stated.in 39.6 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) 2000-2003 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
11 12
# state into the database. Also watches for invalid transitions, timeouts, 
# 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 31
my $nodereboot = "$TB/bin/node_reboot";
my $power = "$TB/bin/power";
Robert Ricci's avatar
Robert Ricci committed
32 33 34 35 36 37

$| = 1;

use event;
use libdb;
use libtestbed;
38
use TimeoutQueue;
Robert Ricci's avatar
Robert Ricci committed
39
use Getopt::Std;
40
#use strict;
Robert Ricci's avatar
Robert Ricci committed
41
use English;
Mac Newbold's avatar
Mac Newbold committed
42 43
use POSIX;			# for strftime, and sigprocmask and friends
use Fcntl;			# file constants for pidfile
Mac Newbold's avatar
Mac Newbold committed
44 45 46 47 48
use Sys::Syslog;
# Important note about syslog: It defaults to using an inet socket,
# but 'syslogd -s' (the default) doesn't listen for one. So either
# run syslogd without -s, or use setlogsock('unix') before openlog.
# (To get setlocksock: 'use Sys::Syslog qw(:DEFAULT setlogsock);' )
Robert Ricci's avatar
Robert Ricci committed
49

Mac Newbold's avatar
Mac Newbold committed
50 51 52
# Do lots of db retries before we fail and die
$libdb::DBQUERY_MAXTRIES = 5;

53 54 55 56 57
# Set up some notification throttling
my $mailgap = 15;		# in seconds
my $lastmail = time() - $mailgap + 2; # Send a digest of startup msgs after 2s.
my %msgs = ();

Robert Ricci's avatar
Robert Ricci committed
58 59 60
# Number of iterations (roughly, seconds) after which we'll reload 
# information from the database. This is so we don't end up with information
# that's _too_ out of sync.
61
my $reload_time = 600;
62
my $last_reload = time;
Robert Ricci's avatar
Robert Ricci committed
63 64 65 66

# Process command-line arguments

sub usage {
Mac Newbold's avatar
Mac Newbold committed
67
    print << "END";
68 69 70
Usage: $0 [-h] [-d] [-s server] [-p port]
-h              This message
-d              Turn on debugging output, and don't go into the background
Robert Ricci's avatar
Robert Ricci committed
71 72
-s server       Use specified server, instead of this site's bossnode
-p port	        Use specified port
73
Send SIGHUP to reload database state, or SIGUSR1 to restart completely.
Robert Ricci's avatar
Robert Ricci committed
74 75 76
END
}

Mac Newbold's avatar
Mac Newbold committed
77
# Only root should run this - it won't work when run as a user...
78
# (Or, let an admin run it if it isn't the real one in /usr/testbed/ )
79
if ($UID && ( $TB eq $REALTB || ! TBAdmin($UID) ) ) {
Mac Newbold's avatar
Mac Newbold committed
80 81 82
    die("Only root can run this script!\n");
}

83
my @args = @ARGV;    # save a copy for restart before we mess with them.
Robert Ricci's avatar
Robert Ricci committed
84
my %opt = ();
85
getopts("ds:p:h",\%opt);
Robert Ricci's avatar
Robert Ricci committed
86

Mac Newbold's avatar
Mac Newbold committed
87 88 89 90 91 92
if ($opt{h}) {
    exit &usage;
}
if (@ARGV) {
    exit &usage;
}
Robert Ricci's avatar
Robert Ricci committed
93

94
my ($server,$port,$debug);
Mac Newbold's avatar
Mac Newbold committed
95 96 97 98 99 100 101 102 103 104 105 106 107
if ($opt{s}) {
    $server = $opt{s};
} else {
    $server = $BOSSNODE;
}
if ($opt{p}) {
    $port = $opt{p};
}
if ($opt{d}) {
    $debug = 1;
} else {
    $debug = 0;
}
Robert Ricci's avatar
Robert Ricci committed
108

109
# Grab some constants into variables
110
my $TBANYMODE    = TBDB_NODEOPMODE_ANY;
111 112 113 114 115 116
my $TBRESET      = TBDB_TBCONTROL_RESET;
my $TBRELOADDONE = TBDB_TBCONTROL_RELOADDONE;
my $TBTIMEOUT    = TBDB_TBCONTROL_TIMEOUT;
my $TBNOTIMEOUT  = TBDB_NO_STATE_TIMEOUT;
my $TBNODESTATE  = TBDB_TBEVENT_NODESTATE;
my $TBNODEOPMODE = TBDB_TBEVENT_NODEOPMODE;
117 118 119 120 121 122
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;
123
my $TBISUP       = TBDB_NODESTATE_ISUP;
124 125 126 127
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;
128

129 130 131
# This only gets used here, so it isn't in a lib constant.
my $TBFREENODE = "FREENODE";

Mac Newbold's avatar
Mac Newbold committed
132
my $pidfile;
133
if ( $TB eq $REALTB ) {
Mac Newbold's avatar
Mac Newbold committed
134 135
    $pidfile = "/var/run/stated.pid";
} else {
136
    $pidfile = "$TB/locks/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
137
}
Mac Newbold's avatar
Mac Newbold committed
138 139
debug("Using pidfile $pidfile\n");
if (-e $pidfile) {
Mac Newbold's avatar
Mac Newbold committed
140 141 142 143 144 145 146 147 148 149 150
    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: $? $!\n");
Mac Newbold's avatar
Mac Newbold committed
151
}
Robert Ricci's avatar
Robert Ricci committed
152
# Background
153
if (!$debug) {
Mac Newbold's avatar
Mac Newbold committed
154 155 156 157
    # We use syslog, so redirect the output to nothing
    if (TBBackGround("/dev/null")) {
	exit(0);
    }
Robert Ricci's avatar
Robert Ricci committed
158
}
Mac Newbold's avatar
Mac Newbold committed
159 160
# set up syslog
openlog("stated","pid","user");
Mac Newbold's avatar
Mac Newbold committed
161 162 163 164 165 166
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
my $lockfile=$pidfile;
Robert Ricci's avatar
Robert Ricci committed
167

