stated.in 57.9 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 42

$| = 1;

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

53 54 55 56 57
# 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
58
# Number of iterations (roughly, seconds) after which we'll reload
Robert Ricci's avatar
Robert Ricci committed
59 60
# information from the database. This is so we don't end up with information
# that's _too_ out of sync.
61
my $reload_time = 600;
62
my $last_reload = time;
Robert Ricci's avatar
Robert Ricci committed
63

Mike Hibler's avatar
Mike Hibler committed
64 65 66
# 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.
67
my $maxpasttimeout = -(7 * 24 * 60 * 60);	# 1 week
Mike Hibler's avatar
Mike Hibler committed
68

69 70 71
# Handling of SECVIOLATIONS, eventually controlled by a sitevar
my $soft_secviolation = 1;

72 73 74
# Command line opts.
my $dbtag = "";
my $debug = 0;
75 76
my $server = "localhost";
my $port   = @BOSSEVENTPORT@;
77 78
my $lockfile;
my $pidfile;
79
my $eventchild;
80

Robert Ricci's avatar
Robert Ricci committed
81 82 83
# Process command-line arguments

sub usage {
Mac Newbold's avatar
Mac Newbold committed
84
    print << "END";
85
Usage: $0 [-h] [-d] [-s server] [-p port] [-t dbtag]
86
-h              This message
87 88
-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
89 90
-s server       Use specified server, instead of this site's bossnode
-p port	        Use specified port
91
Send SIGHUP to reload database state, or SIGUSR1 to restart completely.
Robert Ricci's avatar
Robert Ricci committed
92
END
93
    exit(1);
Robert Ricci's avatar
Robert Ricci committed
94 95
}

Mac Newbold's avatar
Mac Newbold committed
96
# Only root should run this - it won't work when run as a user...
97
# (Or, let an admin run it if it isn't the real one in /usr/testbed/ )
98
if ($UID && ( $TB eq $REALTB || ! TBAdmin($UID) ) ) {
Mac Newbold's avatar
Mac Newbold committed
99 100 101
    die("Only root can run this script!\n");
}

102
my @args = @ARGV;    # save a copy for restart before we mess with them.
Robert Ricci's avatar
Robert Ricci committed
103
my %opt = ();
104
if (!getopts("ds:p:ht:l",\%opt)) { usage(); }
Robert Ricci's avatar
Robert Ricci committed
105

Mac Newbold's avatar
Mac Newbold committed
106 107 108 109 110 111
if ($opt{h}) {
    exit &usage;
}
if (@ARGV) {
    exit &usage;
}
Robert Ricci's avatar
Robert Ricci committed
112

Mac Newbold's avatar
Mac Newbold committed
113 114 115 116 117 118
if ($opt{s}) {
    $server = $opt{s};
}
if ($opt{p}) {
    $port = $opt{p};
}
119 120 121
if ($opt{t}) {
    $dbtag = $opt{t};
}
Mac Newbold's avatar
Mac Newbold committed
122 123 124
if ($opt{d}) {
    $debug = 1;
}
Robert Ricci's avatar
Robert Ricci committed
125

126
# Grab some constants into variables
127
my $TBANYMODE    = TBDB_NODEOPMODE_ANY;
128
my $TBPXERESET   = TBDB_TBCONTROL_PXERESET;
129 130
my $TBRESET      = TBDB_TBCONTROL_RESET;
my $TBTIMEOUT    = TBDB_TBCONTROL_TIMEOUT;
Mac Newbold's avatar
Mac Newbold committed
131 132 133
my $PXEBOOT      = TBDB_TBCONTROL_PXEBOOT;
my $BOOTING      = TBDB_TBCONTROL_BOOTING;
my $CHECKGENISUP = TBDB_TBCONTROL_CHECKGENISUP;
134 135 136
my $TBNOTIMEOUT  = TBDB_NO_STATE_TIMEOUT;
my $TBNODESTATE  = TBDB_TBEVENT_NODESTATE;
my $TBNODEOPMODE = TBDB_TBEVENT_NODEOPMODE;
137 138 139 140 141 142
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;
143
my $TBISUP       = TBDB_NODESTATE_ISUP;
144 145 146
my $PXEWAIT      = TBDB_NODESTATE_PXEWAIT;
my $PXEWAKEUP    = TBDB_NODESTATE_PXEWAKEUP;
my $PXEBOOTING   = TBDB_NODESTATE_PXEBOOTING;
147 148
my $TBRELOADDONEV1    = TBDB_TBCONTROL_RELOADDONE;
my $TBRELOADDONEV2    = TBDB_TBCONTROL_RELOADDONE_V2;
149 150 151 152
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;
153

