stated.in 53.2 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);
Mac Newbold's avatar
Mac Newbold committed
50

51 52 53 54 55
# 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
56
# Number of iterations (roughly, seconds) after which we'll reload
Robert Ricci's avatar
Robert Ricci committed
57 58
# information from the database. This is so we don't end up with information
# that's _too_ out of sync.
59
my $reload_time = 600;
60
my $last_reload = time;
Robert Ricci's avatar
Robert Ricci committed
61

62 63 64
# Handling of SECVIOLATIONS, eventually controlled by a sitevar
my $soft_secviolation = 1;

65 66 67
# Command line opts.
my $dbtag = "";
my $debug = 0;
68 69
my $server = "localhost";
my $port   = @BOSSEVENTPORT@;
70 71
my $lockfile;
my $pidfile;
72
my $eventchild;
73

Robert Ricci's avatar
Robert Ricci committed
74 75 76
# Process command-line arguments

sub usage {
Mac Newbold's avatar
Mac Newbold committed
77
    print << "END";
78
Usage: $0 [-h] [-d] [-s server] [-p port] [-t dbtag]
79
-h              This message
80 81
-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
82 83
-s server       Use specified server, instead of this site's bossnode
-p port	        Use specified port
84
Send SIGHUP to reload database state, or SIGUSR1 to restart completely.
Robert Ricci's avatar
Robert Ricci committed
85
END
86
    exit(1);
Robert Ricci's avatar
Robert Ricci committed
87 88
}

Mac Newbold's avatar
Mac Newbold committed
89
# Only root should run this - it won't work when run as a user...
90
# (Or, let an admin run it if it isn't the real one in /usr/testbed/ )
91
if ($UID && ( $TB eq $REALTB || ! TBAdmin($UID) ) ) {
Mac Newbold's avatar
Mac Newbold committed
92 93 94
    die("Only root can run this script!\n");
}

95
my @args = @ARGV;    # save a copy for restart before we mess with them.
Robert Ricci's avatar
Robert Ricci committed
96
my %opt = ();
97
if (!getopts("ds:p:ht:l",\%opt)) { usage(); }
Robert Ricci's avatar
Robert Ricci committed
98

Mac Newbold's avatar
Mac Newbold committed
99 100 101 102 103 104
if ($opt{h}) {
    exit &usage;
}
if (@ARGV) {
    exit &usage;
}
Robert Ricci's avatar
Robert Ricci committed
105

Mac Newbold's avatar
Mac Newbold committed
106 107 108 109 110 111
if ($opt{s}) {
    $server = $opt{s};
}
if ($opt{p}) {
    $port = $opt{p};
}
112 113 114
if ($opt{t}) {
    $dbtag = $opt{t};
}
Mac Newbold's avatar
Mac Newbold committed
115 116 117
if ($opt{d}) {
    $debug = 1;
}
Robert Ricci's avatar
Robert Ricci committed
118

119
# Grab some constants into variables
120
my $TBANYMODE    = TBDB_NODEOPMODE_ANY;
121 122
my $TBRESET      = TBDB_TBCONTROL_RESET;
my $TBTIMEOUT    = TBDB_TBCONTROL_TIMEOUT;
Mac Newbold's avatar
Mac Newbold committed
123 124 125
my $PXEBOOT      = TBDB_TBCONTROL_PXEBOOT;
my $BOOTING      = TBDB_TBCONTROL_BOOTING;
my $CHECKGENISUP = TBDB_TBCONTROL_CHECKGENISUP;
126 127 128
my $TBNOTIMEOUT  = TBDB_NO_STATE_TIMEOUT;
my $TBNODESTATE  = TBDB_TBEVENT_NODESTATE;
my $TBNODEOPMODE = TBDB_TBEVENT_NODEOPMODE;
129 130 131 132 133 134
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;
135
my $TBISUP       = TBDB_NODESTATE_ISUP;
136 137 138
my $PXEWAIT      = TBDB_NODESTATE_PXEWAIT;
my $PXEWAKEUP    = TBDB_NODESTATE_PXEWAKEUP;
my $PXEBOOTING   = TBDB_NODESTATE_PXEBOOTING;
139 140
my $TBRELOADDONEV1    = TBDB_TBCONTROL_RELOADDONE;
my $TBRELOADDONEV2    = TBDB_TBCONTROL_RELOADDONE_V2;
141 142 143 144
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;
145

