stated.in 46.3 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) 2000-2005 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
Mac Newbold's avatar
Mac Newbold committed
11
# state into the database. Also watches for invalid transitions, timeouts,
12
# 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 $apod = "$TB/sbin/apod";
34
my $TBLOG = "@TBLOGFACIL@";
Robert Ricci's avatar
Robert Ricci committed
35 36 37 38 39 40

$| = 1;

use event;
use libdb;
use libtestbed;
41
use TimeoutQueue;
Robert Ricci's avatar
Robert Ricci committed
42
use Getopt::Std;
43
#use strict;
Robert Ricci's avatar
Robert Ricci committed
44
use English;
Mac Newbold's avatar
Mac Newbold committed
45 46
use POSIX;			# for strftime, and sigprocmask and friends
use Fcntl;			# file constants for pidfile
Mac Newbold's avatar
Mac Newbold committed
47 48 49 50 51
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
52

Mac Newbold's avatar
Mac Newbold committed
53
# Do lots of db retries before we fail and die
54
$libdb::DBQUERY_MAXTRIES = 100;
Mac Newbold's avatar
Mac Newbold committed
55

56 57 58 59 60
# 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 = ();

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

67 68 69 70
# Command line opts.
my $dbtag = "";
my $debug = 0;
my $nolog = 0;
71 72
my $server = "localhost";
my $port   = @BOSSEVENTPORT@;
73 74 75
my $lockfile;
my $pidfile;

Robert Ricci's avatar
Robert Ricci committed
76 77 78
# Process command-line arguments

sub usage {
Mac Newbold's avatar
Mac Newbold committed
79
    print << "END";
80
Usage: $0 [-h] [-d] [-s server] [-p port] [-t dbtag]
81
-h              This message
82 83 84
-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
85 86
-s server       Use specified server, instead of this site's bossnode
-p port	        Use specified port
87
Send SIGHUP to reload database state, or SIGUSR1 to restart completely.
Robert Ricci's avatar
Robert Ricci committed
88
END
89
    exit(1);
Robert Ricci's avatar
Robert Ricci committed
90 91
}

Mac Newbold's avatar
Mac Newbold committed
92
# Only root should run this - it won't work when run as a user...
93
# (Or, let an admin run it if it isn't the real one in /usr/testbed/ )
94
if ($UID && ( $TB eq $REALTB || ! TBAdmin($UID) ) ) {
Mac Newbold's avatar
Mac Newbold committed
95 96 97
    die("Only root can run this script!\n");
}

98
my @args = @ARGV;    # save a copy for restart before we mess with them.
Robert Ricci's avatar
Robert Ricci committed
99
my %opt = ();
100
if (!getopts("ds:p:ht:l",\%opt)) { usage(); }
Robert Ricci's avatar
Robert Ricci committed
101

Mac Newbold's avatar
Mac Newbold committed
102 103 104 105 106 107
if ($opt{h}) {
    exit &usage;
}
if (@ARGV) {
    exit &usage;
}
Robert Ricci's avatar
Robert Ricci committed
108

Mac Newbold's avatar
Mac Newbold committed
109 110 111 112 113 114
if ($opt{s}) {
    $server = $opt{s};
}
if ($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
my $TBRESET      = TBDB_TBCONTROL_RESET;
my $TBTIMEOUT    = TBDB_TBCONTROL_TIMEOUT;
Mac Newbold's avatar
Mac Newbold committed
131 132 133
my $PXEBOOT      = TBDB_TBCONTROL_PXEBOOT;
my $BOOTING      = TBDB_TBCONTROL_BOOTING;
my $CHECKGENISUP = TBDB_TBCONTROL_CHECKGENISUP;
134 135 136
my $TBNOTIMEOUT  = TBDB_NO_STATE_TIMEOUT;
my $TBNODESTATE  = TBDB_TBEVENT_NODESTATE;
my $TBNODEOPMODE = TBDB_TBEVENT_NODEOPMODE;
137 138 139 140 141 142
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;
143
my $TBISUP       = TBDB_NODESTATE_ISUP;
144 145 146
my $PXEWAIT      = TBDB_NODESTATE_PXEWAIT;
my $PXEWAKEUP    = TBDB_NODESTATE_PXEWAKEUP;
my $PXEBOOTING   = TBDB_NODESTATE_PXEBOOTING;
147 148
my $TBRELOADDONEV1    = TBDB_TBCONTROL_RELOADDONE;
my $TBRELOADDONEV2    = TBDB_TBCONTROL_RELOADDONE_V2;
149 150 151 152
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;
153

