stated.in 47.5 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

143 144 145 146
# Protos.
sub debug(@);
sub fatal($);
sub notify($;$);
147
sub info($);
148 149 150 151 152 153 154 155
sub getTimeouts();
sub getValid();
sub getModeTrans();
sub getTriggers();
sub readStates(;@);
sub handleCtrlEvent($$);
sub reload();
    
156 157 158
if (!$debug) {
    if ( $TB eq $REALTB ) {
	$pidfile = "/var/run/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
159
    } else {
160
	$pidfile = "$TB/locks/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
161
    }
162
    debug("Using pidfile $pidfile\n");
Mac Newbold's avatar
Mac Newbold committed
163

164 165 166 167 168 169 170 171 172 173 174 175 176 177 178
    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: $? $!\n");
    }
    # Background
179
    if (TBBackGround($LOGFILE)) {
Mac Newbold's avatar
Mac Newbold committed
180 181
	exit(0);
    }
182
    TBdbfork();
Robert Ricci's avatar
Robert Ricci committed
183
}
184 185 186 187 188 189 190 191 192

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
193

194 195 196
# Change my $0 so that it is easier to see in a ps/top
$0 = "$0";

Robert Ricci's avatar
Robert Ricci committed
197
my $URL = "elvin://$server";
Mac Newbold's avatar
Mac Newbold committed
198 199 200
if ($port) {
    $URL .= ":$port";
}
Robert Ricci's avatar
Robert Ricci committed
201

Mac Newbold's avatar
Mac Newbold committed
202
# Connect to the event system, and subscribe the the events we want
Robert Ricci's avatar
Robert Ricci committed
203
my $handle = event_register($URL,0);
Mac Newbold's avatar
Mac Newbold committed
204 205 206
if (!$handle) {
    fatal("Unable to register with event system\n");
}
Robert Ricci's avatar
Robert Ricci committed
207 208

my $tuple = address_tuple_alloc();
Mac Newbold's avatar
Mac Newbold committed
209 210 211
if (!$tuple) {
    fatal("Could not allocate an address tuple\n");
}
Robert Ricci's avatar
Robert Ricci committed
212

213 214 215
%$tuple = ( objtype => join(",",
			    $TBNODESTATE, $TBNODEOPMODE,
			    $TBCONTROL, $TBCOMMAND) );
216

Robert Ricci's avatar
Robert Ricci committed
217
if (!event_subscribe($handle,\&handleEvent,$tuple)) {
Mac Newbold's avatar
Mac Newbold committed
218
    fatal("Could not subscribe to events\n");
Robert Ricci's avatar
Robert Ricci committed
219 220 221 222
}

# Read in the pre-existing node states, and timeout and valid transition
# information from the database
223 224 225 226
my %timeouts  = getTimeouts();
my %valid     = getValid();
my %modeTrans = getModeTrans();
my %triggers  = getTriggers();
227
my %nodes     = readStates();
228
my %timeouttag= ();
229
if ($debug) { qshow(); }
Robert Ricci's avatar
Robert Ricci committed
230 231 232

# Gets set if a reload of state from the database should happen.
my $do_reload = 0;
233 234
my $sigrestart= 0;
my $sigcleanup= 0;
235
my $exiting   = 0;
Robert Ricci's avatar
Robert Ricci committed
236 237 238 239

# 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.
240
$SIG{HUP}  = sub { info("SIGHUP - Reloading DB state\n"); $do_reload = 1; };
241
$SIG{USR2} = sub { info("SIGUSR2 - Reopening logfile\n"); ReOpenLog($LOGFILE); };
242

Mac Newbold's avatar
Mac Newbold committed
243
# Set up other signals.
244 245 246 247 248
$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
249

250 251 252
# Track if I handled an event or not
my $event_count = 0;

253 254
# Control how long I block while waiting for events
my $blockwait=0;
255
my $nextdeadline=0;
256 257
my $mailqueue=0;