154 155
# Special PXEBOOT state machine that all local nodes use.
my $PXEKERNEL	 = "PXEKERNEL";
156

157 158
# Even special-er SECUREBOOT state machine that local nodes may use.
my $SECUREBOOT	 = "SECUREBOOT";
159

160 161 162 163
# Protos.
sub debug(@);
sub fatal($);
sub notify($;$);
164
sub info($);
165 166 167 168 169 170 171
sub getTimeouts();
sub getValid();
sub getModeTrans();
sub getTriggers();
sub readStates(;@);
sub handleCtrlEvent($$);
sub reload();
172 173
sub StartEvents();
sub PollEvents($$);
174
    
175 176 177
if (!$debug) {
    if ( $TB eq $REALTB ) {
	$pidfile = "/var/run/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
178
    } else {
179
	$pidfile = "$TB/locks/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
180
    }
181
    debug("Using pidfile $pidfile\n");
Mac Newbold's avatar
Mac Newbold committed
182

183 184 185 186 187 188 189 190 191 192 193 194 195 196 197
    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: $? $!\n");
    }
    # Background
198
    if (TBBackGround($LOGFILE)) {
Mac Newbold's avatar
Mac Newbold committed
199 200
	exit(0);
    }
201
    TBdbfork();
Robert Ricci's avatar
Robert Ricci committed
202
}
203 204 205 206 207 208 209 210 211

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
212

213 214 215
# Change my $0 so that it is easier to see in a ps/top
$0 = "$0";

216 217 218 219 220
#
# Start up the event system interface.
#
if (StartEvents() != 0) {
    fatal("Error starting events");
Robert Ricci's avatar
Robert Ricci committed
221
}
222 223
# We want to exit on any warning. 
$SIG{__WARN__} = sub { print STDERR $_[0]; exit(-1); };
Robert Ricci's avatar
Robert Ricci committed
224 225 226

# Read in the pre-existing node states, and timeout and valid transition
# information from the database
227 228 229 230
my %timeouts  = getTimeouts();
my %valid     = getValid();
my %modeTrans = getModeTrans();
my %triggers  = getTriggers();
231
my %nodes     = readStates();
232
my %timeouttag= ();
233
if ($debug) { qshow(); }
Robert Ricci's avatar
Robert Ricci committed
234 235 236

# Gets set if a reload of state from the database should happen.
my $do_reload = 0;
237
my $do_reopen = 0;
238 239
my $sigrestart= 0;
my $sigcleanup= 0;
240
my $exiting   = 0;
Robert Ricci's avatar
Robert Ricci committed
241 242 243 244

# 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.
245
$SIG{HUP}  = sub { info("SIGHUP - Reloading DB state\n"); $do_reload = 1; };
246
$SIG{USR2} = sub { info("SIGUSR2 - Reopening logfile\n"); $do_reopen = 1; };
247

Mac Newbold's avatar
Mac Newbold committed
248
# Set up other signals.
249 250 251 252 253
$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
254

255 256 257
# Track if I handled an event or not
my $event_count = 0;

258 259
# Control how long I block while waiting for events
my $blockwait=0;
260
my $nextdeadline=0;
261 262
my $mailqueue=0;

263 264 265 266 267
notify("Stated starting up\n");

