stated.in 57.2 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 $osselect = "$TB/bin/os_select";
30
my $nodereboot = "$TB/bin/node_reboot";
31
my $rebootlog  = "$TB/log/nodereboot.log";
32
my $power = "$TB/bin/power";
33
my $apod = "$TB/sbin/apod";
34
my $TBLOG = "@TBLOGFACIL@";
35
my $LOGFILE = "$TB/log/stated.log";
Robert Ricci's avatar
Robert Ricci committed
36 37 38 39 40 41

$| = 1;

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

52 53 54 55 56
# 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
57
# Number of iterations (roughly, seconds) after which we'll reload
Robert Ricci's avatar
Robert Ricci committed
58 59
# information from the database. This is so we don't end up with information
# that's _too_ out of sync.
60
my $reload_time = 600;
61
my $last_reload = time;
Robert Ricci's avatar
Robert Ricci committed
62

Mike Hibler's avatar
Mike Hibler committed
63 64 65
# 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.
66
my $maxpasttimeout = -(7 * 24 * 60 * 60);	# 1 week
Mike Hibler's avatar
Mike Hibler committed
67

68 69 70
# Handling of SECVIOLATIONS, eventually controlled by a sitevar
my $soft_secviolation = 1;

71 72 73
# Command line opts.
my $dbtag = "";
my $debug = 0;
74 75
my $server = "localhost";
my $port   = @BOSSEVENTPORT@;
76 77
my $lockfile;
my $pidfile;
78
my $eventchild;
79

Robert Ricci's avatar
Robert Ricci committed
80 81 82
# Process command-line arguments

sub usage {
Mac Newbold's avatar
Mac Newbold committed
83
    print << "END";
84
Usage: $0 [-h] [-d] [-s server] [-p port] [-t dbtag]
85
-h              This message
86 87
-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
88 89
-s server       Use specified server, instead of this site's bossnode
-p port	        Use specified port
90
Send SIGHUP to reload database state, or SIGUSR1 to restart completely.
Robert Ricci's avatar
Robert Ricci committed
91
END
92
    exit(1);
Robert Ricci's avatar
Robert Ricci committed
93 94
}

Mac Newbold's avatar
Mac Newbold committed
95
# Only root should run this - it won't work when run as a user...
96
# (Or, let an admin run it if it isn't the real one in /usr/testbed/ )
97
if ($UID && ( $TB eq $REALTB || ! TBAdmin($UID) ) ) {
Mac Newbold's avatar
Mac Newbold committed
98 99 100
    die("Only root can run this script!\n");
}

101
my @args = @ARGV;    # save a copy for restart before we mess with them.
Robert Ricci's avatar
Robert Ricci committed
102
my %opt = ();
103
if (!getopts("ds:p:ht:l",\%opt)) { usage(); }
Robert Ricci's avatar
Robert Ricci committed
104

Mac Newbold's avatar
Mac Newbold committed
105 106 107 108 109 110
if ($opt{h}) {
    exit &usage;
}
if (@ARGV) {
    exit &usage;
}
Robert Ricci's avatar
Robert Ricci committed
111

Mac Newbold's avatar
Mac Newbold committed
112 113 114 115 116 117
if ($opt{s}) {
    $server = $opt{s};
}
if ($opt{p}) {
    $port = $opt{p};
}
118 119 120
if ($opt{t}) {
    $dbtag = $opt{t};
}
Mac Newbold's avatar
Mac Newbold committed
121 122 123
if ($opt{d}) {
    $debug = 1;
}
Robert Ricci's avatar
Robert Ricci committed
124

125
# Grab some constants into variables
126
my $TBANYMODE    = TBDB_NODEOPMODE_ANY;
127 128
my $TBRESET      = TBDB_TBCONTROL_RESET;
my $TBTIMEOUT    = TBDB_TBCONTROL_TIMEOUT;
Mac Newbold's avatar
Mac Newbold committed
129 130 131
my $PXEBOOT      = TBDB_TBCONTROL_PXEBOOT;
my $BOOTING      = TBDB_TBCONTROL_BOOTING;
my $CHECKGENISUP = TBDB_TBCONTROL_CHECKGENISUP;
132 133 134
my $TBNOTIMEOUT  = TBDB_NO_STATE_TIMEOUT;
my $TBNODESTATE  = TBDB_TBEVENT_NODESTATE;
my $TBNODEOPMODE = TBDB_TBEVENT_NODEOPMODE;
135 136 137 138 139 140
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;
141
my $TBISUP       = TBDB_NODESTATE_ISUP;
142 143 144
my $PXEWAIT      = TBDB_NODESTATE_PXEWAIT;
my $PXEWAKEUP    = TBDB_NODESTATE_PXEWAKEUP;
my $PXEBOOTING   = TBDB_NODESTATE_PXEBOOTING;
145 146
my $TBRELOADDONEV1    = TBDB_TBCONTROL_RELOADDONE;
my $TBRELOADDONEV2    = TBDB_TBCONTROL_RELOADDONE_V2;
147 148 149 150
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;
151

