stated.in 40.5 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) 2000-2003 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
my $TBCONTROL    = TBDB_TBEVENT_CONTROL;
my $TBCOMMAND    = TBDB_TBEVENT_COMMAND;
119
my $TBFAILED     = TBDB_TBEVENT_TBFAILED;
120 121 122 123
my $TBREBOOT     = TBDB_COMMAND_REBOOT;
my $TBPOWEROFF   = TBDB_COMMAND_POWEROFF;
my $TBPOWERON    = TBDB_COMMAND_POWERON;
my $TBPOWERCYCLE = TBDB_COMMAND_POWERCYCLE;
124
my $TBISUP       = TBDB_NODESTATE_ISUP;
125 126 127 128
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;
129

130 131 132
# This only gets used here, so it isn't in a lib constant.
my $TBFREENODE = "FREENODE";

Mac Newbold's avatar
Mac Newbold committed
133
my $pidfile;
134
if ( $TB eq $REALTB ) {
Mac Newbold's avatar
Mac Newbold committed
135 136
    $pidfile = "/var/run/stated.pid";
} else {
137
    $pidfile = "$TB/locks/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
138
}
Mac Newbold's avatar
Mac Newbold committed
139 140
debug("Using pidfile $pidfile\n");
if (-e $pidfile) {
Mac Newbold's avatar
Mac Newbold committed
141 142 143 144 145 146 147 148 149 150 151
    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: $? $!\n");
Mac Newbold's avatar
Mac Newbold committed
152
}
Robert Ricci's avatar
Robert Ricci committed
153
# Background
154
if (!$debug) {
Mac Newbold's avatar
Mac Newbold committed
155 156 157 158
    # We use syslog, so redirect the output to nothing
    if (TBBackGround("/dev/null")) {
	exit(0);
    }
Robert Ricci's avatar
Robert Ricci committed
159
}
Mac Newbold's avatar
Mac Newbold committed
160 161
# set up syslog
openlog("stated","pid","user");
Mac Newbold's avatar
Mac Newbold committed
162 163 164 165 166 167
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
168

169 170 171
# Change my $0 so that it is easier to see in a ps/top
$0 = "$0";

Robert Ricci's avatar
Robert Ricci committed
172
my $URL = "elvin://$server";
Mac Newbold's avatar
Mac Newbold committed
173 174 175
if ($port) {
    $URL .= ":$port";
}
Robert Ricci's avatar
Robert Ricci committed
176 177 178

# 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
179 180 181
if (!$handle) {
    fatal("Unable to register with event system\n");
}
Robert Ricci's avatar
Robert Ricci committed
182 183

my $tuple = address_tuple_alloc();
Mac Newbold's avatar
Mac Newbold committed
184 185 186
if (!$tuple) {
    fatal("Could not allocate an address tuple\n");
}
Robert Ricci's avatar
Robert Ricci committed
187

188 189 190
%$tuple = ( objtype => join(",",
			    $TBNODESTATE, $TBNODEOPMODE,
			    $TBCONTROL, $TBCOMMAND) );
191

Robert Ricci's avatar
Robert Ricci committed
192
if (!event_subscribe($handle,\&handleEvent,$tuple)) {
Mac Newbold's avatar
Mac Newbold committed
193
    fatal("Could not subscribe to events\n");
Robert Ricci's avatar
Robert Ricci committed
194 195 196 197
}

# Read in the pre-existing node states, and timeout and valid transition
# information from the database
198 199 200 201
my %timeouts  = getTimeouts();
my %valid     = getValid();
my %modeTrans = getModeTrans();
my %triggers  = getTriggers();
202
my %nodes     = readStates();
203
my %timeouttag= ();
204
if ($debug) { qshow(); }
Robert Ricci's avatar
Robert Ricci committed
205 206 207

# Gets set if a reload of state from the database should happen.
my $do_reload = 0;
208 209
my $sigrestart= 0;
my $sigcleanup= 0;
Robert Ricci's avatar
Robert Ricci committed
210 211 212 213

# 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.
214 215
$SIG{HUP}  = sub { info("SIGHUP - Reloading DB state\n"); $do_reload = 1; };

Mac Newbold's avatar
Mac Newbold committed
216
# Set up other signals.
217 218 219 220 221 222 223
$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
224