258 259 260 261 262
notify("Stated starting up\n");

sub process_event_queue() {
    $event_count=0;
    my $lastcount=-1;
263 264 265 266 267 268 269 270 271 272 273 274 275
    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;
276 277
	    } else {
		$wait = 0;
278 279 280 281 282
	    }
	}
    } else {
	# mail is waiting. Only block until it is time to send it.
	$wait = $lastmail + $mailgap - $now;
Mac Newbold's avatar
Mac Newbold committed
283
	debug("Now $now, mailgap $mailgap, last $lastmail ==> wait $wait\n");
284 285 286
    }
    if ($wait < 0) { debug("Wait was $wait!\n"); $wait=0; }
    my $finish = $now + $wait;
Mac Newbold's avatar
Mac Newbold committed
287 288
    while (($event_count != $lastcount || $wait > 0) &&
	   !($sigrestart || $sigcleanup || $do_reload)) {
289
	$lastcount = $event_count;
Mac Newbold's avatar
Mac Newbold committed
290 291
	# Don't block if we got a signal!
	if ($wait<=0 || $sigrestart || $sigcleanup || $do_reload) {
292 293
	    event_poll($handle);
	} else {
294
	    #debug("Using blocking event poll - $wait seconds\n");
295 296 297 298 299
	    # timeout param is in milliseconds, so multiply
	    event_poll_blocking($handle, $wait*1000);
	    $now = time();
	    # subtract seconds elapsed from my wait time
	    $wait = $finish - $now;
300
	    #debug("Finished blocking event poll - $wait seconds remain\n");
301
	    if ($event_count > 0 &&
Mac Newbold's avatar
Mac Newbold committed
302 303
		(qsize() > 0 || $mailqueue ||
		 $sigrestart || $sigcleanup || $do_reload)) {
304 305 306 307 308 309 310
		$blockwait=0;
		$wait=0;
		#debug("Cancelling wait - timeouts/msgs waiting, or HUP'd\n");
		#debug("---End Blocking Wait ---\n");
	    }
	}
	#debug("Wait is $wait\n");
311 312
    }
    if ($event_count > 0) {
Mac Newbold's avatar
Mac Newbold committed
313
	debug("Handled $event_count event(s).\n");
314 315
    }
}
Robert Ricci's avatar
Robert Ricci committed
316

317
# Now, we just poll for events, and watch for timeouts
Robert Ricci's avatar
Robert Ricci committed
318
while (1) {
Mac Newbold's avatar
Mac Newbold committed
319
    my $now = time();
320 321 322 323
    my ($deadline,$node);

    # Check for nodes that have passed their timeout
    if (!qhead($deadline,$node)) {
324
	info("HEAD: $node in ".($deadline-$now).", queue=".qsize()."\n");
325 326
	while ($now >= $deadline && $node ne "") {
	    qpop($deadline,$node);
327
	    info("POP: $node in ".($deadline-$now).", queue=".qsize()."\n");
328
	    handleCtrlEvent($node,$TBTIMEOUT);
329 330 331 332
	    if (0) { qshow(); }
	    if (qhead($deadline,$node)) {
		$deadline=0; $node="";
	    }
333
	}
334 335
    } else {
	$deadline=0;
336
    }
337
    $nextdeadline = $deadline;
338

339 340 341 342
    if (qsize()==0) {
	$blockwait=1;
	debug("---Blocking wait okay---\n");
    }
Mac Newbold's avatar
Mac Newbold committed
343

Mac Newbold's avatar
Mac Newbold committed
344 345 346 347
    if ($do_reload || ($now - $last_reload > $reload_time)) {
	reload();
	$do_reload = 0;
    }
Mac Newbold's avatar
Mac Newbold committed
348

Mac Newbold's avatar
Mac Newbold committed
349 350
    # Send any messages in the queue if it is time
    notify("",1);
Mac Newbold's avatar
Mac Newbold committed
351

352 353 354
    if ($sigrestart) { restart(); }
    if ($sigcleanup) { cleanup(); }