154 155
# Special PXEBOOT state machine that all local nodes use.
my $PXEKERNEL	 = "PXEKERNEL";
156

157 158 159
if (!$debug) {
    if ( $TB eq $REALTB ) {
	$pidfile = "/var/run/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
160
    } else {
161
	$pidfile = "$TB/locks/stated.pid";
Mac Newbold's avatar
Mac Newbold committed
162
    }
163
    debug("Using pidfile $pidfile\n");
Mac Newbold's avatar
Mac Newbold committed
164

165 166 167 168 169 170 171 172 173 174 175 176 177 178 179
    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
180 181 182 183
    # We use syslog, so redirect the output to nothing
    if (TBBackGround("/dev/null")) {
	exit(0);
    }
Robert Ricci's avatar
Robert Ricci committed
184
}
Mac Newbold's avatar
Mac Newbold committed
185
# set up syslog
186 187 188 189 190 191 192 193 194 195 196 197
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
198

199 200 201
# Change my $0 so that it is easier to see in a ps/top
$0 = "$0";

Robert Ricci's avatar
Robert Ricci committed
202
my $URL = "elvin://$server";
Mac Newbold's avatar
Mac Newbold committed
203 204 205
if ($port) {
    $URL .= ":$port";
}
Robert Ricci's avatar
Robert Ricci committed
206

Mac Newbold's avatar
Mac Newbold committed
207
# Connect to the event system, and subscribe the the events we want
Robert Ricci's avatar
Robert Ricci committed
208
my $handle = event_register($URL,0);
Mac Newbold's avatar
Mac Newbold committed
209 210 211
if (!$handle) {
    fatal("Unable to register with event system\n");
}
Robert Ricci's avatar
Robert Ricci committed
212 213

my $tuple = address_tuple_alloc();
Mac Newbold's avatar
Mac Newbold committed
214 215 216
if (!$tuple) {
    fatal("Could not allocate an address tuple\n");
}
Robert Ricci's avatar
Robert Ricci committed
217

218 219 220
%$tuple = ( objtype => join(",",
			    $TBNODESTATE, $TBNODEOPMODE,
			    $TBCONTROL, $TBCOMMAND) );
221

Robert Ricci's avatar
Robert Ricci committed
222
if (!event_subscribe($handle,\&handleEvent,$tuple)) {
Mac Newbold's avatar
Mac Newbold committed
223
    fatal("Could not subscribe to events\n");
Robert Ricci's avatar
Robert Ricci committed
224 225 226 227
}

# Read in the pre-existing node states, and timeout and valid transition
# information from the database
228 229 230 231
my %timeouts  = getTimeouts();
my %valid     = getValid();
my %modeTrans = getModeTrans();
my %triggers  = getTriggers();
232
my %nodes     = readStates();
233
my %timeouttag= ();
234
if ($debug) { qshow(); }
Robert Ricci's avatar
Robert Ricci committed
235 236 237

# Gets set if a reload of state from the database should happen.
my $do_reload = 0;
238 239
my $sigrestart= 0;
my $sigcleanup= 0;
Robert Ricci's avatar
Robert Ricci committed
240 241 242 243

# 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.
244 245
$SIG{HUP}  = sub { info("SIGHUP - Reloading DB state\n"); $do_reload = 1; };

Mac Newbold's avatar
Mac Newbold committed
246
# Set up other signals.
247 248 249 250 251 252 253
$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
254

255 256 257
# Track if I handled an event or not
my $event_count = 0;

258 259
# Control how long I block while waiting for events
my $blockwait=0;
260
my $nextdeadline=0;
261 262
my $mailqueue=0;