168 169 170
# Change my $0 so that it is easier to see in a ps/top
$0 = "$0";

Robert Ricci's avatar
Robert Ricci committed
171
my $URL = "elvin://$server";
Mac Newbold's avatar
Mac Newbold committed
172 173 174
if ($port) {
    $URL .= ":$port";
}
Robert Ricci's avatar
Robert Ricci committed
175 176 177

# Connect to the event system, and subscribe the the events we want 
my $handle = event_register($URL,0);
Mac Newbold's avatar
Mac Newbold committed
178 179 180
if (!$handle) {
    fatal("Unable to register with event system\n");
}
Robert Ricci's avatar
Robert Ricci committed
181 182

my $tuple = address_tuple_alloc();
Mac Newbold's avatar
Mac Newbold committed
183 184 185
if (!$tuple) {
    fatal("Could not allocate an address tuple\n");
}
Robert Ricci's avatar
Robert Ricci committed
186

187 188 189
%$tuple = ( objtype => join(",",
			    $TBNODESTATE, $TBNODEOPMODE,
			    $TBCONTROL, $TBCOMMAND) );
190

Robert Ricci's avatar
Robert Ricci committed
191
if (!event_subscribe($handle,\&handleEvent,$tuple)) {
Mac Newbold's avatar
Mac Newbold committed
192
    fatal("Could not subscribe to events\n");
Robert Ricci's avatar
Robert Ricci committed
193 194 195 196
}

# Read in the pre-existing node states, and timeout and valid transition
# information from the database
197 198 199 200
my %timeouts  = getTimeouts();
my %valid     = getValid();
my %modeTrans = getModeTrans();
my %triggers  = getTriggers();
201
my %nodes     = readStates();
202
my %timeouttag= ();
203
if ($debug) { qshow(); }
Robert Ricci's avatar
Robert Ricci committed
204 205 206

# Gets set if a reload of state from the database should happen.
my $do_reload = 0;
207 208
my $sigrestart= 0;
my $sigcleanup= 0;
Robert Ricci's avatar
Robert Ricci committed
209 210 211 212

# 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.
213 214
$SIG{HUP}  = sub { info("SIGHUP - Reloading DB state\n"); $do_reload = 1; };

Mac Newbold's avatar
Mac Newbold committed
215
# Set up other signals.
216 217 218 219 220 221 222
$SIG{USR1} = \&restart_wrap;
$SIG{USR2} = \&cleanup_wrap;
$SIG{INT}  = \&cleanup_wrap;
$SIG{QUIT} = \&cleanup_wrap;
$SIG{ABRT} = \&cleanup_wrap;
$SIG{TERM} = \&cleanup_wrap;
$SIG{KILL} = \&cleanup_wrap;
Robert Ricci's avatar
Robert Ricci committed
223

224 225 226
# Track if I handled an event or not
my $event_count = 0;

227 228
# Control how long I block while waiting for events
my $blockwait=0;
229
my $nextdeadline=0;
230 231
my $mailqueue=0;

232 233 234 235 236
notify("Stated starting up\n");

sub process_event_queue() {
    $event_count=0;
    my $lastcount=-1;
237 238 239 240 241 242 243 244 245 246 247 248 249
    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;
250 251
	    } else {
		$wait = 0;
252 253 254 255 256
	    }
	}
    } else {
	# mail is waiting. Only block until it is time to send it.
	$wait = $lastmail + $mailgap - $now;
Mac Newbold's avatar
Mac Newbold committed
257
	debug("Now $now, mailgap $mailgap, last $lastmail ==> wait $wait\n");
258 259 260
    }
    if ($wait < 0) { debug("Wait was $wait!\n"); $wait=0; }
    my $finish = $now + $wait;
Mac Newbold's avatar
Mac Newbold committed
261 262
    while (($event_count != $lastcount || $wait > 0) &&
	   !($sigrestart || $sigcleanup || $do_reload)) {
263
	$lastcount = $event_count;
Mac Newbold's avatar
Mac Newbold committed
264 265
	# Don't block if we got a signal!
	if ($wait<=0 || $sigrestart || $sigcleanup || $do_reload) {
266 267 268 269 270 271 272 273
	    event_poll($handle);
	} else {
	    debug("Using blocking event poll - $wait seconds\n");
	    # timeout param is in milliseconds, so multiply
	    event_poll_blocking($handle, $wait*1000);
	    $now = time();
	    # subtract seconds elapsed from my wait time
	    $wait = $finish - $now;
Mac Newbold's avatar
Mac Newbold committed
274
	    debug("Finished blocking event poll - $wait seconds remain\n");
275
	    if ($event_count > 0 &&
Mac Newbold's avatar
Mac Newbold committed
276 277
		(qsize() > 0 || $mailqueue ||
		 $sigrestart || $sigcleanup || $do_reload)) {
278 279 280 281 282 283 284
		$blockwait=0;
		$wait=0;
		#debug("Cancelling wait - timeouts/msgs waiting, or HUP'd\n");
		#debug("---End Blocking Wait ---\n");
	    }
	}
	#debug("Wait is $wait\n");
285 286
    }
    if ($event_count > 0) {
Mac Newbold's avatar
Mac Newbold committed
287
	debug("Handled $event_count event(s).\n");
288 289
    }
}
Robert Ricci's avatar
Robert Ricci committed
290

291
# Now, we just poll for events, and watch for timeouts
Robert Ricci's avatar
Robert Ricci committed
292
while (1) {
Mac Newbold's avatar
Mac Newbold committed
293
    my $now = time();
294 295 296 297
    my ($deadline,$node);