sub process_event_queue() {
    $event_count=0;
    my $lastcount=-1;
268 269
    my $wait;
    my $now = time();
Mike Hibler's avatar
Mike Hibler committed
270
    debug("Polling at $now - mq=$mailqueue bw=$blockwait ndl=$nextdeadline\n");
271 272 273 274 275 276 277 278 279 280
    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;
281 282
	    } else {
		$wait = 0;
283 284 285 286 287
	    }
	}
    } else {
	# mail is waiting. Only block until it is time to send it.
	$wait = $lastmail + $mailgap - $now;
Mac Newbold's avatar
Mac Newbold committed
288
	debug("Now $now, mailgap $mailgap, last $lastmail ==> wait $wait\n");
289 290 291
    }
    if ($wait < 0) { debug("Wait was $wait!\n"); $wait=0; }
    my $finish = $now + $wait;
Mac Newbold's avatar
Mac Newbold committed
292
    while (($event_count != $lastcount || $wait > 0) &&
293
	   !($sigrestart || $sigcleanup || $do_reload || $do_reopen)) {
294
	$lastcount = $event_count;
Mac Newbold's avatar
Mac Newbold committed
295
	# Don't block if we got a signal!
296
	if ($wait<=0 || $sigrestart || $sigcleanup || $do_reload || $do_reload){
297
	    PollEvents(0, 0);
298
	} else {
299
	    #debug("Using blocking event poll - $wait seconds\n");
300
	    # timeout param is in milliseconds, so multiply
301
	    PollEvents(1, $wait);
302 303 304
	    $now = time();
	    # subtract seconds elapsed from my wait time
	    $wait = $finish - $now;
305
	    #debug("Finished blocking event poll - $wait seconds remain\n");
306
	    if ($event_count > 0 &&
Mac Newbold's avatar
Mac Newbold committed
307
		(qsize() > 0 || $mailqueue ||
308
		 $sigrestart || $sigcleanup || $do_reload || $do_reopen)) {
309 310 311 312 313 314 315
		$blockwait=0;
		$wait=0;
		#debug("Cancelling wait - timeouts/msgs waiting, or HUP'd\n");
		#debug("---End Blocking Wait ---\n");
	    }
	}
	#debug("Wait is $wait\n");
316 317
    }
    if ($event_count > 0) {
Mac Newbold's avatar
Mac Newbold committed
318
	debug("Handled $event_count event(s).\n");
319 320
    }
}
Robert Ricci's avatar
Robert Ricci committed
321

322
# Now, we just poll for events, and watch for timeouts
Robert Ricci's avatar
Robert Ricci committed
323
while (1) {
Mac Newbold's avatar
Mac Newbold committed
324
    my $now = time();
325 326 327 328
    my ($deadline,$node);

    # Check for nodes that have passed their timeout
    if (!qhead($deadline,$node)) {
329
	info("HEAD: $node in ".($deadline-$now).", queue=".qsize()."\n");
330 331
	while ($now >= $deadline && $node ne "") {
	    qpop($deadline,$node);
332
	    info("POP: $node in ".($deadline-$now).", queue=".qsize()."\n");
333 334 335 336 337 338 339 340 341 342 343 344 345
	    #
	    # If the node is no longer in the nodes array, it was most
	    # likely a dynamic virtual node which is now gone. Ignore.
	    # Need to look at reload() to see if we can catch this earlier. 
	    #
	    if (exists($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{$node}));
	    }
346 347 348 349
	    if (0) { qshow(); }
	    if (qhead($deadline,$node)) {
		$deadline=0; $node="";
	    }
350
	}
351 352
    } else {
	$deadline=0;
353
    }
354
    $nextdeadline = $deadline;
355

Mac Newbold's avatar
Mac Newbold committed
356 357 358 359
    if ($do_reload || ($now - $last_reload > $reload_time)) {
	reload();
	$do_reload = 0;
    }
Mac Newbold's avatar
Mac Newbold committed
360

Mike Hibler's avatar
Mike Hibler committed
361 362 363 364 365
    if (qsize()==0) {
	$blockwait=1;
	debug("---Blocking wait okay---\n");
    }

Mac Newbold's avatar
Mac Newbold committed
366 367
    # Send any messages in the queue if it is time
    notify("",1);
Mac Newbold's avatar
Mac Newbold committed
368

369 370
    if ($sigrestart) { restart(); }
    if ($sigcleanup) { cleanup(); }
371 372 373 374 375 376 377
    if ($do_reopen)  {
	ReOpenLog($LOGFILE);
	if (! kill('USR2', $eventchild)) {
	    fatal("Could not signal(USR2) event reader child\n");
	}
	$do_reopen = 0;
    }
378
    process_event_queue;
Robert Ricci's avatar
Robert Ricci committed
379 380
}

Mac Newbold's avatar
Mac Newbold committed
381 382
exit(0);

Robert Ricci's avatar
Robert Ricci committed
383
# Read the current states of nodes from the database
384
sub readStates(;@) {
Mac Newbold's avatar
Mac Newbold committed
385
    my %oldnodes = @_;
Mike Hibler's avatar
Mike Hibler committed
386
    my $now = time();
Mac Newbold's avatar
Mac Newbold committed
387 388

    # Guard against undefined variable warnings
389
    if (! %oldnodes) {
Mac Newbold's avatar
Mac Newbold committed
390 391
	%oldnodes = ();
    }
392

