stated.in 20.8 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 4 5 6 7
#
# EMULAB-COPYRIGHT
# Copyright (c) 2000-2002 University of Utah and the Flux Group.
# 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@";
Robert Ricci's avatar
Robert Ricci committed
23
my $BOSSNODE = "@BOSSNODE@";
24
my $TBOPS = '@TBSTATEDEMAIL@';
25
my $TBDBNAME = "@TBDBNAME@";
26
my $osselect = "$TB/bin/os_select";
Robert Ricci's avatar
Robert Ricci committed
27 28 29 30 31 32 33

$| = 1;

use event;
use libdb;
use libtestbed;
use Getopt::Std;
34
#use strict;
Robert Ricci's avatar
Robert Ricci committed
35
use English;
36 37
use POSIX; # for strftime, and sigprocmask and friends
use Sys::Syslog;
38
use Fcntl; # file constants for pidfile
Robert Ricci's avatar
Robert Ricci committed
39 40 41 42

# 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.
43
my $reload_time = 3600;
Robert Ricci's avatar
Robert Ricci committed
44 45 46 47 48

# Process command-line arguments

sub usage {
	print << "END";
49 50 51
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
52 53
-s server       Use specified server, instead of this site's bossnode
-p port	        Use specified port
54
Send SIGHUP to reload database state, or SIGUSR1 to restart completely.
Robert Ricci's avatar
Robert Ricci committed
55 56 57
END
}

58
my @args = @ARGV; # save a copy for restart before we mess with them.
Robert Ricci's avatar
Robert Ricci committed
59
my %opt = ();
60
getopts("ds:p:h",\%opt);
Robert Ricci's avatar
Robert Ricci committed
61 62 63 64

if ($opt{h}) { exit &usage; }
if (@ARGV) { exit &usage; }

65
my ($server,$port,$debug);
Robert Ricci's avatar
Robert Ricci committed
66 67 68 69
if ($opt{s}) { $server = $opt{s}; } else { $server = $BOSSNODE; }
if ($opt{p}) { $port = $opt{p}; }
if ($opt{d}) { $debug = 1; } else { $debug = 0; }

70 71 72 73 74 75 76 77 78
# Grab some constants into variables
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;
my $TBCONTROL    = TBDB_TBEVENT_TBCONTROL;

Mac Newbold's avatar
Mac Newbold committed
79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94
my $pidfile;
$pidfile = "$TB/locks/stated.pid";
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: $? $!\n");
}
Robert Ricci's avatar
Robert Ricci committed
95
# Background
96 97 98 99 100
if (!$debug) {
  # We use syslog, so redirect the output to nothing
  if (TBBackGround("/dev/null")) { exit(0); }
  # set up syslog
  openlog("stated","pid","user");
Robert Ricci's avatar
Robert Ricci committed
101
}
Mac Newbold's avatar
Mac Newbold committed
102 103 104 105 106 107
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
108

109 110 111
# Change my $0 so that it is easier to see in a ps/top
$0 = "$0";

Robert Ricci's avatar
Robert Ricci committed
112 113 114 115 116
my $URL = "elvin://$server";
if ($port) { $URL .= ":$port"; }

# 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
117
if (!$handle) { fatal("Unable to register with event system\n"); }
Robert Ricci's avatar
Robert Ricci committed
118 119

my $tuple = address_tuple_alloc();
Mac Newbold's avatar
Mac Newbold committed
120
if (!$tuple) { fatal("Could not allocate an address tuple\n"); }
Robert Ricci's avatar
Robert Ricci committed
121

122 123
%$tuple = ( objtype => join(",",$TBNODESTATE,$TBNODEOPMODE,$TBCONTROL) );

Robert Ricci's avatar
Robert Ricci committed
124
if (!event_subscribe($handle,\&handleEvent,$tuple)) {
Mac Newbold's avatar
Mac Newbold committed
125
	fatal("Could not subscribe to events\n");
Robert Ricci's avatar
Robert Ricci committed
126 127 128 129
}