152 153
# Special PXEBOOT state machine that all local nodes use.
my $PXEKERNEL	 = "PXEKERNEL";
154

155 156
# Even special-er SECUREBOOT state machine that local nodes may use.
my $SECUREBOOT	 = "SECUREBOOT";
157

158 159 160 161
# Protos.
sub debug(@);
sub fatal($);
sub notify($;$);
162
sub info($);
163 164 165 166 167 168 169
sub getTimeouts();
sub getValid();
sub getModeTrans();
sub getTriggers();
sub readStates(;@);
sub handleCtrlEvent($$);
sub reload();
170 171
sub StartEvents();
sub PollEvents($$);
172
    
173 174 175
if (!$debug) {
    if ( $TB eq $REALTB ) {
	$pidfile = "/var/run/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
176
    } else {
177
	$pidfile = "$TB/locks/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
178
    }
179
    debug("Using pidfile $pidfile\n");
Mac Newbold's avatar
Mac Newbold committed
180

181 182 183 184 185 186 187 188 189 190 191 192 193 194 195
    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: $? $!\n");
    }
    # Background
196
    if (TBBackGround($LOGFILE)) {
Mac Newbold's avatar
Mac Newbold committed
197 198
	exit(0);
    }
199
    TBdbfork();
Robert Ricci's avatar
Robert Ricci committed
200
}
201 202 203 204 205 206 207 208 209

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
210

211 212 213
# Change my $0 so that it is easier to see in a ps/top
$0 = "$0";

214 215 216 217 218
#
# Start up the event system interface.
#
if (StartEvents() != 0) {
    fatal("Error starting events");
Robert Ricci's avatar
Robert Ricci committed
219
}
220 221
# We want to exit on any warning. 
$SIG{__WARN__} = sub { print STDERR $_[0]; exit(-1); };
Robert Ricci's avatar
Robert Ricci committed
222 223 224

# Read in the pre-existing node states, and timeout and valid transition
# information from the database
225 226 227 228
my %timeouts  = getTimeouts();
my %valid     = getValid();
my %modeTrans = getModeTrans();
my %triggers  = getTriggers();
229
my %nodes     = readStates();
230
my %timeouttag= ();
231
if ($debug) { qshow(); }
Robert Ricci's avatar
Robert Ricci committed
232 233 234

# Gets set if a reload of state from the database should happen.
my $do_reload = 0;
235
my $do_reopen = 0;
236 237
my $sigrestart= 0;
my $sigcleanup= 0;
238
my $exiting   = 0;
Robert Ricci's avatar
Robert Ricci committed
239 240 241 242

# 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.
243
$SIG{HUP}  = sub { info("SIGHUP - Reloading DB state\n"); $do_reload = 1; };
244
$SIG{USR2} = sub { info("SIGUSR2 - Reopening logfile\n"); $do_reopen = 1; };
245

Mac Newbold's avatar
Mac Newbold committed
246
# Set up other signals.
247 248 249 250 251
$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
252

253 254 255
# Track if I handled an event or not
my $event_count = 0;

256 257
# Control how long I block while waiting for events
my $blockwait=0;
258
my $nextdeadline=0;
259 260
my $mailqueue=0;

261 262 263 264 265
notify("Stated starting up\n");

