committed
235 236 237

# Gets set if a reload of state from the database should happen.
my $do_reload = 0;
238
my $do_reopen = 0;
239 240
my $sigrestart= 0;
my $sigcleanup= 0;
241
my $exiting   = 0;
Robert Ricci's avatar
Robert Ricci committed
242 243 244 245

# Make the daemon reload database state on a sighup - but I'm worried
# about what would happen if we tried to do this mid-loop. So, we'll
# just set a flag and do it when we're done with our current pass.
246
$SIG{HUP}  = sub { info("SIGHUP - Reloading DB state\n"); $do_reload = 1; };
247
$SIG{USR2} = sub { info("SIGUSR2 - Reopening logfile\n"); $do_reopen = 1; };
248

Mac Newbold's avatar
Mac Newbold committed
249
# Set up other signals.
250 251 252 253 254
$SIG{USR1} = \&restart_wrap;
$SIG{INT}  = \&cleanup_wrap;
$SIG{QUIT} = \&cleanup_wrap;
$SIG{ABRT} = \&cleanup_wrap;
$SIG{TERM} = \&cleanup_wrap;
Robert Ricci's avatar
Robert Ricci committed
255

256 257 258
# Track if I handled an event or not
my $event_count = 0;

259 260
# Control how long I block while waiting for events
my $blockwait=0;
261
my $nextdeadline=0;
262 263
my $mailqueue=0;

264 265 266 267 268
notify("Stated starting up\n");

sub process_event_queue() {
    $event_count=0;
    my $lastcount=-1;
269 270
    my $wait;
    my $now = time();
Mike Hibler's avatar
Mike Hibler committed
271
    debug("Polling at $now - mq=$mailqueue bw=$blockwait ndl=$nextdeadline\n");
272 273 274 275 276 277 278 279 280 281
    if ( $mailqueue == 0) {
	# no messages waiting...
	if ($blockwait) {
	    # we can wait a long time - nothing else will happen
	    # until we get an event, or get woken up by a signal
	    $wait = 600;
	} else {
	    # only wait until the next deadline...
	    if ($nextdeadline > 0) {
		$wait = $nextdeadline - $now;
282 283
	    } else {
		$wait = 0;
284 285 286 287 288
	    }
	}
    } else {
	# mail is waiting. Only block until it is time to send it.
	$wait = $lastmail + $mailgap - $now;
Mac Newbold's avatar
Mac Newbold committed
289
	debug("Now $now, mailgap $mailgap, last $lastmail ==> wait $wait\n");
290 291 292
    }
    if ($wait < 0) { debug("Wait was $wait!\n"); $wait=0; }
    my $finish = $now + $wait;
Mac Newbold's avatar
Mac Newbold committed
293
    while (($event_count != $lastcount || $wait > 0) &&
294
	   !($sigrestart || $sigcleanup || $do_reload || $do_reopen)) {
295
	$lastcount = $event_count;
Mac Newbold's avatar
Mac Newbold committed
296
	# Don't block if we got a signal!
297
	if ($wait<=0 || $sigrestart || $sigcleanup || $do_reload || $do_reload){
298
	    PollEvents(0, 0);
299
	} else {
300
	    #debug("Using blocking event poll - $wait seconds\n");
301
	    # timeout param is in milliseconds, so multiply
302
	    PollEvents(1, $wait);
303 304 305
	    $now = time();
	    # subtract seconds elapsed from my wait time
	    $wait = $finish - $now;
306
	    #debug("Finished blocking event poll - $wait seconds remain\n");
307
	    if ($event_count > 0 &&
Mac Newbold's avatar
Mac Newbold committed
308
		(qsize() > 0 || $mailqueue ||
309
		 $sigrestart || $sigcleanup || $do_reload || $do_reopen)) {
310 311 312 313 314 315 316
		$blockwait=0;
		$wait=0;
		#debug("Cancelling wait - timeouts/msgs waiting, or HUP'd\n");
		#debug("---End Blocking Wait ---\n");
	    }
	}
	#debug("Wait is $wait\n");
317 318
    }
    if ($event_count > 0) {
Mac Newbold's avatar
Mac Newbold committed
319
	debug("Handled $event_count event(s).\n");
320 321
    }
}
Robert Ricci's avatar
Robert Ricci committed
322

323
# Now, we just poll for events, and watch for timeouts
Robert Ricci's avatar
Robert Ricci committed
324
while (1) {
Mac Newbold's avatar
Mac Newbold committed
325
    my $now = time();
326 327 328 329
    my ($deadline,$node);

    # Check for nodes that have passed their timeout
    if (!qhead($deadline,$node)) {
330
	info("HEAD: $node in ".($deadline-$now).", queue=".qsize()."\n");
331 332
	while ($now >= $deadline && $node ne "") {
	    qpop($deadline,$node);
333
	    info("POP: $node in ".($deadline-$now).", queue=".qsize()."\n");
334 335 336 337 338 339 340 341 342 343 344 345 346
	    #
	    # If the node is no longer in the nodes array, it was most
	    # likely a dynamic virtual node which is now gone. Ignore.
	    # Need to look at reload() to see if we can catch this earlier. 
	    #
	    if (exists($nodes{$node})) {
		handleCtrlEvent($node,$TBTIMEOUT);
	    }
	    else {
		info("POP: $node is no longer in the nodes array. Skipping\n");
		delete($timeout_tag{$node})
		    if (exists($timeout_tag{$node}));
	    }
347 348 349 350
	    if (0) { qshow(); }
	    if (qhead($deadline,$node)) {
		$deadline=0; $node="";
	    }
351
	}
352 353
    } else {
	$deadline=0;
354
    }
355
    $nextdeadline = $deadline;
356

