stated.in 26.7 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 $REALTBOPS = "@TBOPSEMAIL@";
26
my $TBDBNAME = "@TBDBNAME@";
27
my $osselect = "$TB/bin/os_select";
Robert Ricci's avatar
Robert Ricci committed
28 29 30 31 32 33 34

$| = 1;

use event;
use libdb;
use libtestbed;
use Getopt::Std;
35
#use strict;
Robert Ricci's avatar
Robert Ricci committed
36
use English;
37
use POSIX; # for strftime, and sigprocmask and friends
38
use Fcntl; # file constants for pidfile
Mac Newbold's avatar
Mac Newbold committed
39 40 41 42 43
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
44

Mac Newbold's avatar
Mac Newbold committed
45 46 47
# Do lots of db retries before we fail and die
$libdb::DBQUERY_MAXTRIES = 5;

Robert Ricci's avatar
Robert Ricci committed
48 49 50
# 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.
51 52
my $reload_time = 600;
my $last_reload = 0;
Robert Ricci's avatar
Robert Ricci committed
53 54 55 56 57

# Process command-line arguments

sub usage {
	print << "END";
58 59 60
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
61 62
-s server       Use specified server, instead of this site's bossnode
-p port	        Use specified port
63
Send SIGHUP to reload database state, or SIGUSR1 to restart completely.
Robert Ricci's avatar
Robert Ricci committed
64 65 66
END
}

67
my @args = @ARGV; # save a copy for restart before we mess with them.
Robert Ricci's avatar
Robert Ricci committed
68
my %opt = ();
69
getopts("ds:p:h",\%opt);
Robert Ricci's avatar
Robert Ricci committed
70 71 72 73

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

74
my ($server,$port,$debug);
Robert Ricci's avatar
Robert Ricci committed
75 76 77 78
if ($opt{s}) { $server = $opt{s}; } else { $server = $BOSSNODE; }
if ($opt{p}) { $port = $opt{p}; }
if ($opt{d}) { $debug = 1; } else { $debug = 0; }

79 80 81 82 83 84 85 86
# 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;
87
my $TB_OSID_MBKERNEL = TB_OSID_MBKERNEL;
88

Mac Newbold's avatar
Mac Newbold committed
89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104
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: $? $!\n");
}
Robert Ricci's avatar
Robert Ricci committed
105
# Background
106 107 108
if (!$debug) {
  # We use syslog, so redirect the output to nothing
  if (TBBackGround("/dev/null")) { exit(0); }
Robert Ricci's avatar
Robert Ricci committed
109
}
Mac Newbold's avatar
Mac Newbold committed
110 111
# set up syslog
openlog("stated","pid","user");
Mac Newbold's avatar
Mac Newbold committed
112 113 114 115 116 117
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
118

119 120 121
# Change my $0 so that it is easier to see in a ps/top
$0 = "$0";

Robert Ricci's avatar
Robert Ricci committed
122 123 124 125 126
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
127
if (!$handle) { fatal("Unable to register with event system\n"); }
Robert Ricci's avatar
Robert Ricci committed
128 129

my $tuple = address_tuple_alloc();
Mac Newbold's avatar
Mac Newbold committed
130
if (!$tuple) { fatal("Could not allocate an address tuple\n"); }
Robert Ricci's avatar
Robert Ricci committed
131

132 133
%$tuple = ( objtype => join(",",$TBNODESTATE,$TBNODEOPMODE,$TBCONTROL) );

Robert Ricci's avatar
Robert Ricci committed
134
if (!event_subscribe($handle,\&handleEvent,$tuple)) {
Mac Newbold's avatar
Mac Newbold committed
135
	fatal("Could not subscribe to events\n");
Robert Ricci's avatar
Robert Ricci committed
136 137 138 139
}