sub process_event_queue() {
    $event_count=0;
    my $lastcount=-1;
266 267
    my $wait;
    my $now = time();
Mike Hibler's avatar
Mike Hibler committed
268
    debug("Polling at $now - mq=$mailqueue bw=$blockwait ndl=$nextdeadline\n");
269 270 271 272 273 274 275 276 277 278
    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;
279 280
	    } else {
		$wait = 0;
281 282 283 284 285
	    }
	}
    } else {
	# mail is waiting. Only block until it is time to send it.
	$wait = $lastmail + $mailgap - $now;
Mac Newbold's avatar
Mac Newbold committed
286
	debug("Now $now, mailgap $mailgap, last $lastmail ==> wait $wait\n");
287 288 289
    }
    if ($wait < 0) { debug("Wait was $wait!\n"); $wait=0; }
    my $finish = $now + $wait;
Mac Newbold's avatar
Mac Newbold committed
290
    while (($event_count != $lastcount || $wait > 0) &&
291
	   !($sigrestart || $sigcleanup || $do_reload || $do_reopen)) {
292
	$lastcount = $event_count;
Mac Newbold's avatar
Mac Newbold committed
293
	# Don't block if we got a signal!
294
	if ($wait<=0 || $sigrestart || $sigcleanup || $do_reload || $do_reload){
295
	    PollEvents(0, 0);
296
	} else {
297
	    #debug("Using blocking event poll - $wait seconds\n");
298
	    # timeout param is in milliseconds, so multiply
299
	    PollEvents(1, $wait);
300 301 302
	    $now = time();
	    # subtract seconds elapsed from my wait time
	    $wait = $finish - $now;
303
	    #debug("Finished blocking event poll - $wait seconds remain\n");
304
	    if ($event_count > 0 &&
Mac Newbold's avatar
Mac Newbold committed
305
		(qsize() > 0 || $mailqueue ||
306
		 $sigrestart || $sigcleanup || $do_reload || $do_reopen)) {
307 308 309 310 311 312 313
		$blockwait=0;
		$wait=0;
		#debug("Cancelling wait - timeouts/msgs waiting, or HUP'd\n");
		#debug("---End Blocking Wait ---\n");
	    }
	}
	#debug("Wait is $wait\n");
314 315
    }
    if ($event_count > 0) {
Mac Newbold's avatar
Mac Newbold committed
316
	debug("Handled $event_count event(s).\n");
317 318
    }
}
Robert Ricci's avatar
Robert Ricci committed
319

320
# Now, we just poll for events, and watch for timeouts
Robert Ricci's avatar
Robert Ricci committed
321
while (1) {
Mac Newbold's avatar
Mac Newbold committed
322
    my $now = time();
323 324 325 326
    my ($deadline,$node);

    # Check for nodes that have passed their timeout
    if (!qhead($deadline,$node)) {
327
	info("HEAD: $node in ".($deadline-$now).", queue=".qsize()."\n");
328 329
	while ($now >= $deadline && $node ne "") {
	    qpop($deadline,$node);
330
	    info("POP: $node in ".($deadline-$now).", queue=".qsize()."\n");
331 332 333 334 335 336 337 338 339 340 341 342 343
	    #
	    # If the node is no longer in the nodes array, it was most
	    # likely a dynamic virtual node which is now gone. Ignore.
	    # Need to look at reload() to see if we can catch this earlier. 
	    #
	    if (exists($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{$node}));
	    }
344 345 346 347
	    if (0) { qshow(); }
	    if (qhead($deadline,$node)) {
		$deadline=0; $node="";
	    }
348
	}
349 350
    } else {
	$deadline=0;
351
    }
352
    $nextdeadline = $deadline;
353

Mac Newbold's avatar
Mac Newbold committed
354 355 356 357
    if ($do_reload || ($now - $last_reload > $reload_time)) {
	reload();
	$do_reload = 0;
    }
Mac Newbold's avatar
Mac Newbold committed
358

Mike Hibler's avatar
Mike Hibler committed
359 360 361 362 363
    if (qsize()==0) {
	$blockwait=1;
	debug("---Blocking wait okay---\n");
    }

Mac Newbold's avatar
Mac Newbold committed
364 365
    # Send any messages in the queue if it is time
    notify("",1);
Mac Newbold's avatar
Mac Newbold committed
366

367 368
    if ($sigrestart) { restart(); }
    if ($sigcleanup) { cleanup(); }
369 370 371 372 373 374 375
    if ($do_reopen)  {
	ReOpenLog($LOGFILE);
	if (! kill('USR2', $eventchild)) {
	    fatal("Could not signal(USR2) event reader child\n");
	}
	$do_reopen = 0;
    }
376
    process_event_queue;
Robert Ricci's avatar
Robert Ricci committed
377 378
}

Mac Newbold's avatar
Mac Newbold committed
379 380
exit(0);

Robert Ricci's avatar
Robert Ricci committed
381
# Read the current states of nodes from the database
382
sub readStates(;@) {
Mac Newbold's avatar
Mac Newbold committed
383
    my %oldnodes = @_;
Mike Hibler's avatar
Mike Hibler committed
384
    my $now = time();
Mac Newbold's avatar
Mac Newbold committed
385 386