355
    process_event_queue;
Robert Ricci's avatar
Robert Ricci committed
356 357
}

Mac Newbold's avatar
Mac Newbold committed
358 359
exit(0);

Robert Ricci's avatar
Robert Ricci committed
360
# Read the current states of nodes from the database
361
sub readStates(;@) {
Mac Newbold's avatar
Mac Newbold committed
362 363 364
    my %oldnodes = @_;

    # Guard against undefined variable warnings
365
    if (! %oldnodes) {
Mac Newbold's avatar
Mac Newbold committed
366 367
	%oldnodes = ();
    }
368

Mac Newbold's avatar
Mac Newbold committed
369 370 371
    #debug("readStates called\n");
    my $result = DBQueryFatal("SELECT node_id, eventstate, " .
			      "state_timestamp, op_mode, " .
372
			      "op_mode_timestamp, stated_tag FROM nodes ".
373
			      "where node_id not like 'sh%'");
Mac Newbold's avatar
Mac Newbold committed
374 375

    my %nodes;
376
    while (my ($node_id, $state, $timestamp, $mode, $mode_timestamp, $tag)
Mac Newbold's avatar
Mac Newbold committed
377
	   = $result->fetchrow()) {
378 379 380 381 382 383 384 385 386 387 388
	$nodes{$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
389

390
	#
Mac Newbold's avatar
Mac Newbold committed
391 392 393 394
	# 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.)
395
	#
Mac Newbold's avatar
Mac Newbold committed
396 397 398 399 400
	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};
401
	} else {
Mac Newbold's avatar
Mac Newbold committed
402 403 404 405
	    $nodes{$node_id}{state}          = $state;
	    $nodes{$node_id}{timestamp}      = $timestamp;
	    $nodes{$node_id}{mode}           = $mode;
	    $nodes{$node_id}{mode_timestamp} = $mode_timestamp;
406 407 408
	    $nodes{$node_id}{notified}       = 0;
	    $nodes{$node_id}{timedout}       = 0;
	    $nodes{$node_id}{noretry}        = 0;
409
	    # Is there a timeout? If so, set it up!
410 411 412
	    if ($dbtag eq $nodes{$node_id}{"tag"}) {
		setTimeout($mode,$state,$node_id,$timestamp);
	    }
413
	}
Mac Newbold's avatar
Mac Newbold committed
414 415
    }
    return %nodes;
Robert Ricci's avatar
Robert Ricci committed
416 417 418 419 420 421
}

#
# Read timeouts for various states from the database
#
sub getTimeouts() {
Mac Newbold's avatar
Mac Newbold committed
422 423 424
    #debug("getTimeouts called\n");
    my $result = DBQueryFatal("SELECT op_mode, state, timeout, action " .
			      "FROM state_timeouts");
Robert Ricci's avatar
Robert Ricci committed
425

Mac Newbold's avatar
Mac Newbold committed
426 427 428 429 430
    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
431 432 433 434 435 436
}

#
# Read the list of valid state transitions from the database
#
sub getValid() {
Mac Newbold's avatar
Mac Newbold committed
437 438 439
    #debug("getValid called\n");
    my $result = DBQueryFatal("SELECT op_mode, state1, state2 " .
			      "FROM state_transitions");
Robert Ricci's avatar
Robert Ricci committed
440

Mac Newbold's avatar
Mac Newbold committed
441 442 443 444 445
    my %valid;
    while (my ($mode,$state1, $state2) = $result->fetchrow()) {
	$valid{$mode}{$state1}{$state2} = 1;
    }
    return %valid;
Robert Ricci's avatar
Robert Ricci committed
446 447
}

448 449 450 451
#
# Read the list of valid mode transitions from the database
#
sub getModeTrans() {
Mac Newbold's avatar
Mac Newbold committed
452
    #debug("getModeTrans called\n");
Mac Newbold's avatar
Mac Newbold committed
453
    my $result =
Mac Newbold's avatar
Mac Newbold committed
454 455 456 457 458 459 460 461 462 463 464
      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;
465
	}
Mac Newbold's avatar
Mac Newbold committed
466 467
    }
    return %modeTrans;
468 469 470 471 472 473
}