# Read in the pre-existing node states, and timeout and valid transition
# information from the database
140 141 142 143 144
my %nodes     = readStates();
my %timeouts  = getTimeouts();
my %valid     = getValid();
my %modeTrans = getModeTrans();
my %triggers  = getTriggers();
Robert Ricci's avatar
Robert Ricci committed
145 146 147 148

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

Mac Newbold's avatar
Mac Newbold committed
149 150 151 152 153 154
# Set when I've got a child pinging a node for me.
my $have_children = 0;

# Keep track of which pid was pinging which node.
my %children= (); # $children($child} = "pcXXX";

Robert Ricci's avatar
Robert Ricci committed
155 156 157
# 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.
158 159
$SIG{HUP}  = sub { info("SIGHUP - Reloading DB state\n"); $do_reload = 1; };

Mac Newbold's avatar
Mac Newbold committed
160
# Set up other signals.
161 162 163
$SIG{USR1} = \&restart;
$SIG{USR2} = \&cleanup;
$SIG{INT}  = \&cleanup;
164 165 166 167
$SIG{QUIT} = \&cleanup;
$SIG{ABRT} = \&cleanup;
$SIG{TERM} = \&cleanup;
$SIG{KILL} = \&cleanup;
Robert Ricci's avatar
Robert Ricci committed
168

Mac Newbold's avatar
Mac Newbold committed
169
# Set up some notification throttling
170
my $mailgap = 15; # in seconds
171
my $lastmail = time - $mailgap + 2; # Send a digest of startup msgs after 2s.
Mac Newbold's avatar
Mac Newbold committed
172 173
my %msgs = ();

174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189
# Track if I handled an event or not
my $event_count = 0;

notify("Stated starting up\n");

sub process_event_queue() {
    $event_count=0;
    my $lastcount=-1;
    while ($event_count != $lastcount) {
	$lastcount = $event_count;
	event_poll($handle);
    }
    if ($event_count > 0) {
      debug("Handled $event_count event(s).\n");
    }
}
Robert Ricci's avatar
Robert Ricci committed
190

191
# Now, we just poll for events, and watch for timeouts
Robert Ricci's avatar
Robert Ricci committed
192
while (1) {
193
	process_event_queue;
Robert Ricci's avatar
Robert Ricci committed
194 195 196 197 198
	my $now = time();
	#
	# Look for nodes that have passed their timeout
	#
	while (my ($node, $value) = each %nodes) {
199 200 201 202 203 204 205 206 207 208 209 210 211 212 213
	    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
214 215
	}

216
	if ($do_reload || ($now - $last_reload > $reload_time)) {
Robert Ricci's avatar
Robert Ricci committed
217 218 219 220
		reload();
		$do_reload = 0;
	}

Mac Newbold's avatar
Mac Newbold committed
221 222 223 224 225 226 227 228
	if ($have_children) {
	  # Check for kids that have finished
	  handleChild();
	}

	# Send any messages in the queue if it is time
	notify("",1);

Robert Ricci's avatar
Robert Ricci committed
229 230 231
	sleep(1);
}

Mac Newbold's avatar
Mac Newbold committed
232 233
exit(0);

Robert Ricci's avatar
Robert Ricci committed
234
# Read the current states of nodes from the database
235 236 237
sub readStates(;@) {
    	my %oldnodes = @_;

238 239 240 241 242 243 244
	#
	# Guard against undefined variable warnings
	#
	if (! defined(%oldnodes)) {
		%oldnodes = ();
	}