    # Guard against undefined variable warnings
387
    if (! %oldnodes) {
Mac Newbold's avatar
Mac Newbold committed
388 389
	%oldnodes = ();
    }
390

Mac Newbold's avatar
Mac Newbold committed
391 392 393
    #debug("readStates called\n");
    my $result = DBQueryFatal("SELECT node_id, eventstate, " .
			      "state_timestamp, op_mode, " .
394
			      "op_mode_timestamp, stated_tag FROM nodes ".
395
			      "where node_id not like 'sh%'");
Mac Newbold's avatar
Mac Newbold committed
396 397

    my %nodes;
398
    while (my ($node_id, $state, $timestamp, $mode, $mode_timestamp, $tag)
Mac Newbold's avatar
Mac Newbold committed
399
	   = $result->fetchrow()) {
400 401 402 403 404 405 406 407 408 409 410
	$nodes{$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
411

412
	#
Mac Newbold's avatar
Mac Newbold committed
413 414 415 416
	# 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.)
417
	#
Mac Newbold's avatar
Mac Newbold committed
418 419 420 421 422
	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};
423
	} else {
Mac Newbold's avatar
Mac Newbold committed
424 425 426 427
	    $nodes{$node_id}{state}          = $state;
	    $nodes{$node_id}{timestamp}      = $timestamp;
	    $nodes{$node_id}{mode}           = $mode;
	    $nodes{$node_id}{mode_timestamp} = $mode_timestamp;
428 429 430
	    $nodes{$node_id}{notified}       = 0;
	    $nodes{$node_id}{timedout}       = 0;
	    $nodes{$node_id}{noretry}        = 0;
431
	    # Is there a timeout? If so, set it up!
Mike Hibler's avatar
Mike Hibler committed
432
	    if (defined($timestamp) && $dbtag eq $nodes{$node_id}{"tag"}) {
433
		setTimeout($mode,$state,$node_id,$timestamp);
434
	    }
435
	}
Mac Newbold's avatar
Mac Newbold committed
436 437
    }
    return %nodes;
Robert Ricci's avatar
Robert Ricci committed
438 439 440 441 442 443
}

#
# Read timeouts for various states from the database
#
sub getTimeouts() {
Mac Newbold's avatar
Mac Newbold committed
444 445 446
    #debug("getTimeouts called\n");
    my $result = DBQueryFatal("SELECT op_mode, state, timeout, action " .
			      "FROM state_timeouts");
Robert Ricci's avatar
Robert Ricci committed
447

Mac Newbold's avatar
Mac Newbold committed
448 449 450 451 452
    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
453 454 455 456 457 458
}

#
# Read the list of valid state transitions from the database
#
sub getValid() {
Mac Newbold's avatar
Mac Newbold committed
459 460 461
    #debug("getValid called\n");
    my $result = DBQueryFatal("SELECT op_mode, state1, state2 " .
			      "FROM state_transitions");
Robert Ricci's avatar
Robert Ricci committed
462

Mac Newbold's avatar
Mac Newbold committed
463 464 465 466 467
    my %valid;
    while (my ($mode,$state1, $state2) = $result->fetchrow()) {
	$valid{$mode}{$state1}{$state2} = 1;
    }
    return %valid;
Robert Ricci's avatar
Robert Ricci committed
468 469
}

470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487
#
# 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;
}

488 489 490 491
#
# Read the list of valid mode transitions from the database
#
sub getModeTrans() {
Mac Newbold's avatar
Mac Newbold committed
492
    #debug("getModeTrans called\n");
Mac Newbold's avatar
Mac Newbold committed
493
    my $result =
Mac Newbold's avatar
Mac Newbold committed
494 495 496 497 498 499 500 501 502 503 504
      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;
505
	}
Mac Newbold's avatar
Mac Newbold committed
506 507
    }
    return %modeTrans;
508 509 510 511 512 513
}

#
# Read the list of states which trigger an action
#
sub getTriggers() {
514
    debug("getTriggers called\n");
Mac Newbold's avatar
Mac Newbold committed
515

516 517
    debug("anymode ==> '$TBANYMODE'\n");

Mac Newbold's avatar
Mac Newbold committed
518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538
    # 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 $result =
539
      DBQueryFatal("SELECT op_mode, state, `trigger` " .
540 541
		   "FROM state_triggers where node_id='$TBANYMODE' ".
		   "order by op_mode,state");
Mac Newbold's avatar
Mac Newbold committed
542
    my %t;
543
    while (my ($mode, $state, $trig) = $result->fetchrow()) {
Mac Newbold's avatar
Mac Newbold committed
544 545 546
	my @trigs = split(/\s*,\s*/,$trig);
	$t{"$mode:$state"} = \@trigs;
	debug("trig($mode:$state)\t => ".join(',',@trigs)."\n");
547 548
    }