Mac Newbold's avatar
Mac Newbold committed
357 358 359 360
    if ($do_reload || ($now - $last_reload > $reload_time)) {
	reload();
	$do_reload = 0;
    }
Mac Newbold's avatar
Mac Newbold committed
361

Mike Hibler's avatar
Mike Hibler committed
362 363 364 365 366
    if (qsize()==0) {
	$blockwait=1;
	debug("---Blocking wait okay---\n");
    }

Mac Newbold's avatar
Mac Newbold committed
367 368
    # Send any messages in the queue if it is time
    notify("",1);
Mac Newbold's avatar
Mac Newbold committed
369

370 371
    if ($sigrestart) { restart(); }
    if ($sigcleanup) { cleanup(); }
372 373 374 375 376 377 378
    if ($do_reopen)  {
	ReOpenLog($LOGFILE);
	if (! kill('USR2', $eventchild)) {
	    fatal("Could not signal(USR2) event reader child\n");
	}
	$do_reopen = 0;
    }
379
    process_event_queue;
Robert Ricci's avatar
Robert Ricci committed
380 381
}

Mac Newbold's avatar
Mac Newbold committed
382 383
exit(0);

Robert Ricci's avatar
Robert Ricci committed
384
# Read the current states of nodes from the database
385
sub readStates(;@) {
Mac Newbold's avatar
Mac Newbold committed
386
    my %oldnodes = @_;
Mike Hibler's avatar
Mike Hibler committed
387
    my $now = time();
Mac Newbold's avatar
Mac Newbold committed
388 389

    # Guard against undefined variable warnings
390
    if (! %oldnodes) {
Mac Newbold's avatar
Mac Newbold committed
391 392
	%oldnodes = ();
    }
393

Mac Newbold's avatar
Mac Newbold committed
394 395 396
    #debug("readStates called\n");
    my $result = DBQueryFatal("SELECT node_id, eventstate, " .
			      "state_timestamp, op_mode, " .
397
			      "op_mode_timestamp, stated_tag FROM nodes ".
398
			      "where node_id not like 'sh%'");
Mac Newbold's avatar
Mac Newbold committed
399 400

    my %nodes;
401
    while (my ($node_id, $state, $timestamp, $mode, $mode_timestamp, $tag)
Mac Newbold's avatar
Mac Newbold committed
402
	   = $result->fetchrow()) {
403 404 405 406 407 408 409 410 411 412 413
	$nodes{$node_id}{"tag"} = (defined($tag) ? $tag : "");

	if ($dbtag ne "" && $dbtag eq $nodes{$node_id}{"tag"}) {
	    info("This stated will work on $node_id\n");
	}
	if ($dbtag eq "" && $dbtag ne $nodes{$node_id}{"tag"}) {
	    info("This stated will *NOT* work on $node_id\n");
	}
	if ($dbtag ne $nodes{$node_id}{"tag"}) {
	    remTimeout($node_id);
	}
Mac Newbold's avatar
Mac Newbold committed
414

415
	#
Mac Newbold's avatar
Mac Newbold committed
416 417 418 419
	# If there's an entry in oldnodes for this node, and it
	# hasn't changed state or time, use the old entry (so that
	# we don't lose information about which nodes we've already
	# notified the ops about, etc.)
420
	#
Mac Newbold's avatar
Mac Newbold committed
421 422 423 424 425
	if ($oldnodes{$node_id} && $state && $timestamp &&
	    ($oldnodes{$node_id}{state} eq $state) &&
	    ($oldnodes{$node_id}{mode} eq $mode) &&
	    ($oldnodes{$node_id}{timestamp} == $timestamp)) {
	    $nodes{$node_id} = $oldnodes{$node_id};
426
	} else {
Mac Newbold's avatar
Mac Newbold committed
427 428 429 430
	    $nodes{$node_id}{state}          = $state;
	    $nodes{$node_id}{timestamp}      = $timestamp;
	    $nodes{$node_id}{mode}           = $mode;
	    $nodes{$node_id}{mode_timestamp} = $mode_timestamp;
431 432 433
	    $nodes{$node_id}{notified}       = 0;
	    $nodes{$node_id}{timedout}       = 0;
	    $nodes{$node_id}{noretry}        = 0;
434
	    # Is there a timeout? If so, set it up!
Mike Hibler's avatar
Mike Hibler committed
435
	    if (defined($timestamp) && $dbtag eq $nodes{$node_id}{"tag"}) {
436
		setTimeout($mode,$state,$node_id,$timestamp);
437
	    }
438
	}
Mac Newbold's avatar
Mac Newbold committed
439 440
    }
    return %nodes;
Robert Ricci's avatar
Robert Ricci committed
441 442 443 444 445 446
}

#
# Read timeouts for various states from the database
#
sub getTimeouts() {
Mac Newbold's avatar
Mac Newbold committed
447 448 449
    #debug("getTimeouts called\n");
    my $result = DBQueryFatal("SELECT op_mode, state, timeout, action " .
			      "FROM state_timeouts");
Robert Ricci's avatar
Robert Ricci committed
450