245
	#debug("readStates called\n");
Robert Ricci's avatar
Robert Ricci committed
246
	my $result = DBQueryFatal("SELECT node_id, eventstate, " .
247 248
				  "state_timestamp, op_mode, " .
				  "op_mode_timestamp FROM nodes");
Robert Ricci's avatar
Robert Ricci committed
249 250

	my %nodes;
251 252
	while (my ($node_id, $state, $timestamp, $mode, $mode_timestamp)
			= $result->fetchrow()) {
253 254 255 256 257 258
	    	#
		# If there's an entry in oldnodes for this node, and it
		# hasn't changed state or time, use the old entry (so that
		# we don't lose information about which nodes we've already
		# notified the ops about, etc.)
		#
259 260
		if ($oldnodes{$node_id} && $state && $timestamp &&
		   ($oldnodes{$node_id}{state} eq $state) &&
261
		   ($oldnodes{$node_id}{mode} eq $mode) &&
262
		   ($oldnodes{$node_id}{timestamp} == $timestamp)) {
263 264
		       $nodes{$node_id} = $oldnodes{$node_id};
		} else { 
265 266 267 268
		    $nodes{$node_id}{state}          = $state;
		    $nodes{$node_id}{timestamp}      = $timestamp;
		    $nodes{$node_id}{mode}           = $mode;
		    $nodes{$node_id}{mode_timestamp} = $mode_timestamp;
269
		}
Robert Ricci's avatar
Robert Ricci committed
270 271 272 273 274 275 276 277
	}
	return %nodes;
}

#
# Read timeouts for various states from the database
#
sub getTimeouts() {
278
	#debug("getTimeouts called\n");
279
	my $result = DBQueryFatal("SELECT op_mode, state, timeout, action " .
Robert Ricci's avatar
Robert Ricci committed
280 281 282
				  "FROM state_timeouts");

	my %timeouts;
283 284
	while (my ($op_mode, $state, $timeout, $action) = $result->fetchrow()) {
		$timeouts{$op_mode}{$state} = [ $timeout, $action ];
Robert Ricci's avatar
Robert Ricci committed
285 286 287 288 289 290 291 292
	}
	return %timeouts;
}

#
# Read the list of valid state transitions from the database
#
sub getValid() {
293
	#debug("getValid called\n");
294
	my $result = DBQueryFatal("SELECT op_mode, state1, state2 " .
Robert Ricci's avatar
Robert Ricci committed
295 296 297
				  "FROM state_transitions");

	my %valid;
298 299
	while (my ($mode,$state1, $state2) = $result->fetchrow()) {
		$valid{$mode}{$state1}{$state2} = 1;
Robert Ricci's avatar
Robert Ricci committed
300 301 302 303
	}
	return %valid;
}

304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340
#
# 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
341 342 343 344 345
#
# Gets called for every event that we recieve
#
sub handleEvent($$$) {
	my ($handle,$notification,$data) = @_;
346
	my $objtype = event_notification_get_objtype($handle,$notification);
Robert Ricci's avatar
Robert Ricci committed
347 348 349
	my $objname = event_notification_get_objname($handle,$notification);
	my $eventtype = event_notification_get_eventtype($handle,$notification);

350
	$event_count++;
351
	info("Got an event: ($objtype,$objname,$eventtype)\n");
352

353 354 355 356
	#
	# Check to see if another instance is supposed to be handling this node
	#
	if (!checkDBRedirect($objname)) {
357
	    info("Got an event for node $objname, which isn't mine\n");
358 359 360
	    return;
	}

361 362
	SWITCH: for ($objtype) {

363
	    (/$TBNODESTATE/) && do {
364 365 366
		stateTransition($objname,$eventtype);
		last;
	    };
367
	    (/$TBNODEOPMODE/) && do {
368
		opModeTransition($objname,$eventtype);
369 370 371 372
		notify("Use of deprecated event TBNODEOPMODE:\n".
		       "$objname->$eventtype\n");
		last;
	    };
373
	    (/$TBCONTROL/) && do {
374
		handleCtrlEvent($objname,$eventtype);
375 376 377 378 379 380 381 382 383
		last;
	    };

	}

}

sub stateTransition($$) {

384
    my ($node,$newstate) = @_;
Robert Ricci's avatar
Robert Ricci committed
385