# Read in the pre-existing node states, and timeout and valid transition
# information from the database
130 131 132 133 134
my %nodes     = readStates();
my %timeouts  = getTimeouts();
my %valid     = getValid();
my %modeTrans = getModeTrans();
my %triggers  = getTriggers();
Robert Ricci's avatar
Robert Ricci committed
135 136 137 138 139 140 141

# Gets set if a reload of state from the database should happen.
my $do_reload = 0;

# 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.
142 143 144 145 146 147 148
$SIG{HUP}  = sub { info("SIGHUP - Reloading DB state\n"); $do_reload = 1; };

# Set up other signals. Unblock USR1 in case we're here because of a
# restart, and it still think's we're handling USR1.
$SIG{USR1} = \&restart;
$SIG{USR2} = \&cleanup;
$SIG{INT}  = \&cleanup;
149 150 151 152
$SIG{QUIT} = \&cleanup;
$SIG{ABRT} = \&cleanup;
$SIG{TERM} = \&cleanup;
$SIG{KILL} = \&cleanup;
Robert Ricci's avatar
Robert Ricci committed
153

154
info("stated starting up\n");
Robert Ricci's avatar
Robert Ricci committed
155

156
# Now, we just poll for events, and watch for timeouts
Robert Ricci's avatar
Robert Ricci committed
157 158 159 160 161 162 163 164
my $iterations = 0;
while (1) {
	event_poll($handle);
	my $now = time();
	#
	# Look for nodes that have passed their timeout
	#
	while (my ($node, $value) = each %nodes) {
165 166 167 168 169 170 171 172 173 174 175 176 177 178 179
	    my $state    = $value->{state};
	    my $mode     = $value->{mode};
	    my $time     = $value->{timestamp};
	    my $notified = $value->{notified};
	    my ($timeout,$action);
	    if ($mode && $state && $timeouts{$mode} &&
		$timeouts{$mode}{$state}) {
		($timeout, $action) = @{$timeouts{$mode}{$state}};
	    }
	    if ((!$notified) && $time && $timeout && 
		$timeout!= $TBNOTIMEOUT && (($time + $timeout) < $now)) {
		handleCtrlEvent($node,$TBTIMEOUT);
		$value->{notified} = 1;
		
	    }
Robert Ricci's avatar
Robert Ricci committed
180 181 182 183 184 185 186 187 188 189 190 191 192
	}

	if ($do_reload || ($iterations > $reload_time)) {
		reload();
		$do_reload = 0;
		$iterations = 0;
	}

	$iterations++;
	sleep(1);
}

# Read the current states of nodes from the database
193 194 195
sub readStates(;@) {
    	my %oldnodes = @_;

196 197 198 199 200 201 202
	#
	# Guard against undefined variable warnings
	#
	if (! defined(%oldnodes)) {
		%oldnodes = ();
	}

203
	#debug("readStates called\n");
Robert Ricci's avatar
Robert Ricci committed
204
	my $result = DBQueryFatal("SELECT node_id, eventstate, " .
205 206
				  "state_timestamp, op_mode, " .
				  "op_mode_timestamp FROM nodes");
Robert Ricci's avatar
Robert Ricci committed
207 208

	my %nodes;
209 210
	while (my ($node_id, $state, $timestamp, $mode, $mode_timestamp)
			= $result->fetchrow()) {
211 212 213 214 215 216
	    	#
		# If there's an entry in oldnodes for this node, and it
		# hasn't changed state or time, use the old entry (so that
		# we don't lose information about which nodes we've already
		# notified the ops about, etc.)
		#
217 218
		if ($oldnodes{$node_id} && $state && $timestamp &&
		   ($oldnodes{$node_id}{state} eq $state) &&
219
		   ($oldnodes{$node_id}{mode} eq $mode) &&
220
		   ($oldnodes{$node_id}{timestamp} == $timestamp)) {
221 222
		       $nodes{$node_id} = $oldnodes{$node_id};
		} else { 
223 224 225 226
		    $nodes{$node_id}{state}          = $state;
		    $nodes{$node_id}{timestamp}      = $timestamp;
		    $nodes{$node_id}{mode}           = $mode;
		    $nodes{$node_id}{mode_timestamp} = $mode_timestamp;
227
		}
Robert Ricci's avatar
Robert Ricci committed
228 229 230 231 232 233 234 235
	}
	return %nodes;
}