Mac Newbold's avatar
Mac Newbold committed
451 452 453 454 455
    my %timeouts;
    while (my ($op_mode, $state, $timeout, $action) = $result->fetchrow()) {
	$timeouts{$op_mode}{$state} = [ $timeout, $action ];
    }
    return %timeouts;
Robert Ricci's avatar
Robert Ricci committed
456 457 458 459 460 461
}

#
# Read the list of valid state transitions from the database
#
sub getValid() {
Mac Newbold's avatar
Mac Newbold committed
462 463 464
    #debug("getValid called\n");
    my $result = DBQueryFatal("SELECT op_mode, state1, state2 " .
			      "FROM state_transitions");
Robert Ricci's avatar
Robert Ricci committed
465

Mac Newbold's avatar
Mac Newbold committed
466 467 468 469 470
    my %valid;
    while (my ($mode,$state1, $state2) = $result->fetchrow()) {
	$valid{$mode}{$state1}{$state2} = 1;
    }
    return %valid;
Robert Ricci's avatar
Robert Ricci committed
471 472
}

473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490
#
# Check if a transition is valid. Actually, it is a check for an invalid
# transition.
#
sub IsValidTransition($$$)
{
    my ($mode, $oldstate, $newstate) = @_;
    my $isvalid = 1;

    $isvalid = 0
	if ($valid{$mode} &&
	    ($valid{$mode}{$oldstate} || $valid{$mode}{"*"}) &&
	    ! ($valid{$mode}{$oldstate}{$newstate} ||
	       $valid{$mode}{"*"}{$newstate}));

    return $isvalid;
}

491 492 493 494
#
# Read the list of valid mode transitions from the database
#
sub getModeTrans() {
Mac Newbold's avatar
Mac Newbold committed
495
    #debug("getModeTrans called\n");
Mac Newbold's avatar
Mac Newbold committed
496
    my $result =
Mac Newbold's avatar
Mac Newbold committed
497 498 499 500 501 502 503 504 505 506 507
      DBQueryFatal("SELECT op_mode1, state1, op_mode2, state2 " .
		   "FROM mode_transitions order by op_mode1,state1");

    my %modeTrans;
    while (my ($mode1,$state1, $mode2, $state2) = $result->fetchrow()) {
	if (!defined($modeTrans{"$mode1:$state1"})) {
	    $modeTrans{"$mode1:$state1"}= ["$mode2:$state2"];
	} else {
	    my @l = @{$modeTrans{"$mode1:$state1"}};
	    push(@l, "$mode2:$state2");
	    $modeTrans{"$mode1:$state1"}= \@l;
508
	}
Mac Newbold's avatar
Mac Newbold committed
509 510
    }
    return %modeTrans;
511 512 513 514 515 516
}

#
# Read the list of states which trigger an action
#
sub getTriggers() {
517
    debug("getTriggers called\n");
Mac Newbold's avatar
Mac Newbold committed
518

519 520
    debug("anymode ==> '$TBANYMODE'\n");

Mac Newbold's avatar
Mac Newbold committed
521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541
    # A note about triggers:
    #
    # "per-node" triggers only affect their specific node in a
    # particular mode/state, and are run first of all. "global"
    # triggers are triggers for a given mode/state that affect all
    # nodes, and are run after any per-node triggers. "Any-mode"
    # triggers are tied to a state, and occur in that state in any
    # mode. The any-mode triggers are over-ridden by global triggers,
    # and if an "Any-mode" trigger for state XYZ exists as well as a
    # global trigger for mode FOOBAR state XYZ, then when I arrive in
    # XYZ any per-node triggers will be run. Then, if I'm in mode
    # FOOBAR, only the global trigger will run. If I'm in any other
    # mode, only the any-mode trigger will run.

    # (our "*" is stored as $TBANYMODE)
    # Per-node triggers have a specific node_id
    # Global triggers have "*" as the node_id
    # Any-mode triggers have "*" as the mode, and can be global or per-node

    # Grab global triggers (including any-mode)
    my $result =
542
      DBQueryFatal("SELECT op_mode, state, `trigger` " .
543 544
		   "FROM state_triggers where node_id='$TBANYMODE' ".
		   "order by op_mode,state");
Mac Newbold's avatar
Mac Newbold committed
545
    my %t;
546
    while (my ($mode, $state, $trig) = $result->fetchrow()) {
Mac Newbold's avatar
Mac Newbold committed
547 548 549
	my @trigs = split(/\s*,\s*/,$trig);
	$t{"$mode:$state"} = \@trigs;
	debug("trig($mode:$state)\t => ".join(',',@trigs)."\n");
550 551
    }

Mac Newbold's avatar
Mac Newbold committed
552 553
    # Grab per-node triggers (including any-mode)
    $result =
554
      DBQueryFatal("SELECT node_id, op_mode, state, `trigger` " .
555 556 557 558 559 560
		   "FROM state_triggers where node_id!='$TBANYMODE' ".
		   "order by op_mode,state");
    while (my ($n, $mode, $state, $trig) = $result->fetchrow()) {
	my @trigs = split(/\s*,\s*/,$trig);
	$t{"$n:$mode:$state"} = \@trigs;
	debug("trig($n:$mode:$state)\t => ".join(',',@trigs)."\n");
Mac Newbold's avatar
Mac Newbold committed
561
    }
562