    # Check for nodes that have passed their timeout
    if (!qhead($deadline,$node)) {
298
	info("HEAD: $node in ".($deadline-$now).", queue=".qsize()."\n");
299 300
	while ($now >= $deadline && $node ne "") {
	    qpop($deadline,$node);
301
	    info("POP: $node in ".($deadline-$now).", queue=".qsize()."\n");
302
	    handleCtrlEvent($node,$TBTIMEOUT);
303 304 305 306
	    if (0) { qshow(); }
	    if (qhead($deadline,$node)) {
		$deadline=0; $node="";
	    }
307
	}
308 309
    } else {
	$deadline=0;
310
    }
311
    $nextdeadline = $deadline;
312

313 314 315 316 317
    if (qsize()==0) {
	$blockwait=1;
	debug("---Blocking wait okay---\n");
    }
		
Mac Newbold's avatar
Mac Newbold committed
318 319 320 321
    if ($do_reload || ($now - $last_reload > $reload_time)) {
	reload();
	$do_reload = 0;
    }
Mac Newbold's avatar
Mac Newbold committed
322

Mac Newbold's avatar
Mac Newbold committed
323 324
    # Send any messages in the queue if it is time
    notify("",1);
Mac Newbold's avatar
Mac Newbold committed
325

326 327 328
    if ($sigrestart) { restart(); }
    if ($sigcleanup) { cleanup(); }

329
    process_event_queue;
Robert Ricci's avatar
Robert Ricci committed
330 331
}

Mac Newbold's avatar
Mac Newbold committed
332 333
exit(0);

Robert Ricci's avatar
Robert Ricci committed
334
# Read the current states of nodes from the database
335
sub readStates(;@) {
Mac Newbold's avatar
Mac Newbold committed
336 337 338 339 340 341
    my %oldnodes = @_;

    # Guard against undefined variable warnings
    if (! defined(%oldnodes)) {
	%oldnodes = ();
    }
342

Mac Newbold's avatar
Mac Newbold committed
343 344 345
    #debug("readStates called\n");
    my $result = DBQueryFatal("SELECT node_id, eventstate, " .
			      "state_timestamp, op_mode, " .
346 347
			      "op_mode_timestamp FROM nodes ".
			      "where node_id not like 'sh%'");
Mac Newbold's avatar
Mac Newbold committed
348 349 350 351

    my %nodes;
    while (my ($node_id, $state, $timestamp, $mode, $mode_timestamp)
	   = $result->fetchrow()) {
352
	#
Mac Newbold's avatar
Mac Newbold committed
353 354 355 356
	# If there's an entry in oldnodes for this node, and it
	# hasn't changed state or time, use the old entry (so that
	# we don't lose information about which nodes we've already
	# notified the ops about, etc.)
357
	#
Mac Newbold's avatar
Mac Newbold committed
358 359 360 361 362
	if ($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};
363
	} else {
Mac Newbold's avatar
Mac Newbold committed
364 365 366 367
	    $nodes{$node_id}{state}          = $state;
	    $nodes{$node_id}{timestamp}      = $timestamp;
	    $nodes{$node_id}{mode}           = $mode;
	    $nodes{$node_id}{mode_timestamp} = $mode_timestamp;
368 369 370
	    $nodes{$node_id}{notified}       = 0;
	    $nodes{$node_id}{timedout}       = 0;
	    $nodes{$node_id}{noretry}        = 0;
371 372
	    # Is there a timeout? If so, set it up!
	    setTimeout($mode,$state,$node_id,$timestamp);
373
	}
Mac Newbold's avatar
Mac Newbold committed
374 375
    }
    return %nodes;
Robert Ricci's avatar
Robert Ricci committed
376 377 378 379 380 381
}

#
# Read timeouts for various states from the database
#
sub getTimeouts() {
Mac Newbold's avatar
Mac Newbold committed
382 383 384
    #debug("getTimeouts called\n");
    my $result = DBQueryFatal("SELECT op_mode, state, timeout, action " .
			      "FROM state_timeouts");
Robert Ricci's avatar
Robert Ricci committed
385

Mac Newbold's avatar
Mac Newbold committed
386 387 388 389 390
    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
391 392 393 394 395 396
}

#
# Read the list of valid state transitions from the database
#
sub getValid() {
Mac Newbold's avatar
Mac Newbold committed
397 398 399
    #debug("getValid called\n");
    my $result = DBQueryFatal("SELECT op_mode, state1, state2 " .
			      "FROM state_transitions");
Robert Ricci's avatar
Robert Ricci committed
400

Mac Newbold's avatar
Mac Newbold committed
401 402 403 404 405
    my %valid;
    while (my ($mode,$state1, $state2) = $result->fetchrow()) {
	$valid{$mode}{$state1}{$state2} = 1;
    }
    return %valid;
Robert Ricci's avatar
Robert Ricci committed
406 407
}

408 409 410 411
#
# Read the list of valid mode transitions from the database
#
sub getModeTrans() {
Mac Newbold's avatar
Mac Newbold committed
412 413 414 415 416 417 418 419 420 421 422 423 424
    #debug("getModeTrans called\n");
    my $result = 
      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;
425
	}
Mac Newbold's avatar
Mac Newbold committed
426 427
    }
    return %modeTrans;
428 429 430 431 432 433
}

#
# Read the list of states which trigger an action
#
sub getTriggers() {
434 435 436 437 438
    debug("getTriggers called\n");
    
    debug("anymode ==> '$TBANYMODE'\n");

    # Grab global triggers
Mac Newbold's avatar
Mac Newbold committed
439 440
    my $result = 
      DBQueryFatal("SELECT op_mode, state, trigger " .
441 442
		   "FROM state_triggers where node_id='$TBANYMODE' ".
		   "order by op_mode,state");
Mac Newbold's avatar
Mac Newbold committed
443
    my %t;
444
    while (my ($mode, $state, $trig) = $result->fetchrow()) {
Mac Newbold's avatar
Mac Newbold committed
445
	$t{"$mode:$state"} = $trig;
446 447 448 449 450 451 452 453 454 455 456 457
	debug("trig($mode:$state)\t => $trig\n");
    }

    # Grab per-node triggers
    $result = 
      DBQueryFatal("SELECT node_id, op_mode, state, trigger " .
		   "FROM state_triggers where node_id!='$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
458
    }
459