Mac Newbold's avatar
Mac Newbold committed
549 550
    # Grab per-node triggers (including any-mode)
    $result =
551
      DBQueryFatal("SELECT node_id, op_mode, state, `trigger` " .
552 553 554 555 556 557
		   "FROM state_triggers where node_id!='$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
558
    }
559

Mac Newbold's avatar
Mac Newbold committed
560 561
    debug(hash_recurse(%t));

Mac Newbold's avatar
Mac Newbold committed
562
    return %t;
563 564
}

Robert Ricci's avatar
Robert Ricci committed
565 566 567 568
#
# Gets called for every event that we recieve
#
sub handleEvent($$$) {
569
    my ($objtype,$objname,$eventtype) = @_;
Mac Newbold's avatar
Mac Newbold committed
570
    $event_count++;
Mac Newbold's avatar
Mac Newbold committed
571

572
    #
Mac Newbold's avatar
Mac Newbold committed
573 574 575
    # 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.
576
    #
577
    if ($dbtag eq "") {
578 579
	debug("Got an event: ($objtype,$objname,$eventtype)\n");
    }
Mac Newbold's avatar
Mac Newbold committed
580 581 582 583

    #
    # Check to see if another instance is supposed to be handling this node
    #
584 585
    if ($objtype ne $TBCOMMAND) {
	my $node = $objname;
Mac Newbold's avatar
Mac Newbold committed
586

587 588 589 590 591 592 593
	#
	# 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
594
	    # churning via reload(). Bad.
595
	    if (! defined($nodes{$node})) {
596 597 598 599
		notify("Got $objtype/$eventtype for nonexistent $node!\n");
		return;
	    }
	}
Mac Newbold's avatar
Mac Newbold committed
600

601 602
	#
	# If a stated_tag was specified on the command line, ignore those
Mac Newbold's avatar
Mac Newbold committed
603
	# nodes that do not match.
604
	#
Mac Newbold's avatar
Mac Newbold committed
605
	#debug("dbtag='$dbtag', node $node='".$nodes{$node}{"tag"}."'\n");
606 607 608 609 610 611 612 613 614 615
	if ($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
616
    }
Mac Newbold's avatar
Mac Newbold committed
617 618 619 620 621

    #
    # 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.)
    #
622 623 624
    if ($dbtag ne "") {
	debug("Got an event: ($objtype,$objname,$eventtype)\n");
    }
Mac Newbold's avatar
Mac Newbold committed
625 626 627

 SWITCH: for ($objtype) {

Mac Newbold's avatar
Mac Newbold committed
628 629 630 631 632 633 634 635 636 637 638 639 640 641
	(/$TBNODESTATE/) && do {
	    stateTransition($objname,$eventtype);
	    last;
	};
	(/$TBNODEOPMODE/) && do {
	    opModeTransition($objname,$eventtype);
	    notify("Use of deprecated event TBNODEOPMODE:\n".
		   "$objname->$eventtype\n");
	    last;
	};
	(/$TBCONTROL/) && do {
	    handleCtrlEvent($objname,$eventtype);
	    last;
	};
642 643 644 645
	(/$TBCOMMAND/) && do {
	    handleCommand($objname,$eventtype);
	    last;
	};
646

Mac Newbold's avatar
Mac Newbold committed
647
    }
648 649 650 651 652

}

sub stateTransition($$) {

653
    my ($node,$newstate) = @_;
Robert Ricci's avatar
Robert Ricci committed
654

655 656
    # Check for invalid transitions
    my ($oldstate, $mode);
657 658 659
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

660
    if ($oldstate && $mode && !IsValidTransition($mode,$oldstate,$newstate)) {
661
	notify("Invalid transition for node $node from $mode/$oldstate " .
662
	       "to $newstate\n");
663
	