Mac Newbold's avatar
Mac Newbold committed
393 394 395
    #debug("readStates called\n");
    my $result = DBQueryFatal("SELECT node_id, eventstate, " .
			      "state_timestamp, op_mode, " .
396
			      "op_mode_timestamp, stated_tag FROM nodes ".
397
			      "where node_id not like 'sh%'");
Mac Newbold's avatar
Mac Newbold committed
398 399

    my %nodes;
400
    while (my ($node_id, $state, $timestamp, $mode, $mode_timestamp, $tag)
Mac Newbold's avatar
Mac Newbold committed
401
	   = $result->fetchrow()) {
402 403 404 405 406 407 408 409 410 411 412
	$nodes{$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
413

414
	#
Mac Newbold's avatar
Mac Newbold committed
415 416 417 418
	# 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.)
419
	#
Mac Newbold's avatar
Mac Newbold committed
420 421 422 423 424
	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};
425
	} else {
Mac Newbold's avatar
Mac Newbold committed
426 427 428 429
	    $nodes{$node_id}{state}          = $state;
	    $nodes{$node_id}{timestamp}      = $timestamp;
	    $nodes{$node_id}{mode}           = $mode;
	    $nodes{$node_id}{mode_timestamp} = $mode_timestamp;
430 431 432
	    $nodes{$node_id}{notified}       = 0;
	    $nodes{$node_id}{timedout}       = 0;
	    $nodes{$node_id}{noretry}        = 0;
433
	    # Is there a timeout? If so, set it up!
Mike Hibler's avatar
Mike Hibler committed
434
	    if (defined($timestamp) && $dbtag eq $nodes{$node_id}{"tag"}) {
435
		setTimeout($mode,$state,$node_id,$timestamp);
436
	    }
437
	}
Mac Newbold's avatar
Mac Newbold committed
438 439
    }
    return %nodes;
Robert Ricci's avatar
Robert Ricci committed
440 441 442 443 444 445
}

#
# Read timeouts for various states from the database
#
sub getTimeouts() {
Mac Newbold's avatar
Mac Newbold committed
446 447 448
    #debug("getTimeouts called\n");
    my $result = DBQueryFatal("SELECT op_mode, state, timeout, action " .
			      "FROM state_timeouts");
Robert Ricci's avatar
Robert Ricci committed
449

Mac Newbold's avatar
Mac Newbold committed
450 451 452 453 454
    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
455 456 457 458 459 460
}

#
# Read the list of valid state transitions from the database
#
sub getValid() {
Mac Newbold's avatar
Mac Newbold committed
461 462 463
    #debug("getValid called\n");
    my $result = DBQueryFatal("SELECT op_mode, state1, state2 " .
			      "FROM state_transitions");
Robert Ricci's avatar
Robert Ricci committed
464

Mac Newbold's avatar
Mac Newbold committed
465 466 467 468 469
    my %valid;
    while (my ($mode,$state1, $state2) = $result->fetchrow()) {
	$valid{$mode}{$state1}{$state2} = 1;
    }
    return %valid;
Robert Ricci's avatar
Robert Ricci committed
470 471
}

472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489
#
# 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;
}

490 491 492 493
#
# Read the list of valid mode transitions from the database
#
sub getModeTrans() {
Mac Newbold's avatar
Mac Newbold committed
494
    #debug("getModeTrans called\n");
Mac Newbold's avatar
Mac Newbold committed
495
    my $result =
Mac Newbold's avatar
Mac Newbold committed
496 497 498 499 500 501 502 503 504 505 506
      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;
507
	}
Mac Newbold's avatar
Mac Newbold committed
508 509
    }
    return %modeTrans;
510 511 512 513 514 515
}

#
# Read the list of states which trigger an action
#
sub getTriggers() {
516
    debug("getTriggers called\n");
Mac Newbold's avatar
Mac Newbold committed
517

518 519
    debug("anymode ==> '$TBANYMODE'\n");

Mac Newbold's avatar
Mac Newbold committed
520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540
    # 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 $result =
541
      DBQueryFatal("SELECT op_mode, state, `trigger` " .
542 543
		   "FROM state_triggers where node_id='$TBANYMODE' ".
		   "order by op_mode,state");
Mac Newbold's avatar
Mac Newbold committed
544
    my %t;
545
    while (my ($mode, $state, $trig) = $result->fetchrow()) {
Mac Newbold's avatar
Mac Newbold committed
546 547 548
	my @trigs = split(/\s*,\s*/,$trig);
	$t{"$mode:$state"} = \@trigs;
	debug("trig($mode:$state)\t => ".join(',',@trigs)."\n");
549 550
    }