146 147
# Special PXEBOOT state machine that all local nodes use.
my $PXEKERNEL	 = "PXEKERNEL";
148

149 150
# Even special-er SECUREBOOT state machine that local nodes may use.
my $SECUREBOOT	 = "SECUREBOOT";
151

152 153 154 155
# Protos.
sub debug(@);
sub fatal($);
sub notify($;$);
156
sub info($);
157 158 159 160 161 162 163
sub getTimeouts();
sub getValid();
sub getModeTrans();
sub getTriggers();
sub readStates(;@);
sub handleCtrlEvent($$);
sub reload();
164 165
sub StartEvents();
sub PollEvents($$);
166
    
167 168 169
if (!$debug) {
    if ( $TB eq $REALTB ) {
	$pidfile = "/var/run/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
170
    } else {
171
	$pidfile = "$TB/locks/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
172
    }
173
    debug("Using pidfile $pidfile\n");
Mac Newbold's avatar
Mac Newbold committed
174

175 176 177 178 179 180 181 182 183 184 185 186 187 188 189
    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: $? $!\n");
    }
    # Background
190
    if (TBBackGround($LOGFILE)) {
Mac Newbold's avatar
Mac Newbold committed
191 192
	exit(0);
    }
193
    TBdbfork();
Robert Ricci's avatar
Robert Ricci committed
194
}
195 196 197 198 199 200 201 202 203

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
204

205 206 207
# Change my $0 so that it is easier to see in a ps/top
$0 = "$0";

208 209 210 211 212
#
# Start up the event system interface.
#
if (StartEvents() != 0) {
    fatal("Error starting events");
Robert Ricci's avatar
Robert Ricci committed
213
}
214 215
# We want to exit on any warning. 
$SIG{__WARN__} = sub { print STDERR $_[0]; exit(-1); };
Robert Ricci's avatar
Robert Ricci committed
216 217 218

# Read in the pre-existing node states, and timeout and valid transition
# information from the database
219 220 221 222
my %timeouts  = getTimeouts();
my %valid     = getValid();
my %modeTrans = getModeTrans();
my %triggers  = getTriggers();
223
my %nodes     = readStates();
224
my %timeouttag= ();
225
if ($debug) { qshow(); }
Robert Ricci's avatar
Robert Ricci committed
226 227 228

# Gets set if a reload of state from the database should happen.
my $do_reload = 0;
229 230
my $sigrestart= 0;
my $sigcleanup= 0;
231
my $exiting   = 0;
Robert Ricci's avatar
Robert Ricci committed
232 233 234 235

# 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.
236
$SIG{HUP}  = sub { info("SIGHUP - Reloading DB state\n"); $do_reload = 1; };
237
$SIG{USR2} = sub { info("SIGUSR2 - Reopening logfile\n"); ReOpenLog($LOGFILE); };
238

Mac Newbold's avatar
Mac Newbold committed
239
# Set up other signals.
240 241 242 243 244
$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
245

246 247 248
# Track if I handled an event or not
my $event_count = 0;

249 250
# Control how long I block while waiting for events
my $blockwait=0;
251
my $nextdeadline=0;
252 253
my $mailqueue=0;

254 255 256 257 258
notify("Stated starting up\n");

sub process_event_queue() {
    $event_count=0;
    my $lastcount=-1;
259 260 261 262 263 264 265 266 267 268 269 270 271
    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;
272 273
	    } else {
		$wait = 0;
274 275 276 277 278
	    }
	}
    } else {
	# mail is waiting. Only block until it is time to send it.
	$wait = $lastmail + $mailgap - $now;
Mac Newbold's avatar
Mac Newbold committed
279
	debug("Now $now, mailgap $mailgap, last $lastmail ==> wait $wait\n");
280 281 282
    }
    if ($wait < 0) { debug("Wait was $wait!\n"); $wait=0; }
    my $finish = $now + $wait;
Mac Newbold's avatar
Mac Newbold committed
283 284
    while (($event_count != $lastcount || $wait > 0) &&
	   !($sigrestart || $sigcleanup || $do_reload)) {
285
	$lastcount = $event_count;
Mac Newbold's avatar
Mac Newbold committed
286 287
	# Don't block if we got a signal!
	if ($wait<=0 || $sigrestart || $sigcleanup || $do_reload) {
288
	    PollEvents(0, 0);
289
	} else {
290
	    #debug("Using blocking event poll - $wait seconds\n");
291
	    # timeout param is in milliseconds, so multiply
292
	    PollEvents(1, $wait*1000);
293 294 295
	    $now = time();
	    # subtract seconds elapsed from my wait time
	    $wait = $finish - $now;
296
	    #debug("Finished blocking event poll - $wait seconds remain\n");
297
	    if ($event_count > 0 &&
Mac Newbold's avatar
Mac Newbold committed
298 299
		(qsize() > 0 || $mailqueue ||
		 $sigrestart || $sigcleanup || $do_reload)) {
300 301 302 303 304 305 306
		$blockwait=0;
		$wait=0;
		#debug("Cancelling wait - timeouts/msgs waiting, or HUP'd\n");
		#debug("---End Blocking Wait ---\n");
	    }
	}
	#debug("Wait is $wait\n");
307 308
    }
    if ($event_count > 0) {
Mac Newbold's avatar
Mac Newbold committed
309
	debug("Handled $event_count event(s).\n");
310 311
    }
}
Robert Ricci's avatar
Robert Ricci committed
312