664 665 666 667 668 669
	#
	# 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) {
670 671 672
            $newstate = TBDB_NODESTATE_SECVIOLATION;
            notify("Moving $node to $newstate because it's in $mode\n");
        }
673 674 675 676 677 678
    }
    #
    # Nodes that are in the SECVIOLATION state are not allowed to leave!
    #
    if ($oldstate eq TBDB_NODESTATE_SECVIOLATION &&
	$newstate ne TBDB_NODESTATE_SECVIOLATION) {
679 680 681 682 683 684 685
	#
	# Allow transitions to SHUTDOWN.
	# This allows someone to reboot a node in the SECVIOLATION state
	# getting it back to MINIMAL/SHUTDOWN.
	#
	# XXX DEBUG ONLY!
	# 
686 687 688 689
	if ($soft_secviolation &&
	    ($newstate eq TBDB_NODESTATE_SHUTDOWN ||
	     $newstate eq TBDB_NODESTATE_GPXEBOOTING)) {
	    notify("$node allowed to transition: SECVIOLATION => $newstate\n");
690 691 692 693
	} else {
	    notify("$node tried to leave SECVIOLATION (to $newstate)\n");
	    $newstate = TBDB_NODESTATE_SECVIOLATION;
	}
694
    }
Robert Ricci's avatar
Robert Ricci committed
695

696 697 698 699
    my $now = time();
    $nodes{$node}{state}     = $newstate;
    $nodes{$node}{timestamp} = $now;
    $nodes{$node}{notified}  = 0;
700

701 702 703
    info("$node: $mode/$oldstate => $mode/$newstate\n");
    DBQueryFatal("UPDATE nodes SET eventstate='$newstate', " .
		 "state_timestamp='$now' WHERE node_id='$node'");
704

705 706 707 708
    # 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} =~ /^$TBCOMMAND:/) {
709
        debug("TimeoutTag = '$timeout_tag{$node}'\n");
710
	my ($str,$cmd) = split(":",$timeout_tag{$node});
711
	debug("str=$str\tcmd=$cmd\tTBREBOOT=$TBREBOOT\tstate=$newstate\n");
712
	if ($cmd eq $TBREBOOT) {
713
	    if ($newstate eq TBDB_NODESTATE_SHUTDOWN ) {
714 715 716
		info("$node: $TBREBOOT success\n");
		# Timeout will get cleared below by setTimeout call
	    } else {
717 718
		notify("$node: $TBREBOOT in progress, but got state ".
		       "$newstate instead of ".TBDB_NODESTATE_SHUTDOWN."!\n");
719 720 721 722 723
	    }
	#} elsif ($cmd eq $FOO ) {
	    # Add more here...
	} else {
	    notify("$node: Unknown command timeout '$timeout_tag{$node}' ".
724
		   "found at $mode/$newstate\n");
725 726 727
	}
    }

728 729 730 731 732
    #
    # 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.
    #
733 734
    setTimeout($mode,$newstate,$node,$now);

Mac Newbold's avatar
Mac Newbold committed
735 736 737 738 739 740 741
    # 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 = $_;
742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758
	    /^SCRIPT:([-\w\/]+)$/ && 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
759
	    /^$PXEBOOT$/ && do {
760
		#
Mac Newbold's avatar
Mac Newbold committed
761 762 763 764 765 766 767 768 769 770 771 772
		# 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.
773
		#
Mac Newbold's avatar
Mac Newbold committed
774 775 776 777
		# 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.
778
		#
Mac Newbold's avatar
Mac Newbold committed
779 780 781 782 783
		debug("Running $PXEBOOT trigger\n");
		if ($mode ne $PXEKERNEL) {
		    info("$node: Forcing mode transition into $PXEKERNEL!\n");
		    opModeTransition($node, $PXEKERNEL, 1);
		    $mode=$PXEKERNEL;
784
		}
Mac Newbold's avatar
Mac Newbold committed
785 786
		next;
	    };
787 788 789 790 791 792
	    /^$SECUREBOOT$/ && 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.
793
		#
794 795 796 797
		# 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.
798
		#
799 800 801 802 803
		debug("Running $SECUREBOOT trigger\n");
		my $query_result =
		    DBQueryWarn("select next_op_mode from nodes ".
				"where node_id='$node'");
		my ($nextmode) = $query_result->fetchrow();
804 805 806 807 808
		if (!$nextmode) {
		    $nextmode = $mode;
		}
		if ($nextmode ne TBDB_NODEOPMODE_SECURELOAD ||
		    $oldstate ne TBDB_NODESTATE_SHUTDOWN) {
809 810 811 812 813 814
		    $nextmode = TBDB_NODEOPMODE_SECUREBOOT;
		}
		if ($mode ne $nextmode) {
		    info("$node: Forcing mode transition to $nextmode!\n");
		    opModeTransition($node, $nextmode, 1);
		    $mode=$nextmode;
815 816 817
		}
		next;
	    };
Mac Newbold's avatar
Mac Newbold committed
818 819 820 821 822
	    /^$BOOTING$/ && do {
		#
		# See if we are in the right mode/osid.
		#
		my ($bootosid,$bootopmode) = TBBootWhat($node, $debug);
823