263 264 265 266 267
notify("Stated starting up\n");

sub process_event_queue() {
    $event_count=0;
    my $lastcount=-1;
268 269 270 271 272 273 274 275 276 277 278 279 280
    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;
281 282
	    } else {
		$wait = 0;
283 284 285 286 287
	    }
	}
    } else {
	# mail is waiting. Only block until it is time to send it.
	$wait = $lastmail + $mailgap - $now;
Mac Newbold's avatar
Mac Newbold committed
288
	debug("Now $now, mailgap $mailgap, last $lastmail ==> wait $wait\n");
289 290 291
    }
    if ($wait < 0) { debug("Wait was $wait!\n"); $wait=0; }
    my $finish = $now + $wait;
Mac Newbold's avatar
Mac Newbold committed
292 293
    while (($event_count != $lastcount || $wait > 0) &&
	   !($sigrestart || $sigcleanup || $do_reload)) {
294
	$lastcount = $event_count;
Mac Newbold's avatar
Mac Newbold committed
295 296
	# Don't block if we got a signal!
	if ($wait<=0 || $sigrestart || $sigcleanup || $do_reload) {
297 298
	    event_poll($handle);
	} else {
299
	    #debug("Using blocking event poll - $wait seconds\n");
300 301 302 303 304
	    # timeout param is in milliseconds, so multiply
	    event_poll_blocking($handle, $wait*1000);
	    $now = time();
	    # subtract seconds elapsed from my wait time
	    $wait = $finish - $now;
305
	    #debug("Finished blocking event poll - $wait seconds remain\n");
306
	    if ($event_count > 0 &&
Mac Newbold's avatar
Mac Newbold committed
307 308
		(qsize() > 0 || $mailqueue ||
		 $sigrestart || $sigcleanup || $do_reload)) {
309 310 311 312 313 314 315
		$blockwait=0;
		$wait=0;
		#debug("Cancelling wait - timeouts/msgs waiting, or HUP'd\n");
		#debug("---End Blocking Wait ---\n");
	    }
	}
	#debug("Wait is $wait\n");
316 317
    }
    if ($event_count > 0) {
Mac Newbold's avatar
Mac Newbold committed
318
	debug("Handled $event_count event(s).\n");
319 320
    }
}
Robert Ricci's avatar
Robert Ricci committed
321

322
# Now, we just poll for events, and watch for timeouts
Robert Ricci's avatar
Robert Ricci committed
323
while (1) {
Mac Newbold's avatar
Mac Newbold committed
324
    my $now = time();
325 326 327 328
    my ($deadline,$node);

    # Check for nodes that have passed their timeout
    if (!qhead($deadline,$node)) {
329
	info("HEAD: $node in ".($deadline-$now).", queue=".qsize()."\n");
330 331
	while ($now >= $deadline && $node ne "") {
	    qpop($deadline,$node);
332
	    info("POP: $node in ".($deadline-$now).", queue=".qsize()."\n");
333
	    handleCtrlEvent($node,$TBTIMEOUT);
334 335 336 337
	    if (0) { qshow(); }
	    if (qhead($deadline,$node)) {
		$deadline=0; $node="";
	    }
338
	}
339 340
    } else {
	$deadline=0;
341
    }
342
    $nextdeadline = $deadline;
343

344 345 346 347
    if (qsize()==0) {
	$blockwait=1;
	debug("---Blocking wait okay---\n");
    }
Mac Newbold's avatar
Mac Newbold committed
348

Mac Newbold's avatar
Mac Newbold committed
349 350 351 352
    if ($do_reload || ($now - $last_reload > $reload_time)) {
	reload();
	$do_reload = 0;
    }
Mac Newbold's avatar
Mac Newbold committed
353

Mac Newbold's avatar
Mac Newbold committed
354 355
    # Send any messages in the queue if it is time
    notify("",1);
Mac Newbold's avatar
Mac Newbold committed
356

357 358 359
    if ($sigrestart) { restart(); }
    if ($sigcleanup) { cleanup(); }

360
    process_event_queue;
Robert Ricci's avatar
Robert Ricci committed
361 362
}

Mac Newbold's avatar
Mac Newbold committed
363 364
exit(0);

Robert Ricci's avatar
Robert Ricci committed
365
# Read the current states of nodes from the database
366
sub readStates(;@) {
Mac Newbold's avatar
Mac Newbold committed
367 368 369 370 371 372
    my %oldnodes = @_;