313
# Now, we just poll for events, and watch for timeouts
Robert Ricci's avatar
Robert Ricci committed
314
while (1) {
Mac Newbold's avatar
Mac Newbold committed
315
    my $now = time();
316 317 318 319
    my ($deadline,$node);

    # Check for nodes that have passed their timeout
    if (!qhead($deadline,$node)) {
320
	info("HEAD: $node in ".($deadline-$now).", queue=".qsize()."\n");
321 322
	while ($now >= $deadline && $node ne "") {
	    qpop($deadline,$node);
323
	    info("POP: $node in ".($deadline-$now).", queue=".qsize()."\n");
324
	    handleCtrlEvent($node,$TBTIMEOUT);
325 326 327 328
	    if (0) { qshow(); }
	    if (qhead($deadline,$node)) {
		$deadline=0; $node="";
	    }
329
	}
330 331
    } else {
	$deadline=0;
332
    }
333
    $nextdeadline = $deadline;
334

335 336 337 338
    if (qsize()==0) {
	$blockwait=1;
	debug("---Blocking wait okay---\n");
    }
Mac Newbold's avatar
Mac Newbold committed
339

Mac Newbold's avatar
Mac Newbold committed
340 341 342 343
    if ($do_reload || ($now - $last_reload > $reload_time)) {
	reload();
	$do_reload = 0;
    }
Mac Newbold's avatar
Mac Newbold committed
344

Mac Newbold's avatar
Mac Newbold committed
345 346
    # Send any messages in the queue if it is time
    notify("",1);
Mac Newbold's avatar
Mac Newbold committed
347

348 349 350
    if ($sigrestart) { restart(); }
    if ($sigcleanup) { cleanup(); }

351
    process_event_queue;
Robert Ricci's avatar
Robert Ricci committed
352 353
}

Mac Newbold's avatar
Mac Newbold committed
354 355
exit(0);

Robert Ricci's avatar
Robert Ricci committed
356
# Read the current states of nodes from the database
357
sub readStates(;@) {
Mac Newbold's avatar
Mac Newbold committed
358 359 360
    my %oldnodes = @_;

    # Guard against undefined variable warnings
361
    if (! %oldnodes) {
Mac Newbold's avatar
Mac Newbold committed
362 363
	%oldnodes = ();
    }
364

Mac Newbold's avatar
Mac Newbold committed
365 366 367
    #debug("readStates called\n");
    my $result = DBQueryFatal("SELECT node_id, eventstate, " .
			      "state_timestamp, op_mode, " .
368
			      "op_mode_timestamp, stated_tag FROM nodes ".
369
			      "where node_id not like 'sh%'");
Mac Newbold's avatar
Mac Newbold committed
370 371

    my %nodes;
372
    while (my ($node_id, $state, $timestamp, $mode, $mode_timestamp, $tag)
Mac Newbold's avatar
Mac Newbold committed
373
	   = $result->fetchrow()) {
374 375 376 377 378 379 380 381 382 383 384
	$nodes{$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
385