Mac Newbold's avatar
Mac Newbold committed
563 564
    debug(hash_recurse(%t));

Mac Newbold's avatar
Mac Newbold committed
565
    return %t;
566 567
}

Robert Ricci's avatar
Robert Ricci committed
568 569 570 571
#
# Gets called for every event that we recieve
#
sub handleEvent($$$) {
572
    my ($objtype,$objname,$eventtype) = @_;
Mac Newbold's avatar
Mac Newbold committed
573
    $event_count++;
Mac Newbold's avatar
Mac Newbold committed
574

575
    #
Mac Newbold's avatar
Mac Newbold committed
576 577 578
    # For readability, only do this on the main stated. This will print all
    # events, which gets cumbersome with debugging versions, so we'll print
    # only applicable events in debug versions, after we decide they apply.
579
    #
580
    if ($dbtag eq "") {
581 582
	debug("Got an event: ($objtype,$objname,$eventtype)\n");
    }
Mac Newbold's avatar
Mac Newbold committed
583 584 585 586

    #
    # Check to see if another instance is supposed to be handling this node
    #
587 588
    if ($objtype ne $TBCOMMAND) {
	my $node = $objname;
Mac Newbold's avatar
Mac Newbold committed
589

590 591 592 593 594 595 596
	#
	# If we have never seen this node, reload.
	#
	if (! defined($nodes{$node})) {
	    reload();

	    # Still not defined, someone screwed up! This could end up
Mac Newbold's avatar
Mac Newbold committed
597
	    # churning via reload(). Bad.
598
	    if (! defined($nodes{$node})) {
599 600 601 602
		notify("Got $objtype/$eventtype for nonexistent $node!\n");
		return;
	    }
	}
Mac Newbold's avatar
Mac Newbold committed
603

604 605
	#
	# If a stated_tag was specified on the command line, ignore those
Mac Newbold's avatar
Mac Newbold committed
606
	# nodes that do not match.
607
	#
Mac Newbold's avatar
Mac Newbold committed
608
	#debug("dbtag='$dbtag', node $node='".$nodes{$node}{"tag"}."'\n");
609 610 611 612 613 614 615 616 617 618
	if ($dbtag ne $nodes{$node}{"tag"}) {
	    # Record when main stated ignores a node.
	    info("Got $objtype/$eventtype for $node, which is not mine\n")
		if ($dbtag eq "");
	    return;
	}
	if (!checkDBRedirect($node)) {
	    info("Got $objtype/$eventtype for $node, which is not mine\n");
	    return;
	}
Mac Newbold's avatar
Mac Newbold committed
619
    }
Mac Newbold's avatar
Mac Newbold committed
620 621 622 623 624

    #
    # If this is a debugging version, then this event is for one of my
    # nodes, so I can print out the event now. (Main version prints earlier.)
    #
625 626 627
    if ($dbtag ne "") {
	debug("Got an event: ($objtype,$objname,$eventtype)\n");
    }
Mac Newbold's avatar
Mac Newbold committed
628 629 630

 SWITCH: for ($objtype) {

Mac Newbold's avatar
Mac Newbold committed
631 632 633 634 635 636 637 638 639 640 641 642 643 644
	(/$TBNODESTATE/) && do {
	    stateTransition($objname,$eventtype);
	    last;
	};
	(/$TBNODEOPMODE/) && do {
	    opModeTransition($objname,$eventtype);
	    notify("Use of deprecated event TBNODEOPMODE:\n".
		   "$objname->$eventtype\n");
	    last;
	};
	(/$TBCONTROL/) && do {
	    handleCtrlEvent($objname,$eventtype);
	    last;
	};
645 646 647 648
	(/$TBCOMMAND/) && do {
	    handleCommand($objname,$eventtype);
	    last;
	};
649

Mac Newbold's avatar
Mac Newbold committed
650
    }
651 652 653 654 655

}

sub stateTransition($$) {

656
    my ($node,$newstate) = @_;
Robert Ricci's avatar
Robert Ricci committed
657

658 659
    # Check for invalid transitions
    my ($oldstate, $mode);
660 661 662
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

663
    if ($oldstate && $mode && !IsValidTransition($mode,$oldstate,$newstate)) {
664
	notify("Invalid transition for node $node from $mode/$oldstate " .
665
	       "to $newstate\n");
666
	
667 668 669 670 671 672
	#
	# Machines in the secure boot path are not allowed to jump
	# willy-nilly into unknown states.
	#
        if ($mode eq TBDB_NODEOPMODE_SECUREBOOT ||
	    $mode eq TBDB_NODEOPMODE_SECURELOAD) {
673 674 675
            $newstate = TBDB_NODESTATE_SECVIOLATION;
            notify("Moving $node to $newstate because it's in $mode\n");
        }
676 677 678 679 680 681
    }
    #
    # Nodes that are in the SECVIOLATION state are not allowed to leave!
    #
    if ($oldstate eq TBDB_NODESTATE_SECVIOLATION &&
	$newstate ne TBDB_NODESTATE_SECVIOLATION) {
682
	#
683 684 685
	# Allow transitions to SHUTDOWN/POWEROFF.
	# This allows someone to reboot (turn-off and back on) a node in
	# the SECVIOLATION state getting it back to MINIMAL/SHUTDOWN.
686 687 688
	#
	# XXX DEBUG ONLY!
	# 
689 690
	if ($soft_secviolation &&
	    ($newstate eq TBDB_NODESTATE_SHUTDOWN ||
691
	     $newstate eq TBDB_NODESTATE_POWEROFF ||
692 693
	     $newstate eq TBDB_NODESTATE_GPXEBOOTING)) {
	    notify("$node allowed to transition: SECVIOLATION => $newstate\n");
694 695 696 697
	} else {
	    notify("$node tried to leave SECVIOLATION (to $newstate)\n");
	    $newstate = TBDB_NODESTATE_SECVIOLATION;
	}
698
    }
Robert Ricci's avatar
Robert Ricci committed
699