225 226 227
# Track if I handled an event or not
my $event_count = 0;

228 229
# Control how long I block while waiting for events
my $blockwait=0;
230
my $nextdeadline=0;
231 232
my $mailqueue=0;

233 234 235 236 237
notify("Stated starting up\n");

sub process_event_queue() {
    $event_count=0;
    my $lastcount=-1;
238 239 240 241 242 243 244 245 246 247 248 249 250
    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;
251 252
	    } else {
		$wait = 0;
253 254 255 256 257
	    }
	}
    } else {
	# mail is waiting. Only block until it is time to send it.
	$wait = $lastmail + $mailgap - $now;
Mac Newbold's avatar
Mac Newbold committed
258
	debug("Now $now, mailgap $mailgap, last $lastmail ==> wait $wait\n");
259 260 261
    }
    if ($wait < 0) { debug("Wait was $wait!\n"); $wait=0; }
    my $finish = $now + $wait;
Mac Newbold's avatar
Mac Newbold committed
262 263
    while (($event_count != $lastcount || $wait > 0) &&
	   !($sigrestart || $sigcleanup || $do_reload)) {
264
	$lastcount = $event_count;
Mac Newbold's avatar
Mac Newbold committed
265 266
	# Don't block if we got a signal!
	if ($wait<=0 || $sigrestart || $sigcleanup || $do_reload) {
267 268 269 270 271 272 273 274
	    event_poll($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
275
	    debug("Finished blocking event poll - $wait seconds remain\n");
276
	    if ($event_count > 0 &&
Mac Newbold's avatar
Mac Newbold committed
277 278
		(qsize() > 0 || $mailqueue ||
		 $sigrestart || $sigcleanup || $do_reload)) {
279 280 281 282 283 284 285
		$blockwait=0;
		$wait=0;
		#debug("Cancelling wait - timeouts/msgs waiting, or HUP'd\n");
		#debug("---End Blocking Wait ---\n");
	    }
	}
	#debug("Wait is $wait\n");
286 287
    }
    if ($event_count > 0) {
Mac Newbold's avatar
Mac Newbold committed
288
	debug("Handled $event_count event(s).\n");
289 290
    }
}
Robert Ricci's avatar
Robert Ricci committed
291

292
# Now, we just poll for events, and watch for timeouts
Robert Ricci's avatar
Robert Ricci committed
293
while (1) {
Mac Newbold's avatar
Mac Newbold committed
294
    my $now = time();
295 296 297 298
    my ($deadline,$node);

    # Check for nodes that have passed their timeout
    if (!qhead($deadline,$node)) {
299
	debug("HEAD: $node in ".($deadline-$now).", queue=".qsize()."\n");
300 301
	while ($now >= $deadline && $node ne "") {
	    qpop($deadline,$node);
302
	    debug("POP: $node in ".($deadline-$now).", queue=".qsize()."\n");
303
	    handleCtrlEvent($node,$TBTIMEOUT);
304 305 306 307
	    if (0) { qshow(); }
	    if (qhead($deadline,$node)) {
		$deadline=0; $node="";
	    }
308
	}
309 310
    } else {
	$deadline=0;
311
    }
312
    $nextdeadline = $deadline;
313

314 315 316 317 318
    if (qsize()==0) {
	$blockwait=1;
	debug("---Blocking wait okay---\n");
    }
		
Mac Newbold's avatar
Mac Newbold committed
319 320 321 322
    if ($do_reload || ($now - $last_reload > $reload_time)) {
	reload();
	$do_reload = 0;
    }
Mac Newbold's avatar
Mac Newbold committed
323

Mac Newbold's avatar
Mac Newbold committed
324 325
    # Send any messages in the queue if it is time
    notify("",1);
Mac Newbold's avatar
Mac Newbold committed
326

327 328 329
    if ($sigrestart) { restart(); }
    if ($sigcleanup) { cleanup(); }

330
    process_event_queue;
Robert Ricci's avatar
Robert Ricci committed
331 332
}

Mac Newbold's avatar
Mac Newbold committed
333 334
exit(0);