386
	#
Mac Newbold's avatar
Mac Newbold committed
387 388 389 390
	# 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.)
391
	#
Mac Newbold's avatar
Mac Newbold committed
392 393 394 395 396
	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};
397
	} else {
Mac Newbold's avatar
Mac Newbold committed
398 399 400 401
	    $nodes{$node_id}{state}          = $state;
	    $nodes{$node_id}{timestamp}      = $timestamp;
	    $nodes{$node_id}{mode}           = $mode;
	    $nodes{$node_id}{mode_timestamp} = $mode_timestamp;
402 403 404
	    $nodes{$node_id}{notified}       = 0;
	    $nodes{$node_id}{timedout}       = 0;
	    $nodes{$node_id}{noretry}        = 0;
405
	    # Is there a timeout? If so, set it up!
406 407 408
	    if ($dbtag eq $nodes{$node_id}{"tag"}) {
		setTimeout($mode,$state,$node_id,$timestamp);
	    }
409
	}
Mac Newbold's avatar
Mac Newbold committed
410 411
    }
    return %nodes;
Robert Ricci's avatar
Robert Ricci committed
412 413 414 415 416 417
}

#
# Read timeouts for various states from the database
#
sub getTimeouts() {
Mac Newbold's avatar
Mac Newbold committed
418 419 420
    #debug("getTimeouts called\n");
    my $result = DBQueryFatal("SELECT op_mode, state, timeout, action " .
			      "FROM state_timeouts");
Robert Ricci's avatar
Robert Ricci committed
421

Mac Newbold's avatar
Mac Newbold committed
422 423 424 425 426
    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
427 428 429 430 431 432
}

#
# Read the list of valid state transitions from the database
#
sub getValid() {
Mac Newbold's avatar
Mac Newbold committed
433 434 435
    #debug("getValid called\n");
    my $result = DBQueryFatal("SELECT op_mode, state1, state2 " .
			      "FROM state_transitions");
Robert Ricci's avatar
Robert Ricci committed
436

Mac Newbold's avatar
Mac Newbold committed
437 438 439 440 441
    my %valid;
    while (my ($mode,$state1, $state2) = $result->fetchrow()) {
	$valid{$mode}{$state1}{$state2} = 1;
    }
    return %valid;
Robert Ricci's avatar
Robert Ricci committed
442 443
}

444 445 446 447
#
# Read the list of valid mode transitions from the database
#
sub getModeTrans() {
Mac Newbold's avatar
Mac Newbold committed
448
    #debug("getModeTrans called\n");
Mac Newbold's avatar
Mac Newbold committed
449
    my $result =
Mac Newbold's avatar
Mac Newbold committed
450 451 452 453 454 455 456 457 458 459 460
      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;
461
	}
Mac Newbold's avatar
Mac Newbold committed
462 463
    }
    return %modeTrans;
464 465 466 467 468 469
}

#
# Read the list of states which trigger an action
#
sub getTriggers() {
470
    debug("getTriggers called\n");
Mac Newbold's avatar
Mac Newbold committed
471

472 473
    debug("anymode ==> '$TBANYMODE'\n");

Mac Newbold's avatar
Mac Newbold committed
474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494
    # 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 $result =
495
      DBQueryFatal("SELECT op_mode, state, `trigger` " .
496 497
		   "FROM state_triggers where node_id='$TBANYMODE' ".
		   "order by op_mode,state");
Mac Newbold's avatar
Mac Newbold committed
498
    my %t;
499
    while (my ($mode, $state, $trig) = $result->fetchrow()) {
Mac Newbold's avatar
Mac Newbold committed
500 501 502
	my @trigs = split(/\s*,\s*/,$trig);
	$t{"$mode:$state"} = \@trigs;
	debug("trig($mode:$state)\t => ".join(',',@trigs)."\n");
503 504
    }

Mac Newbold's avatar
Mac Newbold committed
505 506
    # Grab per-node triggers (including any-mode)
    $result =
507
      DBQueryFatal("SELECT node_id, op_mode, state, `trigger` " .
508 509 510 511 512 513
		   "FROM state_triggers where node_id!='$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
514
    }
515

Mac Newbold's avatar
Mac Newbold committed
516 517
    debug(hash_recurse(%t));

Mac Newbold's avatar
Mac Newbold committed
518
    return %t;
519 520
}

Robert Ricci's avatar
Robert Ricci committed
521 522 523 524
#
# Gets called for every event that we recieve
#
sub handleEvent($$$) {
525
    my ($objtype,$objname,$eventtype) = @_;
Mac Newbold's avatar
Mac Newbold committed
526
    $event_count++;
Mac Newbold's avatar
Mac Newbold committed
527