386 387 388 389 390 391 392 393
    # 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();
394
	if ($nodes{$node}) {
395 396
	    $oldstate = $nodes{$node}{state};
	    $mode = $nodes{$node}{mode};
Robert Ricci's avatar
Robert Ricci committed
397
	} else {
398
	    notify("Got an event for a node ($node) I don't know about\n");
Robert Ricci's avatar
Robert Ricci committed
399
	}
400 401 402 403
    }
    if ($oldstate && $mode && $valid{$mode} && $valid{$mode}{$oldstate} &&
	!$valid{$mode}{$oldstate}{$newstate}) {
	notify("Invalid transition for node $node from $mode/$oldstate " .
404
	       "to $newstate\n");
405
    }
Robert Ricci's avatar
Robert Ricci committed
406

407 408 409 410
    my $now = time();
    $nodes{$node}{state}     = $newstate;
    $nodes{$node}{timestamp} = $now;
    $nodes{$node}{notified}  = 0;
411

412 413 414
    info("$node: $mode/$oldstate => $mode/$newstate\n");
    DBQueryFatal("UPDATE nodes SET eventstate='$newstate', " .
		 "state_timestamp='$now' WHERE node_id='$node'");
415

416 417 418
    # Check if this is TBDB_NODESTATE_BOOTING , which has actions
    if ($newstate eq TBDB_NODESTATE_BOOTING) {
	my $osid = TBBootWhat($node,$debug);
419 420
	my $os_op_mode = os_opmode($osid);
	info("$node: Current OS is '$osid', OS mode is '$os_op_mode'\n");
421
	DBQueryFatal("UPDATE nodes SET osid='$osid' WHERE node_id='$node'");
422
	if ($os_op_mode ne $mode) {
423 424
	    my $str = "Node $node is running OS '$osid' but in mode '$mode' ".
	      "instead of mode '$os_op_mode'!\n";
425 426 427
	    # For now, only force if we're going into reload mode, so we
	    # don't get stuck looping in reloading.
	    if ($os_op_mode eq "RELOAD") {
428 429 430 431 432 433 434 435 436
		DBQueryFatal("UPDATE nodes SET op_mode='$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($str);
	}
437 438
	checkGenISUP($node);
    }
439

440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458
    # 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,$trig);
		next;
	    };
459
	    notify("Unknown trigger '$trig' for $node in $mode/$newstate!\n");
460 461
	}
    }
462

463 464 465 466 467 468 469 470 471 472 473
    # 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"); }
    }
}
474

475
sub opModeTransition($$) {
Mac Newbold's avatar
Mac Newbold committed
476

477
    my ($node,$newmode) = @_;
Mac Newbold's avatar
Mac Newbold committed
478

479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505
    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 {
506
	    notify("Invalid mode transition for $node from ".
507 508 509
		   "$mode/$oldstate to $newmode!\n");
	}
    } else {
510
	notify("Invalid mode transition for $node from $mode/$oldstate: ".
511 512 513 514 515 516 517 518 519 520
	       "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
521

522 523 524 525 526 527 528 529 530 531 532 533 534 535
    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
536 537 538 539
	    my $result = DBQueryFatal("SELECT pxe_boot_path, def_boot_osid ".
				      "FROM nodes where node_id='$node'");
	    my ($pxepath,$osid) = $result->fetchrow();

540 541 542 543 544 545 546 547 548 549
	    # 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.

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

550 551
	    $pxepath = "-p ".$pxepath;
	    if ($pxepath eq "-p ") { $pxepath="PXEBOOT"; };
Mac Newbold's avatar
Mac Newbold committed
552
	    my $cmd = "$osselect -m $pxepath $node";
553 554 555
	    system($cmd) and
		notify("$node/$event: Couldn't clear next_pxe_boot_path\n".
		       "\tcmd=$cmd\n\t*** $!\n");
556

Mac Newbold's avatar
Mac Newbold committed
557
	    info("Performed RESET for $node to $osid/$pxepath\n");
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
	    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 $action." : "").
583
		   "\n");
584 585
	    next;
	};
586
	notify("$node: Unknown CtrlEvent: $event\n");
587 588
    }
}
Robert Ricci's avatar
Robert Ricci committed
589