#
# Read timeouts for various states from the database
#
sub getTimeouts() {
236
	#debug("getTimeouts called\n");
237
	my $result = DBQueryFatal("SELECT op_mode, state, timeout, action " .
Robert Ricci's avatar
Robert Ricci committed
238 239 240
				  "FROM state_timeouts");

	my %timeouts;
241 242
	while (my ($op_mode, $state, $timeout, $action) = $result->fetchrow()) {
		$timeouts{$op_mode}{$state} = [ $timeout, $action ];
Robert Ricci's avatar
Robert Ricci committed
243 244 245 246 247 248 249 250
	}
	return %timeouts;
}

#
# Read the list of valid state transitions from the database
#
sub getValid() {
251
	#debug("getValid called\n");
252
	my $result = DBQueryFatal("SELECT op_mode, state1, state2 " .
Robert Ricci's avatar
Robert Ricci committed
253 254 255
				  "FROM state_transitions");

	my %valid;
256 257
	while (my ($mode,$state1, $state2) = $result->fetchrow()) {
		$valid{$mode}{$state1}{$state2} = 1;
Robert Ricci's avatar
Robert Ricci committed
258 259 260 261
	}
	return %valid;
}

262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298
#
# Read the list of valid mode transitions from the database
#
sub getModeTrans() {
	#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;
	    }
	}
	return %modeTrans;
}

#
# Read the list of states which trigger an action
#
sub getTriggers() {
	#debug("getTriggers called\n");
	my $result = 
	    DBQueryFatal("SELECT op_mode, state, trigger " .
			 "FROM state_triggers order by op_mode,state");
	my %t;
	while (my ($mode,$state, $trig) = $result->fetchrow()) {
	    $t{"$mode:$state"} = $trig;
	}
	return %t;
}

Robert Ricci's avatar
Robert Ricci committed
299 300 301 302 303
#
# Gets called for every event that we recieve
#
sub handleEvent($$$) {
	my ($handle,$notification,$data) = @_;
304
	my $objtype = event_notification_get_objtype($handle,$notification);
Robert Ricci's avatar
Robert Ricci committed
305 306 307
	my $objname = event_notification_get_objname($handle,$notification);
	my $eventtype = event_notification_get_eventtype($handle,$notification);

308
	info("Got an event: ($objtype,$objname,$eventtype)\n");
309

310 311 312 313
	#
	# Check to see if another instance is supposed to be handling this node
	#
	if (!checkDBRedirect($objname)) {
314
	    info("Got an event for node $objname, which isn't mine\n");
315 316 317
	    return;
	}

318 319
	SWITCH: for ($objtype) {

320
	    (/$TBNODESTATE/) && do {
321 322 323
		stateTransition($objname,$eventtype);
		last;
	    };
324
	    (/$TBNODEOPMODE/) && do {
325
		opModeTransition($objname,$eventtype);
326 327 328 329
		notify("Use of deprecated event TBNODEOPMODE:\n".
		       "$objname->$eventtype\n");
		last;
	    };
330
	    (/$TBCONTROL/) && do {
331
		handleCtrlEvent($objname,$eventtype);
332 333 334 335 336 337 338 339 340
		last;
	    };

	}

}

sub stateTransition($$) {

341
    my ($node,$newstate) = @_;
Robert Ricci's avatar
Robert Ricci committed
342

