stated.in 49.5 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) 2000-2010 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
Mac Newbold's avatar
Mac Newbold committed
48

49 50 51 52 53
# 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
54
# Number of iterations (roughly, seconds) after which we'll reload
Robert Ricci's avatar
Robert Ricci committed
55 56
# information from the database. This is so we don't end up with information
# that's _too_ out of sync.
57
my $reload_time = 600;
58
my $last_reload = time;
Robert Ricci's avatar
Robert Ricci committed
59

60 61 62
# Command line opts.
my $dbtag = "";
my $debug = 0;
63 64
my $server = "localhost";
my $port   = @BOSSEVENTPORT@;
65 66 67
my $lockfile;
my $pidfile;

Robert Ricci's avatar
Robert Ricci committed
68 69 70
# Process command-line arguments

sub usage {
Mac Newbold's avatar
Mac Newbold committed
71
    print << "END";
72
Usage: $0 [-h] [-d] [-s server] [-p port] [-t dbtag]
73
-h              This message
74 75
-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
76 77
-s server       Use specified server, instead of this site's bossnode
-p port	        Use specified port
78
Send SIGHUP to reload database state, or SIGUSR1 to restart completely.
Robert Ricci's avatar
Robert Ricci committed
79
END
80
    exit(1);
Robert Ricci's avatar
Robert Ricci committed
81 82
}

Mac Newbold's avatar
Mac Newbold committed
83
# Only root should run this - it won't work when run as a user...
84
# (Or, let an admin run it if it isn't the real one in /usr/testbed/ )
85
if ($UID && ( $TB eq $REALTB || ! TBAdmin($UID) ) ) {
Mac Newbold's avatar
Mac Newbold committed
86 87 88
    die("Only root can run this script!\n");
}

89
my @args = @ARGV;    # save a copy for restart before we mess with them.
Robert Ricci's avatar
Robert Ricci committed
90
my %opt = ();
91
if (!getopts("ds:p:ht:l",\%opt)) { usage(); }
Robert Ricci's avatar
Robert Ricci committed
92

Mac Newbold's avatar
Mac Newbold committed
93 94 95 96 97 98
if ($opt{h}) {
    exit &usage;
}
if (@ARGV) {
    exit &usage;
}
Robert Ricci's avatar
Robert Ricci committed
99

Mac Newbold's avatar
Mac Newbold committed
100 101 102 103 104 105
if ($opt{s}) {
    $server = $opt{s};
}
if ($opt{p}) {
    $port = $opt{p};
}
106 107 108
if ($opt{t}) {
    $dbtag = $opt{t};
}
Mac Newbold's avatar
Mac Newbold committed
109 110 111
if ($opt{d}) {
    $debug = 1;
}
Robert Ricci's avatar
Robert Ricci committed
112

113
# Grab some constants into variables
114
my $TBANYMODE    = TBDB_NODEOPMODE_ANY;
115 116
my $TBRESET      = TBDB_TBCONTROL_RESET;
my $TBTIMEOUT    = TBDB_TBCONTROL_TIMEOUT;
Mac Newbold's avatar
Mac Newbold committed
117 118 119
my $PXEBOOT      = TBDB_TBCONTROL_PXEBOOT;
my $BOOTING      = TBDB_TBCONTROL_BOOTING;
my $CHECKGENISUP = TBDB_TBCONTROL_CHECKGENISUP;
120 121 122
my $TBNOTIMEOUT  = TBDB_NO_STATE_TIMEOUT;
my $TBNODESTATE  = TBDB_TBEVENT_NODESTATE;
my $TBNODEOPMODE = TBDB_TBEVENT_NODEOPMODE;
123 124 125 126 127 128
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;
129
my $TBISUP       = TBDB_NODESTATE_ISUP;
130 131 132
my $PXEWAIT      = TBDB_NODESTATE_PXEWAIT;
my $PXEWAKEUP    = TBDB_NODESTATE_PXEWAKEUP;
my $PXEBOOTING   = TBDB_NODESTATE_PXEBOOTING;
133 134
my $TBRELOADDONEV1    = TBDB_TBCONTROL_RELOADDONE;
my $TBRELOADDONEV2    = TBDB_TBCONTROL_RELOADDONE_V2;
135 136 137 138
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;
139

140 141
# Special PXEBOOT state machine that all local nodes use.
my $PXEKERNEL	 = "PXEKERNEL";
142

143 144
# Even special-er SECUREBOOT state machine that local nodes may use.
my $SECUREBOOT	 = "SECUREBOOT";
145