Mac Newbold's avatar
Mac Newbold committed
824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889
		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($node);
		next;
	    };
890 891 892 893
	    /^CHECKPORTREG$/ && do {
		CheckPortRegistration($node);
		next;
	    };
894
	    /^$TBRESET$/ && do {
895 896 897
		# We successfully booted, so clear some flags
		$nodes{$node}{noretry}   = 0;
		$nodes{$node}{timedout}  = 0;
898 899 900 901
		# 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) = $r->fetchrow();
902 903 904 905 906 907 908
		if (! (defined($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 $defosid) {
909 910 911 912
		    handleCtrlEvent($node,$trig);
		}
		next;
	    };
913
	    (/^$TBRELOADDONEV1$/ || /^$TBRELOADDONEV2$/) && do {
914 915 916
		handleCtrlEvent($node,$trig);
		next;
	    };
917 918 919 920 921 922 923 924
	    /^$TBISUP$/ && do {
		info("$node: Triggered $TBISUP\n");
		EventSendWarn(host      => $BOSSNODE ,
			      objtype   => TBDB_TBEVENT_NODESTATE ,
			      eventtype => TBDB_NODESTATE_ISUP ,
			      objname   => $node);
		next;
	    };
925
	    (/^$TBREBOOT$/ || /^$TBPOWERCYCLE$/ || /^$TBPOWEROFF$/) && do {
926
		handleCommand($node,$trig);
927 928 929
		next;
	    };
            (/^EMAILNOTIFY$/) && do {
930 931
		my $msg = "$node entered state $mode/$newstate from " .
		    "$mode/$oldstate";
932
		my $dest = $REALTBOPS;
933
		if ($newstate eq TBDB_NODESTATE_SECVIOLATION) {
934 935 936 937 938 939 940 941
		    if ($soft_secviolation) {
			$msg .= "\n\nNode $node was allowed to continue.\n";
			$dest = $TBOPS;
		    } else {
			$msg .= "\n\nNode $node has been powered off.\n" .
			        "You must address the cause of the violation ".
				"and reset the eventstate before powering on.";
		    }
942
		}
943 944
		SENDMAIL($dest, "STATED: $node entered state $newstate",
			 $msg, "Stated Daemon <".$TBOPS.">");
945
		next;
946
            };
947 948 949 950 951 952 953 954 955 956 957 958
	    /^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;
	    };
959
	    notify("Unknown trigger '$trig' for $node in $mode/$newstate!\n");
960
	}
Mac Newbold's avatar
Mac Newbold committed
961 962 963 964 965 966 967
	# 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);
	}
968
    }
969

970 971 972 973 974 975 976 977
    # 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
978 979 980
	} else {
	    debug("No next mode.\n");
	}
981 982
    }
}
983

984
sub opModeTransition($$;$) {
Mac Newbold's avatar
Mac Newbold committed
985

986 987
    my ($node,$newmode,$force) = @_;
    if (!defined($force)) { $force = 0; }
Mac Newbold's avatar
Mac Newbold committed
988

989
    info("$node: Mode change to $newmode requested ($force)\n");
Mac Newbold's avatar
Mac Newbold committed
990

991 992
    # Check for invalid transitions
    my ($oldstate, $mode, $nextstate);
993 994 995
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

996
    if (defined($modeTrans{"$mode:$oldstate"}) || $force) {
997 998 999 1000 1001
	#
	# If there is a next state associated with mode:oldstate, use it.
	# Otherwise we just remain in the old state.
	#
	if (defined($modeTrans{"$mode:$oldstate"})) {
1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012
	    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{$newmode};
1013 1014 1015 1016
		if ($force) {
		    info("$node: opMode force changing state along with mode\n");
		}
	    } elsif (!$force) {
1017 1018 1019
		notify("Invalid mode transition for $node from ".
		       "$mode/$oldstate to $newmode!\n");
	    }
1020 1021
	}
    } else {
1022
	notify("Invalid mode transition for $node from $mode/$oldstate: ".
1023 1024
	       "Not a valid mode transition state!\n");
    }
Mac Newbold's avatar
Mac Newbold committed
1025 1026 1027
    if (!$nextstate) {
	$nextstate=$oldstate;
    }
Mac Newbold's avatar
Mac Newbold committed
1028

1029 1030 1031 1032 1033 1034
    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
1035

1036 1037 1038 1039 1040
    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
1041 1042 1043 1044