343 344 345 346 347 348 349 350
    # 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();
351
	if ($nodes{$node}) {
352 353
	    $oldstate = $nodes{$node}{state};
	    $mode = $nodes{$node}{mode};
Robert Ricci's avatar
Robert Ricci committed
354
	} else {
355
	    notify("Got an event for a node ($node) I don't know about\n");
Robert Ricci's avatar
Robert Ricci committed
356
	}
357 358 359 360
    }
    if ($oldstate && $mode && $valid{$mode} && $valid{$mode}{$oldstate} &&
	!$valid{$mode}{$oldstate}{$newstate}) {
	notify("Invalid transition for node $node from $mode/$oldstate " .
361
	       "to $newstate\n".`/usr/testbed/sbin/inuse | grep $node\  `);
362
    }
Robert Ricci's avatar
Robert Ricci committed
363

364 365 366 367
    my $now = time();
    $nodes{$node}{state}     = $newstate;
    $nodes{$node}{timestamp} = $now;
    $nodes{$node}{notified}  = 0;
368

369 370 371
    info("$node: $mode/$oldstate => $mode/$newstate\n");
    DBQueryFatal("UPDATE nodes SET eventstate='$newstate', " .
		 "state_timestamp='$now' WHERE node_id='$node'");
372

373 374 375 376 377
    # Check if this is TBDB_NODESTATE_BOOTING , which has actions
    if ($newstate eq TBDB_NODESTATE_BOOTING) {
	my $osid = TBBootWhat($node,$debug);
	info("$node: Current OS is '$osid'\n");
	DBQueryFatal("UPDATE nodes SET osid='$osid' WHERE node_id='$node'");
378

379 380
	checkGenISUP($node);
    }
381

382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403
    # Check if this state is the "successful boot" state
    if (defined($triggers{"$mode:$newstate"})) {
	my @trigs = split(/\s*,\s*/,$triggers{"$mode:$newstate"});
	foreach ( @trigs ) {
	    my $trig = $_;
	    /^$TBRESET$/ && do {
		# 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,$trig);
		next;
	    };
	    notify("Unknown trigger '$trig' in $mode/$newstate!\n");
	}
    }
404

405 406 407 408 409 410 411 412 413 414 415 416
    # 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);
	} else { debug("No next mode.\n"); }
    }
    
}
417

418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463
sub opModeTransition($$) {
    
    my ($node,$newmode) = @_;
    
    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 ($node) I don't know about\n");
	}
    }
    if (defined($modeTrans{"$mode:$oldstate"})) {
	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{$newmode};
	} else {
	    notify("Invalid mode transition from ".
		   "$mode/$oldstate to $newmode!\n");
	}
    } else {
	notify("Invalid mode transition from $mode/$oldstate: ".
	       "Not a valid mode transition state!\n");
    }
    if (!$nextstate) { $nextstate=$oldstate; }

    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
464

465 466 467 468 469 470 471 472 473 474 475 476 477 478
    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'");
}

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

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

    foreach ($event) {
	/^$TBRESET$/ && do {
Mac Newbold's avatar
Mac Newbold committed
479 480 481 482
	    my $result = DBQueryFatal("SELECT pxe_boot_path, def_boot_osid ".
				      "FROM nodes where node_id='$node'");
	    my ($pxepath,$osid) = $result->fetchrow();

483 484
	    $pxepath = "-p ".$pxepath;
	    if ($pxepath eq "-p ") { $pxepath="PXEBOOT"; };
Mac Newbold's avatar
Mac Newbold committed
485
	    my $cmd = "$osselect -m $pxepath $node";
486 487 488
	    system($cmd) and
		notify("$node/$event: Couldn't clear next_pxe_boot_path\n".
		       "\tcmd=$cmd\n\t*** $!\n");
489

490 491 492 493
	    $cmd = "$osselect $osid $node";
	    system($cmd) and
		notify("$node/$event: Couldn't clear next_boot_*\n".
		       "\tcmd=$cmd\n\t*** $!\n");
494