700 701 702 703
    my $now = time();
    $nodes{$node}{state}     = $newstate;
    $nodes{$node}{timestamp} = $now;
    $nodes{$node}{notified}  = 0;
704

705 706 707
    info("$node: $mode/$oldstate => $mode/$newstate\n");
    DBQueryFatal("UPDATE nodes SET eventstate='$newstate', " .
		 "state_timestamp='$now' WHERE node_id='$node'");
708

709 710 711 712
    # Before we set the timeout (overwriting any current ones), we need
    # to check if we had a pending command
    if (qfind($node) &&
	$timeout_tag{$node} =~ /^$TBCOMMAND:/) {
713
        debug("TimeoutTag = '$timeout_tag{$node}'\n");
714
	my ($str,$cmd) = split(":",$timeout_tag{$node});
715
	debug("str=$str\tcmd=$cmd\tTBREBOOT=$TBREBOOT\tstate=$newstate\n");
716
	if ($cmd eq $TBREBOOT) {
717
	    if ($newstate eq TBDB_NODESTATE_SHUTDOWN ) {
718 719 720
		info("$node: $TBREBOOT success\n");
		# Timeout will get cleared below by setTimeout call
	    } else {
721 722
		notify("$node: $TBREBOOT in progress, but got state ".
		       "$newstate instead of ".TBDB_NODESTATE_SHUTDOWN."!\n");
723 724 725 726 727
	    }
	#} elsif ($cmd eq $FOO ) {
	    # Add more here...
	} else {
	    notify("$node: Unknown command timeout '$timeout_tag{$node}' ".
728
		   "found at $mode/$newstate\n");
729 730 731
	}
    }

732 733 734 735 736
    #
    # Check if this state has a timeout, and if so, put it in the queue.
    # Note that any opmode transition below will replace (or remove) this
    # timeout if appropriate.
    #
737 738
    setTimeout($mode,$newstate,$node,$now);

Mac Newbold's avatar
Mac Newbold committed
739 740 741 742 743 744 745
    # Check if this state has any triggers
    my @nodetrigs = GetNodeTriggerList($node,$mode,$newstate,1);
    my @trigs = GetNodeTriggerList($node,$mode,$newstate);
    if (@trigs > 0) {
	debug("Running triggers: ".join("/",@trigs)."\n");
	foreach ( @trigs) {
	    my $trig = $_;
746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762
	    /^SCRIPT:([-\w\/]+)$/ && do {
		#
		# Run a script. No arguments at the moment.
		#
		my $script = $1;

		#
		# If the path is absolute, run it. Otherwise it has to
		# come from the sbin directory.
		#
		if (! ($script =~ /^\//)) {
		    $script = "$TB/sbin/$script";
		}
		info("$node: Running $script.\n");
		system("$script &");
		next;
	    };
Mac Newbold's avatar
Mac Newbold committed
763
	    /^$PXEBOOT$/ && do {
764
		#
Mac Newbold's avatar
Mac Newbold committed
765 766 767 768 769 770 771 772 773 774 775 776
		# See if we jumped into the PXEBOOT kernel. Bootinfo
		# will send PXEBOOTING every time a node contacts it,
		# which is our indicator that the node is in the first
		# phase of booting. At this point we want to switch
		# state machines since the entire boot process is
		# governed by a single state machine that is
		# independent of the OS that the node will eventually
		# boot.  Rather then encode that in each state
		# machine, we use a special machine with a defined
		# entrypoint (PXEBOOTING) and a defined exitpoint
		# (BOOTING). See below for where we jump back out of
		# this state machine.
777
		#
Mac Newbold's avatar
Mac Newbold committed
778 779 780 781
		# Jumped in. We need to change the opmode so that
		# the state transitions are legal. We do not
		# bother to save the old opmode since we can
		# figure it out later when we leave.
782
		#
Mac Newbold's avatar
Mac Newbold committed
783 784 785 786 787
		debug("Running $PXEBOOT trigger\n");
		if ($mode ne $PXEKERNEL) {
		    info("$node: Forcing mode transition into $PXEKERNEL!\n");
		    opModeTransition($node, $PXEKERNEL, 1);
		    $mode=$PXEKERNEL;
788
		}
Mac Newbold's avatar
Mac Newbold committed
789 790
		next;
	    };
791 792 793 794 795 796
	    /^$SECUREBOOT$/ && do {
		#
		# Force machine into the SECUREBOOT/LOAD op_mode.
		# Currently triggered by receipt of GPXEBOOTING state.
		# This could come from any state as it just indicates that
		# a machine with a gPXE dongle has rebooted.
797
		#
798 799 800 801
		# To differentiate BOOT from LOAD:
		# if next_op_mode is SECURELOAD, goto SECURELOAD
		# else if mode/state is SECURELOAD/SHUTDOWN, goto SECURELOAD
		# else goto SECUREBOOT.
802
		#
803 804 805 806 807
		debug("Running $SECUREBOOT trigger\n");
		my $query_result =
		    DBQueryWarn("select next_op_mode from nodes ".
				"where node_id='$node'");
		my ($nextmode) = $query_result->fetchrow();
808 809 810 811 812
		if (!$nextmode) {
		    $nextmode = $mode;
		}
		if ($nextmode ne TBDB_NODEOPMODE_SECURELOAD ||
		    $oldstate ne TBDB_NODESTATE_SHUTDOWN) {
813 814 815 816 817 818
		    $nextmode = TBDB_NODEOPMODE_SECUREBOOT;
		}
		if ($mode ne $nextmode) {
		    info("$node: Forcing mode transition to $nextmode!\n");
		    opModeTransition($node, $nextmode, 1);
		    $mode=$nextmode;
819 820 821
		}
		next;
	    };
Mac Newbold's avatar
Mac Newbold committed
822 823 824 825 826
	    /^$BOOTING$/ && do {
		#
		# See if we are in the right mode/osid.
		#
		my ($bootosid,$bootopmode) = TBBootWhat($node, $debug);
827