    # Guard against undefined variable warnings
    if (! defined(%oldnodes)) {
	%oldnodes = ();
    }
373

Mac Newbold's avatar
Mac Newbold committed
374 375 376
    #debug("readStates called\n");
    my $result = DBQueryFatal("SELECT node_id, eventstate, " .
			      "state_timestamp, op_mode, " .
377
			      "op_mode_timestamp, stated_tag FROM nodes ".
378
			      "where node_id not like 'sh%'");
Mac Newbold's avatar
Mac Newbold committed
379 380

    my %nodes;
381
    while (my ($node_id, $state, $timestamp, $mode, $mode_timestamp, $tag)
Mac Newbold's avatar
Mac Newbold committed
382
	   = $result->fetchrow()) {
383 384 385 386 387 388 389 390 391 392 393
	$nodes{$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);
	}
Mac Newbold's avatar
Mac Newbold committed
394

395
	#
Mac Newbold's avatar
Mac Newbold committed
396 397 398 399
	# 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.)
400
	#
Mac Newbold's avatar
Mac Newbold committed
401 402 403 404 405
	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};
406
	} else {
Mac Newbold's avatar
Mac Newbold committed
407 408 409 410
	    $nodes{$node_id}{state}          = $state;
	    $nodes{$node_id}{timestamp}      = $timestamp;
	    $nodes{$node_id}{mode}           = $mode;
	    $nodes{$node_id}{mode_timestamp} = $mode_timestamp;
411 412 413
	    $nodes{$node_id}{notified}       = 0;
	    $nodes{$node_id}{timedout}       = 0;
	    $nodes{$node_id}{noretry}        = 0;
414
	    # Is there a timeout? If so, set it up!
415 416 417
	    if ($dbtag eq $nodes{$node_id}{"tag"}) {
		setTimeout($mode,$state,$node_id,$timestamp);
	    }
418
	}
Mac Newbold's avatar
Mac Newbold committed
419 420
    }
    return %nodes;
Robert Ricci's avatar
Robert Ricci committed
421 422 423 424 425 426
}

#
# Read timeouts for various states from the database
#
sub getTimeouts() {
Mac Newbold's avatar
Mac Newbold committed
427 428 429
    #debug("getTimeouts called\n");
    my $result = DBQueryFatal("SELECT op_mode, state, timeout, action " .
			      "FROM state_timeouts");
Robert Ricci's avatar
Robert Ricci committed
430

Mac Newbold's avatar
Mac Newbold committed
431 432 433 434 435
    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
436 437 438 439 440 441
}

#
# Read the list of valid state transitions from the database
#
sub getValid() {
Mac Newbold's avatar
Mac Newbold committed
442 443 444
    #debug("getValid called\n");
    my $result = DBQueryFatal("SELECT op_mode, state1, state2 " .
			      "FROM state_transitions");
Robert Ricci's avatar
Robert Ricci committed
445

Mac Newbold's avatar
Mac Newbold committed
446 447 448 449 450
    my %valid;
    while (my ($mode,$state1, $state2) = $result->fetchrow()) {
	$valid{$mode}{$state1}{$state2} = 1;
    }
    return %valid;
Robert Ricci's avatar
Robert Ricci committed
451 452
}

453 454 455 456
#
# Read the list of valid mode transitions from the database
#
sub getModeTrans() {
Mac Newbold's avatar
Mac Newbold committed
457
    #debug("getModeTrans called\n");
Mac Newbold's avatar
Mac Newbold committed
458
    my $result =
Mac Newbold's avatar
Mac Newbold committed
459 460 461 462 463 464 465 466 467 468 469
      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;
470
	}
Mac Newbold's avatar
Mac Newbold committed
471 472
    }
    return %modeTrans;
473 474 475 476 477 478
}

#
# Read the list of states which trigger an action
#
sub getTriggers() {
479
    debug("getTriggers called\n");
Mac Newbold's avatar
Mac Newbold committed
480

481 482
    debug("anymode ==> '$TBANYMODE'\n");

Mac Newbold's avatar
Mac Newbold committed
483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503
    # A note about triggers:
    #
    # "per-node" triggers only affect their specific node in a
    # particular mode/state, and are run first of all. "global"
    # triggers are triggers for a given mode/state that affect all
    # nodes, and are run after any per-node triggers. "Any-mode"
    # triggers are tied to a state, and occur in that state in any
    # mode. The any-mode triggers are over-ridden by global triggers,
    # and if an "Any-mode" trigger for state XYZ exists as well as a
    # global trigger for mode FOOBAR state XYZ, then when I arrive in
    # XYZ any per-node triggers will be run. Then, if I'm in mode
    # FOOBAR, only the global trigger will run. If I'm in any other
    # mode, only the any-mode trigger will run.