#
# Read the list of states which trigger an action
#
sub getTriggers() {
474
    debug("getTriggers called\n");
Mac Newbold's avatar
Mac Newbold committed
475

476 477
    debug("anymode ==> '$TBANYMODE'\n");

Mac Newbold's avatar
Mac Newbold committed
478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498
    # 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 $result =
499
      DBQueryFatal("SELECT op_mode, state, `trigger` " .
500 501
		   "FROM state_triggers where node_id='$TBANYMODE' ".
		   "order by op_mode,state");
Mac Newbold's avatar
Mac Newbold committed
502
    my %t;
503
    while (my ($mode, $state, $trig) = $result->fetchrow()) {
Mac Newbold's avatar
Mac Newbold committed
504 505 506
	my @trigs = split(/\s*,\s*/,$trig);
	$t{"$mode:$state"} = \@trigs;
	debug("trig($mode:$state)\t => ".join(',',@trigs)."\n");
507 508
    }

Mac Newbold's avatar
Mac Newbold committed
509 510
    # Grab per-node triggers (including any-mode)
    $result =
511
      DBQueryFatal("SELECT node_id, op_mode, state, `trigger` " .
512 513 514 515 516 517
		   "FROM state_triggers where node_id!='$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
518
    }
519

Mac Newbold's avatar
Mac Newbold committed
520 521
    debug(hash_recurse(%t));

Mac Newbold's avatar
Mac Newbold committed
522
    return %t;
523 524
}

Robert Ricci's avatar
Robert Ricci committed
525 526 527 528
#
# Gets called for every event that we recieve
#
sub handleEvent($$$) {
Mac Newbold's avatar
Mac Newbold committed
529 530 531 532 533 534
    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
535

536
    #
Mac Newbold's avatar
Mac Newbold committed
537 538 539
    # 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.
540
    #
541
    if ($dbtag eq "") {
542 543
	debug("Got an event: ($objtype,$objname,$eventtype)\n");
    }
Mac Newbold's avatar
Mac Newbold committed
544 545 546 547

    #
    # Check to see if another instance is supposed to be handling this node
    #
548 549
    if ($objtype ne $TBCOMMAND) {
	my $node = $objname;
Mac Newbold's avatar
Mac Newbold committed
550

551 552 553 554 555 556 557
	#
	# 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
558
	    # churning via reload(). Bad.
559
	    if (! defined($nodes{$node})) {
560 561 562 563
		notify("Got $objtype/$eventtype for nonexistent $node!\n");
		return;
	    }
	}
Mac Newbold's avatar
Mac Newbold committed
564

565 566
	#
	# If a stated_tag was specified on the command line, ignore those
Mac Newbold's avatar
Mac Newbold committed
567
	# nodes that do not match.
568
	#
Mac Newbold's avatar
Mac Newbold committed
569
	#debug("dbtag='$dbtag', node $node='".$nodes{$node}{"tag"}."'\n");
570 571 572 573 574 575 576 577 578 579
	if ($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
580
    }
Mac Newbold's avatar
Mac Newbold committed
581 582 583 584 585

    #
    # 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.)
    #
586 587 588
    if ($dbtag ne "") {
	debug("Got an event: ($objtype,$objname,$eventtype)\n");
    }
Mac Newbold's avatar
Mac Newbold committed
589 590 591

 SWITCH: for ($objtype) {

Mac Newbold's avatar
Mac Newbold committed
592 593 594 595 596 597 598 599 600 601 602 603 604 605
	(/$TBNODESTATE/) && do {
	    stateTransition($objname,$eventtype);
	    last;
	};
	(/$TBNODEOPMODE/) && do {
	    opModeTransition($objname,$eventtype);
	    notify("Use of deprecated event TBNODEOPMODE:\n".
		   "$objname->$eventtype\n");
	    last;
	};
	(/$TBCONTROL/) && do {
	    handleCtrlEvent($objname,$eventtype);
	    last;
	};
606 607 608 609
	(/$TBCOMMAND/) && do {
	    handleCommand($objname,$eventtype);
	    last;
	};
610