Mac Newbold's avatar
Mac Newbold committed
828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893
		info("$node: BootWhat says $bootosid (mode $bootopmode).\n");
		DBQueryFatal("update nodes set osid='$bootosid' ".
			     "where node_id='$node'");

		if ($bootopmode ne $mode) {
		    if ($mode eq $PXEKERNEL) {
			#
			# If we came from PXE boot, then we have to
			# jump out of the PXEKERNEL state machine into
			# whatever state machine is current for the
			# node. Since we came through bootinfo, we
			# know that the node is doing what it is
			# supposed to, and that this change matches
			# what the node is booting.
			#
			info("$node: Forcing mode transition out of $PXEKERNEL!\n");
			opModeTransition($node, $bootopmode, 1);
			$mode=$bootopmode;
		    }
		    elsif ($oldstate eq TBDB_NODESTATE_ISUP) {
			#
			# Skipped SHUTDOWN, which could result in a
			# missed opmode transition. Can this really
			# happen anymore?
			#
			info("$node: Came from ISUP! ".
			     "Checking for mode transition\n");
			my $query_result =
			  DBQueryWarn("select next_op_mode from nodes ".
				      "where node_id='$node'");
			my ($nextmode) = $query_result->fetchrow();
			if ($nextmode) {
			    info("$node: Forcing mode transition!\n");
			    opModeTransition($node, $nextmode, 1);
			    $mode=$nextmode;
			}
		    }
		    else {
			my $str = "$node is running $bootosid, but in ".
			  "mode $mode\ninstead of mode $bootopmode!\n";
			
			if ($bootopmode eq "RELOAD") {
			    #
			    # For now, only force if we're going into
			    # reload mode, so we don't get stuck
			    # looping in reloading.  Can this happen
			    # anymore?
			    #
			    DBQueryFatal("UPDATE nodes SET ".
					 "op_mode='$bootopmode', ".
					 "op_mode_timestamp=".
					 "unix_timestamp(now()) ".
					 "WHERE node_id='$node'");
			    $nodes{$node}{mode} = $bootopmode;
			    $nodes{$node}{mode_timestamp} = $now;
			    $str .= "Forced op_mode to $bootopmode.\n";
			}
			notify($str);
		    }
		}
		next;
	    };
	    /^$CHECKGENISUP$/ && do {
		checkGenISUP($node);
		next;
	    };
894 895 896 897
	    /^CHECKPORTREG$/ && do {
		CheckPortRegistration($node);
		next;
	    };
898 899 900 901 902 903
	    /^$TBPXERESET$/ && do {
		# We successfully booted, so reset one-shot PXEboot if any
		# Check if we really need to do a reset
		handleCtrlEvent($node,$trig);
		next;
	    };
904
	    /^$TBRESET$/ && do {
905 906 907
		# We successfully booted, so clear some flags
		$nodes{$node}{noretry}   = 0;
		$nodes{$node}{timedout}  = 0;
908 909 910 911
		# Check if we really need to do a reset
		my $r = DBQueryWarn("select osid,def_boot_osid from nodes ".
				    "where node_id='$node'");
		my ($osid,$defosid) = $r->fetchrow();
912 913 914 915 916 917 918
		if (! (defined($osid) && defined($defosid))) {
		    info("$node: osid not defined\n")
			if (!defined($osid));
		    info("$node: def_boot_osid not defined\n")
			if (!defined($defosid));
		}
		elsif ($osid ne $defosid) {
919 920
		    handleCtrlEvent($node,$trig);
		}
921 922 923 924 925
		# XXX tmp hack
		elsif ($mode eq "WIMRELOAD") {
		    info("$node: RESET in WIMRELOAD, but osid=defosid=$osid\n");
		    handleCtrlEvent($node,$trig);
		}
926 927
		next;
	    };
928
	    (/^$TBRELOADDONEV1$/ || /^$TBRELOADDONEV2$/) && do {
929 930 931
		handleCtrlEvent($node,$trig);
		next;
	    };
932 933 934 935 936 937 938 939
	    /^$TBISUP$/ && do {
		info("$node: Triggered $TBISUP\n");
		EventSendWarn(host      => $BOSSNODE ,
			      objtype   => TBDB_TBEVENT_NODESTATE ,
			      eventtype => TBDB_NODESTATE_ISUP ,
			      objname   => $node);
		next;
	    };
940
	    (/^$TBREBOOT$/ || /^$TBPOWERCYCLE$/ || /^$TBPOWEROFF$/) && do {
941
		handleCommand($node,$trig);
942 943 944
		next;
	    };
            (/^EMAILNOTIFY$/) && do {
945 946
		my $msg = "$node entered state $mode/$newstate from " .
		    "$mode/$oldstate";
947
		my $dest = $REALTBOPS;
948
		if ($newstate eq TBDB_NODESTATE_SECVIOLATION) {
949 950 951 952 953 954 955 956
		    if ($soft_secviolation) {
			$msg .= "\n\nNode $node was allowed to continue.\n";
			$dest = $TBOPS;
		    } else {
			$msg .= "\n\nNode $node has been powered off.\n" .
			        "You must address the cause of the violation ".
				"and reset the eventstate before powering on.";
		    }
957
		}
958 959
		SENDMAIL($dest, "STATED: $node entered state $newstate",
			 $msg, "Stated Daemon <".$TBOPS.">");
960
		next;
961
            };
962 963 964 965 966 967 968 969 970 971 972 973
	    /^RELOADOLDMFS$/ && do {
		my $frisbee_osid = TBNodeDiskloadOSID($node);
		my $frisbee_name = DBQuerySingleFatal("select osname from os_info where osid=$frisbee_osid");
		my $msg = 
		    ("Attempted to load multiple images on $node using an old Frisbee MFS.\n".
		     "To make this work please update the $frisbee_name MFS image.\n");
		SENDMAIL($REALTBOPS,
			 "$frisbee_name Needs Updating",
			 $msg,
			 "Stated Daemon <".$TBOPS.">");
		next;
	    };
974
	    notify("Unknown trigger '$trig' for $node in $mode/$newstate!\n");
975
	}
Mac Newbold's avatar
Mac Newbold committed
976 977 978 979 980 981 982
	# Clear any of the node triggers that we ran.
	# (Don't clear all of them, because some of the triggers we ran
	# may have caused others to be set, and we don't want to nuke them.)
	if (@nodetrigs > 0) {
	    debug("Clearing node triggers: ".join("/",@nodetrigs)."\n");
	    ClearNodeTrigger($node,$mode,$newstate,@nodetrigs);
	}
983
    }
984