Mac Newbold's avatar
Mac Newbold committed
460
    return %t;
461 462
}

Robert Ricci's avatar
Robert Ricci committed
463 464 465 466
#
# Gets called for every event that we recieve
#
sub handleEvent($$$) {
Mac Newbold's avatar
Mac Newbold committed
467 468 469 470 471 472 473 474 475 476 477
    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);

    $event_count++;
    debug("Got an event: ($objtype,$objname,$eventtype)\n");

    #
    # Check to see if another instance is supposed to be handling this node
    #
478
    if ($objtype ne $TBCOMMAND && !checkDBRedirect($objname)) {
Mac Newbold's avatar
Mac Newbold committed
479 480 481
	info("Got an event for node $objname, which isn't mine\n");
	return;
    }
482

Mac Newbold's avatar
Mac Newbold committed
483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498
  SWITCH: for ($objtype) {
	
	(/$TBNODESTATE/) && do {
	    stateTransition($objname,$eventtype);
	    last;
	};
	(/$TBNODEOPMODE/) && do {
	    opModeTransition($objname,$eventtype);
	    notify("Use of deprecated event TBNODEOPMODE:\n".
		   "$objname->$eventtype\n");
	    last;
	};
	(/$TBCONTROL/) && do {
	    handleCtrlEvent($objname,$eventtype);
	    last;
	};
499 500 501 502
	(/$TBCOMMAND/) && do {
	    handleCommand($objname,$eventtype);
	    last;
	};
503

Mac Newbold's avatar
Mac Newbold committed
504
    }
505 506 507 508 509

}

sub stateTransition($$) {

510
    my ($node,$newstate) = @_;
Robert Ricci's avatar
Robert Ricci committed
511

512 513 514 515 516 517 518 519
    # Check for invalid transitions
    my ($oldstate, $mode);
    if ($nodes{$node}) {
	$oldstate = $nodes{$node}{state};
	$mode = $nodes{$node}{mode};
    } else {
	# Try reloading the cache once before we give up on this node
	reload();
520
	if ($nodes{$node}) {
521 522
	    $oldstate = $nodes{$node}{state};
	    $mode = $nodes{$node}{mode};
Robert Ricci's avatar
Robert Ricci committed
523
	} else {
524
	    notify("Got an event for a node ($node) I don't know about\n");
Robert Ricci's avatar
Robert Ricci committed
525
	}
526 527 528 529
    }
    if ($oldstate && $mode && $valid{$mode} && $valid{$mode}{$oldstate} &&
	!$valid{$mode}{$oldstate}{$newstate}) {
	notify("Invalid transition for node $node from $mode/$oldstate " .
530
	       "to $newstate\n");
531
    }
Robert Ricci's avatar
Robert Ricci committed
532

533 534 535 536
    my $now = time();
    $nodes{$node}{state}     = $newstate;
    $nodes{$node}{timestamp} = $now;
    $nodes{$node}{notified}  = 0;
537

538 539 540
    info("$node: $mode/$oldstate => $mode/$newstate\n");
    DBQueryFatal("UPDATE nodes SET eventstate='$newstate', " .
		 "state_timestamp='$now' WHERE node_id='$node'");
541

542 543 544 545
    # 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} =~ /^$TBCOMMAND:/) {
546
        debug("TimeoutTag = '$timeout_tag{$node}'\n");
547
	my ($str,$cmd) = split(":",$timeout_tag{$node});
548
	debug("str=$str\tcmd=$cmd\tTBREBOOT=$TBREBOOT\tstate=$newstate\n");
549
	if ($cmd eq $TBREBOOT) {
550
	    if ($newstate eq TBDB_NODESTATE_SHUTDOWN ) {
551 552 553
		info("$node: $TBREBOOT success\n");
		# Timeout will get cleared below by setTimeout call
	    } else {
554 555
		notify("$node: $TBREBOOT in progress, but got state ".
		       "$newstate instead of ".TBDB_NODESTATE_SHUTDOWN."!\n");
556 557 558 559 560
	    }
	#} elsif ($cmd eq $FOO ) {
	    # Add more here...
	} else {
	    notify("$node: Unknown command timeout '$timeout_tag{$node}' ".
561
		   "found at $mode/$newstate\n");
562 563 564
	}
    }

565 566 567
    # Check if this state has a timeout, and if so, put it in the queue
    setTimeout($mode,$newstate,$node,$now);

568 569
    # Check if this is TBDB_NODESTATE_BOOTING , which has actions
    if ($newstate eq TBDB_NODESTATE_BOOTING) {
570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587
	# If I skipped shutdown, and came to booting directly from isup,
	# check for a mode transition so I don't miss one...
	if ($oldstate eq TBDB_NODESTATE_ISUP) {
	    info("$node: Came from ISUP! Checking for mode transition\n");
	    my $r = DBQueryWarn("select next_op_mode from nodes ".
				"where node_id='$node'");
	    my ($nextmode) = $r->fetchrow();
	    if ($nextmode) {
		# Force the transition even though it is illegal
		info("$node: Forcing mode transition!\n");
		opModeTransition($node,$nextmode,1);
		$mode=$nextmode;
	    } else {
		debug("No next mode.\n");
	    }
	}

	# Check if I'm in the right mode
588
	my $osid = TBBootWhat($node,$debug);
589 590
	my $os_op_mode = os_opmode($osid);
	info("$node: Current OS is '$osid', OS mode is '$os_op_mode'\n");
591
	DBQueryFatal("UPDATE nodes SET osid='$osid' WHERE node_id='$node'");
592
	if ($os_op_mode ne $mode) {
593 594
	    my $str = "Node $node is running OS '$osid' but in mode '$mode' ".
	      "instead of mode '$os_op_mode'!\n";
595 596 597
	    # For now, only force if we're going into reload mode, so we
	    # don't get stuck looping in reloading.
	    if ($os_op_mode eq "RELOAD") {
598 599 600 601 602 603 604 605 606
		DBQueryFatal("UPDATE nodes SET op_mode='$os_op_mode', ".
			     "op_mode_timestamp=unix_timestamp(now()) ".
			     "WHERE node_id='$node'");
		$nodes{$node}{mode} = $os_op_mode;
		$nodes{$node}{mode_timestamp} = $now;
		$str .= "Forced op_mode to '$os_op_mode'.\n";
	    }
	    notify($str);
	}
607 608
	checkGenISUP($node);
    }
609

