stated.in 42.6 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) 2000-2004 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
my $nodereboot = "$TB/bin/node_reboot";
31
my $rebootlog  = "$TB/log/nodereboot.log";
32
my $power = "$TB/bin/power";
33
my $TBLOG = "@TBLOGFACIL@";
Robert Ricci's avatar
Robert Ricci committed
34 35 36 37 38 39

$| = 1;

use event;
use libdb;
use libtestbed;
40
use TimeoutQueue;
Robert Ricci's avatar
Robert Ricci committed
41
use Getopt::Std;
42
#use strict;
Robert Ricci's avatar
Robert Ricci committed
43
use English;
Mac Newbold's avatar
Mac Newbold committed
44 45
use POSIX;			# for strftime, and sigprocmask and friends
use Fcntl;			# file constants for pidfile
Mac Newbold's avatar
Mac Newbold committed
46 47 48 49 50
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
51

Mac Newbold's avatar
Mac Newbold committed
52 53 54
# Do lots of db retries before we fail and die
$libdb::DBQUERY_MAXTRIES = 5;

55 56 57 58 59
# 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
60 61 62
# 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.
63
my $reload_time = 600;
64
my $last_reload = time;
Robert Ricci's avatar
Robert Ricci committed
65

66 67 68 69 70 71 72 73
# Command line opts.
my $dbtag = "";
my $debug = 0;
my $nolog = 0;
my ($server,$port);
my $lockfile;
my $pidfile;

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

sub usage {
Mac Newbold's avatar
Mac Newbold committed
77
    print << "END";
78
Usage: $0 [-h] [-d] [-s server] [-p port] [-t dbtag]
79
-h              This message
80 81 82
-d              Turn on debugging output, and do not go into the background
-l              Do not use syslog; send output to stderr. Use with -d only
-t tag          Use only those nodes with matching tag in nodes table
Robert Ricci's avatar
Robert Ricci committed
83 84
-s server       Use specified server, instead of this site's bossnode
-p port	        Use specified port
85
Send SIGHUP to reload database state, or SIGUSR1 to restart completely.
Robert Ricci's avatar
Robert Ricci committed
86
END
87
    exit(1);
Robert Ricci's avatar
Robert Ricci committed
88 89
}

Mac Newbold's avatar
Mac Newbold committed
90
# Only root should run this - it won't work when run as a user...
91
# (Or, let an admin run it if it isn't the real one in /usr/testbed/ )
92
if ($UID && ( $TB eq $REALTB || ! TBAdmin($UID) ) ) {
Mac Newbold's avatar
Mac Newbold committed
93 94 95
    die("Only root can run this script!\n");
}

96
my @args = @ARGV;    # save a copy for restart before we mess with them.
Robert Ricci's avatar
Robert Ricci committed
97
my %opt = ();
98
if (!getopts("ds:p:ht:l",\%opt)) { usage(); }
Robert Ricci's avatar
Robert Ricci committed
99

Mac Newbold's avatar
Mac Newbold committed
100 101 102 103 104 105
if ($opt{h}) {
    exit &usage;
}
if (@ARGV) {
    exit &usage;
}
Robert Ricci's avatar
Robert Ricci committed
106

Mac Newbold's avatar
Mac Newbold committed
107 108 109 110 111 112 113 114
if ($opt{s}) {
    $server = $opt{s};
} else {
    $server = $BOSSNODE;
}
if ($opt{p}) {
    $port = $opt{p};
}
115 116 117 118 119 120 121 122
if ($opt{l}) {
    usage()
	if (! $opt{d});
    $nolog = 1;
}
if ($opt{t}) {
    $dbtag = $opt{t};
}
Mac Newbold's avatar
Mac Newbold committed
123 124 125
if ($opt{d}) {
    $debug = 1;
}
Robert Ricci's avatar
Robert Ricci committed
126

127
# Grab some constants into variables
128
my $TBANYMODE    = TBDB_NODEOPMODE_ANY;
129 130 131 132 133 134
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;
135 136 137 138 139 140
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;
141
my $TBISUP       = TBDB_NODESTATE_ISUP;
142 143 144
my $PXEWAIT      = TBDB_NODESTATE_PXEWAIT;
my $PXEWAKEUP    = TBDB_NODESTATE_PXEWAKEUP;
my $PXEBOOTING   = TBDB_NODESTATE_PXEBOOTING;
145 146 147 148
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;
149

150 151
# Special PXEBOOT state machine that all local nodes use.
my $PXEKERNEL	 = "PXEKERNEL";
152