590 591 592 593
#
# Check if we need to generate an ISUP
#
sub checkGenISUP($) {
Mac Newbold's avatar
Mac Newbold committed
594 595 596 597 598 599 600 601 602 603
  my ($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();
  }
Robert Ricci's avatar
Robert Ricci committed
604

Mac Newbold's avatar
Mac Newbold committed
605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628
  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
  }

  # If os will send ISUP on its own, do nothing here.
  if ($can{"isup"}) {
    debug("$node: Will send own ISUP\n"); return 0;
  }

  # 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\n");
    my $pid = fork();
    if ($pid) {
      $children{$pid}= $node;
      $have_children = 1;
      # don't wait, return and go on with life
      return 0;
    } else {
      info("Forked process $$ to ping $node\n");
629
      $lockfile = ""; # Don't clean up my pidfile, since I'm a child
Mac Newbold's avatar
Mac Newbold committed
630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672

      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(4);
      };

      alarm $maxtime;

      my $n=0;
      while ( $n <= ($maxtime/$wait)) {
	my $status=system("/sbin/ping -c 1 -t 1 $node > /dev/null");
	my $rv = $status >> 8;
	my $sig = $status & 127;
	debug("checkGenISUP: ping returned $status ($rv / $sig)\n");
	if ($rv==0) {
	  # we got a response
	  info("$node: ping response received... sending ISUP\n");
	  # We want to send an ISUP event for the node. But we
	  # can't just call stateTransition, because that will
	  # only make the change happen in our state, not the real
	  # stated that we forked from. We do this by exit(0)
	  # and the handleChild function will pick it up and
	  # send the ISUP for us.
	  exit(0);
	} elsif ($rv==2) {
	  # no response
	  debug("$node: No ping response, waiting $wait seconds...\n");
	} elsif ($rv==68) {
	  # ping: cannot resolve $node: Unknown host
	  notify("$node: ping couldn't resolve $node!\n");
	  exit(2);
	} else {
	  notify("$node: checkGenISUP ping returned $rv!\n");
	  exit(3);
673
	}
Mac Newbold's avatar
Mac Newbold committed
674 675 676 677 678 679
	sleep $wait;
	$n+=1;
      }
      notify("$node: Sent $n pings in ".($n*$wait)." seconds ".
	     "with no response.\n");
      exit(1);
680
    }
Mac Newbold's avatar
Mac Newbold committed
681 682 683 684 685 686
  }
  # 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
687 688
}

689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715
# 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....
	#
716 717

	#debug("checkDBRedirect: $node => $testdb (I'm $TBDBNAME)\n");
718 719
	if ((!$testdb && ($TBDBNAME eq "tbdb")) ||
	    ($testdb && ($testdb eq $TBDBNAME))) {
720
	  return 1;
721
	} else {
722
	  return 0;
723 724 725
	}
}

Robert Ricci's avatar
Robert Ricci committed
726 727
# Reload state from the database
sub reload() {
728 729
  debug("Reloading state from database\n");
  $last_reload = time();
Mac Newbold's avatar
Mac Newbold committed
730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765
  %nodes     = readStates(%nodes);
  %timeouts  = getTimeouts();
  %valid     = getValid();
  %modeTrans = getModeTrans();
  %triggers  = getTriggers();
}