    # (our "*" is stored as $TBANYMODE)
    # Per-node triggers have a specific node_id
    # Global triggers have "*" as the node_id
    # Any-mode triggers have "*" as the mode, and can be global or per-node

    # Grab global triggers (including any-mode)
    my $result =
Mac Newbold's avatar
Mac Newbold committed
504
      DBQueryFatal("SELECT op_mode, state, trigger " .
505 506
		   "FROM state_triggers where node_id='$TBANYMODE' ".
		   "order by op_mode,state");
Mac Newbold's avatar
Mac Newbold committed
507
    my %t;
508
    while (my ($mode, $state, $trig) = $result->fetchrow()) {
Mac Newbold's avatar
Mac Newbold committed
509 510 511
	my @trigs = split(/\s*,\s*/,$trig);
	$t{"$mode:$state"} = \@trigs;
	debug("trig($mode:$state)\t => ".join(',',@trigs)."\n");
512 513
    }

Mac Newbold's avatar
Mac Newbold committed
514 515
    # Grab per-node triggers (including any-mode)
    $result =
516 517 518 519 520 521 522
      DBQueryFatal("SELECT node_id, op_mode, state, trigger " .
		   "FROM state_triggers where node_id!='$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
523
    }
524

Mac Newbold's avatar
Mac Newbold committed
525 526
    debug(hash_recurse(%t));

Mac Newbold's avatar
Mac Newbold committed
527
    return %t;
528 529
}

Robert Ricci's avatar
Robert Ricci committed
530 531 532 533
#
# Gets called for every event that we recieve
#
sub handleEvent($$$) {
Mac Newbold's avatar
Mac Newbold committed
534 535 536 537 538 539
    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++;
Mac Newbold's avatar
Mac Newbold committed
540

541
    #
Mac Newbold's avatar
Mac Newbold committed
542 543 544
    # For readability, only do this on the main stated. This will print all
    # events, which gets cumbersome with debugging versions, so we'll print
    # only applicable events in debug versions, after we decide they apply.
545
    #
546
    if ($dbtag eq "") {
547 548
	debug("Got an event: ($objtype,$objname,$eventtype)\n");
    }
Mac Newbold's avatar
Mac Newbold committed
549 550 551 552

    #
    # Check to see if another instance is supposed to be handling this node
    #
553 554
    if ($objtype ne $TBCOMMAND) {
	my $node = $objname;
Mac Newbold's avatar
Mac Newbold committed
555

556 557 558 559 560 561 562
	#
	# If we have never seen this node, reload.
	#
	if (! defined($nodes{$node})) {
	    reload();

	    # Still not defined, someone screwed up! This could end up
Mac Newbold's avatar
Mac Newbold committed
563
	    # churning via reload(). Bad.
564
	    if (! defined($nodes{$node})) {
565 566 567 568
		notify("Got $objtype/$eventtype for nonexistent $node!\n");
		return;
	    }
	}
Mac Newbold's avatar
Mac Newbold committed
569

570 571
	#
	# If a stated_tag was specified on the command line, ignore those
Mac Newbold's avatar
Mac Newbold committed
572
	# nodes that do not match.
573
	#
Mac Newbold's avatar
Mac Newbold committed
574
	#debug("dbtag='$dbtag', node $node='".$nodes{$node}{"tag"}."'\n");
575 576 577 578 579 580 581 582 583 584
	if ($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
585
    }
Mac Newbold's avatar
Mac Newbold committed
586 587 588 589 590

    #
    # If this is a debugging version, then this event is for one of my
    # nodes, so I can print out the event now. (Main version prints earlier.)
    #
591 592 593
    if ($dbtag ne "") {
	debug("Got an event: ($objtype,$objname,$eventtype)\n");
    }
Mac Newbold's avatar
Mac Newbold committed
594 595 596

 SWITCH: for ($objtype) {

Mac Newbold's avatar
Mac Newbold committed
597 598 599 600 601 602 603 604 605 606 607 608 609 610
	(/$TBNODESTATE/) && do {
	    stateTransition($objname,$eventtype);
	    last;
	};
	(/$TBNODEOPMODE/) && do {
	    opModeTransition($objname,$eventtype);
	    notify("Use of deprecated event TBNODEOPMODE:\n".
		   "$objname->$eventtype\n");
	    last;
	};
	(/$TBCONTROL/) && do {
	    handleCtrlEvent($objname,$eventtype);
	    last;
	};
611 612 613 614
	(/$TBCOMMAND/) && do {
	    handleCommand($objname,$eventtype);
	    last;
	};
615