Robert Ricci's avatar
Robert Ricci committed
335
# Read the current states of nodes from the database
336
sub readStates(;@) {
Mac Newbold's avatar
Mac Newbold committed
337 338 339 340 341 342
    my %oldnodes = @_;

    # Guard against undefined variable warnings
    if (! defined(%oldnodes)) {
	%oldnodes = ();
    }
343

Mac Newbold's avatar
Mac Newbold committed
344 345 346
    #debug("readStates called\n");
    my $result = DBQueryFatal("SELECT node_id, eventstate, " .
			      "state_timestamp, op_mode, " .
347 348
			      "op_mode_timestamp FROM nodes ".
			      "where node_id not like 'sh%'");
Mac Newbold's avatar
Mac Newbold committed
349 350 351 352

    my %nodes;
    while (my ($node_id, $state, $timestamp, $mode, $mode_timestamp)
	   = $result->fetchrow()) {
353
	#
Mac Newbold's avatar
Mac Newbold committed
354 355 356 357
	# If there's an entry in oldnodes for this node, and it
	# hasn't changed state or time, use the old entry (so that
	# we don't lose information about which nodes we've already
	# notified the ops about, etc.)
358
	#
Mac Newbold's avatar
Mac Newbold committed
359 360 361 362 363
	if ($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};
364
	} else {
Mac Newbold's avatar
Mac Newbold committed
365 366 367 368
	    $nodes{$node_id}{state}          = $state;
	    $nodes{$node_id}{timestamp}      = $timestamp;
	    $nodes{$node_id}{mode}           = $mode;
	    $nodes{$node_id}{mode_timestamp} = $mode_timestamp;
369 370 371
	    $nodes{$node_id}{notified}       = 0;
	    $nodes{$node_id}{timedout}       = 0;
	    $nodes{$node_id}{noretry}        = 0;
372
	    $nodes{$node_id}{rebooting}      = 0;
373 374
	    # Is there a timeout? If so, set it up!
	    setTimeout($mode,$state,$node_id,$timestamp);
375
	}
Mac Newbold's avatar
Mac Newbold committed
376 377
    }
    return %nodes;
Robert Ricci's avatar
Robert Ricci committed
378 379 380 381 382 383
}

#
# Read timeouts for various states from the database
#
sub getTimeouts() {
Mac Newbold's avatar
Mac Newbold committed
384 385 386
    #debug("getTimeouts called\n");
    my $result = DBQueryFatal("SELECT op_mode, state, timeout, action " .
			      "FROM state_timeouts");
Robert Ricci's avatar
Robert Ricci committed
387

Mac Newbold's avatar
Mac Newbold committed
388 389 390 391 392
    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
393 394 395 396 397 398
}

#
# Read the list of valid state transitions from the database
#
sub getValid() {
Mac Newbold's avatar
Mac Newbold committed
399 400 401
    #debug("getValid called\n");
    my $result = DBQueryFatal("SELECT op_mode, state1, state2 " .
			      "FROM state_transitions");
Robert Ricci's avatar
Robert Ricci committed
402

Mac Newbold's avatar
Mac Newbold committed
403 404 405 406 407
    my %valid;
    while (my ($mode,$state1, $state2) = $result->fetchrow()) {
	$valid{$mode}{$state1}{$state2} = 1;
    }
    return %valid;
Robert Ricci's avatar
Robert Ricci committed
408 409
}

410 411 412 413
#
# Read the list of valid mode transitions from the database
#
sub getModeTrans() {
Mac Newbold's avatar
Mac Newbold committed
414 415 416 417 418 419 420 421 422 423 424 425 426
    #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;
427
	}
Mac Newbold's avatar
Mac Newbold committed
428 429
    }
    return %modeTrans;
430 431 432 433 434 435
}

#
# Read the list of states which trigger an action
#
sub getTriggers() {
436 437 438 439 440
    debug("getTriggers called\n");
    
    debug("anymode ==> '$TBANYMODE'\n");

    # Grab global triggers
Mac Newbold's avatar
Mac Newbold committed
441 442
    my $result = 
      DBQueryFatal("SELECT op_mode, state, trigger " .
443 444
		   "FROM state_triggers where node_id='$TBANYMODE' ".
		   "order by op_mode,state");
Mac Newbold's avatar
Mac Newbold committed
445
    my %t;
446
    while (my ($mode, $state, $trig) = $result->fetchrow()) {
Mac Newbold's avatar
Mac Newbold committed
447
	$t{"$mode:$state"} = $trig;
448 449 450 451 452 453 454 455 456 457 458 459
	debug("trig($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
460
    }
461