153 154 155
if (!$debug) {
    if ( $TB eq $REALTB ) {
	$pidfile = "/var/run/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
156
    } else {
157
	$pidfile = "$TB/locks/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
158
    }
159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175
    debug("Using pidfile $pidfile\n");
    
    if (-e $pidfile) {
	my $otherpid = `cat $pidfile`;
	my $running = `ps -auxww | grep $otherpid | grep -v grep`;
	if ($running ne "") {
	    fatal("Lockfile $pidfile exists, and process $otherpid appears ".
		  "to be running.\n");
	} else {
	    notify("Lockfile exists, but process $otherpid appears to be dead".
		   "\n".
		   "Removing lock file...\n");
	}
	system("rm $pidfile") &&
	    fatal("Couldn't remove $pidfile: $? $!\n");
    }
    # Background
Mac Newbold's avatar
Mac Newbold committed
176 177 178 179
    # We use syslog, so redirect the output to nothing
    if (TBBackGround("/dev/null")) {
	exit(0);
    }
Robert Ricci's avatar
Robert Ricci committed
180
}
Mac Newbold's avatar
Mac Newbold committed
181
# set up syslog
182 183 184 185 186 187 188 189 190 191 192 193
if (! $nolog) {
    openlog("stated","pid",$TBLOG);
}

if (defined($pidfile)) {
    sysopen(PIDFILE, $pidfile, O_WRONLY | O_EXCL | O_CREAT) ||
	fatal("Couldn't create '$pidfile': $? $!\n");
    print PIDFILE "$$";
    close PIDFILE;
    # If I make it to here, I'll need to clean up the lock file
    $lockfile = $pidfile;
}
Robert Ricci's avatar
Robert Ricci committed
194

195 196 197
# Change my $0 so that it is easier to see in a ps/top
$0 = "$0";

Robert Ricci's avatar
Robert Ricci committed
198
my $URL = "elvin://$server";
Mac Newbold's avatar
Mac Newbold committed
199 200 201
if ($port) {
    $URL .= ":$port";
}
Robert Ricci's avatar
Robert Ricci committed
202 203 204

# 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
205 206 207
if (!$handle) {
    fatal("Unable to register with event system\n");
}
Robert Ricci's avatar
Robert Ricci committed
208 209

my $tuple = address_tuple_alloc();
Mac Newbold's avatar
Mac Newbold committed
210 211 212
if (!$tuple) {
    fatal("Could not allocate an address tuple\n");
}
Robert Ricci's avatar
Robert Ricci committed
213

214 215 216
%$tuple = ( objtype => join(",",
			    $TBNODESTATE, $TBNODEOPMODE,
			    $TBCONTROL, $TBCOMMAND) );
217

Robert Ricci's avatar
Robert Ricci committed
218
if (!event_subscribe($handle,\&handleEvent,$tuple)) {
Mac Newbold's avatar
Mac Newbold committed
219
    fatal("Could not subscribe to events\n");
Robert Ricci's avatar
Robert Ricci committed
220 221 222 223
}

# Read in the pre-existing node states, and timeout and valid transition
# information from the database
224 225 226 227
my %timeouts  = getTimeouts();
my %valid     = getValid();
my %modeTrans = getModeTrans();
my %triggers  = getTriggers();
228
my %nodes     = readStates();
229
my %timeouttag= ();
230
if ($debug) { qshow(); }
Robert Ricci's avatar
Robert Ricci committed
231 232 233

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

# Make the daemon reload database state on a sighup - but I'm worried
# about what would happen if we tried to do this mid-loop. So, we'll
# just set a flag and do it when we're done with our current pass.
240 241
$SIG{HUP}  = sub { info("SIGHUP - Reloading DB state\n"); $do_reload = 1; };

Mac Newbold's avatar
Mac Newbold committed
242
# Set up other signals.
243 244 245 246 247 248 249
$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
250

251 252 253
# Track if I handled an event or not
my $event_count = 0;

254 255
# Control how long I block while waiting for events
my $blockwait=0;
256
my $nextdeadline=0;
257 258
my $mailqueue=0;

259 260 261 262 263
notify("Stated starting up\n");

sub process_event_queue() {
    $event_count=0;
    my $lastcount=-1;
264 265 266 267 268 269 270 271 272 273 274 275 276
    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;
277 278
	    } else {
		$wait = 0;
279 280 281 282 283
	    }
	}
    } else {
	# mail is waiting. Only block until it is time to send it.
	$wait = $lastmail + $mailgap - $now;
Mac Newbold's avatar
Mac Newbold committed
284
	debug("Now $now, mailgap $mailgap, last $lastmail ==> wait $wait\n");
285 286 287
    }
    if ($wait < 0) { debug("Wait was $wait!\n"); $wait=0; }
    my $finish = $now + $wait;
Mac Newbold's avatar
Mac Newbold committed
288 289
    while (($event_count != $lastcount || $wait > 0) &&
	   !($sigrestart || $sigcleanup || $do_reload)) {
290
	$lastcount = $event_count;
Mac Newbold's avatar
Mac Newbold committed
291 292
	# Don't block if we got a signal!
	if ($wait<=0 || $sigrestart || $sigcleanup || $do_reload) {
293 294
	    event_poll($handle);
	} else {
295
	    #debug("Using blocking event poll - $wait seconds\n");
296 297 298 299 300
	    # timeout param is in milliseconds, so multiply
	    event_poll_blocking($handle, $wait*1000);
	    $now = time();
	    # subtract seconds elapsed from my wait time
	    $wait = $finish - $now;
301
	    #debug("Finished blocking event poll - $wait seconds remain\n");
302
	    if ($event_count > 0 &&
Mac Newbold's avatar
Mac Newbold committed
303 304
		(qsize() > 0 || $mailqueue ||
		 $sigrestart || $sigcleanup || $do_reload)) {
305 306 307 308 309 310 311
		$blockwait=0;
		$wait=0;
		#debug("Cancelling wait - timeouts/msgs waiting, or HUP'd\n");
		#debug("---End Blocking Wait ---\n");
	    }
	}
	#debug("Wait is $wait\n");
312 313
    }
    if ($event_count > 0) {
Mac Newbold's avatar
Mac Newbold committed
314
	debug("Handled $event_count event(s).\n");
315 316
    }
}
Robert Ricci's avatar
Robert Ricci committed
317