528
    #
Mac Newbold's avatar
Mac Newbold committed
529 530 531
    # 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.
532
    #
533
    if ($dbtag eq "") {
534 535
	debug("Got an event: ($objtype,$objname,$eventtype)\n");
    }
Mac Newbold's avatar
Mac Newbold committed
536 537 538 539

    #
    # Check to see if another instance is supposed to be handling this node
    #
540 541
    if ($objtype ne $TBCOMMAND) {
	my $node = $objname;
Mac Newbold's avatar
Mac Newbold committed
542

543 544 545 546 547 548 549
	#
	# 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
550
	    # churning via reload(). Bad.
551
	    if (! defined($nodes{$node})) {
552 553 554 555
		notify("Got $objtype/$eventtype for nonexistent $node!\n");
		return;
	    }
	}
Mac Newbold's avatar
Mac Newbold committed
556

557 558
	#
	# If a stated_tag was specified on the command line, ignore those
Mac Newbold's avatar
Mac Newbold committed
559
	# nodes that do not match.
560
	#
Mac Newbold's avatar
Mac Newbold committed
561
	#debug("dbtag='$dbtag', node $node='".$nodes{$node}{"tag"}."'\n");
562 563 564 565 566 567 568 569 570 571
	if ($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
572
    }
Mac Newbold's avatar
Mac Newbold committed
573 574 575 576 577

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

 SWITCH: for ($objtype) {

Mac Newbold's avatar
Mac Newbold committed
584 585 586 587 588 589 590 591 592 593 594 595 596 597
	(/$TBNODESTATE/) && do {
	    stateTransition($objname,$eventtype);
	    last;
	};
	(/$TBNODEOPMODE/) && do {
	    opModeTransition($objname,$eventtype);
	    notify("Use of deprecated event TBNODEOPMODE:\n".
		   "$objname->$eventtype\n");
	    last;
	};
	(/$TBCONTROL/) && do {
	    handleCtrlEvent($objname,$eventtype);
	    last;
	};
598 599 600 601
	(/$TBCOMMAND/) && do {
	    handleCommand($objname,$eventtype);
	    last;
	};
602

Mac Newbold's avatar
Mac Newbold committed
603
    }
604 605 606 607 608

}

sub stateTransition($$) {

609
    my ($node,$newstate) = @_;
Robert Ricci's avatar
Robert Ricci committed
610

611 612
    # Check for invalid transitions
    my ($oldstate, $mode);
613 614 615
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

616 617 618
    if ($oldstate && $mode && $valid{$mode} && $valid{$mode}{$oldstate} &&
	!$valid{$mode}{$oldstate}{$newstate}) {
	notify("Invalid transition for node $node from $mode/$oldstate " .
619
	       "to $newstate\n");
620 621 622 623 624 625
	#
	# Machines in the secure boot path are not allowed to jump
	# willy-nilly into unknown states.
	#
        if ($mode eq TBDB_NODEOPMODE_SECUREBOOT ||
	    $mode eq TBDB_NODEOPMODE_SECURELOAD) {
626 627 628
            $newstate = TBDB_NODESTATE_SECVIOLATION;
            notify("Moving $node to $newstate because it's in $mode\n");
        }
629 630 631 632 633 634
    }
    #
    # Nodes that are in the SECVIOLATION state are not allowed to leave!
    #
    if ($oldstate eq TBDB_NODESTATE_SECVIOLATION &&
	$newstate ne TBDB_NODESTATE_SECVIOLATION) {
635 636 637 638 639 640 641 642 643 644 645 646 647
	#
	# Allow transitions to SHUTDOWN.
	# This allows someone to reboot a node in the SECVIOLATION state
	# getting it back to MINIMAL/SHUTDOWN.
	#
	# XXX DEBUG ONLY!
	# 
	if ($soft_secviolation && $newstate eq TBDB_NODESTATE_SHUTDOWN) {
	    notify("$node allowed to transition: SECVIOLATION => SHUTDOWN\n");
	} else {
	    notify("$node tried to leave SECVIOLATION (to $newstate)\n");
	    $newstate = TBDB_NODESTATE_SECVIOLATION;
	}
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   => $node);
		next;
	    };
868
	    (/^$TBREBOOT$/ || /^$TBPOWERCYCLE$/ || /^$TBPOWEROFF$/) && do {
869
		handleCommand($node,$trig);
870 871 872
		next;
	    };
            (/^EMAILNOTIFY$/) && do {
873 874
		my $msg = "$node entered state $mode/$newstate from " .
		    "$mode/$oldstate";
875
		my $dest = $REALTBOPS;
876
		if ($newstate eq TBDB_NODESTATE_SECVIOLATION) {
877 878 879 880 881 882 883 884
		    if ($soft_secviolation) {
			$msg .= "\n\nNode $node was allowed to continue.\n";
			$dest = $TBOPS;
		    } else {
			$msg .= "\n\nNode $node has been powered off.\n" .
			        "You must address the cause of the violation ".
				"and reset the eventstate before powering on.";
		    }
885
		}
886 887
		SENDMAIL($dest, "STATED: $node entered state $newstate",
			 $msg, "Stated Daemon <".$TBOPS.">");
888
		next;
889
            };
890 891 892 893 894 895 896 897 898 899 900 901
	    /^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;
	    };
902
	    notify("Unknown trigger '$trig' for $node in $mode/$newstate!\n");
903
	}
Mac Newbold's avatar
Mac Newbold committed
904 905 906 907 908 909 910
	# 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);
	}
911
    }
912