Mac Newbold's avatar
Mac Newbold committed
611
    }
612 613 614 615 616

}

sub stateTransition($$) {

617
    my ($node,$newstate) = @_;
Robert Ricci's avatar
Robert Ricci committed
618

619 620
    # Check for invalid transitions
    my ($oldstate, $mode);
621 622 623
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

624 625 626
    if ($oldstate && $mode && $valid{$mode} && $valid{$mode}{$oldstate} &&
	!$valid{$mode}{$oldstate}{$newstate}) {
	notify("Invalid transition for node $node from $mode/$oldstate " .
627
	       "to $newstate\n");
628
    }
Robert Ricci's avatar
Robert Ricci committed
629

630 631 632 633
    my $now = time();
    $nodes{$node}{state}     = $newstate;
    $nodes{$node}{timestamp} = $now;
    $nodes{$node}{notified}  = 0;
634

635 636 637
    info("$node: $mode/$oldstate => $mode/$newstate\n");
    DBQueryFatal("UPDATE nodes SET eventstate='$newstate', " .
		 "state_timestamp='$now' WHERE node_id='$node'");
638

639 640 641 642
    # 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} =~ /^$TBCOMMAND:/) {
643
        debug("TimeoutTag = '$timeout_tag{$node}'\n");
644
	my ($str,$cmd) = split(":",$timeout_tag{$node});
645
	debug("str=$str\tcmd=$cmd\tTBREBOOT=$TBREBOOT\tstate=$newstate\n");
646
	if ($cmd eq $TBREBOOT) {
647
	    if ($newstate eq TBDB_NODESTATE_SHUTDOWN ) {
648 649 650
		info("$node: $TBREBOOT success\n");
		# Timeout will get cleared below by setTimeout call
	    } else {
651 652
		notify("$node: $TBREBOOT in progress, but got state ".
		       "$newstate instead of ".TBDB_NODESTATE_SHUTDOWN."!\n");
653 654 655 656 657
	    }
	#} elsif ($cmd eq $FOO ) {
	    # Add more here...
	} else {
	    notify("$node: Unknown command timeout '$timeout_tag{$node}' ".
658
		   "found at $mode/$newstate\n");
659 660 661
	}
    }

662 663 664 665 666
    #
    # 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.
    #
667 668
    setTimeout($mode,$newstate,$node,$now);

Mac Newbold's avatar
Mac Newbold committed
669 670 671 672 673 674 675
    # 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 = $_;
676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692
	    /^SCRIPT:([-\w\/]+)$/ && 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
693
	    /^$PXEBOOT$/ && do {
694
		#
Mac Newbold's avatar
Mac Newbold committed
695 696 697 698 699 700 701 702 703 704 705 706
		# 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.
707
		#
Mac Newbold's avatar
Mac Newbold committed
708 709 710 711
		# 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.
712
		#
Mac Newbold's avatar
Mac Newbold committed
713 714 715 716 717
		debug("Running $PXEBOOT trigger\n");
		if ($mode ne $PXEKERNEL) {
		    info("$node: Forcing mode transition into $PXEKERNEL!\n");
		    opModeTransition($node, $PXEKERNEL, 1);
		    $mode=$PXEKERNEL;
718
		}
Mac Newbold's avatar
Mac Newbold committed
719 720 721 722 723 724 725
		next;
	    };
	    /^$BOOTING$/ && do {
		#
		# See if we are in the right mode/osid.
		#
		my ($bootosid,$bootopmode) = TBBootWhat($node, $debug);
726