318
# Now, we just poll for events, and watch for timeouts
Robert Ricci's avatar
Robert Ricci committed
319
while (1) {
Mac Newbold's avatar
Mac Newbold committed
320
    my $now = time();
321 322 323 324
    my ($deadline,$node);

    # Check for nodes that have passed their timeout
    if (!qhead($deadline,$node)) {
325
	info("HEAD: $node in ".($deadline-$now).", queue=".qsize()."\n");
326 327
	while ($now >= $deadline && $node ne "") {
	    qpop($deadline,$node);
328
	    info("POP: $node in ".($deadline-$now).", queue=".qsize()."\n");
329
	    handleCtrlEvent($node,$TBTIMEOUT);
330 331 332 333
	    if (0) { qshow(); }
	    if (qhead($deadline,$node)) {
		$deadline=0; $node="";
	    }
334
	}
335 336
    } else {
	$deadline=0;
337
    }
338
    $nextdeadline = $deadline;
339

340 341 342 343 344
    if (qsize()==0) {
	$blockwait=1;
	debug("---Blocking wait okay---\n");
    }
		
Mac Newbold's avatar
Mac Newbold committed
345 346 347 348
    if ($do_reload || ($now - $last_reload > $reload_time)) {
	reload();
	$do_reload = 0;
    }
Mac Newbold's avatar
Mac Newbold committed
349

Mac Newbold's avatar
Mac Newbold committed
350 351
    # Send any messages in the queue if it is time
    notify("",1);
Mac Newbold's avatar
Mac Newbold committed
352

353 354 355
    if ($sigrestart) { restart(); }
    if ($sigcleanup) { cleanup(); }

356
    process_event_queue;
Robert Ricci's avatar
Robert Ricci committed
357 358
}

Mac Newbold's avatar
Mac Newbold committed
359 360
exit(0);

Robert Ricci's avatar
Robert Ricci committed
361
# Read the current states of nodes from the database
362
sub readStates(;@) {
Mac Newbold's avatar
Mac Newbold committed
363 364 365 366 367 368
    my %oldnodes = @_;

    # Guard against undefined variable warnings
    if (! defined(%oldnodes)) {
	%oldnodes = ();
    }
369

Mac Newbold's avatar
Mac Newbold committed
370 371 372
    #debug("readStates called\n");
    my $result = DBQueryFatal("SELECT node_id, eventstate, " .
			      "state_timestamp, op_mode, " .
373
			      "op_mode_timestamp, stated_tag FROM nodes ".
374
			      "where node_id not like 'sh%'");
Mac Newbold's avatar
Mac Newbold committed
375 376

    my %nodes;
377
    while (my ($node_id, $state, $timestamp, $mode, $mode_timestamp, $tag)
Mac Newbold's avatar
Mac Newbold committed
378
	   = $result->fetchrow()) {
379 380 381 382 383 384 385 386 387 388 389 390
	$nodes{$node_id}{"tag"} = (defined($tag) ? $tag : "");

	if ($dbtag ne "" && $dbtag eq $nodes{$node_id}{"tag"}) {
	    info("This stated will work on $node_id\n");
	}
	if ($dbtag eq "" && $dbtag ne $nodes{$node_id}{"tag"}) {
	    info("This stated will *NOT* work on $node_id\n");
	}
	if ($dbtag ne $nodes{$node_id}{"tag"}) {
	    remTimeout($node_id);
	}
	
391
	#
Mac Newbold's avatar
Mac Newbold committed
392 393 394 395
	# 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.)
396
	#
Mac Newbold's avatar
Mac Newbold committed
397 398 399 400 401
	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};
402
	} else {
Mac Newbold's avatar
Mac Newbold committed
403 404 405 406
	    $nodes{$node_id}{state}          = $state;
	    $nodes{$node_id}{timestamp}      = $timestamp;
	    $nodes{$node_id}{mode}           = $mode;
	    $nodes{$node_id}{mode_timestamp} = $mode_timestamp;
407 408 409
	    $nodes{$node_id}{notified}       = 0;
	    $nodes{$node_id}{timedout}       = 0;
	    $nodes{$node_id}{noretry}        = 0;
410
	    # Is there a timeout? If so, set it up!
411 412 413
	    if ($dbtag eq $nodes{$node_id}{"tag"}) {
		setTimeout($mode,$state,$node_id,$timestamp);
	    }
414
	}
Mac Newbold's avatar
Mac Newbold committed
415 416
    }
    return %nodes;
Robert Ricci's avatar
Robert Ricci committed
417 418 419 420 421 422
}