146 147 148 149
# Protos.
sub debug(@);
sub fatal($);
sub notify($;$);
150
sub info($);
151 152 153 154 155 156 157 158
sub getTimeouts();
sub getValid();
sub getModeTrans();
sub getTriggers();
sub readStates(;@);
sub handleCtrlEvent($$);
sub reload();
    
159 160 161
if (!$debug) {
    if ( $TB eq $REALTB ) {
	$pidfile = "/var/run/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
162
    } else {
163
	$pidfile = "$TB/locks/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
164
    }
165
    debug("Using pidfile $pidfile\n");
Mac Newbold's avatar
Mac Newbold committed
166

167 168 169 170 171 172 173 174 175 176 177 178 179 180 181
    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: $? $!\n");
    }
    # Background
182
    if (TBBackGround($LOGFILE)) {
Mac Newbold's avatar
Mac Newbold committed
183 184
	exit(0);
    }
185
    TBdbfork();
Robert Ricci's avatar
Robert Ricci committed
186
}
187 188 189 190 191 192 193 194 195

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
196

197 198 199
# Change my $0 so that it is easier to see in a ps/top
$0 = "$0";

Robert Ricci's avatar
Robert Ricci committed
200
my $URL = "elvin://$server";
Mac Newbold's avatar
Mac Newbold committed
201 202 203
if ($port) {
    $URL .= ":$port";
}
Robert Ricci's avatar
Robert Ricci committed
204

Mac Newbold's avatar
Mac Newbold committed
205
# Connect to the event system, and subscribe the the events we want
Robert Ricci's avatar
Robert Ricci committed
206
my $handle = event_register($URL,0);
Mac Newbold's avatar
Mac Newbold committed
207 208 209
if (!$handle) {
    fatal("Unable to register with event system\n");
}
Robert Ricci's avatar
Robert Ricci committed
210 211

my $tuple = address_tuple_alloc();
Mac Newbold's avatar
Mac Newbold committed
212 213 214
if (!$tuple) {
    fatal("Could not allocate an address tuple\n");
}
Robert Ricci's avatar
Robert Ricci committed
215

216 217 218
%$tuple = ( objtype => join(",",
			    $TBNODESTATE, $TBNODEOPMODE,
			    $TBCONTROL, $TBCOMMAND) );
219

Robert Ricci's avatar
Robert Ricci committed
220
if (!event_subscribe($handle,\&handleEvent,$tuple)) {
Mac Newbold's avatar
Mac Newbold committed
221
    fatal("Could not subscribe to events\n");
Robert Ricci's avatar
Robert Ricci committed
222 223 224 225
}

# Read in the pre-existing node states, and timeout and valid transition
# information from the database
226 227 228 229
my %timeouts  = getTimeouts();
my %valid     = getValid();
my %modeTrans = getModeTrans();
my %triggers  = getTriggers();
230
my %nodes     = readStates();
231
my %timeouttag= ();
232
if ($debug) { qshow(); }
Robert Ricci's avatar
Robert Ricci committed
233 234 235

# Gets set if a reload of state from the database should happen.
my $do_reload = 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
$SIG{USR2} = sub { info("SIGUSR2 - Reopening logfile\n"); ReOpenLog($LOGFILE); };
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;
266 267 268 269 270 271 272 273 274 275 276 277 278
    my $wait;
    my $now = time();
    debug("Polling - mq=$mailqueue bw=$blockwait\n");
    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 + $wait;
Mac Newbold's avatar
Mac Newbold committed
290 291
    while (($event_count != $lastcount || $wait > 0) &&
	   !($sigrestart || $sigcleanup || $do_reload)) {
292
	$lastcount = $event_count;
Mac Newbold's avatar
Mac Newbold committed
293 294
	# Don't block if we got a signal!
	if ($wait<=0 || $sigrestart || $sigcleanup || $do_reload) {
295 296
	    event_poll($handle);
	} else {
297
	    #debug("Using blocking event poll - $wait seconds\n");
298 299 300 301 302
	    # timeout param is in milliseconds, so multiply
	    event_poll_blocking($handle, $wait*1000);
	    $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
305 306
		(qsize() > 0 || $mailqueue ||
		 $sigrestart || $sigcleanup || $do_reload)) {
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-$now).", queue=".qsize()."\n");
331
	    handleCtrlEvent($node,$TBTIMEOUT);
332 333 334 335
	    if (0) { qshow(); }
	    if (qhead($deadline,$node)) {
		$deadline=0; $node="";
	    }
336
	}
337 338
    } else {
	$deadline=0;
339
    }
340
    $nextdeadline = $deadline;
341