Mac Newbold's avatar
Mac Newbold committed
551 552
    # Grab per-node triggers (including any-mode)
    $result =
553
      DBQueryFatal("SELECT node_id, op_mode, state, `trigger` " .
554 555 556 557 558 559
		   "FROM state_triggers where node_id!='$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
560
    }
561

Mac Newbold's avatar
Mac Newbold committed
562 563
    debug(hash_recurse(%t));

Mac Newbold's avatar
Mac Newbold committed
564
    return %t;
565 566
}

Robert Ricci's avatar
Robert Ricci committed
567 568 569 570
#
# Gets called for every event that we recieve
#
sub handleEvent($$$) {
571
    my ($objtype,$objname,$eventtype) = @_;
Mac Newbold's avatar
Mac Newbold committed
572
    $event_count++;
Mac Newbold's avatar
Mac Newbold committed
573

574
    #
Mac Newbold's avatar
Mac Newbold committed
575 576 577
    # 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.
578
    #
579
    if ($dbtag eq "") {
580 581
	debug("Got an event: ($objtype,$objname,$eventtype)\n");
    }
Mac Newbold's avatar
Mac Newbold committed
582 583 584 585

    #
    # Check to see if another instance is supposed to be handling this node
    #
586 587
    if ($objtype ne $TBCOMMAND) {
	my $node = $objname;
Mac Newbold's avatar
Mac Newbold committed
588

589 590 591 592 593 594 595
	#
	# 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
596
	    # churning via reload(). Bad.
597
	    if (! defined($nodes{$node})) {
598 599 600 601
		notify("Got $objtype/$eventtype for nonexistent $node!\n");
		return;
	    }
	}
Mac Newbold's avatar
Mac Newbold committed
602

603 604
	#
	# If a stated_tag was specified on the command line, ignore those
Mac Newbold's avatar
Mac Newbold committed
605
	# nodes that do not match.
606
	#
Mac Newbold's avatar
Mac Newbold committed
607
	#debug("dbtag='$dbtag', node $node='".$nodes{$node}{"tag"}."'\n");
608 609 610 611 612 613 614 615 616 617
	if ($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
618
    }
Mac Newbold's avatar
Mac Newbold committed
619 620 621 622 623

    #
    # 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.)
    #
624 625 626
    if ($dbtag ne "") {
	debug("Got an event: ($objtype,$objname,$eventtype)\n");
    }
Mac Newbold's avatar
Mac Newbold committed
627 628 629

 SWITCH: for ($objtype) {

Mac Newbold's avatar
Mac Newbold committed
630 631 632 633 634 635 636 637 638 639 640 641 642 643
	(/$TBNODESTATE/) && do {
	    stateTransition($objname,$eventtype);
	    last;
	};
	(/$TBNODEOPMODE/) && do {
	    opModeTransition($objname,$eventtype);
	    notify("Use of deprecated event TBNODEOPMODE:\n".
		   "$objname->$eventtype\n");
	    last;
	};
	(/$TBCONTROL/) && do {
	    handleCtrlEvent($objname,$eventtype);
	    last;
	};
644 645 646 647
	(/$TBCOMMAND/) && do {
	    handleCommand($objname,$eventtype);
	    last;
	};
648

Mac Newbold's avatar
Mac Newbold committed
649
    }
650 651 652 653 654

}

sub stateTransition($$) {

655
    my ($node,$newstate) = @_;
Robert Ricci's avatar
Robert Ricci committed
656

657 658
    # Check for invalid transitions
    my ($oldstate, $mode);
659 660 661
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

662
    if ($oldstate && $mode && !IsValidTransition($mode,$oldstate,$newstate)) {
663
	notify("Invalid transition for node $node from $mode/$oldstate " .
664
	       "to $newstate\n");
665
	
666 667 668 669 670 671
	#
	# 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) {
672 673 674
            $newstate = TBDB_NODESTATE_SECVIOLATION;
            notify("Moving $node to $newstate because it's in $mode\n");
        }
675 676 677 678 679 680
    }
    #
    # Nodes that are in the SECVIOLATION state are not allowed to leave!
    #
    if ($oldstate eq TBDB_NODESTATE_SECVIOLATION &&
	$newstate ne TBDB_NODESTATE_SECVIOLATION) {
681 682 683 684 685 686 687
	#
	# Allow transitions to SHUTDOWN.
	# This allows someone to reboot a node in the SECVIOLATION state
	# getting it back to MINIMAL/SHUTDOWN.
	#
	# XXX DEBUG ONLY!
	# 
688 689 690 691
	if ($soft_secviolation &&
	    ($newstate eq TBDB_NODESTATE_SHUTDOWN ||
	     $newstate eq TBDB_NODESTATE_GPXEBOOTING)) {
	    notify("$node allowed to transition: SECVIOLATION => $newstate\n");
692 693 694 695
	} else {
	    notify("$node tried to leave SECVIOLATION (to $newstate)\n");
	    $newstate = TBDB_NODESTATE_SECVIOLATION;
	}
696
    }
Robert Ricci's avatar
Robert Ricci committed
697