610 611 612 613 614
    # Check if this state has any triggers
    my @nodetrigs = GetNodeTriggerList($node,$mode,$newstate);
    if (defined($triggers{"$mode:$newstate"}) ||
        (@nodetrigs > 0) ) {
	# check for global triggers
615
	my @trigs = split(/\s*,\s*/,$triggers{"$mode:$newstate"});
616 617 618 619
	# Run all the triggers
	debug("Running triggers. Global=".join("/",@trigs).
	      "   node=".join("/",@nodetrigs)."\n");
	foreach ( @trigs , @nodetrigs) {
620 621
	    my $trig = $_;
	    /^$TBRESET$/ && do {
622 623 624
		# We successfully booted, so clear some flags
		$nodes{$node}{noretry}   = 0;
		$nodes{$node}{timedout}  = 0;
625 626 627 628 629 630 631 632 633 634 635 636 637
		# 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,$trig);
		}
		next;
	    };
	    /^$TBRELOADDONE$/ && do {
		handleCtrlEvent($node,$trig);
		next;
	    };
638 639 640 641
	    /^$TBFREENODE$/ && do {
		handleCtrlEvent($node,$trig);
		next;
	    };
642 643 644 645 646 647 648 649
	    /^$TBISUP$/ && do {
		info("$node: Triggered $TBISUP\n");
		EventSendWarn(host      => $BOSSNODE ,
			      objtype   => TBDB_TBEVENT_NODESTATE ,
			      eventtype => TBDB_NODESTATE_ISUP ,
			      objname   => $node);
		next;
	    };
650
	    notify("Unknown trigger '$trig' for $node in $mode/$newstate!\n");
651
	}
652 653 654
	# Clear any of the node triggers that we ran
	debug("Clearing node triggers: ".join("/",@nodetrigs)."\n");
	ClearNodeTrigger($node,$mode,$newstate,@nodetrigs);
655
    }
656

657 658 659 660 661 662 663 664
    # 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
665 666 667
	} else {
	    debug("No next mode.\n");
	}
668 669
    }
}
670

671
sub opModeTransition($$;$) {
Mac Newbold's avatar
Mac Newbold committed
672

673 674
    my ($node,$newmode,$force) = @_;
    if (!defined($force)) { $force = 0; }
Mac Newbold's avatar
Mac Newbold committed
675

676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691
    info("$node: Mode change to $newmode requested\n");
    # Check for invalid transitions
    my ($oldstate, $mode, $nextstate);
    if ($nodes{$node}) {
	$oldstate = $nodes{$node}{state};
	$mode = $nodes{$node}{mode};
    } else {
	# Try reloading the cache once before we give up on this node
	reload();
	if ($nodes{$node}) {
	    $oldstate = $nodes{$node}{state};
	    $mode = $nodes{$node}{mode};
	} else {
	    notify("Got an event for a node ($node) I don't know about\n");
	}
    }
692
    if (defined($modeTrans{"$mode:$oldstate"}) || $force) {
693
	if (!$force) {
694 695 696 697 698 699 700 701 702 703 704 705 706 707 708
	    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");
	    }
709 710
	}
    } else {
711
	notify("Invalid mode transition for $node from $mode/$oldstate: ".
712 713
	       "Not a valid mode transition state!\n");
    }
Mac Newbold's avatar
Mac Newbold committed
714 715 716
    if (!$nextstate) {
	$nextstate=$oldstate;
    }
Mac Newbold's avatar
Mac Newbold committed
717

718 719 720 721 722 723
    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
724

725 726 727 728 729
    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
730 731 732 733

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

734 735 736 737
}

sub handleCtrlEvent($$) {
    my ($node,$event) = @_;
738

739
    info("CtrlEvent: $node, $event\n");
740

741 742
    foreach ($event) {
	/^$TBRESET$/ && do {
Mac Newbold's avatar
Mac Newbold committed
743 744 745
	    my $result = DBQueryFatal("SELECT pxe_boot_path, def_boot_osid ".
				      "FROM nodes where node_id='$node'");
	    my ($pxepath,$osid) = $result->fetchrow();
746

747 748 749 750
	    # Important note on ordering here:
	    # Because setting a normal osid resets pxe path to PXEBOOT,
	    # We need to read it out first, then set the osid, then set
	    # the pxepath back to its original value at the end.
751

752 753
	    $cmd = "$osselect $osid $node";
	    system($cmd) and
Mac Newbold's avatar
Mac Newbold committed
754 755
	      notify("$node/$event: Couldn't clear next_boot_*\n".
		     "\tcmd=$cmd\n\t*** $!\n");
756

757
	    $pxepath = "-p ".$pxepath;
Mac Newbold's avatar
Mac Newbold committed
758 759 760 761
	    if ($pxepath eq "-p ") {
		$pxepath="PXEBOOT";
	    }
	    ;
Mac Newbold's avatar
Mac Newbold committed
762
	    my $cmd = "$osselect -m $pxepath $node";
763
	    system($cmd) and
Mac Newbold's avatar
Mac Newbold committed
764 765
	      notify("$node/$event: Couldn't clear next_pxe_boot_path\n".
		     "\tcmd=$cmd\n\t*** $!\n");
766