913 914 915 916 917 918 919 920
    # 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
921 922 923
	} else {
	    debug("No next mode.\n");
	}
924 925
    }
}
926

927
sub opModeTransition($$;$) {
Mac Newbold's avatar
Mac Newbold committed
928

929 930
    my ($node,$newmode,$force) = @_;
    if (!defined($force)) { $force = 0; }
Mac Newbold's avatar
Mac Newbold committed
931

932
    info("$node: Mode change to $newmode requested ($force)\n");
Mac Newbold's avatar
Mac Newbold committed
933

934 935
    # Check for invalid transitions
    my ($oldstate, $mode, $nextstate);
936 937 938
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

939
    if (defined($modeTrans{"$mode:$oldstate"}) || $force) {
940
	if (!$force) {
941 942 943 944 945 946 947 948 949 950 951 952 953 954 955
	    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");
	    }
956 957
	}
    } else {
958
	notify("Invalid mode transition for $node from $mode/$oldstate: ".
959 960
	       "Not a valid mode transition state!\n");
    }
Mac Newbold's avatar
Mac Newbold committed
961 962 963
    if (!$nextstate) {
	$nextstate=$oldstate;
    }
Mac Newbold's avatar
Mac Newbold committed
964

965 966 967 968 969 970
    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
971

972 973 974 975 976
    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
977 978 979 980

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

981 982 983 984
}

sub handleCtrlEvent($$) {
    my ($node,$event) = @_;
985

986
    info("CtrlEvent: $node, $event\n");
987

988 989
    foreach ($event) {
	/^$TBRESET$/ && do {
990 991
	    #
	    # Clear next_boot_path with os_select.
Mac Newbold's avatar
Mac Newbold committed
992
	    #
993
	    $cmd = "$osselect -d -c -1 $node";
994
	    system($cmd) and
995
		notify("$node/$event: Could not clear next_boot_path!\n");
Mac Newbold's avatar
Mac Newbold committed
996 997

	    info("Performed $TBRESET for $node\n");
998 999
	    next;
	};
1000
	(/^$TBRELOADDONEV1$/ || /^$TBRELOADDONEV2$/) && do {
1001 1002 1003 1004 1005 1006 1007
	    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'");
1008
		DBQueryFatal("delete from reserved where node_id='$node'");
1009 1010
		TBSetNodeHistory($node, TB_NODEHISTORY_OP_FREE,
				 $UID, $pid, $eid);
1011
		info("Released $node from $pid/$eid\n");
1012
	    }
1013 1014 1015 1016 1017
	    if ($event eq $TBRELOADDONEV2) {
		info("Sending an apod to $node\n");
		system("$apod $node") == 0 or
		    notify("Could not apod $node after $TBRELOADDONEV2!\n");
	    }
1018 1019 1020
	    next;
	};
	/^$TBTIMEOUT$/ && do {
1021 1022 1023 1024
	    my ($mode,$state) = split(":",$timeout_tag{$node});
	    delete($timeout_tag{$node});
	    my $curstate = $nodes{$node}{state};
	    my $curmode = $nodes{$node}{mode};
1025
	    my ($timeout,$action);
1026 1027 1028 1029 1030
	    if (!defined($nodes{$node}{notified})) {
		$nodes{$node}{notified}=0;
	    }
	    $nodes{$node}{notified}++;
	    my $notified = $nodes{$node}{notified};
1031 1032
	    $nodes{$node}{timedout}++;
	    my $timedout = $nodes{$node}{timedout};
1033 1034 1035 1036
	    if ($mode && $state && $timeouts{$mode} &&
		$timeouts{$mode}{$state}) {
		($timeout, $action) = @{$timeouts{$mode}{$state}};
	    }
1037 1038
	    if ($mode eq $TBCOMMAND) {
		# It is a command, not a true state
1039
		if ($action eq $TBTIMEOUTCMDRETRY) {
1040
		    # Retry the command
1041
		    notify("$node: Command $state, retry #$timedout\n");
1042
		    # notify in case we get in a retry loop...
1043
		    handleCommand($node,$state,$timedout,1);
1044 1045 1046 1047 1048 1049
		} else {
		    notify("$node: Unknown timeout action for ".
			   "$mode/$state: '$action'\n");
		}
		next;
	    }
1050 1051 1052 1053 1054 1055 1056