698 699 700 701
    my $now = time();
    $nodes{$node}{state}     = $newstate;
    $nodes{$node}{timestamp} = $now;
    $nodes{$node}{notified}  = 0;
702

703 704 705
    info("$node: $mode/$oldstate => $mode/$newstate\n");
    DBQueryFatal("UPDATE nodes SET eventstate='$newstate', " .
		 "state_timestamp='$now' WHERE node_id='$node'");
706

707 708 709 710
    # 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} =~ /^$TBCOMMAND:/) {
711
        debug("TimeoutTag = '$timeout_tag{$node}'\n");
712
	my ($str,$cmd) = split(":",$timeout_tag{$node});
713
	debug("str=$str\tcmd=$cmd\tTBREBOOT=$TBREBOOT\tstate=$newstate\n");
714
	if ($cmd eq $TBREBOOT) {
715
	    if ($newstate eq TBDB_NODESTATE_SHUTDOWN ) {
716 717 718
		info("$node: $TBREBOOT success\n");
		# Timeout will get cleared below by setTimeout call
	    } else {
719 720
		notify("$node: $TBREBOOT in progress, but got state ".
		       "$newstate instead of ".TBDB_NODESTATE_SHUTDOWN."!\n");
721 722 723 724 725
	    }
	#} elsif ($cmd eq $FOO ) {
	    # Add more here...
	} else {
	    notify("$node: Unknown command timeout '$timeout_tag{$node}' ".
726
		   "found at $mode/$newstate\n");
727 728 729
	}
    }

730 731 732 733 734
    #
    # 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.
    #
735 736
    setTimeout($mode,$newstate,$node,$now);

Mac Newbold's avatar
Mac Newbold committed
737 738 739 740 741 742 743
    # 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 = $_;
744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760
	    /^SCRIPT:([-\w\/]+)$/ && 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
761
	    /^$PXEBOOT$/ && do {
762
		#
Mac Newbold's avatar
Mac Newbold committed
763 764 765 766 767 768 769 770 771 772 773 774
		# 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.
775
		#
Mac Newbold's avatar
Mac Newbold committed
776 777 778 779
		# 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.
780
		#
Mac Newbold's avatar
Mac Newbold committed
781 782 783 784 785
		debug("Running $PXEBOOT trigger\n");
		if ($mode ne $PXEKERNEL) {
		    info("$node: Forcing mode transition into $PXEKERNEL!\n");
		    opModeTransition($node, $PXEKERNEL, 1);
		    $mode=$PXEKERNEL;
786
		}
Mac Newbold's avatar
Mac Newbold committed
787 788
		next;
	    };
789 790 791 792 793 794
	    /^$SECUREBOOT$/ && 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.
795
		#
796 797 798 799
		# 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.
800
		#
801 802 803 804 805
		debug("Running $SECUREBOOT trigger\n");
		my $query_result =
		    DBQueryWarn("select next_op_mode from nodes ".
				"where node_id='$node'");
		my ($nextmode) = $query_result->fetchrow();
806 807 808 809 810
		if (!$nextmode) {
		    $nextmode = $mode;
		}
		if ($nextmode ne TBDB_NODEOPMODE_SECURELOAD ||
		    $oldstate ne TBDB_NODESTATE_SHUTDOWN) {
811 812 813 814 815 816
		    $nextmode = TBDB_NODEOPMODE_SECUREBOOT;
		}
		if ($mode ne $nextmode) {
		    info("$node: Forcing mode transition to $nextmode!\n");
		    opModeTransition($node, $nextmode, 1);
		    $mode=$nextmode;
817 818 819
		}
		next;
	    };
Mac Newbold's avatar
Mac Newbold committed
820 821 822 823 824
	    /^$BOOTING$/ && do {
		#
		# See if we are in the right mode/osid.
		#
		my ($bootosid,$bootopmode) = TBBootWhat($node, $debug);
825