Mac Newbold's avatar
Mac Newbold committed
767
	    info("Performed RESET for $node to $osid/$pxepath\n");
768 769 770 771 772 773 774 775 776 777
	    next;
	};
	/^$TBRELOADDONE$/ && do {
	    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'");
778 779 780 781
		AddNodeTrigger($node, $TBANYMODE, TBDB_NODESTATE_ISUP,
			       $TBFREENODE)
		  && notify("$node: Couldn't add trigger $TBFREENODE!\n");
		info("Set up freeing of $node from $pid/$eid\n");
782 783 784
	    }
	    next;
	};
785 786 787 788 789 790 791 792
	/^$TBFREENODE$/ && do {
	    # Don't need pid/eid, but we should put it in the log
	    my ($pid,$eid);
	    NodeidToExp($node,\$pid,\$eid);
	    DBQueryFatal("delete from reserved where node_id='$node'");
	    info("Released $node from $pid/$eid\n");
	    next;
	};
793
	/^$TBTIMEOUT$/ && do {
794 795 796 797
	    my ($mode,$state) = split(":",$timeout_tag{$node});
	    delete($timeout_tag{$node});
	    my $curstate = $nodes{$node}{state};
	    my $curmode = $nodes{$node}{mode};
798
	    my ($timeout,$action);
799 800 801 802 803
	    if (!defined($nodes{$node}{notified})) {
		$nodes{$node}{notified}=0;
	    }
	    $nodes{$node}{notified}++;
	    my $notified = $nodes{$node}{notified};
804 805
	    $nodes{$node}{timedout}++;
	    my $timedout = $nodes{$node}{timedout};
806 807 808 809
	    if ($mode && $state && $timeouts{$mode} &&
		$timeouts{$mode}{$state}) {
		($timeout, $action) = @{$timeouts{$mode}{$state}};
	    }
810 811
	    if ($mode eq $TBCOMMAND) {
		# It is a command, not a true state
812
		if ($action eq $TBTIMEOUTCMDRETRY) {
813
		    # Retry the command
814
		    notify("$node: Command $state, retry #$timedout\n");
815
		    # notify in case we get in a retry loop...
816
		    handleCommand($node,$state,$timedout,1);
817 818 819 820 821 822
		} else {
		    notify("$node: Unknown timeout action for ".
			   "$mode/$state: '$action'\n");
		}
		next;
	    }
823 824 825 826 827 828
	    info("Node $node has timed out in state $mode/$state".
		 ($action ne "" ? "\n\tRequested action $action." : "").
		 "\n");

	    foreach ($action) {
		/^$TBTIMEOUTREBOOT/ && do {
829 830 831 832 833
		    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;
834
		    } else {
835 836 837 838 839 840
			# 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);
841 842 843 844 845 846 847
		    }
		    last; };
		/^$TBTIMEOUTNOTIFY/ && do {
		    notify("Node $node has timed out in state $mode/$state\n");
		    last; };
		notify("$node: Unknown Timeout Action: $action\n");
	    }
848 849
	    next;
	};
850
	notify("$node: Unknown CtrlEvent: $event\n");
851 852
    }
}
Robert Ricci's avatar
Robert Ricci committed
853

854 855
sub handleCommand($$;$$) {
    my ($params,$command,$retry,$force) = @_;
856
    if (!defined($retry)) { $retry=0; }
857
    if (!defined($force)) { $force=0; }
858 859 860 861 862 863 864 865

    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.

866
    if ($command eq $TBREBOOT && $retry >=4) {
867 868
	announce("Node $params has tried rebooting $retry times and has \n".
		 "still not been successful. Please look into it soon.\n".
869 870 871
	"" );#	 "In the meantime, $params will be powered off.\n");
	# Just return...
	return 0;
872
	# change my command to poweroff.
873
	#$command = $TBPOWEROFF;
874 875 876 877 878 879
    }

    foreach ($command) {
	/^$TBREBOOT$/ && do {
	    # For reboot, the params is a comma-separated list of nodes
	    my @nodes = split(",",$params);
880 881 882 883 884 885 886 887

	    # 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");
888 889
		    if (($nodes{$node}{state} ne TBDB_NODESTATE_ISUP) &&
			(!$nodes{$node}{noretry}) ) {
890
			# This node shouldn't be rebooted now...
891 892
			# XXX Send feedback here somehow!

893 894 895 896 897 898 899 900 901 902
			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; }
	    }

903 904 905 906 907
	    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.

908 909 910 911 912
	    my $cmd = "$nodereboot -r $nodelist";
	    my $redir = " 2>&1 >> /usr/testbed/log/nodereboot.log &";
	    debug("$cmd $redir\n");
	    system("date $redir");
	    system($cmd.$redir) and
913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934
	      notify("$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};
935
	    info("Sending power $func for nodes: $nodelist\n");
936 937 938 939
	    # Permissions were checked in order to send the message,
	    # so we don't need to do any fancy stuff here.

	    my $cmd = "$power $func $nodelist &";
940 941
	    debug("$cmd\n") and
	    system($cmd) and
942 943 944 945 946 947 948 949 950 951
	      notify("$params/$command: ".
		     "Command '$cmd' failed, error $?: $!\n");

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

952 953 954 955
#
# Check if we need to generate an ISUP
#
sub checkGenISUP($) {
Mac Newbold's avatar
Mac Newbold committed
956 957 958 959 960 961 962 963 964 965
    my ($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
966

Mac Newbold's avatar
Mac Newbold committed
967 968 969 970 971 972
    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
973

Mac Newbold's avatar
Mac Newbold committed
974 975 976 977 978
    # If os will send ISUP on its own, do nothing here.
    if ($can{"isup"}) {
	debug("$node: Will send own ISUP\n"); 
	return 0;
    }
Mac Newbold's avatar
Mac Newbold committed
979

Mac Newbold's avatar
Mac Newbold committed
980 981 982 983 984 985
    # 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;
986
    }
Mac Newbold's avatar
Mac Newbold committed
987

Mac Newbold's avatar
Mac Newbold committed
988 989 990 991 992 993 994 995
    # If os doesn't support ping or isup, stated sends ISUP just after 
    # 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   => TBDB_TBEVENT_NODESTATE ,
		  eventtype => TBDB_NODESTATE_ISUP ,
		  objname   => $node);
Robert Ricci's avatar
Robert Ricci committed
996 997
}