Mac Newbold's avatar
Mac Newbold committed
616
    }
617 618 619 620 621

}

sub stateTransition($$) {

622
    my ($node,$newstate) = @_;
Robert Ricci's avatar
Robert Ricci committed
623

624 625
    # Check for invalid transitions
    my ($oldstate, $mode);
626 627 628
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

629 630 631
    if ($oldstate && $mode && $valid{$mode} && $valid{$mode}{$oldstate} &&
	!$valid{$mode}{$oldstate}{$newstate}) {
	notify("Invalid transition for node $node from $mode/$oldstate " .
632
	       "to $newstate\n");
633
    }
Robert Ricci's avatar
Robert Ricci committed
634

635 636 637 638
    my $now = time();
    $nodes{$node}{state}     = $newstate;
    $nodes{$node}{timestamp} = $now;
    $nodes{$node}{notified}  = 0;
639

640 641 642
    info("$node: $mode/$oldstate => $mode/$newstate\n");
    DBQueryFatal("UPDATE nodes SET eventstate='$newstate', " .
		 "state_timestamp='$now' WHERE node_id='$node'");
643

644 645 646 647
    # 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} =~ /^$TBCOMMAND:/) {
648
        debug("TimeoutTag = '$timeout_tag{$node}'\n");
649
	my ($str,$cmd) = split(":",$timeout_tag{$node});
650
	debug("str=$str\tcmd=$cmd\tTBREBOOT=$TBREBOOT\tstate=$newstate\n");
651
	if ($cmd eq $TBREBOOT) {
652
	    if ($newstate eq TBDB_NODESTATE_SHUTDOWN ) {
653 654 655
		info("$node: $TBREBOOT success\n");
		# Timeout will get cleared below by setTimeout call
	    } else {
656 657
		notify("$node: $TBREBOOT in progress, but got state ".
		       "$newstate instead of ".TBDB_NODESTATE_SHUTDOWN."!\n");
658 659 660 661 662
	    }
	#} elsif ($cmd eq $FOO ) {
	    # Add more here...
	} else {
	    notify("$node: Unknown command timeout '$timeout_tag{$node}' ".
663
		   "found at $mode/$newstate\n");
664 665 666
	}
    }

667 668 669 670 671
    #
    # 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.
    #
672 673
    setTimeout($mode,$newstate,$node,$now);

Mac Newbold's avatar
Mac Newbold committed
674 675 676 677 678 679 680
    # Check if this state has any triggers
    my @nodetrigs = GetNodeTriggerList($node,$mode,$newstate,1);
    my @trigs = GetNodeTriggerList($node,$mode,$newstate);
    if (@trigs > 0) {
	debug("Running triggers: ".join("/",@trigs)."\n");
	foreach ( @trigs) {
	    my $trig = $_;
681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697
	    /^SCRIPT:([-\w\/]+)$/ && do {
		#
		# Run a script. No arguments at the moment.
		#
		my $script = $1;

		#
		# If the path is absolute, run it. Otherwise it has to
		# come from the sbin directory.
		#
		if (! ($script =~ /^\//)) {
		    $script = "$TB/sbin/$script";
		}
		info("$node: Running $script.\n");
		system("$script &");
		next;
	    };
Mac Newbold's avatar
Mac Newbold committed
698
	    /^$PXEBOOT$/ && do {
699
		#
Mac Newbold's avatar
Mac Newbold committed
700 701 702 703 704 705 706 707 708 709 710 711
		# 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.
712
		#
Mac Newbold's avatar
Mac Newbold committed
713 714 715 716
		# 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.
717
		#
Mac Newbold's avatar
Mac Newbold committed
718 719 720 721 722
		debug("Running $PXEBOOT trigger\n");
		if ($mode ne $PXEKERNEL) {
		    info("$node: Forcing mode transition into $PXEKERNEL!\n");
		    opModeTransition($node, $PXEKERNEL, 1);
		    $mode=$PXEKERNEL;
723
		}
Mac Newbold's avatar
Mac Newbold committed
724 725 726 727 728 729 730
		next;
	    };
	    /^$BOOTING$/ && do {
		#
		# See if we are in the right mode/osid.
		#
		my ($bootosid,$bootopmode) = TBBootWhat($node, $debug);
731