342 343 344 345
    if (qsize()==0) {
	$blockwait=1;
	debug("---Blocking wait okay---\n");
    }
Mac Newbold's avatar
Mac Newbold committed
346

Mac Newbold's avatar
Mac Newbold committed
347 348 349 350
    if ($do_reload || ($now - $last_reload > $reload_time)) {
	reload();
	$do_reload = 0;
    }
Mac Newbold's avatar
Mac Newbold committed
351

Mac Newbold's avatar
Mac Newbold committed
352 353
    # Send any messages in the queue if it is time
    notify("",1);
Mac Newbold's avatar
Mac Newbold committed
354

355 356 357
    if ($sigrestart) { restart(); }
    if ($sigcleanup) { cleanup(); }

358
    process_event_queue;
Robert Ricci's avatar
Robert Ricci committed
359 360
}

Mac Newbold's avatar
Mac Newbold committed
361 362
exit(0);

Robert Ricci's avatar
Robert Ricci committed
363
# Read the current states of nodes from the database
364
sub readStates(;@) {
Mac Newbold's avatar
Mac Newbold committed
365 366 367
    my %oldnodes = @_;

    # Guard against undefined variable warnings
368
    if (! %oldnodes) {
Mac Newbold's avatar
Mac Newbold committed
369 370
	%oldnodes = ();
    }
371

Mac Newbold's avatar
Mac Newbold committed
372 373 374
    #debug("readStates called\n");
    my $result = DBQueryFatal("SELECT node_id, eventstate, " .
			      "state_timestamp, op_mode, " .
375
			      "op_mode_timestamp, stated_tag FROM nodes ".
376
			      "where node_id not like 'sh%'");
Mac Newbold's avatar
Mac Newbold committed
377 378

    my %nodes;
379
    while (my ($node_id, $state, $timestamp, $mode, $mode_timestamp, $tag)
Mac Newbold's avatar
Mac Newbold committed
380
	   = $result->fetchrow()) {
381 382 383 384 385 386 387 388 389 390 391
	$nodes{$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
392

393
	#
Mac Newbold's avatar
Mac Newbold committed
394 395 396 397
	# 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.)
398
	#
Mac Newbold's avatar
Mac Newbold committed
399 400 401 402 403
	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};
404
	} else {
Mac Newbold's avatar
Mac Newbold committed
405 406 407 408
	    $nodes{$node_id}{state}          = $state;
	    $nodes{$node_id}{timestamp}      = $timestamp;
	    $nodes{$node_id}{mode}           = $mode;
	    $nodes{$node_id}{mode_timestamp} = $mode_timestamp;
409 410 411
	    $nodes{$node_id}{notified}       = 0;
	    $nodes{$node_id}{timedout}       = 0;
	    $nodes{$node_id}{noretry}        = 0;
412
	    # Is there a timeout? If so, set it up!
413 414 415
	    if ($dbtag eq $nodes{$node_id}{"tag"}) {
		setTimeout($mode,$state,$node_id,$timestamp);
	    }
416
	}
Mac Newbold's avatar
Mac Newbold committed
417 418
    }
    return %nodes;
Robert Ricci's avatar
Robert Ricci committed
419 420 421 422 423 424
}

#
# Read timeouts for various states from the database
#
sub getTimeouts() {
Mac Newbold's avatar
Mac Newbold committed
425 426 427
    #debug("getTimeouts called\n");
    my $result = DBQueryFatal("SELECT op_mode, state, timeout, action " .
			      "FROM state_timeouts");
Robert Ricci's avatar
Robert Ricci committed
428

Mac Newbold's avatar
Mac Newbold committed
429 430 431 432 433
    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
434 435 436 437 438 439
}

#
# Read the list of valid state transitions from the database
#
sub getValid() {
Mac Newbold's avatar
Mac Newbold committed
440 441 442
    #debug("getValid called\n");
    my $result = DBQueryFatal("SELECT op_mode, state1, state2 " .
			      "FROM state_transitions");
Robert Ricci's avatar
Robert Ricci committed
443

Mac Newbold's avatar
Mac Newbold committed
444 445 446 447 448
    my %valid;
    while (my ($mode,$state1, $state2) = $result->fetchrow()) {
	$valid{$mode}{$state1}{$state2} = 1;
    }
    return %valid;
Robert Ricci's avatar
Robert Ricci committed
449 450
}