Mac Newbold's avatar
Mac Newbold committed
727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792
		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($node);
		next;
	    };
793 794 795 796
	    /^CHECKPORTREG$/ && do {
		CheckPortRegistration($node);
		next;
	    };
797
	    /^$TBRESET$/ && do {
798 799 800
		# We successfully booted, so clear some flags
		$nodes{$node}{noretry}   = 0;
		$nodes{$node}{timedout}  = 0;
801 802 803 804 805 806 807 808 809
		# 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,$trig);
		}
		next;
	    };
810
	    (/^$TBRELOADDONEV1$/ || /^$TBRELOADDONEV2$/) && do {
811 812 813
		handleCtrlEvent($node,$trig);
		next;
	    };
814 815 816 817 818 819 820 821
	    /^$TBISUP$/ && do {
		info("$node: Triggered $TBISUP\n");
		EventSendWarn(host      => $BOSSNODE ,
			      objtype   => TBDB_TBEVENT_NODESTATE ,
			      eventtype => TBDB_NODESTATE_ISUP ,
			      objname   => $node);
		next;
	    };
822 823 824 825
	    (/^$TBPOWERCYCLE$/) && do {
		handleCommand($node,$TBPOWERCYCLE);
		next;
	    };
826 827 828 829 830 831 832 833 834 835 836 837
	    /^RELOADOLDMFS$/ && 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;
	    };
838
	    notify("Unknown trigger '$trig' for $node in $mode/$newstate!\n");
839
	}
Mac Newbold's avatar
Mac Newbold committed
840 841 842 843 844 845 846
	# 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);
	}
847
    }
848

849 850 851 852 853 854 855 856
    # 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
857 858 859
	} else {
	    debug("No next mode.\n");
	}
860 861
    }
}
862

863
sub opModeTransition($$;$) {
Mac Newbold's avatar
Mac Newbold committed
864

865 866
    my ($node,$newmode,$force) = @_;
    if (!defined($force)) { $force = 0; }
Mac Newbold's avatar
Mac Newbold committed
867

868
    info("$node: Mode change to $newmode requested ($force)\n");
Mac Newbold's avatar
Mac Newbold committed
869

870 871
    # Check for invalid transitions
    my ($oldstate, $mode, $nextstate);
872 873 874
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

875
    if (defined($modeTrans{"$mode:$oldstate"}) || $force) {
876
	if (!$force) {
877 878 879 880 881 882 883 884 885 886 887 888 889 890 891
	    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");
	    }
892 893
	}
    } else {
894
	notify("Invalid mode transition for $node from $mode/$oldstate: ".
895 896
	       "Not a valid mode transition state!\n");
    }
Mac Newbold's avatar
Mac Newbold committed
897 898 899
    if (!$nextstate) {
	$nextstate=$oldstate;
    }
Mac Newbold's avatar
Mac Newbold committed
900

901 902 903 904 905 906
    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
907

908 909 910 911 912
    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
913 914 915 916

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

917 918 919 920
}

sub handleCtrlEvent($$) {
    my ($node,$event) = @_;
921

922
    info("CtrlEvent: $node, $event\n");
923

924 925
    foreach ($event) {
	/^$TBRESET$/ && do {
926 927
	    #
	    # Clear next_boot_path with os_select.
Mac Newbold's avatar
Mac Newbold committed
928
	    #
929
	    $cmd = "$osselect -d -c -1 $node";
930
	    system($cmd) and
931
		notify("$node/$event: Could not clear next_boot_path!\n");
Mac Newbold's avatar
Mac Newbold committed
932 933