Mac Newbold's avatar
Mac Newbold committed
462
    return %t;
463 464
}

Robert Ricci's avatar
Robert Ricci committed
465 466 467 468
#
# Gets called for every event that we recieve
#
sub handleEvent($$$) {
Mac Newbold's avatar
Mac Newbold committed
469 470 471 472 473 474 475 476 477 478 479
    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
    #
480
    if ($objtype ne $TBCOMMAND && !checkDBRedirect($objname)) {
Mac Newbold's avatar
Mac Newbold committed
481 482 483
	info("Got an event for node $objname, which isn't mine\n");
	return;
    }
484

Mac Newbold's avatar
Mac Newbold committed
485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500
  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,$eventtype);
	    last;
	};
501 502 503 504
	(/$TBCOMMAND/) && do {
	    handleCommand($objname,$eventtype);
	    last;
	};
505

Mac Newbold's avatar
Mac Newbold committed
506
    }
507 508 509 510 511

}

sub stateTransition($$) {

512
    my ($node,$newstate) = @_;
Robert Ricci's avatar
Robert Ricci committed
513

514 515 516 517 518 519 520 521
    # 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();
522
	if ($nodes{$node}) {
523 524
	    $oldstate = $nodes{$node}{state};
	    $mode = $nodes{$node}{mode};
Robert Ricci's avatar
Robert Ricci committed
525
	} else {
526
	    notify("Got an event for a node ($node) I don't know about\n");
Robert Ricci's avatar
Robert Ricci committed
527
	}
528 529 530 531
    }
    if ($oldstate && $mode && $valid{$mode} && $valid{$mode}{$oldstate} &&
	!$valid{$mode}{$oldstate}{$newstate}) {
	notify("Invalid transition for node $node from $mode/$oldstate " .
532
	       "to $newstate\n");
533
    }
Robert Ricci's avatar
Robert Ricci committed
534

535 536 537 538
    my $now = time();
    $nodes{$node}{state}     = $newstate;
    $nodes{$node}{timestamp} = $now;
    $nodes{$node}{notified}  = 0;
539

540 541 542
    info("$node: $mode/$oldstate => $mode/$newstate\n");
    DBQueryFatal("UPDATE nodes SET eventstate='$newstate', " .
		 "state_timestamp='$now' WHERE node_id='$node'");
543

544 545 546 547
    # 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} =~ /^$TBCOMMAND:/) {
548
        debug("TimeoutTag = '$timeout_tag{$node}'\n");
549
	my ($str,$cmd) = split(":",$timeout_tag{$node});
550
	debug("str=$str\tcmd=$cmd\tTBREBOOT=$TBREBOOT\tstate=$newstate\n");
551
	if ($cmd eq $TBREBOOT) {
552
	    if ($newstate eq TBDB_NODESTATE_SHUTDOWN ) {
553 554 555
		info("$node: $TBREBOOT success\n");
		# Timeout will get cleared below by setTimeout call
	    } else {
556 557
		notify("$node: $TBREBOOT in progress, but got state ".
		       "$newstate instead of ".TBDB_NODESTATE_SHUTDOWN."!\n");
558 559 560 561 562
	    }
	#} elsif ($cmd eq $FOO ) {
	    # Add more here...
	} else {
	    notify("$node: Unknown command timeout '$timeout_tag{$node}' ".
563
		   "found at $mode/$newstate\n");
564 565 566
	}
    }

567 568 569
    # Check if this state has a timeout, and if so, put it in the queue
    setTimeout($mode,$newstate,$node,$now);