451 452 453 454
#
# Read the list of valid mode transitions from the database
#
sub getModeTrans() {
Mac Newbold's avatar
Mac Newbold committed
455
    #debug("getModeTrans called\n");
Mac Newbold's avatar
Mac Newbold committed
456
    my $result =
Mac Newbold's avatar
Mac Newbold committed
457 458 459 460 461 462 463 464 465 466 467
      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;
468
	}
Mac Newbold's avatar
Mac Newbold committed
469 470
    }
    return %modeTrans;
471 472 473 474 475 476
}

#
# Read the list of states which trigger an action
#
sub getTriggers() {
477
    debug("getTriggers called\n");
Mac Newbold's avatar
Mac Newbold committed
478

479 480
    debug("anymode ==> '$TBANYMODE'\n");

Mac Newbold's avatar
Mac Newbold committed
481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501
    # 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 $result =
502
      DBQueryFatal("SELECT op_mode, state, `trigger` " .
503 504
		   "FROM state_triggers where node_id='$TBANYMODE' ".
		   "order by op_mode,state");
Mac Newbold's avatar
Mac Newbold committed
505
    my %t;
506
    while (my ($mode, $state, $trig) = $result->fetchrow()) {
Mac Newbold's avatar
Mac Newbold committed
507 508 509
	my @trigs = split(/\s*,\s*/,$trig);
	$t{"$mode:$state"} = \@trigs;
	debug("trig($mode:$state)\t => ".join(',',@trigs)."\n");
510 511
    }

Mac Newbold's avatar
Mac Newbold committed
512 513
    # Grab per-node triggers (including any-mode)
    $result =
514
      DBQueryFatal("SELECT node_id, op_mode, state, `trigger` " .
515 516 517 518 519 520
		   "FROM state_triggers where node_id!='$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
521
    }
522

Mac Newbold's avatar
Mac Newbold committed
523 524
    debug(hash_recurse(%t));

Mac Newbold's avatar
Mac Newbold committed
525
    return %t;
526 527
}

Robert Ricci's avatar
Robert Ricci committed
528 529 530 531
#
# Gets called for every event that we recieve
#
sub handleEvent($$$) {
Mac Newbold's avatar
Mac Newbold committed
532 533 534 535 536 537
    my ($handle,$notification,$data) = @_;
    my $objtype = event_notification_get_objtype($handle,$notification);
    my $objname = event_notification_get_objname($handle,$notification);
    my $eventtype = event_notification_get_eventtype($handle,$notification);

    $event_count++;
Mac Newbold's avatar
Mac Newbold committed
538

539
    #
Mac Newbold's avatar
Mac Newbold committed
540 541 542
    # 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.
543
    #
544
    if ($dbtag eq "") {
545 546
	debug("Got an event: ($objtype,$objname,$eventtype)\n");
    }
Mac Newbold's avatar
Mac Newbold committed
547 548 549 550

    #
    # Check to see if another instance is supposed to be handling this node
    #
551 552
    if ($objtype ne $TBCOMMAND) {
	my $node = $objname;
Mac Newbold's avatar
Mac Newbold committed
553

554 555 556 557 558 559 560
	#
	# 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
561
	    # churning via reload(). Bad.
562
	    if (! defined($nodes{$node})) {
563 564 565 566
		notify("Got $objtype/$eventtype for nonexistent $node!\n");
		return;
	    }
	}
Mac Newbold's avatar
Mac Newbold committed
567

568 569
	#
	# If a stated_tag was specified on the command line, ignore those
Mac Newbold's avatar
Mac Newbold committed
570
	# nodes that do not match.
571
	#
Mac Newbold's avatar
Mac Newbold committed
572
	#debug("dbtag='$dbtag', node $node='".$nodes{$node}{"tag"}."'\n");
573 574 575 576 577 578 579 580 581 582
	if ($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
583
    }
Mac Newbold's avatar
Mac Newbold committed
584 585 586 587 588

    #
    # 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.)
    #
589 590 591
    if ($dbtag ne "") {
	debug("Got an event: ($objtype,$objname,$eventtype)\n");
    }
Mac Newbold's avatar
Mac Newbold committed
592 593 594