#
# Read timeouts for various states from the database
#
sub getTimeouts() {
Mac Newbold's avatar
Mac Newbold committed
423 424 425
    #debug("getTimeouts called\n");
    my $result = DBQueryFatal("SELECT op_mode, state, timeout, action " .
			      "FROM state_timeouts");
Robert Ricci's avatar
Robert Ricci committed
426

Mac Newbold's avatar
Mac Newbold committed
427 428 429 430 431
    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
432 433 434 435 436 437
}

#
# Read the list of valid state transitions from the database
#
sub getValid() {
Mac Newbold's avatar
Mac Newbold committed
438 439 440
    #debug("getValid called\n");
    my $result = DBQueryFatal("SELECT op_mode, state1, state2 " .
			      "FROM state_transitions");
Robert Ricci's avatar
Robert Ricci committed
441

Mac Newbold's avatar
Mac Newbold committed
442 443 444 445 446
    my %valid;
    while (my ($mode,$state1, $state2) = $result->fetchrow()) {
	$valid{$mode}{$state1}{$state2} = 1;
    }
    return %valid;
Robert Ricci's avatar
Robert Ricci committed
447 448
}

449 450 451 452
#
# Read the list of valid mode transitions from the database
#
sub getModeTrans() {
Mac Newbold's avatar
Mac Newbold committed
453 454 455 456 457 458 459 460 461 462 463 464 465
    #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;
466
	}
Mac Newbold's avatar
Mac Newbold committed
467 468
    }
    return %modeTrans;
469 470 471 472 473 474
}

#
# Read the list of states which trigger an action
#
sub getTriggers() {
475 476 477 478 479
    debug("getTriggers called\n");
    
    debug("anymode ==> '$TBANYMODE'\n");

    # Grab global triggers
Mac Newbold's avatar
Mac Newbold committed
480 481
    my $result = 
      DBQueryFatal("SELECT op_mode, state, trigger " .
482 483
		   "FROM state_triggers where node_id='$TBANYMODE' ".
		   "order by op_mode,state");
Mac Newbold's avatar
Mac Newbold committed
484
    my %t;
485
    while (my ($mode, $state, $trig) = $result->fetchrow()) {
Mac Newbold's avatar
Mac Newbold committed
486
	$t{"$mode:$state"} = $trig;
487 488 489 490 491 492 493 494 495 496 497 498
	debug("trig($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
499
    }
500

Mac Newbold's avatar
Mac Newbold committed
501
    return %t;
502 503
}

Robert Ricci's avatar
Robert Ricci committed
504 505 506 507
#
# Gets called for every event that we recieve
#
sub handleEvent($$$) {
Mac Newbold's avatar
Mac Newbold committed
508 509 510 511 512 513
    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++;
514 515 516 517
    
    #
    # For readability, only do this on the main stated.
    #
518
    if ($dbtag eq "") {
519 520
	debug("Got an event: ($objtype,$objname,$eventtype)\n");
    }
Mac Newbold's avatar
Mac Newbold committed
521 522 523 524

    #
    # Check to see if another instance is supposed to be handling this node
    #
525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555
    if ($objtype ne $TBCOMMAND) {
	my $node = $objname;
	
	#
	# If we have never seen this node, reload.
	#
	if (! defined($nodes{$node})) {
	    reload();

	    # Still not defined, someone screwed up! This could end up
	    # churning via reload(). Bad. 
	    if (defined($nodes{$node})) {
		notify("Got $objtype/$eventtype for nonexistent $node!\n");
		return;
	    }
	}
	
	#
	# If a stated_tag was specified on the command line, ignore those
	# nodes that do not match. 
	#
	if ($dbtag ne $nodes{$node}{"tag"}) {
	    # Record when main stated ignores a node.
	    info("Got $objtype/$eventtype for $node, which is not mine\n")
		if ($dbtag eq "");
	    return;
	}
	if (!checkDBRedirect($node)) {
	    info("Got $objtype/$eventtype for $node, which is not mine\n");
	    return;
	}
Mac Newbold's avatar
Mac Newbold committed
556
    }
557 558 559 560
    # Makes it easier to run a debugging version.
    if ($dbtag ne "") {
	debug("Got an event: ($objtype,$objname,$eventtype)\n");
    }
Mac Newbold's avatar
Mac Newbold committed
561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576
  SWITCH: for ($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,$eventtype);
	    last;
	};
577 578 579 580
	(/$TBCOMMAND/) && do {
	    handleCommand($objname,$eventtype);
	    last;
	};
581