998 999 1000 1001
# Figure out if this node belongs to us (ie. if it's using our database.)
# Returns 1 if it does, 0 if not
sub checkDBRedirect($) {

Mac Newbold's avatar
Mac Newbold committed
1002
    my ($node) = @_;
1003

Mac Newbold's avatar
Mac Newbold committed
1004 1005 1006 1007 1008 1009 1010 1011
    # XXX: I don't want to do this every time, for performance reaons,
    # but we need to make sure that we don't get into an inconsistent 
    # state
    my $result=DBQueryFatal("SELECT testdb FROM nodes as n " .
			    "LEFT JOIN reserved as r ON n.node_id=r.node_id ".
			    "LEFT JOIN experiments as e ON r.pid = e.pid " .
			    "AND r.eid = e.eid " .
			    "WHERE n.node_id = '$node'");
1012

Mac Newbold's avatar
Mac Newbold committed
1013 1014 1015 1016
    if (!$result->num_rows()) {
	notify("Got an event for a node ($node) I don't know about\n");
	return 0;
    }
1017

Mac Newbold's avatar
Mac Newbold committed
1018
    my ($testdb) = $result->fetchrow();
1019

Mac Newbold's avatar
Mac Newbold committed
1020 1021 1022
    # XXX: It's hokey to hardcode tbdb here, but....

    #debug("checkDBRedirect: $node => $testdb (I'm $TBDBNAME)\n");
1023
    if ((!$testdb && ($TBDBNAME eq $REALTBDBNAME)) ||
Mac Newbold's avatar
Mac Newbold committed
1024 1025 1026 1027 1028
	($testdb && ($testdb eq $TBDBNAME))) {
	return 1;
    } else {
	return 0;
    }
1029 1030
}

1031 1032 1033 1034
# Check if this state has a timeout, and if so, put it in the queue
sub setTimeout( $$$$ ) {
    my ($mode,$state,$node,$now) = @_;
    if (0) { print "Original: ($mode,$state,$node,$now)\n"; qshow(); }
1035
    if (defined(qfind($node))) { qdelete($node); delete($timeout_tag{$node}); }
1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046
    if (0) { print "Deleted:\n"; qshow(); }
    if (defined($mode) && defined($state) &&
	defined($timeouts{$mode}) &&
	defined($timeouts{$mode}{$state})) {
	my $deadline = ${$timeouts{$mode}{$state}}[0];
        if (defined($deadline) &&
	    $deadline != $TBNOTIMEOUT) {
	    my $TO = $deadline + $now;
	    debug("Setting timeout for ($node,$mode,$state) at ".
		  "$deadline + $now ($TO)\n");
	    qinsert($TO,$node);
1047
	    $timeout_tag{$node} = "$mode:$state";
1048 1049 1050 1051 1052 1053
	    if (0) { qshow(); }
	}
    }
    if (0) { print "Done:\n"; qshow(); }
}

Robert Ricci's avatar
Robert Ricci committed
1054 1055
# Reload state from the database
sub reload() {
Mac Newbold's avatar
Mac Newbold committed
1056 1057 1058 1059 1060 1061
    debug("Reloading state from database\n");
    $last_reload = time();
    %timeouts  = getTimeouts();
    %valid     = getValid();
    %modeTrans = getModeTrans();
    %triggers  = getTriggers();
1062
    %nodes     = readStates(%nodes);
Robert Ricci's avatar
Robert Ricci committed
1063 1064
}

1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144
#
# Some functions for node triggers
#

# $rv   = AddNodeTrigger($node, $mode, $state, @triglist);
sub AddNodeTrigger( $$$@ ) {
    my ($node, $mode, $state, @trigs) = @_;
    if (@trigs == 0) { return 1; }
    if (defined($triggers{"$node:$mode:$state"})) {
	my %t = ();
	foreach $k (@{$triggers{"$node:$mode:$state"}}) { $t{$k} = 1; }
	my @newtrigs = ();
	foreach $k (@trigs) { if (!defined($t{$k})) { push(@newtrigs,$k); } }
	push(@{$triggers{"$node:$mode:$state"}},@newtrigs);
    } else {
	$triggers{"$node:$mode:$state"} = \@trigs;
    }
    my $triglist = join(",",@{$triggers{"$node:$mode:$state"}});
    DBQueryFatal("replace into state_triggers ".
		 "(node_id,op_mode,state,trigger) values ".
		 "('$node','$mode','$state','$triglist')");
    return 0;
}

# @list = GetNodeTriggerList($node, $mode, $state);
sub GetNodeTriggerList( $$$ ) {
    my ($node, $mode, $state) = @_;
    my @l = ();
    if (defined($triggers{"$node:$mode:$state"})) {
	push(@l,@{$triggers{"$node:$mode:$state"}});
    }
    if (defined($triggers{"$node:$TBANYMODE:$state"})) {
	push(@l,@{$triggers{"$node:$TBANYMODE:$state"}});
    }
    return @l;
}

# $rv   = ClearNodeTrigger($node, $mode, $state, @triglist);
# Note: When not clearing all triggers, ordering is not preserved!
sub ClearNodeTrigger( $$$ ; @ ) {
    my ($node, $mode, $state, @trigs) = @_;
    # We have to keep any triggers that aren't on the list, but the
    # most common case will be that the list they give us is the whole
    # list anyway. So treat that case special.
    my @reallist = GetNodeTriggerList($node,$mode,$state);
    # empty list means clear all...
    if ((@trigs==0) || join(",",sort @reallist) eq join(",",sort @trigs)) {
	# Same list... just nuke the entry
	debug("Clearing all triggers for $node...\n");
	delete($triggers{"$node:$mode:$state"});
	delete($triggers{"$node:$TBANYMODE:$state"});
	DBQueryFatal("delete from state_triggers ".
		     "where node_id='$node' and state='$state' and ".
		     "(op_mode='$mode' or op_mode='$TBANYMODE')");
    } else {
	# Subtract @trigs from @reallist
	my %temptrigs = ();
	foreach $k (@reallist) { $temptrigs{$k} = 1; }
	debug("Reallist = ".join("/",@reallist).", trigs=".
	      join("/",@trigs).".\n");
	foreach $t (@trigs) {
	    if (defined($temptrigs{$t})) {
		delete($temptrigs{$t});
		debug("Clearing $t\n");
	    }
	}
	# Note: This doesn't quite do the right thing with triggers
	# for a fixed mode vs TBANYMODE. So if you start using this
	# code, make sure and debug it first!
	my @newtrigs = keys %temptrigs;
	debug("Newlist = ".join("/",@newtrigs).".\n");
	delete($triggers{"$node:$mode:$state"});
	if (@newtrigs > 0) {
	    AddNodeTrigger($node,$mode,$state,@newtrigs);
	} 
    }