 SWITCH: for ($objtype) {

Mac Newbold's avatar
Mac Newbold committed
595 596 597 598 599 600 601 602 603 604 605 606 607 608
	(/$TBNODESTATE/) && do {
	    stateTransition($objname,$eventtype);
	    last;
	};
	(/$TBNODEOPMODE/) && do {
	    opModeTransition($objname,$eventtype);
	    notify("Use of deprecated event TBNODEOPMODE:\n".
		   "$objname->$eventtype\n");
	    last;
	};
	(/$TBCONTROL/) && do {
	    handleCtrlEvent($objname,$eventtype);
	    last;
	};
609 610 611 612
	(/$TBCOMMAND/) && do {
	    handleCommand($objname,$eventtype);
	    last;
	};
613

Mac Newbold's avatar
Mac Newbold committed
614
    }
615 616 617 618 619

}

sub stateTransition($$) {

620
    my ($node,$newstate) = @_;
Robert Ricci's avatar
Robert Ricci committed
621

622 623
    # Check for invalid transitions
    my ($oldstate, $mode);
624 625 626
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

627 628 629
    if ($oldstate && $mode && $valid{$mode} && $valid{$mode}{$oldstate} &&
	!$valid{$mode}{$oldstate}{$newstate}) {
	notify("Invalid transition for node $node from $mode/$oldstate " .
630
	       "to $newstate\n");
631 632 633 634 635 636
	#
	# Machines in the secure boot path are not allowed to jump
	# willy-nilly into unknown states.
	#
        if ($mode eq TBDB_NODEOPMODE_SECUREBOOT ||
	    $mode eq TBDB_NODEOPMODE_SECURELOAD) {
637 638 639
            $newstate = TBDB_NODESTATE_SECVIOLATION;
            notify("Moving $node to $newstate because it's in $mode\n");
        }
640 641 642 643 644 645 646 647
	#
	# And nodes that are in the SECVIOLATION state already are
	# not allowed to leave!
	#
	elsif ($oldstate eq TBDB_NODESTATE_SECVIOLATION) {
            $newstate = TBDB_NODESTATE_SECVIOLATION;
	    notify("$node tried to leave SECVIOLATION (to $newstate)\n");
	}
648
    }
Robert Ricci's avatar
Robert Ricci committed
649

650 651 652 653
    my $now = time();
    $nodes{$node}{state}     = $newstate;
    $nodes{$node}{timestamp} = $now;
    $nodes{$node}{notified}  = 0;
654

655 656 657
    info("$node: $mode/$oldstate => $mode/$newstate\n");
    DBQueryFatal("UPDATE nodes SET eventstate='$newstate', " .
		 "state_timestamp='$now' WHERE node_id='$node'");
658

659 660 661 662
    # 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} =~ /^$TBCOMMAND:/) {
663
        debug("TimeoutTag = '$timeout_tag{$node}'\n");
664
	my ($str,$cmd) = split(":",$timeout_tag{$node});
665
	debug("str=$str\tcmd=$cmd\tTBREBOOT=$TBREBOOT\tstate=$newstate\n");
666
	if ($cmd eq $TBREBOOT) {
667
	    if ($newstate eq TBDB_NODESTATE_SHUTDOWN ) {
668 669 670
		info("$node: $TBREBOOT success\n");
		# Timeout will get cleared below by setTimeout call
	    } else {
671 672
		notify("$node: $TBREBOOT in progress, but got state ".
		       "$newstate instead of ".TBDB_NODESTATE_SHUTDOWN."!\n");
673 674 675 676 677
	    }
	#} elsif ($cmd eq $FOO ) {
	    # Add more here...
	} else {
	    notify("$node: Unknown command timeout '$timeout_tag{$node}' ".
678
		   "found at $mode/$newstate\n");
679 680 681
	}
    }

682 683 684 685 686
    #
    # 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.
    #
687 688
    setTimeout($mode,$newstate,$node,$now);