985 986 987 988 989 990 991 992
    # Check if this state can trigger a mode transition
    if (defined($modeTrans{"$mode:$newstate"})) {
	info("$node: Checking for mode transition\n");
	my $r = DBQueryWarn("select next_op_mode from nodes ".
			    "where node_id='$node'");
	my ($nextmode) = $r->fetchrow();
	if ($nextmode) {
	    opModeTransition($node,$nextmode);
Mac Newbold's avatar
Mac Newbold committed
993 994 995
	} else {
	    debug("No next mode.\n");
	}
996 997
    }
}
998

999
sub opModeTransition($$;$) {
Mac Newbold's avatar
Mac Newbold committed
1000

1001 1002
    my ($node,$newmode,$force) = @_;
    if (!defined($force)) { $force = 0; }
Mac Newbold's avatar
Mac Newbold committed
1003

1004
    info("$node: Mode change to $newmode requested ($force)\n");
Mac Newbold's avatar
Mac Newbold committed
1005

1006 1007
    # Check for invalid transitions
    my ($oldstate, $mode, $nextstate);
1008 1009 1010
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

1011
    if (defined($modeTrans{"$mode:$oldstate"}) || $force) {
1012 1013 1014 1015 1016
	#
	# If there is a next state associated with mode:oldstate, use it.
	# Otherwise we just remain in the old state.
	#
	if (defined($modeTrans{"$mode:$oldstate"})) {
1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027
	    debug("Mode Transition check:\n");
	    my $translist = join(",",@{$modeTrans{"$mode:$oldstate"}});
	    #debug("translist=$translist\n");
	    #debug("splitlist=".join(", ",split(/[:,]/,$translist))."\n");
	    my %trans = split(/[:,]/,$translist);
	    debug("Valid transitions from $mode/$oldstate are:\n");
	    foreach my $k (sort keys %trans) {
		debug("$k => $trans{$k}\n");
	    }
	    if (defined($trans{$newmode})) {
		$nextstate=$trans{$newmode};
1028 1029 1030 1031
		if ($force) {
		    info("$node: opMode force changing state along with mode\n");
		}
	    } elsif (!$force) {
1032 1033 1034
		notify("Invalid mode transition for $node from ".
		       "$mode/$oldstate to $newmode!\n");
	    }
1035 1036
	}
    } else {
1037
	notify("Invalid mode transition for $node from $mode/$oldstate: ".
1038 1039
	       "Not a valid mode transition state!\n");
    }
Mac Newbold's avatar
Mac Newbold committed
1040 1041 1042
    if (!$nextstate) {
	$nextstate=$oldstate;
    }
Mac Newbold's avatar
Mac Newbold committed
1043

1044 1045 1046 1047 1048 1049
    my $now = time();
    $nodes{$node}{state}     = $nextstate;
    $nodes{$node}{timestamp} = $now;
    $nodes{$node}{mode}           = $newmode;
    $nodes{$node}{mode_timestamp} = $now;
    $nodes{$node}{notified}       = 0;
Mac Newbold's avatar
Mac Newbold committed
1050