Mac Newbold's avatar
Mac Newbold committed
495
	    info("Performed RESET for $node to $osid/$pxepath\n");
496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520
	    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'");
		DBQueryFatal("delete from reserved where node_id='$node'");
		info("Released $node from $pid/$eid\n");
	    }
	    next;
	};
	/^$TBTIMEOUT$/ && do {
	    my $state = $nodes{$node}{state};
	    my $mode = $nodes{$node}{mode};
	    my ($timeout,$action);
	    if ($mode && $state && $timeouts{$mode} &&
		$timeouts{$mode}{$state}) {
		($timeout, $action) = @{$timeouts{$mode}{$state}};
	    }
	    notify("Node $node has timed out in state $mode/$state".
		   ($action ne "" ? "\n\tRequested action $action." : "").
521
		   "\n".`/home/newbold/z/bin/statetime | grep '$node\t'`);
522 523
	    next;
	};
524
	notify("$node: Unknown CtrlEvent: $event\n");
525 526
    }
}
Robert Ricci's avatar
Robert Ricci committed
527

528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546
#
# Check if we need to generate an ISUP
#
sub checkGenISUP($) {
    my ($node) = @_;
    info("$node: Checking ISUP Generation\n");
    my $r = DBQueryWarn("select osfeatures from os_info as o ".
			"left join nodes as n on o.osid=n.osid ".
			"where node_id='$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(); }
   
    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
    }
Robert Ricci's avatar
Robert Ricci committed
547

548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607
    # If os will send ISUP on its own, do nothing here.
    if ($can{"isup"}) { return 0; }

    # If os doesn't support isup but can ping, fork and ping it every
    # 5 seconds and send isup when it pings, or timeout after too long.
    if ($can{"ping"}) {
	my $pid = fork();
	if ($pid) { return 0; } # don't wait, return and go on with life
	else {
	    debug("Forked process to ping $node\n");

	    my $wait=5; # 5 seconds between ping attempts
	    my $maxtime=600; # Set the timer for 10 minutes
	    
	    # XXX : If our maxtime is very different from os_setup's
	    #idea of how long a node should take to reboot, then we've
	    #got a problem, since it might reboot the nodes.

	    # Set an alarm in case it never comes up...
	    local $SIG{ALRM} = sub { 
		notify("$node: checkGenISUP timed out waiting ".
		       "for ping responses after $maxtime seconds.\n");
		exit(); 
	    };
	    
	    alarm $maxtime;

	    my $n=0;
	    while ( $n <= ($maxtime/$wait)) {
		my $rv=system("/sbin/ping -c 1 -t 1 $node > /dev/null") >> 8;
		if ($rv==0) {
		    # we got a response
		    debug("$node: ping response received... sending ISUP\n");
		    stateTransition($node, TBDB_NODESTATE_ISUP);
		    exit();
		} elsif ($rv==2) {
		    # no response
		    debug("No ping response, waiting $wait seconds...\n");
		} elsif ($rv==68) {
		    # ping: cannot resolve $node: Unknown host
		    notify("$node: ping couldn't resolve $node!\n");
		    exit();
		} else {
		    notify("$node: checkGenISUP ping returned $rv!\n");
		    exit();
		}
		sleep $wait;
		$n+=1;
	    }
	    notify("$node: Sent $n pings in ".($n*$wait)." seconds ".
		   "with no response.\n");
	    exit();
	}
    }
    # If os doesn't support ping or isup, stated sets it to ISUP at the
    # same time.
    
    debug("$node: OS doesn't ping - sending ISUP\n");
    stateTransition($node, TBDB_NODESTATE_ISUP);
    
Robert Ricci's avatar
Robert Ricci committed
608 609
}

610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638
#
# 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($) {

	my ($node) = @_;

	#
	# 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'");

	if (!$result->num_rows()) {
		notify("Got an event for a node ($node) I don't know about\n");
		return 0;
	}

	my ($testdb) = $result->fetchrow();

	#
	# XXX: It's hokey to hardcode tbdb here, but....
	#
639 640

	#debug("checkDBRedirect: $node => $testdb (I'm $TBDBNAME)\n");
641 642 643 644 645 646 647 648
	if ((!$testdb && ($TBDBNAME eq "tbdb")) ||
	    ($testdb && ($testdb eq $TBDBNAME))) {
		return 1;
	} else {
		return 0;
	}
}