Mac Newbold's avatar
Mac Newbold committed
689 690 691 692 693 694 695
    # 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 = $_;
696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712
	    /^SCRIPT:([-\w\/]+)$/ && 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
713
	    /^$PXEBOOT$/ && do {
714
		#
Mac Newbold's avatar
Mac Newbold committed
715 716 717 718 719 720 721 722 723 724 725 726
		# 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.
727
		#
Mac Newbold's avatar
Mac Newbold committed
728 729 730 731
		# 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.
732
		#
Mac Newbold's avatar
Mac Newbold committed
733 734 735 736 737
		debug("Running $PXEBOOT trigger\n");
		if ($mode ne $PXEKERNEL) {
		    info("$node: Forcing mode transition into $PXEKERNEL!\n");
		    opModeTransition($node, $PXEKERNEL, 1);
		    $mode=$PXEKERNEL;
738
		}
Mac Newbold's avatar
Mac Newbold committed
739 740
		next;
	    };
741 742 743 744 745 746
	    /^$SECUREBOOT$/ && 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.
747
		#
748 749 750
		# To differentiate BOOT from LOAD we check next_op_mode.
		# It will be set to SECURELOAD when we need to go there,
		# otherwise we force it to SECUREBOOT.
751
		#
752 753 754 755 756 757 758 759 760 761 762 763
		debug("Running $SECUREBOOT trigger\n");
		my $query_result =
		    DBQueryWarn("select next_op_mode from nodes ".
				"where node_id='$node'");
		my ($nextmode) = $query_result->fetchrow();
		if (!$nextmode || $nextmode ne TBDB_NODEOPMODE_SECURELOAD) {
		    $nextmode = TBDB_NODEOPMODE_SECUREBOOT;
		}
		if ($mode ne $nextmode) {
		    info("$node: Forcing mode transition to $nextmode!\n");
		    opModeTransition($node, $nextmode, 1);
		    $mode=$nextmode;
764 765 766
		}
		next;
	    };
Mac Newbold's avatar
Mac Newbold committed
767 768 769 770 771
	    /^$BOOTING$/ && do {
		#
		# See if we are in the right mode/osid.
		#
		my ($bootosid,$bootopmode) = TBBootWhat($node, $debug);
772

Mac Newbold's avatar
Mac Newbold committed
773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838
		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($node);
		next;
	    };
839 840 841 842
	    /^CHECKPORTREG$/ && do {
		CheckPortRegistration($node);
		next;
	    };
843
	    /^$TBRESET$/ && do {
844 845 846
		# We successfully booted, so clear some flags
		$nodes{$node}{noretry}   = 0;
		$nodes{$node}{timedout}  = 0;
847 848 849 850 851 852 853 854 855
		# 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) = $r->fetchrow();
		if ($osid ne $defosid) {
		    handleCtrlEvent($node,$trig);
		}
		next;
	    };
856
	    (/^$TBRELOADDONEV1$/ || /^$TBRELOADDONEV2$/) && do {
857 858 859
		handleCtrlEvent($node,$trig);
		next;
	    };
860 861 862 863 864 865 866 867
	    /^$TBISUP$/ && do {
		info("$node: Triggered $TBISUP\n");
		EventSendWarn(host      => $BOSSNODE ,
			      objtype   => TBDB_TBEVENT_NODESTATE ,
			      eventtype => TBDB_NODESTATE_ISUP ,
			      objname   => $node);
		next;
	    };
868 869
	    (/^$TBPOWERCYCLE$/ || /^$TBPOWEROFF$/) && do {
		handleCommand($node,$trig);
870 871 872 873 874 875 876 877
		next;
	    };
            (/^EMAILNOTIFY$/) && do {
		SENDMAIL($REALTBOPS,
			 "STATED: $node entered state $newstate",
			 "$node entered state $mode/$newstate from " .
                         "$mode/$oldstate",
			 "Stated Daemon <".$TBOPS.">");
878
		next;
879
            };
880 881 882 883 884 885 886 887 888 889 890 891
	    /^RELOADOLDMFS$/ && 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;
	    };
892
	    notify("Unknown trigger '$trig' for $node in $mode/$newstate!\n");
893
	}
Mac Newbold's avatar
Mac Newbold committed
894 895 896 897 898 899 900
	# 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);
	}
901
    }
902

903 904 905 906 907 908 909 910
    # 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
911 912 913
	} else {
	    debug("No next mode.\n");
	}
914 915
    }
}
916

917
sub opModeTransition($$;$) {
Mac Newbold's avatar
Mac Newbold committed
918

919 920
    my ($node,$newmode,$force) = @_;
    if (!defined($force)) { $force = 0; }
Mac Newbold's avatar
Mac Newbold committed
921

922
    info("$node: Mode change to $newmode requested ($force)\n");
Mac Newbold's avatar
Mac Newbold committed
923

924 925
    # Check for invalid transitions
    my ($oldstate, $mode, $nextstate);
926 927 928
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