570 571
    # Check if this is TBDB_NODESTATE_BOOTING , which has actions
    if ($newstate eq TBDB_NODESTATE_BOOTING) {
572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589
	# If I skipped shutdown, and came to booting directly from isup,
	# check for a mode transition so I don't miss one...
	if ($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
590
	my $osid = TBBootWhat($node,$debug);
591 592
	my $os_op_mode = os_opmode($osid);
	info("$node: Current OS is '$osid', OS mode is '$os_op_mode'\n");
593
	DBQueryFatal("UPDATE nodes SET osid='$osid' WHERE node_id='$node'");
594
	if ($os_op_mode ne $mode) {
595 596
	    my $str = "Node $node is running OS '$osid' but in mode '$mode' ".
	      "instead of mode '$os_op_mode'!\n";
597 598 599
	    # For now, only force if we're going into reload mode, so we
	    # don't get stuck looping in reloading.
	    if ($os_op_mode eq "RELOAD") {
600 601 602 603 604 605 606 607 608
		DBQueryFatal("UPDATE nodes SET op_mode='$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($str);
	}
609 610
	checkGenISUP($node);
    }
611

612 613 614 615 616
    # Check if this state has any triggers
    my @nodetrigs = GetNodeTriggerList($node,$mode,$newstate);
    if (defined($triggers{"$mode:$newstate"}) ||
        (@nodetrigs > 0) ) {
	# check for global triggers
617
	my @trigs = split(/\s*,\s*/,$triggers{"$mode:$newstate"});
618 619 620 621
	# Run all the triggers
	debug("Running triggers. Global=".join("/",@trigs).
	      "   node=".join("/",@nodetrigs)."\n");
	foreach ( @trigs , @nodetrigs) {
622 623
	    my $trig = $_;
	    /^$TBRESET$/ && do {
624 625 626
		# We successfully booted, so clear some flags
		$nodes{$node}{noretry}   = 0;
		$nodes{$node}{timedout}  = 0;
627
		$nodes{$node}{rebooting} = 0;
628 629 630 631 632 633 634 635 636 637 638 639 640
		# 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,$trig);
		next;
	    };
641 642 643 644
	    /^$TBFREENODE$/ && do {
		handleCtrlEvent($node,$trig);
		next;
	    };
645 646 647 648 649 650 651 652
	    /^$TBISUP$/ && do {
		info("$node: Triggered $TBISUP\n");
		EventSendWarn(host      => $BOSSNODE ,
			      objtype   => TBDB_TBEVENT_NODESTATE ,
			      eventtype => TBDB_NODESTATE_ISUP ,
			      objname   => $node);
		next;
	    };
653
	    notify("Unknown trigger '$trig' for $node in $mode/$newstate!\n");
654
	}
655 656 657
	# Clear any of the node triggers that we ran
	debug("Clearing node triggers: ".join("/",@nodetrigs)."\n");
	ClearNodeTrigger($node,$mode,$newstate,@nodetrigs);
658
    }
659

660 661 662 663 664 665 666 667
    # 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
668 669 670
	} else {
	    debug("No next mode.\n");
	}
671 672
    }
}
673

674
sub opModeTransition($$;$) {
Mac Newbold's avatar
Mac Newbold committed
675

676 677
    my ($node,$newmode,$force) = @_;
    if (!defined($force)) { $force = 0; }
Mac Newbold's avatar
Mac Newbold committed
678

679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694
    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 ($node) I don't know about\n");
	}
    }
695
    if (defined($modeTrans{"$mode:$oldstate"}) || $force) {
696
	if (!$force) {
697 698 699 700 701 702 703 704 705 706 707 708 709 710 711
	    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");
	    }
712 713
	}
    } else {
714
	notify("Invalid mode transition for $node from $mode/$oldstate: ".
715 716
	       "Not a valid mode transition state!\n");
    }
Mac Newbold's avatar
Mac Newbold committed
717 718 719
    if (!$nextstate) {
	$nextstate=$oldstate;
    }
Mac Newbold's avatar
Mac Newbold committed
720

721 722 723 724 725 726
    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
727

728 729 730 731 732
    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
733 734 735 736

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

737 738 739 740
}

sub handleCtrlEvent($$) {
    my ($node,$event) = @_;
741

742
    info("CtrlEvent: $node, $event\n");
743

744 745
    foreach ($event) {
	/^$TBRESET$/ && do {
Mac Newbold's avatar
Mac Newbold committed
746 747 748
	    my $result = DBQueryFatal("SELECT pxe_boot_path, def_boot_osid ".
				      "FROM nodes where node_id='$node'");
	    my ($pxepath,$osid) = $result->fetchrow();
749