Mac Newbold's avatar
Mac Newbold committed
582
    }
583 584 585 586 587

}

sub stateTransition($$) {

588
    my ($node,$newstate) = @_;
Robert Ricci's avatar
Robert Ricci committed
589

590 591
    # Check for invalid transitions
    my ($oldstate, $mode);
592 593 594
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

595 596 597
    if ($oldstate && $mode && $valid{$mode} && $valid{$mode}{$oldstate} &&
	!$valid{$mode}{$oldstate}{$newstate}) {
	notify("Invalid transition for node $node from $mode/$oldstate " .
598
	       "to $newstate\n");
599
    }
Robert Ricci's avatar
Robert Ricci committed
600

601 602 603 604
    my $now = time();
    $nodes{$node}{state}     = $newstate;
    $nodes{$node}{timestamp} = $now;
    $nodes{$node}{notified}  = 0;
605

606 607 608
    info("$node: $mode/$oldstate => $mode/$newstate\n");
    DBQueryFatal("UPDATE nodes SET eventstate='$newstate', " .
		 "state_timestamp='$now' WHERE node_id='$node'");
609

610 611 612 613
    # 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} =~ /^$TBCOMMAND:/) {
614
        debug("TimeoutTag = '$timeout_tag{$node}'\n");
615
	my ($str,$cmd) = split(":",$timeout_tag{$node});
616
	debug("str=$str\tcmd=$cmd\tTBREBOOT=$TBREBOOT\tstate=$newstate\n");
617
	if ($cmd eq $TBREBOOT) {
618
	    if ($newstate eq TBDB_NODESTATE_SHUTDOWN ) {
619 620 621
		info("$node: $TBREBOOT success\n");
		# Timeout will get cleared below by setTimeout call
	    } else {
622 623
		notify("$node: $TBREBOOT in progress, but got state ".
		       "$newstate instead of ".TBDB_NODESTATE_SHUTDOWN."!\n");
624 625 626 627 628
	    }
	#} elsif ($cmd eq $FOO ) {
	    # Add more here...
	} else {
	    notify("$node: Unknown command timeout '$timeout_tag{$node}' ".
629
		   "found at $mode/$newstate\n");
630 631 632
	}
    }

633 634 635 636 637
    #
    # Check if this state has a timeout, and if so, put it in the queue.
    # Note that any opmode transition below will replace (or remove) this
    # timeout if appropriate.
    #
638 639
    setTimeout($mode,$newstate,$node,$now);

640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666
    #
    # See if we jumped into the PXEBOOT kernel. Bootinfo will send
    # PXEBOOTING every time a node contacts it, which is our indicator that
    # the node is in the first phase of booting. At this point we want to
    # switch state machines since the entire boot process is governed by a
    # single state machine that is independent of the OS that the node will
    # eventually boot.  Rather then encode that in each state machine, we
    # use a special machine with a defined entrypoint (PXEBOOTING) and a
    # defined exitpoint (BOOTING). See below for where we jump back out
    # of this state machine.
    #
    # XXX should this be done with a trigger?
    #
    if ($newstate eq TBDB_NODESTATE_PXEBOOTING) {
	#
	# Jumped in. We need to change the opmode so that the state
	# transitions are legal. We do not bother to save the old opmode
	# since we can figure it out later when we leave.
	#
	if ($mode ne $PXEKERNEL) {
	    info("$node: Forcing mode transition into $PXEKERNEL!\n");
	    opModeTransition($node, $PXEKERNEL, 1);
	    $mode=$PXEKERNEL;
	}
    }

    # Check if this is TBDB_NODESTATE_BOOTING, which has actions.
667
    if ($newstate eq TBDB_NODESTATE_BOOTING) {
668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697
	#
	# See if we are in the right mode/osid. 
	#
	my ($bootosid,$bootopmode) = TBBootWhat($node, $debug);

	info("$node: BootWhat says $bootosid (mode $bootopmode).\n");
	DBQueryFatal("update nodes set osid='$bootosid' ".
		     "where node_id='$node'");
	
	if ($bootopmode ne $mode) {
	    if ($mode eq $PXEKERNEL) {
		#
		# If we came from PXE boot, then we have to jump out of the
		# PXEKERNEL state machine into whatever state machine is
		# current for the node. Since we came through bootinfo,
		# we know that the node is doing what it is supposed to, and
		# that this change matches what the node is booting. 
		#
		info("$node: Forcing mode transition out of $PXEKERNEL!\n");
		opModeTransition($node, $bootopmode, 1);
		$mode=$bootopmode;
	    }
	    elsif ($oldstate eq TBDB_NODESTATE_ISUP) {
		#
		# Skipped SHUTDOWN, which could result in a missed opmode
		# transition. Can this really happen anymore?
		# 
		info("$node: Came from ISUP! Checking for mode transition\n");
		my $query_result =
		    DBQueryWarn("select next_op_mode from nodes ".
698
				"where node_id='$node'");
699 700 701 702 703 704
		my ($nextmode) = $query_result->fetchrow();
		if ($nextmode) {
		    info("$node: Forcing mode transition!\n");
		    opModeTransition($node, $nextmode, 1);
		    $mode=$nextmode;
		}
705
	    }
706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723
	    else {
		my $str = "$node is running $bootosid, but in mode $mode\n".
		    "instead of mode $bootopmode!\n";
		