929
    if (defined($modeTrans{"$mode:$oldstate"}) || $force) {
930
	if (!$force) {
931 932 933 934 935 936 937 938 939 940 941 942 943 944 945
	    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};
	    } else {
		notify("Invalid mode transition for $node from ".
		       "$mode/$oldstate to $newmode!\n");
	    }
946 947
	}
    } else {
948
	notify("Invalid mode transition for $node from $mode/$oldstate: ".
949 950
	       "Not a valid mode transition state!\n");
    }
Mac Newbold's avatar
Mac Newbold committed
951 952 953
    if (!$nextstate) {
	$nextstate=$oldstate;
    }
Mac Newbold's avatar
Mac Newbold committed
954

955 956 957 958 959 960
    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
961

962 963 964 965 966
    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
967 968 969 970

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

971 972 973 974
}

sub handleCtrlEvent($$) {
    my ($node,$event) = @_;
975

976
    info("CtrlEvent: $node, $event\n");
977

978 979
    foreach ($event) {
	/^$TBRESET$/ && do {
980 981
	    #
	    # Clear next_boot_path with os_select.
Mac Newbold's avatar
Mac Newbold committed
982
	    #
983
	    $cmd = "$osselect -d -c -1 $node";
984
	    system($cmd) and
985
		notify("$node/$event: Could not clear next_boot_path!\n");
Mac Newbold's avatar
Mac Newbold committed
986 987

	    info("Performed $TBRESET for $node\n");
988 989
	    next;
	};
990
	(/^$TBRELOADDONEV1$/ || /^$TBRELOADDONEV2$/) && do {
991 992 993 994 995 996 997
	    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'");
998
		DBQueryFatal("delete from reserved where node_id='$node'");
999 1000
		TBSetNodeHistory($node, TB_NODEHISTORY_OP_FREE,
				 $UID, $pid, $eid);
1001
		info("Released $node from $pid/$eid\n");
1002
	    }
1003 1004 1005 1006 1007
	    if ($event eq $TBRELOADDONEV2) {
		info("Sending an apod to $node\n");
		system("$apod $node") == 0 or
		    notify("Could not apod $node after $TBRELOADDONEV2!\n");
	    }
1008 1009 1010
	    next;
	};
	/^$TBTIMEOUT$/ && do {
1011 1012 1013 1014
	    my ($mode,$state) = split(":",$timeout_tag{$node});
	    delete($timeout_tag{$node});
	    my $curstate = $nodes{$node}{state};
	    my $curmode = $nodes{$node}{mode};
1015
	    my ($timeout,$action);
1016 1017 1018 1019 1020
	    if (!defined($nodes{$node}{notified})) {
		$nodes{$node}{notified}=0;
	    }
	    $nodes{$node}{notified}++;
	    my $notified = $nodes{$node}{notified};
1021 1022
	    $nodes{$node}{timedout}++;
	    my $timedout = $nodes{$node}{timedout};
1023 1024 1025 1026
	    if ($mode && $state && $timeouts{$mode} &&
		$timeouts{$mode}{$state}) {
		($timeout, $action) = @{$timeouts{$mode}{$state}};
	    }
1027 1028
	    if ($mode eq $TBCOMMAND) {
		# It is a command, not a true state
1029
		if ($action eq $TBTIMEOUTCMDRETRY) {
1030
		    # Retry the command
1031
		    notify("$node: Command $state, retry #$timedout\n");
1032
		    # notify in case we get in a retry loop...
1033
		    handleCommand($node,$state,$timedout,1);
1034 1035 1036 1037 1038 1039
		} else {
		    notify("$node: Unknown timeout action for ".
			   "$mode/$state: '$action'\n");
		}
		next;
	    }
1040 1041 1042 1043 1044 1045 1046

	    #
	    # 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
1047

1048 1049
		if ($timedout < 3) {
		    #
Mac Newbold's avatar
Mac Newbold committed
1050 1051
		    # Try again.
		    #
1052 1053 1054 1055 1056 1057
		    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
1058 1059
		    # Failed too many times, power cycle instead.
		    #
1060 1061 1062 1063 1064 1065 1066 1067 1068
		    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");
		system("date 2>&1 >> $rebootlog");
		system("$cmd 2>&1 >> $rebootlog &") and
1069
		    notify("$PXEWAKEUP retry: ".
1070 1071 1072 1073 1074
			   "Command '$cmd' failed, error $?: $!\n");

		next;
	    }

1075 1076 1077 1078 1079 1080
	    info("Node $node has timed out in state $mode/$state".
		 ($action ne "" ? "\n\tRequested action $action." : "").
		 "\n");