750 751 752 753
	    # 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.
754

755 756
	    $cmd = "$osselect $osid $node";
	    system($cmd) and
Mac Newbold's avatar
Mac Newbold committed
757 758
	      notify("$node/$event: Couldn't clear next_boot_*\n".
		     "\tcmd=$cmd\n\t*** $!\n");
759

760
	    $pxepath = "-p ".$pxepath;
Mac Newbold's avatar
Mac Newbold committed
761 762 763 764
	    if ($pxepath eq "-p ") {
		$pxepath="PXEBOOT";
	    }
	    ;
Mac Newbold's avatar
Mac Newbold committed
765
	    my $cmd = "$osselect -m $pxepath $node";
766
	    system($cmd) and
Mac Newbold's avatar
Mac Newbold committed
767 768
	      notify("$node/$event: Couldn't clear next_pxe_boot_path\n".
		     "\tcmd=$cmd\n\t*** $!\n");
769

Mac Newbold's avatar
Mac Newbold committed
770
	    info("Performed RESET for $node to $osid/$pxepath\n");
771 772 773 774 775 776 777 778 779 780
	    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'");
781 782 783 784
		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");
785 786 787
	    }
	    next;
	};
788 789 790 791 792 793 794 795
	/^$TBFREENODE$/ && 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;
	};
796
	/^$TBTIMEOUT$/ && do {
797 798 799 800
	    my ($mode,$state) = split(":",$timeout_tag{$node});
	    delete($timeout_tag{$node});
	    my $curstate = $nodes{$node}{state};
	    my $curmode = $nodes{$node}{mode};
801
	    my ($timeout,$action);
802 803 804 805 806
	    if (!defined($nodes{$node}{notified})) {
		$nodes{$node}{notified}=0;
	    }
	    $nodes{$node}{notified}++;
	    my $notified = $nodes{$node}{notified};
807 808
	    $nodes{$node}{timedout}++;
	    my $timedout = $nodes{$node}{timedout};
809 810 811 812
	    if ($mode && $state && $timeouts{$mode} &&
		$timeouts{$mode}{$state}) {
		($timeout, $action) = @{$timeouts{$mode}{$state}};
	    }
813 814
	    if ($mode eq $TBCOMMAND) {
		# It is a command, not a true state
815
		if ($action eq $TBTIMEOUTCMDRETRY) {
816
		    # Retry the command
817
		    notify("$node: Command $state, retry #$timedout\n");
818
		    # notify in case we get in a retry loop...
819
		    handleCommand($node,$state,$timedout,1);
820 821 822 823 824 825
		} else {
		    notify("$node: Unknown timeout action for ".
			   "$mode/$state: '$action'\n");
		}
		next;
	    }
826 827 828 829 830 831
	    info("Node $node has timed out in state $mode/$state".
		 ($action ne "" ? "\n\tRequested action $action." : "").
		 "\n");

	    foreach ($action) {
		/^$TBTIMEOUTREBOOT/ && do {
832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855
		    # If the node is in our control (ie node_reboot),
		    # we want to do something. If it is in the user's
		    # control (went to shutdown without a reboot event),
		    # then we don't want to touch it.
		    if ($nodes{$node}{rebooting}) {
			if ($timedout>1) {
			    # We've tried too many times...
			    # The node has now officially failed to boot
			    notify("Node $node timed out $timedout times!\n".
				   "Giving up until it boots sucessfully.\n");
			    $nodes{$node}{noretry} = 1;
			    info("$node: Sending $TBFAILED $TBREBOOT $node\n");
			    EventSendWarn(host      => $BOSSNODE ,
					  objtype   => $TBFAILED ,
					  eventtype => $TBREBOOT ,
					  objname   => $node);
			} else {
			    # 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);
			}
856
		    } else {
857 858
			info("Node $node timed out in state $mode/$state ".
			     "under user's control - not rebooting\n");
859 860 861 862 863 864 865
		    }
		    last; };
		/^$TBTIMEOUTNOTIFY/ && do {
		    notify("Node $node has timed out in state $mode/$state\n");
		    last; };
		notify("$node: Unknown Timeout Action: $action\n");
	    }
866 867
	    next;
	};
868
	notify("$node: Unknown CtrlEvent: $event\n");
869 870
    }
}
Robert Ricci's avatar
Robert Ricci committed
871

872 873
sub handleCommand($$;$$) {
    my ($params,$command,$retry,$force) = @_;
874
    if (!defined($retry)) { $retry=0; }
875
    if (!defined($force)) { $force=0; }
876 877 878 879 880 881 882 883

    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.