# This gets called to check for forked pids that have finished.  Right
# now we only fork in checkGenISUP to ping nodes.
sub handleChild() {
  my $child = wait;
  if ($child==-1) { return 0; }
  my $node = $children{$child};
  if (!defined($node)) { $node=""; }
  my $stat = $?;
  my $rv = $stat >> 8;
  my $sig = $stat & 127;
  debug("Child = $child, I am $$, children are: (".
	join(",",sort keys %children).")\n");
  if (($node ne "") && ($child!=-1) && ($child!=$$)) {
    info("handleChild: pid $child (node $node), exited $rv (sig $sig)\n");
    delete $children{$child};
    if (($sig == 0) && ($rv == 0)) {
      # Node is pingable, send isup
      info("handleChild: Sending ISUP for $node)\n");
      stateTransition($node, TBDB_NODESTATE_ISUP);
    } else {
      notify("handleChild: Caught a child that failed!\n".
	     "pid $child (node $node), exited $rv (sig $sig)\n");
    }
  }
  if (0+%children == 0) {
    debug("No more children now.\n");
    $have_children=0;
  }
  return 0;
Robert Ricci's avatar
Robert Ricci committed
766 767
}

768 769
sub os_opmode() {
    my $osid = shift || "";
770
    if ($osid eq $TB_OSID_MBKERNEL) { return "MINIMAL"; }
771 772 773 774 775 776 777 778 779
    my $cmd = "select op_mode from os_info where osid='$osid';";
    my $q = DBQueryFatal($cmd);
    if ($q->numrows() < 1) { return ""; }
    my @r = $q->fetchrow_array();
    my $opmode=$r[0];
    debug("OpMode for '$osid' is '$opmode'\n");
    if (defined($opmode) && $opmode ne "") { return $opmode; }
    return "";
}
Mac Newbold's avatar
Mac Newbold committed
780

Robert Ricci's avatar
Robert Ricci committed
781
sub debug(@) {
Mac Newbold's avatar
Mac Newbold committed
782 783 784 785 786 787 788
  if ($debug) { print @_; }
}

sub fatal($) {
  my $msg = shift;
  notify($msg);
  die($msg);
Robert Ricci's avatar
Robert Ricci committed
789 790
}

Mac Newbold's avatar
Mac Newbold committed
791
sub showqueue() {
792
  if ($debug < 2) { return; }
Mac Newbold's avatar
Mac Newbold committed
793 794 795 796 797 798 799 800 801 802
  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");
  }
}

sub notify($;$) {
Mac Newbold's avatar
Mac Newbold committed
803
  my $message = shift;
Mac Newbold's avatar
Mac Newbold committed
804 805 806 807 808 809 810 811 812 813 814 815 816 817
  my $checkonly = shift || 0;
  # Use a timestamp, now that we're throttling mail
  my $tstamp=strftime("%b %e %H:%M:%S",localtime);
  showqueue();
  if (!$checkonly) {
    info($message);
    # Queue up the message
    # (The queue is a hash of lists of timestamps, keyed by message
    if (defined($msgs{$message})) {
      push(@{$msgs{$message}},$tstamp);
    } else {
      $msgs{$message} = [$tstamp];
    }
    showqueue();
Mac Newbold's avatar
Mac Newbold committed
818
  }
Mac Newbold's avatar
Mac Newbold committed
819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845
  my $now = time;
  if ($now - $lastmail >= $mailgap) {
    if ((keys %msgs)>0) {
      debug("SENDING MAILQUEUE\n"."(now $now, lastmail $lastmail, ".
	    ($now-$lastmail).">=$mailgap)\n");
      my $mailbody="";
      my $sep = '-'x5;
      # We're okay to send. Make a digest of all the queued messages.
      foreach my $msg (sort keys %msgs) {
	my @tlist = @{$msgs{$msg}};
	my $count = 0+@tlist;
	$mailbody .= "\n$msg\n";
	if ($count > 1) {
	  my $first = shift @tlist;
	  my $last = pop @tlist;
	  $mailbody .= "($count copies from $first to $last)\n";
	} else {
	  $mailbody .= "($count copy at $tlist[0])\n";
	}
	$mailbody .= "$sep\n";
      }
      # Now reset the mail queue
      %msgs = ();
      showqueue();
      $lastmail = time;
      if (!$debug) {
	SENDMAIL("Stated List <".$TBOPS.">",
846
		 "Stated Messsage",$mailbody,
Mac Newbold's avatar
Mac Newbold committed
847 848 849 850 851 852 853
		 "Stated Daemon <".$TBOPS.">");
      } else {
	debug("notify: Not sending mail in debug mode\n");
	debug("MAIL CONTAINS:\n".$mailbody."\n");
      }
    }
  }  # else do nothing, not time yet
854 855
}