	    foreach ($action) {
		/^$TBTIMEOUTREBOOT/ && do {
1081 1082 1083 1084 1085
		    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;
1086
		    } else {
1087 1088 1089 1090 1091 1092
			# 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);
1093 1094 1095 1096 1097
		    }
		    last; };
		/^$TBTIMEOUTNOTIFY/ && do {
		    notify("Node $node has timed out in state $mode/$state\n");
		    last; };
1098 1099 1100 1101 1102
                /^STATE:([-\w\/]+)$/ && do {
                    my $newstate = $1;
                    # Force the node into a new state
                    stateTransition($node,$newstate);
                    last; };
1103 1104
		notify("$node: Unknown Timeout Action: $action\n");
	    }
1105 1106
	    next;
	};
1107
	notify("$node: Unknown CtrlEvent: $event\n");
1108 1109
    }
}
Robert Ricci's avatar
Robert Ricci committed
1110

1111 1112
sub handleCommand($$;$$) {
    my ($params,$command,$retry,$force) = @_;
1113
    if (!defined($retry)) { $retry=0; }
1114
    if (!defined($force)) { $force=0; }
1115 1116 1117 1118 1119 1120 1121 1122

    info("Command: $params, $command (attempt $retry)\n");

    # XXX - Right now we skip the checkDBRedirect calls for our
    # TBCOMMAND events, since they may have a list of nodes in them.
    # We may need to do it here (while iterating over the list), or
    # make some other fix up in handleEvent.

1123
    if ($command eq $TBREBOOT && $retry >=4) {
1124 1125
	announce("Node $params has tried rebooting $retry times and has \n".
		 "still not been successful. Please look into it soon.\n".
1126 1127 1128
	"" );#	 "In the meantime, $params will be powered off.\n");
	# Just return...
	return 0;
1129
	# change my command to poweroff.
1130
	#$command = $TBPOWEROFF;
1131 1132 1133 1134 1135 1136
    }

    foreach ($command) {
	/^$TBREBOOT$/ && do {
	    # For reboot, the params is a comma-separated list of nodes
	    my @nodes = split(",",$params);
1137 1138 1139 1140 1141 1142 1143 1144

	    # Check if any of these are in the middle of things and
	    # shouldn't be rebooted.
	    if (!$force) {
		foreach $n ( 0 .. $#nodes ) {
		    $node = $nodes[$n];
		    debug("Checking rebooting: $node, $nodes{$node}, ".
			  "$nodes{$node}{state}, $nodes{$node}{noretry}\n");
1145 1146
		    if (($nodes{$node}{state} ne TBDB_NODESTATE_ISUP) &&
			(!$nodes{$node}{noretry}) ) {
1147
			# This node shouldn't be rebooted now...
1148 1149
			# XXX Send feedback here somehow!

1150 1151 1152 1153 1154 1155 1156 1157 1158 1159
			info("$node: Trying to reboot too soon! Skipping.\n");
			# Cut it out of the list
			debug("Nodelist before ==> ".join(" ",@nodes)."\n");
			splice(@nodes,$n,1);
			debug("Nodelist after  ==> ".join(" ",@nodes)."\n");
		    }
		}
		if (@nodes == 0) { next; }
	    }

1160 1161 1162 1163 1164
	    my $nodelist=join(" ",@nodes);
	    info("Rebooting nodes: $nodelist\n");
	    # Permissions were checked in order to send the message,
	    # so we don't need to do any fancy stuff here.

1165
	    my $cmd = "$nodereboot -r $nodelist";
1166
	    my $redir = " 2>&1 >> $rebootlog &";
1167 1168 1169
	    debug("$cmd $redir\n");
	    system("date $redir");
	    system($cmd.$redir) and
1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191
	      notify("$params/$command: ".
		     "Command '$cmd' failed, error $?: $!\n");

	    # Set up a timeout, so we retry if we don't get SHUTDOWN in time
	    foreach $node (@nodes) {
		# Note: This will replace any state timeouts currently in
		# the queue. But here that's okay because we're expecting
		# to see another transition really soon anyway.
		setTimeout($TBCOMMAND,$command,$node,time());
	    }

	    info("Performed $command for $params\n");
	    next;
	};
	(/^$TBPOWEROFF$/ || /^$TBPOWERON$/ || /^$TBPOWERCYCLE$/) && do {
	    # For power, the params is a comma-separated list of nodes
	    my @nodes = split(",",$params);
	    my $nodelist=join(" ",@nodes);
	    my %funcmap = ( $TBPOWERCYCLE => "cycle",
			    $TBPOWERON    => "on",
			    $TBPOWEROFF   => "off");
	    my $func = $funcmap{$command};