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

1045 1046 1047 1048
}

sub handleCtrlEvent($$) {
    my ($node,$event) = @_;
1049

1050
    info("CtrlEvent: $node, $event\n");
1051

1052 1053
    foreach ($event) {
	/^$TBRESET$/ && do {
1054 1055
	    #
	    # Clear next_boot_path with os_select.
Mac Newbold's avatar
Mac Newbold committed
1056
	    #
1057
	    $cmd = "$osselect -d -c -1 $node";
1058
	    system($cmd) and
1059
		notify("$node/$event: Could not clear next_boot_path!\n");
Mac Newbold's avatar
Mac Newbold committed
1060 1061

	    info("Performed $TBRESET for $node\n");
1062 1063
	    next;
	};
1064
	(/^$TBRELOADDONEV1$/ || /^$TBRELOADDONEV2$/) && do {
1065 1066 1067 1068 1069 1070 1071
	    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'");
1072
		DBQueryFatal("delete from reserved where node_id='$node'");
1073 1074
		TBSetNodeHistory($node, TB_NODEHISTORY_OP_FREE,
				 $UID, $pid, $eid);
1075
		info("Released $node from $pid/$eid\n");
1076
	    }
1077 1078 1079 1080 1081
	    if ($event eq $TBRELOADDONEV2) {
		info("Sending an apod to $node\n");
		system("$apod $node") == 0 or
		    notify("Could not apod $node after $TBRELOADDONEV2!\n");
	    }
1082 1083 1084
	    next;
	};
	/^$TBTIMEOUT$/ && do {
1085 1086 1087 1088
	    my ($mode,$state) = split(":",$timeout_tag{$node});
	    delete($timeout_tag{$node});
	    my $curstate = $nodes{$node}{state};
	    my $curmode = $nodes{$node}{mode};
1089
	    my ($timeout,$action);
1090 1091 1092 1093 1094
	    if (!defined($nodes{$node}{notified})) {
		$nodes{$node}{notified}=0;
	    }
	    $nodes{$node}{notified}++;
	    my $notified = $nodes{$node}{notified};
1095 1096
	    $nodes{$node}{timedout}++;
	    my $timedout = $nodes{$node}{timedout};
1097 1098 1099 1100
	    if ($mode && $state && $timeouts{$mode} &&
		$timeouts{$mode}{$state}) {
		($timeout, $action) = @{$timeouts{$mode}{$state}};
	    }
1101 1102
	    if ($mode eq $TBCOMMAND) {
		# It is a command, not a true state
1103
		if ($action eq $TBTIMEOUTCMDRETRY) {
1104
		    # Retry the command
1105
		    notify("$node: Command $state, retry #$timedout\n");
1106
		    # notify in case we get in a retry loop...
1107
		    handleCommand($node,$state,$timedout,1);
1108 1109 1110 1111 1112 1113
		} else {
		    notify("$node: Unknown timeout action for ".
			   "$mode/$state: '$action'\n");
		}
		next;
	    }
1114 1115 1116 1117 1118 1119 1120

	    #
	    # 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
1121

1122 1123
		if ($timedout < 3) {
		    #
Mac Newbold's avatar
Mac Newbold committed
1124 1125
		    # Try again.
		    #
1126 1127 1128 1129 1130 1131
		    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
1132 1133
		    # Failed too many times, power cycle instead.
		    #
1134 1135 1136 1137 1138 1139 1140
		    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");
1141
		system("(date; $cmd) >>$rebootlog 2>&1 &") and
1142
		    notify("$PXEWAKEUP retry: ".
1143 1144 1145 1146 1147
			   "Command '$cmd' failed, error $?: $!\n");

		next;
	    }

1148 1149 1150 1151 1152 1153
	    info("Node $node has timed out in state $mode/$state".
		 ($action ne "" ? "\n\tRequested action $action." : "").
		 "\n");

	    foreach ($action) {
		/^$TBTIMEOUTREBOOT/ && do {
1154 1155 1156 1157 1158
		    if ($timedout>3) {
			# We've tried too many times...
			notify("Node $node has timed out too many times!\n".
			       "Giving up until it boots sucessfully.\n");
			$nodes{$node}{noretry} = 1;
1159
		    } else {
1160 1161 1162 1163 1164 1165
			# XXX Temporary!  For now notify instead of
			# really rebooting, until the timeout/retry
			# stuff is gone from os_setup and os_load
			notify("Node $node has timed out in state ".
			       "$mode/$state - REBOOT requested\n");
			#handleCommand($node,$TBREBOOT,$timedout,1);