856 857 858 859 860 861 862 863 864 865 866 867 868 869 870
sub announce($) {
  my $message = shift;
  my $tstamp=strftime("%b %e %H:%M:%S",localtime);
  notify("ANNOUCEMENT: ".$message."\n\n(Sent to $REALTBOPS)\n");
  $mailbody = "\n$message\n\n$tstamp\n";
  if (!$debug) {
    SENDMAIL($REALTBOPS,
	     "Stated Messsage",$mailbody,
	     "Stated Daemon <".$TBOPS.">");
  } else {
    debug("announce: Not sending mail in debug mode\n");
    debug("MAIL CONTAINS:\n".$mailbody."\n");
  }
}

871
sub info($;$) {
872
    my $message = shift;
873
    my $notice = shift || 0;
Mac Newbold's avatar
Mac Newbold committed
874 875 876 877
    # Use syslog
    my $prio="info";
    if ($notice) { $prio = "notice"; }
    if ($debug) {
878 879 880
      # 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";
Mac Newbold's avatar
Mac Newbold committed
881
      $message = "DEBUG: ".$message;
882
    }
Mac Newbold's avatar
Mac Newbold committed
883
    syslog($prio,$message) || notify("syslog failed: $? $!\n");
884 885 886 887
}

# This gets called if we catch a signal USR1
sub restart {
Mac Newbold's avatar
Mac Newbold committed
888
  info("SIGUSER1 received: Performing final event poll before restarting\n");
889
  process_event_queue;
890 891 892 893
  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"; }
Mac Newbold's avatar
Mac Newbold committed
894
  info("Restarting from '$prog".($params ne "" ? " $params" : "")."'\n");
895 896 897
  if ($handle && event_unregister($handle) == 0) {
    warn "Unable to unregister with event system\n";
  }
Mac Newbold's avatar
Mac Newbold committed
898
  if (defined($lockfile) && $lockfile ne "") { unlink $lockfile; }
899 900 901 902
  if (!defined(sigprocmask(SIG_UNBLOCK, POSIX::SigSet->new(SIGUSR1,SIGHUP)))) {
    notify("sigprocmask: sig unblock failed! $?, $!\n");
    die("\n");
  }
903
  announce("Stated restarted\n");
904 905 906 907
  exec("$prog $params") or 
    do {
      my $msg = "Couldn't restart stated! cmd='$prog $params'\n".
	"Error: ($?) $!\n";
908
      announce($msg);
909 910
      die($msg);
    };
Robert Ricci's avatar
Robert Ricci committed
911 912
}

913 914
# This gets called if we catch a signal (TERM, etc.)
sub cleanup {
915
  notify("Signal received, exiting\n");
916 917
  # now do the normal exit stuff in END {}
  exit(0);
918 919
}

Robert Ricci's avatar
Robert Ricci committed
920 921
# This gets called if we die of 'natural causes' (exit, die, etc.)
END {
922 923 924 925 926 927 928
  if (defined($lockfile) && $lockfile ne "") {
    unlink $lockfile;
    announce("Stated exiting, cleaning up\n");
  } else {
    # Must be a child
    info("Stated child exiting\n");
  }
Mac Newbold's avatar
Mac Newbold committed
929 930
  # clean up Syslog
  closelog();
931 932 933
  if ($handle) {
    if (event_unregister($handle) == 0) {
      die "Unable to unregister with event system\n";
934
    }
935
  }
Robert Ricci's avatar
Robert Ricci committed
936
}
937