Mac Newbold's avatar
Mac Newbold committed
826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891
		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($node);
		next;
	    };
892 893 894 895
	    /^CHECKPORTREG$/ && do {
		CheckPortRegistration($node);
		next;
	    };
896 897 898 899 900 901
	    /^$TBPXERESET$/ && do {
		# We successfully booted, so reset one-shot PXEboot if any
		# Check if we really need to do a reset
		handleCtrlEvent($node,$trig);
		next;
	    };
902
	    /^$TBRESET$/ && do {
903 904 905
		# We successfully booted, so clear some flags
		$nodes{$node}{noretry}   = 0;
		$nodes{$node}{timedout}  = 0;
906 907 908 909
		# 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) = $r->fetchrow();
910 911 912 913 914 915 916
		if (! (defined($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 $defosid) {
917 918 919 920
		    handleCtrlEvent($node,$trig);
		}
		next;
	    };
921
	    (/^$TBRELOADDONEV1$/ || /^$TBRELOADDONEV2$/) && do {
922 923 924
		handleCtrlEvent($node,$trig);
		next;
	    };
925 926 927 928 929 930 931 932
	    /^$TBISUP$/ && do {
		info("$node: Triggered $TBISUP\n");
		EventSendWarn(host      => $BOSSNODE ,
			      objtype   => TBDB_TBEVENT_NODESTATE ,
			      eventtype => TBDB_NODESTATE_ISUP ,
			      objname   => $node);
		next;
	    };
933
	    (/^$TBREBOOT$/ || /^$TBPOWERCYCLE$/ || /^$TBPOWEROFF$/) && do {
934
		handleCommand($node,$trig);
935 936 937
		next;
	    };
            (/^EMAILNOTIFY$/) && do {
938 939
		my $msg = "$node entered state $mode/$newstate from " .
		    "$mode/$oldstate";
940
		my $dest = $REALTBOPS;
941
		if ($newstate eq TBDB_NODESTATE_SECVIOLATION) {
942 943 944 945 946 947 948 949
		    if ($soft_secviolation) {
			$msg .= "\n\nNode $node was allowed to continue.\n";
			$dest = $TBOPS;
		    } else {
			$msg .= "\n\nNode $node has been powered off.\n" .
			        "You must address the cause of the violation ".
				"and reset the eventstate before powering on.";
		    }
950
		}
951 952
		SENDMAIL($dest, "STATED: $node entered state $newstate",
			 $msg, "Stated Daemon <".$TBOPS.">");
953
		next;
954
            };
955 956 957 958 959 960 961 962 963 964 965 966
	    /^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;
	    };
967
	    notify("Unknown trigger '$trig' for $node in $mode/$newstate!\n");
968
	}
Mac Newbold's avatar
Mac Newbold committed
969 970 971 972 973 974 975
	# 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);
	}
976
    }
977

978 979 980 981 982 983 984 985
    # 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
986 987 988
	} else {
	    debug("No next mode.\n");
	}
989 990
    }
}
991

992
sub opModeTransition($$;$) {
Mac Newbold's avatar
Mac Newbold committed
993

994 995
    my ($node,$newmode,$force) = @_;
    if (!defined($force)) { $force = 0; }
Mac Newbold's avatar
Mac Newbold committed
996

997
    info("$node: Mode change to $newmode requested ($force)\n");
Mac Newbold's avatar
Mac Newbold committed
998

999 1000
    # Check for invalid transitions
    my ($oldstate, $mode, $nextstate);
1001 1002 1003
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

1004
    if (defined($modeTrans{"$mode:$oldstate"}) || $force) {
1005 1006 1007 1008 1009
	#
	# If there is a next state associated with mode:oldstate, use it.
	# Otherwise we just remain in the old state.
	#
	if (defined($modeTrans{"$mode:$oldstate"})) {
1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020
	    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{$newmode};
1021 1022 1023 1024
		if ($force) {
		    info("$node: opMode force changing state along with mode\n");
		}
	    } elsif (!$force) {
1025 1026 1027
		notify("Invalid mode transition for $node from ".
		       "$mode/$oldstate to $newmode!\n");
	    }
1028 1029
	}
    } else {
1030
	notify("Invalid mode transition for $node from $mode/$oldstate: ".
1031 1032
	       "Not a valid mode transition state!\n");
    }
Mac Newbold's avatar
Mac Newbold committed
1033 1034 1035
    if (!$nextstate) {
	$nextstate=$oldstate;
    }
Mac Newbold's avatar
Mac Newbold committed
1036

1037 1038 1039 1040 1041 1042
    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
1043

1044 1045 1046 1047 1048
    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
1049 1050 1051 1052