	    info("Performed $TBRESET for $node\n");
934 935
	    next;
	};
936
	(/^$TBRELOADDONEV1$/ || /^$TBRELOADDONEV2$/) && do {
937 938 939 940 941 942 943
	    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'");
944
		DBQueryFatal("delete from reserved where node_id='$node'");
945 946
		TBSetNodeHistory($node, TB_NODEHISTORY_OP_FREE,
				 $UID, $pid, $eid);
947
		info("Released $node from $pid/$eid\n");
948
	    }
949 950 951 952 953
	    if ($event eq $TBRELOADDONEV2) {
		info("Sending an apod to $node\n");
		system("$apod $node") == 0 or
		    notify("Could not apod $node after $TBRELOADDONEV2!\n");
	    }
954 955 956
	    next;
	};
	/^$TBTIMEOUT$/ && do {
957 958 959 960
	    my ($mode,$state) = split(":",$timeout_tag{$node});
	    delete($timeout_tag{$node});
	    my $curstate = $nodes{$node}{state};
	    my $curmode = $nodes{$node}{mode};
961
	    my ($timeout,$action);
962 963 964 965 966
	    if (!defined($nodes{$node}{notified})) {
		$nodes{$node}{notified}=0;
	    }
	    $nodes{$node}{notified}++;
	    my $notified = $nodes{$node}{notified};
967 968
	    $nodes{$node}{timedout}++;
	    my $timedout = $nodes{$node}{timedout};
969 970 971 972
	    if ($mode && $state && $timeouts{$mode} &&
		$timeouts{$mode}{$state}) {
		($timeout, $action) = @{$timeouts{$mode}{$state}};
	    }
973 974
	    if ($mode eq $TBCOMMAND) {
		# It is a command, not a true state
975
		if ($action eq $TBTIMEOUTCMDRETRY) {
976
		    # Retry the command
977
		    notify("$node: Command $state, retry #$timedout\n");
978
		    # notify in case we get in a retry loop...
979
		    handleCommand($node,$state,$timedout,1);
980 981 982 983 984 985
		} else {
		    notify("$node: Unknown timeout action for ".
			   "$mode/$state: '$action'\n");
		}
		next;
	    }
986 987 988 989 990 991 992

	    #
	    # 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
993

994 995
		if ($timedout < 3) {
		    #
Mac Newbold's avatar
Mac Newbold committed
996 997
		    # Try again.
		    #
998 999 1000 1001 1002 1003
		    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
1004 1005
		    # Failed too many times, power cycle instead.
		    #
1006 1007 1008 1009 1010 1011 1012 1013 1014
		    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 >> $rebootlog &") and
1015
		    notify("$PXEWAKEUP retry: ".
1016 1017 1018 1019 1020
			   "Command '$cmd' failed, error $?: $!\n");

		next;
	    }

1021 1022 1023 1024 1025 1026
	    info("Node $node has timed out in state $mode/$state".
		 ($action ne "" ? "\n\tRequested action $action." : "").
		 "\n");

	    foreach ($action) {
		/^$TBTIMEOUTREBOOT/ && do {
1027 1028 1029 1030 1031
		    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;
1032
		    } else {
1033 1034 1035 1036 1037 1038
			# 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);
1039 1040 1041 1042 1043 1044 1045
		    }
		    last; };
		/^$TBTIMEOUTNOTIFY/ && do {
		    notify("Node $node has timed out in state $mode/$state\n");
		    last; };
		notify("$node: Unknown Timeout Action: $action\n");
	    }
1046 1047
	    next;
	};
1048
	notify("$node: Unknown CtrlEvent: $event\n");
1049 1050
    }
}
Robert Ricci's avatar
Robert Ricci committed
1051

1052 1053
sub handleCommand($$;$$) {
    my ($params,$command,$retry,$force) = @_;
1054
    if (!defined($retry)) { $retry=0; }
1055
    if (!defined($force)) { $force=0; }
1056 1057 1058 1059 1060 1061 1062 1063

    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.

1064
    if ($command eq $TBREBOOT && $retry >=4) {
1065 1066
	announce("Node $params has tried rebooting $retry times and has \n".
		 "still not been successful. Please look into it soon.\n".
1067 1068 1069
	"" );#	 "In the meantime, $params will be powered off.\n");
	# Just return...
	return 0;
1070
	# change my command to poweroff.
1071
	#$command = $TBPOWEROFF;
1072 1073 1074 1075 1076 1077
    }

    foreach ($command) {
	/^$TBREBOOT$/ && do {
	    # For reboot, the params is a comma-separated list of nodes
	    my @nodes = split(",",$params);
1078 1079 1080 1081 1082 1083 1084 1085

	    # 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");
1086 1087
		    if (($nodes{$node}{state} ne TBDB_NODESTATE_ISUP) &&
			(!$nodes{$node}{noretry}) ) {
1088
			# This node shouldn't be rebooted now...
1089 1090
			# XXX Send feedback here somehow!