1051 1052 1053 1054 1055
    info("$node: $mode/$oldstate => $newmode/$nextstate\n");
    DBQueryFatal("UPDATE nodes SET eventstate='$nextstate', ".
		 "next_op_mode='', op_mode='$newmode', ".
		 "state_timestamp='$now', ".
		 "op_mode_timestamp='$now' WHERE node_id='$node'");
Mac Newbold's avatar
Mac Newbold committed
1056 1057 1058 1059

    # Check if this state has a timeout, and if so, put it in the queue
    setTimeout($newmode,$nextstate,$node,$now);

1060 1061 1062 1063
}

sub handleCtrlEvent($$) {
    my ($node,$event) = @_;
1064

1065 1066 1067 1068 1069 1070 1071
    # stated should use the node objects.
    my $nodeobj = Node->Lookup($node);
    if (!defined($nodeobj)) {
	notify("handleCtrlEvent: Could not lookup node object for $node!\n");
	return;
    }
    $nodeobj->FlushReserved();
1072
    info("CtrlEvent: $node, $event\n");
1073

1074
    foreach ($event) {
1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089
	/^$TBPXERESET$/ && do {
	    #
	    # Clear next_pxe_boot_path with pxe_select.
	    #
	    # Note that this will recreate the DHCPD config file and HUP it.
	    # It will also ssh over to any subbosses and do the same.
	    # Thus there is lost of potential to get hung or take a long time.
	    #
	    $cmd = "$pxeselect -d -c -1 $node";
	    system($cmd) and
		notify("$node/$event: Could not clear next_pxe_boot_path!\n");

	    info("Performed $TBPXERESET for $node\n");
	    next;
	};
1090
	/^$TBRESET$/ && do {
1091 1092
	    #
	    # Clear next_boot_path with os_select.
Mac Newbold's avatar
Mac Newbold committed
1093
	    #
1094
	    $cmd = "$osselect -d -c -1 $node";
1095
	    system($cmd) and
1096
		notify("$node/$event: Could not clear next_boot_path!\n");
Mac Newbold's avatar
Mac Newbold committed
1097 1098

	    info("Performed $TBRESET for $node\n");
1099 1100
	    next;
	};
1101
	(/^$TBRELOADDONEV1$/ || /^$TBRELOADDONEV2$/) && do {
1102
	    info("Clearing reload info for $node\n");
1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117
	    $nodeobj->ClearCurrentReload();
	    $nodeobj->FlushReserved();
	    my $experiment = $nodeobj->Reservation();
	    if (defined($experiment) &&
		$experiment->pid() eq NODERELOADING_PID &&
		$experiment->eid() eq NODERELOADING_EID) {
		$nodeobj->ClearSchedReload();
		my $reserved_pid = $nodeobj->CheckPreReserve(1);
		if (defined($reserved_pid)) {
		    info("Setting pre reserve for $node to $reserved_pid\n");
		}
		$nodeobj->ClearReservation();
		$nodeobj->SetNodeHistory(TB_NODEHISTORY_OP_FREE,
					 undef, $experiment);
		info("Released $node from $experiment\n");
1118
	    }
1119 1120 1121 1122 1123
	    if ($event eq $TBRELOADDONEV2) {
		info("Sending an apod to $node\n");
		system("$apod $node") == 0 or
		    notify("Could not apod $node after $TBRELOADDONEV2!\n");
	    }
1124 1125 1126
	    next;
	};
	/^$TBTIMEOUT$/ && do {
1127 1128 1129 1130
	    my ($mode,$state) = split(":",$timeout_tag{$node});
	    delete($timeout_tag{$node});
	    my $curstate = $nodes{$node}{state};
	    my $curmode = $nodes{$node}{mode};
1131
	    my ($timeout,$action);
1132 1133 1134 1135 1136
	    if (!defined($nodes{$node}{notified})) {
		$nodes{$node}{notified}=0;
	    }
	    $nodes{$node}{notified}++;
	    my $notified = $nodes{$node}{notified};
1137 1138
	    $nodes{$node}{timedout}++;
	    my $timedout = $nodes{$node}{timedout};
1139 1140 1141 1142
	    if ($mode && $state && $timeouts{$mode} &&
		$timeouts{$mode}{$state}) {
		($timeout, $action) = @{$timeouts{$mode}{$state}};
	    }
1143 1144
	    if ($mode eq $TBCOMMAND) {
		# It is a command, not a true state
1145
		if ($action eq $TBTIMEOUTCMDRETRY) {
1146
		    # Retry the command
1147
		    notify("$node: Command $state, retry #$timedout\n");
1148
		    # notify in case we get in a retry loop...
1149
		    handleCommand($node,$state,$timedout,1);
1150 1151 1152 1153 1154 1155
		} else {
		    notify("$node: Unknown timeout action for ".
			   "$mode/$state: '$action'\n");
		}
		next;
	    }
1156 1157 1158 1159 1160 1161 1162

	    #
	    # Trash. This stuff should not be encoded this way, but I have
	    # no idea how timeouts, TBCOMMAND, and actions interact.
	    #
	    if ($curstate eq $PXEWAKEUP) {
		my $optarg = ($debug ? "-d " : "");
Mac Newbold's avatar
Mac Newbold committed
1163

1164 1165
		if ($timed