		if ($bootopmode eq "RELOAD") {
		    #
		    # For now, only force if we're going into reload mode,
		    # so we don't get stuck looping in reloading.
		    # Can this happen anymore?
		    #
		    DBQueryFatal("UPDATE nodes SET op_mode='$bootopmode', ".
				 "op_mode_timestamp=unix_timestamp(now()) ".
				 "WHERE node_id='$node'");
		    $nodes{$node}{mode} = $bootopmode;
		    $nodes{$node}{mode_timestamp} = $now;
		    $str .= "Forced op_mode to $bootopmode.\n";
		}
		notify($str);
724 725
	    }
	}
726 727
	checkGenISUP($node);
    }
728

729 730 731 732 733
    # Check if this state has any triggers
    my @nodetrigs = GetNodeTriggerList($node,$mode,$newstate);
    if (defined($triggers{"$mode:$newstate"}) ||
        (@nodetrigs > 0) ) {
	# check for global triggers
734
	my @trigs = split(/\s*,\s*/,$triggers{"$mode:$newstate"});
735 736 737 738
	# Run all the triggers
	debug("Running triggers. Global=".join("/",@trigs).
	      "   node=".join("/",@nodetrigs)."\n");
	foreach ( @trigs , @nodetrigs) {
739 740
	    my $trig = $_;
	    /^$TBRESET$/ && do {
741 742 743
		# We successfully booted, so clear some flags
		$nodes{$node}{noretry}   = 0;
		$nodes{$node}{timedout}  = 0;
744 745 746 747 748 749 750 751 752 753 754 755 756
		# 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,$trig);
		next;
	    };
757 758 759 760 761 762 763 764
	    /^$TBISUP$/ && do {
		info("$node: Triggered $TBISUP\n");
		EventSendWarn(host      => $BOSSNODE ,
			      objtype   => TBDB_TBEVENT_NODESTATE ,
			      eventtype => TBDB_NODESTATE_ISUP ,
			      objname   => $node);
		next;
	    };
765
	    notify("Unknown trigger '$trig' for $node in $mode/$newstate!\n");
766
	}
767 768 769
	# Clear any of the node triggers that we ran
	debug("Clearing node triggers: ".join("/",@nodetrigs)."\n");
	ClearNodeTrigger($node,$mode,$newstate,@nodetrigs);
770
    }
771

772 773 774 775 776 777 778 779
    # 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
780 781 782
	} else {
	    debug("No next mode.\n");
	}
783 784
    }
}
785

786
sub opModeTransition($$;$) {
Mac Newbold's avatar
Mac Newbold committed
787

788 789
    my ($node,$newmode,$force) = @_;
    if (!defined($force)) { $force = 0; }
Mac Newbold's avatar
Mac Newbold committed
790

791 792
    info("$node: Mode change to $newmode requested ($force)\n");
    
793 794
    # Check for invalid transitions
    my ($oldstate, $mode, $nextstate);
795 796 797
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

798
    if (defined($modeTrans{"$mode:$oldstate"}) || $force) {
799
	if (!$force) {
800 801 802 803 804 805 806 807 808 809 810 811 812 813 814
	    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");
	    }
815 816
	}
    } else {
817
	notify("Invalid mode transition for $node from $mode/$oldstate: ".
818 819
	       "Not a valid mode transition state!\n");
    }
Mac Newbold's avatar
Mac Newbold committed
820 821 822
    if (!$nextstate) {
	$nextstate=$oldstate;
    }
Mac Newbold's avatar
Mac Newbold committed
823

824 825 826 827 828 829
    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
830

831 832 833 834 835
    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
836 837 838 839

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

840 841 842 843
}

sub handleCtrlEvent($$) {
    my ($node,$event) = @_;
844

845
    info("CtrlEvent: $node, $event\n");
846

847 848
    foreach ($event) {
	/^$TBRESET$/ && do {
849 850 851 852
	    #
	    # Clear next_boot_path with os_select.
	    # 
	    $cmd = "$osselect -d -c -1 $node";
853
	    system($cmd) and
854 855 856
		notify("$node/$event: Could not clear next_boot_path!\n");
	    