1091 1092 1093 1094 1095 1096 1097 1098 1099 1100
			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; }
	    }

1101 1102 1103 1104 1105
	    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.

1106
	    my $cmd = "$nodereboot -r $nodelist";
1107
	    my $redir = " 2>&1 >> $rebootlog &";
1108 1109 1110
	    debug("$cmd $redir\n");
	    system("date $redir");
	    system($cmd.$redir) and
1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132
	      notify("$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};
1133
	    info("Sending power $func for nodes: $nodelist\n");
1134 1135 1136 1137
	    # Permissions were checked in order to send the message,
	    # so we don't need to do any fancy stuff here.

	    my $cmd = "$power $func $nodelist &";
1138 1139
	    debug("$cmd\n") and
	    system($cmd) and
1140 1141 1142 1143 1144 1145 1146 1147 1148 1149
	      notify("$params/$command: ".
		     "Command '$cmd' failed, error $?: $!\n");

	    info("Performed $command for $params\n");
	    next;
	};
	notify("$params: Unknown Command: $command\n");
    }
}

1150 1151 1152 1153
#
# Check if we need to generate an ISUP
#
sub checkGenISUP($) {
Mac Newbold's avatar
Mac Newbold committed
1154 1155 1156 1157 1158 1159 1160 1161 1162 1163
    my ($node) = @_;
    debug("$node: Checking ISUP Generation\n");
    my $r = DBQueryWarn("select osfeatures from nodes as n ".
			"left join os_info as o on o.osid=n.osid ".
			"where node_id='$node' and osfeatures is not null");
    my $osfeatures="";
    # If we don't get anything back, assume it has no features.
    if ($r->num_rows() > 0) {
	($osfeatures) = $r->fetchrow();
    }
Mac Newbold's avatar
Mac Newbold committed
1164

Mac Newbold's avatar
Mac Newbold committed
1165 1166 1167 1168 1169 1170
    my @features = split(",",$osfeatures);
    # Make sure features I care about are defined
    my %can=("ping"=>0, "isup"=>0);
    foreach my $f (@features) {
	$can{"\L$f"}=1;	# make sure it's all lowercase
    }
Mac Newbold's avatar
Mac Newbold committed
1171

Mac Newbold's avatar
Mac Newbold committed
1172 1173
    # If os will send ISUP on its own, do nothing here.
    if ($can{"isup"}) {
Mac Newbold's avatar
Mac Newbold committed
1174
	debug("$node: Will send own ISUP\n");
Mac Newbold's avatar
Mac Newbold committed
1175 1176
	return 0;
    }
Mac Newbold's avatar
Mac Newbold committed
1177

Mac Newbold's avatar
Mac Newbold committed
1178 1179 1180 1181 1182 1183
    # If os doesn't support isup but can ping, fork and ping it every
    # few seconds and send isup when it pings, or timeout after too long.
    if ($can{"ping"}) {
	debug("$node: Needs to be pinged - calling eventping\n");
	system("$TB/sbin/eventping $node &");
	return 0;
1184
    }
Mac Newbold's avatar
Mac Newbold committed
1185

Mac Newbold's avatar
Mac Newbold committed
1186
    # If os doesn't support ping or isup, stated sends ISUP just after
Mac Newbold's avatar
Mac Newbold committed
1187 1188 1189 1190 1191 1192 1193
    # the node gets to BOOTING (a bit early, but the best we can do)

    debug("$node: OS doesn't ping - sending ISUP\n");
    EventSendWarn(host      => $BOSSNODE ,
		  objtype   =&