884
    if ($command eq $TBREBOOT && $retry >1) {
885 886
	announce("Node $params has tried rebooting $retry times and has \n".
		 "still not been successful. Please look into it soon.\n".
887 888 889
	"" );#	 "In the meantime, $params will be powered off.\n");
	# Just return...
	return 0;
890
	# change my command to poweroff.
891
	#$command = $TBPOWEROFF;
892 893 894 895 896 897
    }

    foreach ($command) {
	/^$TBREBOOT$/ && do {
	    # For reboot, the params is a comma-separated list of nodes
	    my @nodes = split(",",$params);
898 899 900 901 902 903 904 905

	    # 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");
906 907
		    if (($nodes{$node}{rebooting}) &&
			 (!$nodes{$node}{noretry}) ) {
908
			# This node shouldn't be rebooted now...
909 910 911 912 913
			info("$node: Sending $TBFAILED $TBREBOOT $node\n");
			EventSendWarn(host      => $BOSSNODE ,
				      objtype   => $TBFAILED ,
				      eventtype => $TBREBOOT ,
				      objname   => $node);
914 915 916 917 918 919 920 921 922 923
			info("$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; }
	    }

924 925 926 927 928
	    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.

929 930
	    $nodes{$node}{rebooting}=1;

931 932 933 934 935
	    my $cmd = "$nodereboot -r $nodelist";
	    my $redir = " 2>&1 >> /usr/testbed/log/nodereboot.log &";
	    debug("$cmd $redir\n");
	    system("date $redir");
	    system($cmd.$redir) and
936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957
	      notify("$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};
958
	    info("Sending power $func for nodes: $nodelist\n");
959 960 961 962
	    # Permissions were checked in order to send the message,
	    # so we don't need to do any fancy stuff here.

	    my $cmd = "$power $func $nodelist &";
963 964
	    debug("$cmd\n") and
	    system($cmd) and
965 966 967 968 969 970 971 972 973 974
	      notify("$params/$command: ".
		     "Command '$cmd' failed, error $?: $!\n");

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

975 976 977 978
#
# Check if we need to generate an ISUP
#
sub checkGenISUP($) {
Mac Newbold's avatar
Mac Newbold committed
979 980 981 982 983 984 985 986 987 988
    my ($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
989

Mac Newbold's avatar
Mac Newbold committed
990 991 992 993 994 995
    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
996

Mac Newbold's avatar
Mac Newbold committed
997 998 999 1000 1001
    # 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
1002

Mac Newbold's avatar
Mac Newbold committed
1003 1004 1005 1006 1007 1008
    # 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;
1009
    }
Mac Newbold's avatar
Mac Newbold committed
1010

Mac Newbold's avatar
Mac Newbold committed
1011 1012 1013 1014 1015 1016 1017 1018
    # 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
1019 1020
}

1021 1022 1023 1024
# 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
1025
    my ($node) = @_;
1026

Mac Newbold's avatar
Mac Newbold committed
1027 1028 1029 1030 1031 1032 1033 1034
    # 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'");
1035

Mac Newbold's avatar
Mac Newbold committed
1036 1037 1038 1039
    if (!$result->num_rows()) {
	notify("Got an event for a node ($node) I don't know about\n");
	return 0;
    }
1040

Mac Newbold's avatar
Mac Newbold committed
1041
    my ($testdb) = $result->fetchrow();
1042

Mac Newbold's avatar
Mac Newbold committed
1043 1044 1045
    # XXX: It's hokey to hardcode tbdb here, but....