	    #
	    # 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
1057

1058 1059
		if ($timedout < 3) {
		    #
Mac Newbold's avatar
Mac Newbold committed
1060 1061
		    # Try again.
		    #
1062 1063 1064 1065 1066 1067
		    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
1068 1069
		    # Failed too many times, power cycle instead.
		    #
1070 1071 1072 1073 1074 1075 1076
		    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");
1077
		system("(date; $cmd) >>$rebootlog 2>&1 &") and
1078
		    notify("$PXEWAKEUP retry: ".
1079 1080 1081 1082 1083
			   "Command '$cmd' failed, error $?: $!\n");

		next;
	    }

1084 1085 1086 1087 1088 1089
	    info("Node $node has timed out in state $mode/$state".
		 ($action ne "" ? "\n\tRequested action $action." : "").
		 "\n");

	    foreach ($action) {
		/^$TBTIMEOUTREBOOT/ && do {
1090 1091 1092 1093 1094
		    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;
1095
		    } else {
1096 1097 1098 1099 1100 1101
			# 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);
1102 1103 1104 1105 1106
		    }
		    last; };
		/^$TBTIMEOUTNOTIFY/ && do {
		    notify("Node $node has timed out in state $mode/$state\n");
		    last; };
1107 1108 1109 1110 1111
                /^STATE:([-\w\/]+)$/ && do {
                    my $newstate = $1;
                    # Force the node into a new state
                    stateTransition($node,$newstate);
                    last; };
1112 1113
		notify("$node: Unknown Timeout Action: $action\n");
	    }
1114 1115
	    next;
	};
1116
	notify("$node: Unknown CtrlEvent: $event\n");
1117 1118
    }
}
Robert Ricci's avatar
Robert Ricci committed
1119

1120 1121
sub handleCommand($$;$$) {
    my ($params,$command,$retry,$force) = @_;
1122
    if (!defined($retry)) { $retry=0; }
1123
    if (!defined($force)) { $force=0; }
1124 1125 1126 1127 1128 1129 1130 1131

    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.

1132
    if ($command eq $TBREBOOT && $retry >=4) {
1133 1134
	announce("Node $params has tried rebooting $retry times and has \n".
		 "still not been successful. Please look into it soon.\n".
1135 1136 1137
	"" );#	 "In the meantime, $params will be powered off.\n");
	# Just return...
	return 0;
1138
	# change my command to poweroff.
1139
	#$command = $TBPOWEROFF;
1140 1141 1142 1143 1144 1145
    }

    foreach ($command) {
	/^$TBREBOOT$/ && do {
	    # For reboot, the params is a comma-separated list of nodes
	    my @nodes = split(",",$params);
1146 1147 1148 1149 1150 1151 1152 1153

	    # 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");
1154
		    if (($nodes{$node}{state} ne TBDB_NODESTATE_ISUP) &&
1155
			($nodes{$node}{state} ne TBDB_NODESTATE_TBFAILED) &&
1156
			(!$nodes{$node}{noretry}) ) {
1157
			# This node shouldn't be rebooted now...
1158 1159
			# XXX Send feedback here somehow!

1160 1161 1162 1163 1164 1165 1166 1167 1168 1169
			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; }
	    }<