	    info("Performed RESET for $node\n");
857 858 859 860 861 862 863 864 865 866
	    next;
	};
	/^$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'");
867 868
		DBQueryFatal("delete from reserved where node_id='$node'");
		info("Released $node from $pid/$eid\n");
869 870 871 872
	    }
	    next;
	};
	/^$TBTIMEOUT$/ && do {
873 874 875 876
	    my ($mode,$state) = split(":",$timeout_tag{$node});
	    delete($timeout_tag{$node});
	    my $curstate = $nodes{$node}{state};
	    my $curmode = $nodes{$node}{mode};
877
	    my ($timeout,$action);
878 879 880 881 882
	    if (!defined($nodes{$node}{notified})) {
		$nodes{$node}{notified}=0;
	    }
	    $nodes{$node}{notified}++;
	    my $notified = $nodes{$node}{notified};
883 884
	    $nodes{$node}{timedout}++;
	    my $timedout = $nodes{$node}{timedout};
885 886 887 888
	    if ($mode && $state && $timeouts{$mode} &&
		$timeouts{$mode}{$state}) {
		($timeout, $action) = @{$timeouts{$mode}{$state}};
	    }
889 890
	    if ($mode eq $TBCOMMAND) {
		# It is a command, not a true state
891
		if ($action eq $TBTIMEOUTCMDRETRY) {
892
		    # Retry the command
893
		    notify("$node: Command $state, retry #$timedout\n");
894
		    # notify in case we get in a retry loop...
895
		    handleCommand($node,$state,$timedout,1);
896 897 898 899 900 901
		} else {
		    notify("$node: Unknown timeout action for ".
			   "$mode/$state: '$action'\n");
		}
		next;
	    }
902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936

	    #
	    # Trash. This stuff should not be encoded this way, but I have
	    # no idea how timeouts, TBCOMMAND, and actions interact.
	    #
	    if ($curstate eq $PXEWAKEUP) {
		my $optarg = ($debug ? "-d " : "");
		
		if ($timedout < 3) {
		    #
		    # Try again. 
		    # 
		    info("Node $node has timed out $timedout times in ".
			 "$PXEWAKEUP!\n".
			 "Sending it a another wakeup command\n");
		}
		else {
		    #
		    # Failed too many times, power cycle instead. 
		    # 
		    notify("Node $node has timed out $timedout times in ".
			   "$PXEWAKEUP!\n".
			   "Sending it a reboot command\n");
		    $optarg .= "-k";
		}
		my $cmd = "$nodereboot -r $optarg $node";
		debug("$cmd\n");
		system("date 2>&1 >> $rebootlog");
		system("$cmd 2>&1 >> $rebootlog &") and
		    notify("PXEWAKEUP retry: ".
			   "Command '$cmd' failed, error $?: $!\n");

		next;
	    }

937 938 939 940 941 942
	    info("Node $node has timed out in state $mode/$state".
		 ($action ne "" ? "\n\tRequested action $action." : "").
		 "\n");

	    foreach ($action) {
		/^$TBTIMEOUTREBOOT/ && do {
943 944 945 946 947
		    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;
948
		    } else {
949 950 951 952 953 954
			# 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);
955 956 957 958 959 960 961
		    }
		    last; };
		/^$TBTIMEOUTNOTIFY/ && do {
		    notify("Node $node has timed out in state $mode/$state\n");
		    last; };
		notify("$node: Unknown Timeout Action: $action\n");
	    }
962 963
	    next;
	};
964
	notify("$node: Unknown CtrlEvent: $event\n");
965 966
    }
}
Robert Ricci's avatar
Robert Ricci committed
967

968 969
sub handleCommand($$;$$) {
    my ($params,$command,$retry,$force) = @_;
970
    if (!defined($retry)) { $retry=0; }
971
    if (!defined($force)) { $force=0; }
972 973 974 975 976 977 978 979

    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.

980
    if ($command eq $TBREBOOT && $retry >=4) {
981 982
	announce("Node $params has tried rebooting $retry times and has \n".
		 "still not been successful. Please look into it soon.\n".
983 984 985
	"" );#	 "In the meantime, $params will be powered off.\n");
	# Just return...
	return 0;
986
	# change my command to poweroff.
987
	#$command = $TBPOWEROFF;
988 989 990 991 992 993
    }

    foreach ($command) {
	/^$TBREBOOT$/ && do {
	    # For reboot, the params is a comma-separated list of nodes
	    my @nodes = split(",",$params);
994 995 996 997 998 999 1000 1001

	    # 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");
1002 1003
		    if (($nodes{$node}{state} ne TBDB_NODESTATE_ISUP) &&
			(!$nodes{$node}{noretry}) ) {
1004
			# This node shouldn't be rebooted now...
1005 1006
			# XXX Send feedback here somehow!

1007 1008 1009 1010 1011 1012 1013 1014 1015 1016
			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; }
	    }

1017 1018 1019 1020 1021
	    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.