Mac Newbold's avatar
Mac Newbold committed
732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797
		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 $query_result =
			  DBQueryWarn("select next_op_mode from nodes ".
				      "where node_id='$node'");
			my ($nextmode) = $query_result->fetchrow();
			if ($nextmode) {
			    info("$node: Forcing mode transition!\n");
			    opModeTransition($node, $nextmode, 1);
			    $mode=$nextmode;
			}
		    }
		    else {
			my $str = "$node is running $bootosid, but in ".
			  "mode $mode\ninstead 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);
		    }
		}
		next;
	    };
	    /^$CHECKGENISUP$/ && do {
		checkGenISUP($node);
		next;
	    };
798
	    /^$TBRESET$/ && do {
799 800 801
		# We successfully booted, so clear some flags
		$nodes{$node}{noretry}   = 0;
		$nodes{$node}{timedout}  = 0;
802 803 804 805 806 807 808 809 810
		# 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,$trig);
		}
		next;
	    };
811
	    (/^$TBRELOADDONEV1$/ || /^$TBRELOADDONEV2$/) && do {
812 813 814
		handleCtrlEvent($node,$trig);
		next;
	    };
815 816 817 818 819 820 821 822
	    /^$TBISUP$/ && do {
		info("$node: Triggered $TBISUP\n");
		EventSendWarn(host      => $BOSSNODE ,
			      objtype   => TBDB_TBEVENT_NODESTATE ,
			      eventtype => TBDB_NODESTATE_ISUP ,
			      objname   => $node);
		next;
	    };
823 824 825 826
	    (/^$TBPOWERCYCLE$/) && do {
		handleCommand($node,$TBPOWERCYCLE);
		next;
	    };
827
	    notify("Unknown trigger '$trig' for $node in $mode/$newstate!\n");
828
	}
Mac Newbold's avatar
Mac Newbold committed
829 830 831 832 833 834 835
	# Clear any of the node triggers that we ran.
	# (Don't clear all of them, because some of the triggers we ran
	# may have caused others to be set, and we don't want to nuke them.)
	if (@nodetrigs > 0) {
	    debug("Clearing node triggers: ".join("/",@nodetrigs)."\n");
	    ClearNodeTrigger($node,$mode,$newstate,@nodetrigs);
	}
836
    }
837

838 839 840 841 842 843 844 845
    # 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
846 847 848
	} else {
	    debug("No next mode.\n");
	}
849 850
    }
}
851

852
sub opModeTransition($$;$) {
Mac Newbold's avatar
Mac Newbold committed
853

854 855
    my ($node,$newmode,$force) = @_;
    if (!defined($force)) { $force = 0; }
Mac Newbold's avatar
Mac Newbold committed
856

857
    info("$node: Mode change to $newmode requested ($force)\n");
Mac Newbold's avatar
Mac Newbold committed
858

859 860
    # Check for invalid transitions
    my ($oldstate, $mode, $nextstate);
861 862 863
    $oldstate = $nodes{$node}{state};
    $mode     = $nodes{$node}{mode};

864
    if (defined($modeTrans{"$mode:$oldstate"}) || $force) {
865
	if (!$force) {
866 867 868 869 870 871 872 873 874 875 876 877 878 879 880
	    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");
	    }
881 882
	}
    } else {
883
	notify("Invalid mode transition for $node from $mode/$oldstate: ".
884 885
	       "Not a valid mode transition state!\n");
    }
Mac Newbold's avatar
Mac Newbold committed
886 887 888
    if (!$nextstate) {
	$nextstate=$oldstate;
    }
Mac Newbold's avatar
Mac Newbold committed
889

890 891 892 893 894 895
    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
896

897 898 899 900 901
    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
902 903 904 905