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

1053 1054 1055 1056
}

sub handleCtrlEvent($$) {
    my ($node,$event) = @_;
1057

1058
    info("CtrlEvent: $node, $event\n");
1059

1060
    foreach ($event) {
1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075
	/^$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 $node\n");
	    next;
	};
1076
	/^$TBRESET$/ && do {
1077 1078
	    #
	    # Clear next_boot_path with os_select.
Mac Newbold's avatar
Mac Newbold committed
1079
	    #
1080
	    $cmd = "$osselect -d -c -1 $node";
1081
	    system($cmd) and
1082
		notify("$node/$event: Could not clear next_boot_path!\n");
Mac Newbold's avatar
Mac Newbold committed
1083 1084

	    info("Performed $TBRESET for $node\n");
1085 1086
	    next;
	};
1087
	(/^$TBRELOADDONEV1$/ || /^$TBRELOADDONEV2$/) && do {
1088 1089 1090 1091 1092 1093 1094
	    info("Clearing reload info for $node\n");
	    DBQueryFatal("delete from current_reloads where node_id='$node'");
	    my ($pid,$eid);
	    NodeidToExp($node,\$pid,\$eid);
	    if (($pid eq NODERELOADING_PID) && ($eid eq NODERELOADING_EID)) {
		DBQueryFatal("delete from scheduled_reloads ".
			     "where node_id='$node'");
1095
		DBQueryFatal("delete from reserved where node_id='$node'");
1096 1097
		TBSetNodeHistory($node, TB_NODEHISTORY_OP_FREE,
				 $UID, $pid, $eid);
1098
		info("Released $node from $pid/$eid\n");
1099
	    }
1100 1101 1102 1103 1104
	    if ($event eq $TBRELOADDONEV2) {
		info("Sending an apod to $node\n");
		system("$apod $node") == 0 or
		    notify("Could not apod $node after $TBRELOADDONEV2!\n");
	    }
1105 1106 1107
	    next;
	};
	/^$TBTIMEOUT$/ && do {
1108 1109 1110 1111
	    my ($mode,$state) = split(":",$timeout_tag{$node});
	    delete($timeout_tag{$node});
	    my $curstate = $nodes{$node}{state};
	    my $curmode = $nodes{$node}{mode};
1112
	    my ($timeout,$action);
1113 1114 1115 1116 1117
	    if (!defined($nodes{$node}{notified})) {
		$nodes{$node}{notified}=0;
	    }
	    $nodes{$node}{notified}++;
	    my $notified = $nodes{$node}{notified};
1118 1119
	    $nodes{$node}{timedout}++;
	    my $timedout = $nodes{$node}{timedout};
1120 1121 1122 1123
	    if ($mode && $state && $timeouts{$mode} &&
		$timeouts{$mode}{$state}) {
		($timeout, $action) = @{$timeouts{$mode}{$state}};
	    }
1124 1125
	    if ($mode eq $TBCOMMAND) {
		# It is a command, not a true state
1126
		if ($action eq $TBTIMEOUTCMDRETRY) {
1127
		    # Retry the command
1128
		    notify("$node: Command $state, retry #$timedout\n");
1129
		    # notify in case we get in a retry loop...
1130
		    handleCommand($node,$state,$timedout,1);
1131 1132 1133 1134 1135 1136
		} else {
		    notify("$node: Unknown timeout action for ".
			   "$mode/$state: '$action'\n");
		}
		next;
	    }
1137 1138 1139 1140 1141 1142 1143

	    #
	    # 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
1144

1145 1146
		if ($timedout < 3) {
		    #
Mac Newbold's avatar
Mac Newbold committed
1147 1148
		    # Try again.
		    #
1149 1150 1151 1152 1153 1154
		    info("Node $node has timed out $timedout times in ".
			 "$PXEWAKEUP!\n".
			 "Sending it a another wakeup command\n");
		}
		else {
		    #
Mac Newbold's avatar
Mac Newbold committed
1155 1156
		    # Failed too many times, power cycle instead.
		    #
1157 1158 1159 1160 1161 1162 1163
		    notify("Node $node has timed out $timedout times in ".
			   "$PXEWAKEUP!\n".
			   "Sending it a reboot command\n");
		    $optarg .= "-k";
		}
		my $cmd = "$nodereboot -r $optarg $node";
		debug("$cmd\n");
1164
		system("(date; $cmd) >>$rebootlog 2>&1 &") and
1165
		    notify("$PXEWAKEUP retry: ".