1022
	    my $cmd = "$nodereboot -r $nodelist";
1023
	    my $redir = " 2>&1 >> $rebootlog &";
1024 1025 1026
	    debug("$cmd $redir\n");
	    system("date $redir");
	    system($cmd.$redir) and
1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048
	      notify("$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};
1049
	    info("Sending power $func for nodes: $nodelist\n");
1050 1051 1052 1053
	    # Permissions were checked in order to send the message,
	    # so we don't need to do any fancy stuff here.

	    my $cmd = "$power $func $nodelist &";
1054 1055
	    debug("$cmd\n") and
	    system($cmd) and
1056 1057 1058 1059 1060 1061 1062 1063 1064 1065
	      notify("$params/$command: ".
		     "Command '$cmd' failed, error $?: $!\n");

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

1066 1067 1068 1069
#
# Check if we need to generate an ISUP
#
sub checkGenISUP($) {
Mac Newbold's avatar
Mac Newbold committed
1070 1071 1072 1073 1074 1075 1076 1077 1078 1079
    my ($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
1080

Mac Newbold's avatar
Mac Newbold committed
1081 1082 1083 1084 1085 1086
    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
1087

Mac Newbold's avatar
Mac Newbold committed
1088 1089 1090 1091 1092
    # 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
1093

Mac Newbold's avatar
Mac Newbold committed
1094 1095 1096 1097 1098 1099
    # 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;
1100
    }
Mac Newbold's avatar
Mac Newbold committed
1101

Mac Newbold's avatar
Mac Newbold committed
1102 1103 1104 1105 1106 1107 1108 1109
    # 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
1110 1111
}

1112 1113 1114 1115
# 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
1116
    my ($node) = @_;
1117

Mac Newbold's avatar
Mac Newbold committed
1118 1119 1120 1121 1122 1123 1124 1125
    # 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'");
1126

Mac Newbold's avatar
Mac Newbold committed
1127 1128 1129 1130
    if (!$result->num_rows()) {
	notify("Got an event for a node ($node) I don't know about\n");
	return 0;
    }
1131

Mac Newbold's avatar
Mac Newbold committed
1132
    my ($testdb) = $result->fetchrow();
1133

Mac Newbold's avatar
Mac Newbold committed
1134 1135 1136
    # XXX: It's hokey to hardcode tbdb here, but....

    #debug("checkDBRedirect: $node => $testdb (I'm $TBDBNAME)\n");
1137
    if ((!$testdb && ($TBDBNAME eq $REALTBDBNAME)) ||
Mac Newbold's avatar
Mac Newbold committed
1138 1139 1140 1141 1142
	($testdb && ($testdb eq $TBDBNAME))) {
	return 1;
    } else {
	return 0;
    }
1143 1144
}

1145 1146 1147 1148
# 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(); }
1149
    if (defined(qfind($node))) { qdelete($node); delete($timeout_tag{$node}); }
1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160
    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);
1161
	    $timeout_tag{$node} = "$mode:$state";
1162 1163 1164 1165 1166 1167
	    if (0) { qshow(); }
	}
    }
    if (0) { print "Done:\n"; qshow(); }
}

1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178
# Remove a timeout.
sub remTimeout ($)
{
    my ($node) = @_;

    if (defined(qfind($node))) {
	qdelete($node);
	delete($timeout_tag{$node});
    }
}

Robert Ricci's avatar
Robert Ricci committed
1179 1180
# Reload state from the database
sub reload() {
Mac Newbold's avatar
Mac Newbold committed
1181 1182 1183 1184 1185 1186
    debug("Reloading state from database\n");
    $last_reload = time();
    %timeouts  = getTimeouts();
    %valid     = getValid();
    %modeTrans = getModeTrans();
    %triggers  = getTriggers();
1187
    %nodes     = readStates(%nodes);
Robert Ricci's avatar
Robert Ricci committed
1188 1189
}

1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269
#
# 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;
}

1270 1271
sub os_opmode() {
    my $osid = shift || "";
Mac Newbold's avatar
Mac Newbold committed
1272 1273 1274
    if ($osid eq $TB_OSID_MBKERNEL) {
	return "MINIMAL";
    }
1275 1276
    my $cmd = "select op_mode from os_info where osid='$osid';";
    my $q = DBQueryFatal($cmd);
Mac Newbold's avatar
Mac Newbold committed
1277 1278 1279
    if ($q->numrows() < 1) {
	return "";
    }
1280 1281 1282
    my @r = $q->fetchrow_array();
    my $opmode=$r[0];
    debug("OpMode for '$osid' is '$opmode'\n");
Mac Newbold's avatar
Mac Newbold committed
1283 1284 1285
    if (defined($opmode) && $opmode ne "") {
	return $opmode;
    }
1286 1287
    return "";
}
Mac Newbold's avatar
Mac Newbold committed
1288

1289 1290 1291 1292
#
# Functions for controlling output/logging, and signal handling
#

Robert Ricci's avatar
Robert Ricci committed
1293
sub debug(@) {
Mac Newbold's avatar
Mac Newbold committed
1294 1295 1296
    if ($debug) {
	print @_;
    }
Mac Newbold's avatar
Mac Newbold committed
1297 1298 1299
}

sub fatal($) {
Mac Newbold's avatar
Mac Newbold committed
1300 1301 1302
    my $msg = shift;
    notify($msg);
    die($msg);
Robert Ricci's avatar
Robert Ricci committed
1303 1304
}

Mac Newbold's avatar
Mac Newbold committed
1305
sub showqueue() {
Mac Newbold's avatar
Mac Newbold committed
1306 1307 1308 1309 1310 1311 1312 1313 1314 1315
    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).