Robert Ricci's avatar
Robert Ricci committed
649 650 651 652
#
# Reload state from the database
#
sub reload() {
653
    	info("Reloading state from database\n");
654 655
	%nodes     = readStates(%nodes);
	%timeouts  = getTimeouts();
Robert Ricci's avatar
Robert Ricci committed
656
	%valid     = getValid();
657 658
	%modeTrans = getModeTrans();
	%triggers  = getTriggers();
Robert Ricci's avatar
Robert Ricci committed
659 660 661
}

sub debug(@) {
Mac Newbold's avatar
Mac Newbold committed
662 663 664 665 666 667 668
  if ($debug) { print @_; }
}

sub fatal($) {
  my $msg = shift;
  notify($msg);
  die($msg);
Robert Ricci's avatar
Robert Ricci committed
669 670 671
}

sub notify($) {
Mac Newbold's avatar
Mac Newbold committed
672 673 674 675 676 677 678 679 680
  my $message = shift;
  if (!$debug) {
    SENDMAIL("Stated List <".$TBOPS.">",
	     "Node State Daemon Messsage",$message,
	     "Stated Daemon <".$TBOPS.">");
  } else {
    print "notify: Not sending mail in debug mode\n";
  }
  info($message);
681 682
}

683
sub info($;$) {
684
    my $message = shift;
685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708
    my $notice = shift || 0;
    if (!$debug) {
      # Use syslog
      my $prio="info";
      if ($notice) { $prio = "notice"; }
      syslog($prio,$message);
    } else {
      # Print out log entries like this:
      # Sep 20 09:36:00 stated[238]: Reloading state from database
      print strftime("%b %e %H:%M:%S",localtime)." stated[$$]: $message";
    }
}

# This gets called if we catch a signal USR1
sub restart {
  my $params = join(" ",@args);
  my $prog = "";
  # If we're started from an abosolute path, use that.
  if ($0 =~ /^\//) { $prog = $0; } else { $prog = "$TB/sbin/stated"; }
  info("SIGUSR1 received, restarting from '$prog".
       ($params ne "" ? " $params" : "")."'\n");
  if ($handle && event_unregister($handle) == 0) {
    warn "Unable to unregister with event system\n";
  }
Mac Newbold's avatar
Mac Newbold committed
709
  if (defined($lockfile) && $lockfile ne "") { unlink $lockfile; }
710 711 712 713 714 715 716 717 718 719 720
  if (!defined(sigprocmask(SIG_UNBLOCK, POSIX::SigSet->new(SIGUSR1,SIGHUP)))) {
    notify("sigprocmask: sig unblock failed! $?, $!\n");
    die("\n");
  }
  exec("$prog $params") or 
    do {
      my $msg = "Couldn't restart stated! cmd='$prog $params'\n".
	"Error: ($?) $!\n";
      notify($msg);
      die($msg);
    };
Robert Ricci's avatar
Robert Ricci committed
721 722
}

723 724
# This gets called if we catch a signal (TERM, etc.)
sub cleanup {
725 726 727
  info("Signal received, exiting\n");
  # now do the normal exit stuff in END {}
  exit(0);
728 729
}

Robert Ricci's avatar
Robert Ricci committed
730 731
# This gets called if we die of 'natural causes' (exit, die, etc.)
END {
732
  info("Exiting, cleaning up\n");
Mac Newbold's avatar
Mac Newbold committed
733
  if (defined($lockfile) && $lockfile ne "") { unlink $lockfile; }
734 735
  # clean up Syslog if we were using it.
  if (!$debug) { closelog(); }
736 737 738
  if ($handle) {
    if (event_unregister($handle) == 0) {
      die "Unable to unregister with event system\n";
739
    }
740
  }
Robert Ricci's avatar
Robert Ricci committed
741
}
742