    #debug("checkDBRedirect: $node => $testdb (I'm $TBDBNAME)\n");
1046
    if ((!$testdb && ($TBDBNAME eq $REALTBDBNAME)) ||
Mac Newbold's avatar
Mac Newbold committed
1047 1048 1049 1050 1051
	($testdb && ($testdb eq $TBDBNAME))) {
	return 1;
    } else {
	return 0;
    }
1052 1053
}

1054 1055 1056 1057
# 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(); }
1058
    if (defined(qfind($node))) { qdelete($node); delete($timeout_tag{$node}); }
1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069
    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);
1070
	    $timeout_tag{$node} = "$mode:$state";
1071 1072 1073 1074 1075 1076
	    if (0) { qshow(); }
	}
    }
    if (0) { print "Done:\n"; qshow(); }
}

Robert Ricci's avatar
Robert Ricci committed
1077 1078
# Reload state from the database
sub reload() {
Mac Newbold's avatar
Mac Newbold committed
1079 1080 1081 1082 1083 1084
    debug("Reloading state from database\n");
    $last_reload = time();
    %timeouts  = getTimeouts();
    %valid     = getValid();
    %modeTrans = getModeTrans();
    %triggers  = getTriggers();
1085
    %nodes     = readStates(%nodes);
Robert Ricci's avatar
Robert Ricci committed
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 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167
#
# 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;
}

1168 1169
sub os_opmode() {
    my $osid = shift || "";
Mac Newbold's avatar
Mac Newbold committed
1170 1171 1172
    if ($osid eq $TB_OSID_MBKERNEL) {
	return "MINIMAL";
    }
1173 1174
    my $cmd = "select op_mode from os_info where osid='$osid';";
    my $q = DBQueryFatal($cmd);
Mac Newbold's avatar
Mac Newbold committed
1175 1176 1177
    if ($q->numrows() < 1) {
	return "";
    }
1178 1179 1180
    my @r = $q->fetchrow_array();
    my $opmode=$r[0];
    debug("OpMode for '$osid' is '$opmode'\n");
Mac Newbold's avatar
Mac Newbold committed
1181 1182 1183
    if (defined($opmode) && $opmode ne "") {
	return $opmode;
    }
1184 1185
    return "";
}
Mac Newbold's avatar
Mac Newbold committed
1186

1187 1188 1189 1190
#
# Functions for controlling output/logging, and signal handling
#

Robert Ricci's avatar
Robert Ricci committed
1191
sub debug(@) {
Mac Newbold's avatar
Mac Newbold committed
1192 1193 1194
    if ($debug) {
	print @_;
    }
Mac Newbold's avatar
Mac Newbold committed
1195 1196 1197
}

sub fatal($) {
Mac Newbold's avatar
Mac Newbold committed
1198 1199 1200
    my $msg = shift;
    notify($msg);
    die($msg);
Robert Ricci's avatar
Robert Ricci committed
1201 1202
}

Mac Newbold's avatar
Mac Newbold committed
1203
sub showqueue() {
Mac Newbold's avatar
Mac Newbold committed
1204 1205 1206 1207 1208 1209 1210 1211 1212 1213
    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
1214 1215 1216
}

sub notify($;$) {
Mac Newbold's avatar
Mac Newbold committed
1217 1218 1219 1220
    my $message = shift;
    my $checkonly = shift || 0;
    # Use a timestamp, now that we're throttling mail
    my $tstamp=strftime("%b %e %H:%M:%S",localtime);
Mac Newbold's avatar
Mac Newbold committed
1221
    showqueue();
Mac Newbold's avatar
Mac Newbold committed
1222 1223
    if (!$checkonly) {
	info($message);
1224
	$mailqueue++;
Mac Newbold's avatar
Mac Newbold committed
1225 1226 1227 1228
	# Queue up the message
	# (The queue is a hash of lists of timestamps, keyed by message
	if (defined($msgs{$message})) {
	    push(@{$msgs{$message}},$tstamp);
Mac Newbold's avatar
Mac Newbold committed
1229
	} else {
Mac Newbold's avatar
Mac Newbold committed
1230
	    $msgs{$message} = [$tstamp];
Mac Newbold's avatar
Mac Newbold committed
1231
	}
Mac Newbold's avatar
Mac Newbold committed
1232
	showqueue();
Mac Newbold's avatar
Mac Newbold committed
1233
    }
Mac Newbold's avatar
Mac Newbold committed
1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256
    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 =