    return 0;
}

1145 1146
sub os_opmode() {
    my $osid = shift || "";
Mac Newbold's avatar
Mac Newbold committed
1147 1148 1149
    if ($osid eq $TB_OSID_MBKERNEL) {
	return "MINIMAL";
    }
1150 1151
    my $cmd = "select op_mode from os_info where osid='$osid';";
    my $q = DBQueryFatal($cmd);
Mac Newbold's avatar
Mac Newbold committed
1152 1153 1154
    if ($q->numrows() < 1) {
	return "";
    }
1155 1156 1157
    my @r = $q->fetchrow_array();
    my $opmode=$r[0];
    debug("OpMode for '$osid' is '$opmode'\n");
Mac Newbold's avatar
Mac Newbold committed
1158 1159 1160
    if (defined($opmode) && $opmode ne "") {
	return $opmode;
    }
1161 1162
    return "";
}
Mac Newbold's avatar
Mac Newbold committed
1163

1164 1165 1166 1167
#
# Functions for controlling output/logging, and signal handling
#

Robert Ricci's avatar
Robert Ricci committed
1168
sub debug(@) {
Mac Newbold's avatar
Mac Newbold committed
1169 1170 1171
    if ($debug) {
	print @_;
    }
Mac Newbold's avatar
Mac Newbold committed
1172 1173 1174
}

sub fatal($) {
Mac Newbold's avatar
Mac Newbold committed
1175 1176 1177
    my $msg = shift;
    notify($msg);
    die($msg);
Robert Ricci's avatar
Robert Ricci committed
1178 1179
}

Mac Newbold's avatar
Mac Newbold committed
1180
sub showqueue() {
Mac Newbold's avatar
Mac Newbold committed
1181 1182 1183 1184 1185 1186 1187 1188 1189 1190
    if ($debug < 2) {
	return;
    }
    if ((keys %msgs) > 0) {
	debug("\nMAILQUEUE:\n");
    }
    foreach $k (sort keys %msgs) {
	my @l = @{$msgs{$k}};
	debug("MSGS:\n$k==> (".(@l+0).",'".join("','",@l)."')\n");
    }
Mac Newbold's avatar
Mac Newbold committed
1191 1192 1193
}

sub notify($;$) {
Mac Newbold's avatar
Mac Newbold committed
1194 1195 1196 1197
    my $message = shift;
    my $checkonly = shift || 0;
    # Use a timestamp, now that we're throttling mail
    my $tstamp=strftime("%b %e %H:%M:%S",localtime);
Mac Newbold's avatar
Mac Newbold committed
1198
    showqueue();
Mac Newbold's avatar
Mac Newbold committed
1199 1200
    if (!$checkonly) {
	info($message);
1201
	$mailqueue++;
Mac Newbold's avatar
Mac Newbold committed
1202 1203 1204 1205
	# Queue up the message
	# (The queue is a hash of lists of timestamps, keyed by message
	if (defined($msgs{$message})) {
	    push(@{$msgs{$message}},$tstamp);
Mac Newbold's avatar
Mac Newbold committed
1206
	} else {
Mac Newbold's avatar
Mac Newbold committed
1207
	    $msgs{$message} = [$tstamp];
Mac Newbold's avatar
Mac Newbold committed
1208
	}
Mac Newbold's avatar
Mac Newbold committed
1209
	showqueue();
Mac Newbold's avatar
Mac Newbold committed
1210
    }
Mac Newbold's avatar
Mac Newbold committed
1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233
    my $now = time;
    if ($now - $lastmail >= $mailgap) {
	if ((keys %msgs)>0) {
	    debug("SENDING MAILQUEUE\n"."(now $now, lastmail $lastmail, ".
		  ($now-$lastmail).">=$mailgap)\n");
	    my $mailbody="";
	    my $sep = '-'x5;
	    # We're okay to send. Make a digest of all the queued messages.
	    foreach my $msg (sort keys %msgs) {
		my @tlist = @{$msgs{$msg}};
		my $count = 0+@tlist;
		$mailbody .= "\n$msg\n";
		if ($count > 1) {
		    my $first = shift @tlist;
		    my $last = pop @tlist;
		    $mailbody .= "($count copies from $first to $last)\n";
		} else {
		    $mailbody .= "($count copy at $tlist[0])\n";
		}
		$mailbody .= "$sep\n";
	    }
	    # Now reset the mail queue
	    %msgs = ();
1234
	    $mailqueue=0;
Mac Newbold's avatar
Mac Newbold committed
1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246
	    showqueue();
	    $lastmail = time;
	    if (!$debug) {
		SENDMAIL("Stated List <".$TBOPS.">",
			 "Stated Messsage",$mailbody,
			 "Stated Daemon <".$TBOPS.">");
	    } else {
		debug("notify: Not sending mail in debug mode\n");
		debug("MAIL CONTAINS:\n".$mailbody."\n");
	    }
	}
    } # else do nothing, not time yet
1247 1248
}

1249
sub announce($) {
Mac Newbold's avatar
Mac Newbold committed
1250 1251 1252 1253 1254 1255 1256 1257