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

906 907 908 909
}

sub handleCtrlEvent($$) {
    my ($node,$event) = @_;
910

911
    info("CtrlEvent: $node, $event\n");
912

913 914
    foreach ($event) {
	/^$TBRESET$/ && do {
915 916
	    #
	    # Clear next_boot_path with os_select.
Mac Newbold's avatar
Mac Newbold committed
917
	    #
918
	    $cmd = "$osselect -d -c -1 $node";
919
	    system($cmd) and
920
		notify("$node/$event: Could not clear next_boot_path!\n");
Mac Newbold's avatar
Mac Newbold committed
921 922

	    info("Performed $TBRESET for $node\n");
923 924
	    next;
	};
925
	(/^$TBRELOADDONEV1$/ || /^$TBRELOADDONEV2$/) && do {
926 927 928 929 930 931 932
	    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'");
933
		DBQueryFatal("delete from reserved where node_id='$node'");
934 935
		TBSetNodeHistory($node, TB_NODEHISTORY_OP_FREE,
				 $UID, $pid, $eid);
936
		info("Released $node from $pid/$eid\n");
937
	    }
938 939 940 941 942
	    if ($event eq $TBRELOADDONEV2) {
		info("Sending an apod to $node\n");
		system("$apod $node") == 0 or
		    notify("Could not apod $node after $TBRELOADDONEV2!\n");
	    }
943 944 945
	    next;
	};
	/^$TBTIMEOUT$/ && do {
946 947 948 949
	    my ($mode,$state) = split(":",$timeout_tag{$node});
	    delete($timeout_tag{$node});
	    my $curstate = $nodes{$node}{state};
	    my $curmode = $nodes{$node}{mode};
950
	    my ($timeout,$action);
951 952 953 954 955
	    if (!defined($nodes{$node}{notified})) {
		$nodes{$node}{notified}=0;
	    }
	    $nodes{$node}{notified}++;
	    my $notified = $nodes{$node}{notified};
956 957
	    $nodes{$node}{timedout}++;
	    my $timedout = $nodes{$node}{timedout};
958 959 960 961
	    if ($mode && $state && $timeouts{$mode} &&
		$timeouts{$mode}{$state}) {
		($timeout, $action) = @{$timeouts{$mode}{$state}};
	    }
962 963
	    if ($mode eq $TBCOMMAND) {
		# It is a command, not a true state
964
		if ($action eq $TBTIMEOUTCMDRETRY) {
965
		    # Retry the command
966
		    notify("$node: Command $state, retry #$timedout\n");
967
		    # notify in case we get in a retry loop...
968
		    handleCommand($node,$state,$timedout,1);
969 970 971 972 973 974
		} else {
		    notify("$node: Unknown timeout action for ".
			   "$mode/$state: '$action'\n");
		}
		next;
	    }
975 976 977 978 979 980 981

	    #
	    # 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 " : "");
Mac Newbold's avatar
Mac Newbold committed
982

983 984
		if ($timedout < 3) {
		    #
Mac Newbold's avatar
Mac Newbold committed
985 986
		    # Try again.
		    #
987 988 989 990 991 992
		    info("Node $node has timed out $timedout times in ".
			 "$PXEWAKEUP!\n".
			 "Sending it a another wakeup command\n");
		}
		else {
		    #
Mac Newbold's avatar
Mac Newbold committed
993 994
		    # Failed too many times, power cycle instead.
		    #
995 996 997 998 999 1000 1001 1002 1003
		    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 >> $rebootlog &") and
1004
		    notify("$PXEWAKEUP retry: ".
1005 1006 1007 1008 1009
			   "Command '$cmd' failed, error $?: $!\n");

		next;
	    }

1010 1011 1012 1013 1014 1015
	    info("Node $node has timed out in state $mode/$state".
		 ($action ne "" ? "\n\tRequested action $action." : "").
		 "\n");

	    foreach ($action) {
		/^$TBTIMEOUTREBOOT/ && do {
1016 1017 1018 1019 1020
		    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;
1021
		    } else {
1022 1023 1024 1025 1026 1027
			# 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);
1028 1029 1030 1031 1032 1033 1034
		    }
		    last; };
		/^$TBTIMEOUTNOTIFY/ && do {
		    notify("Node $node has timed out in state $mode/$state\n");
		    last; };
		notify("$node: Unknown Timeout Action: $action\n");
	    }
1035 1036
	    next;
	};
1037
	notify("$node: Unknown CtrlEvent: $event\n");
1038 1039
    }
}
Robert Ricci's avatar
Robert Ricci committed
1040

1041 1042
sub handleCommand($$;$$) {
    my ($params,$command,$retry,$force) = @_;
1043
    if (!defined($retry)) { $retry=0; }
1044
    if (!defined($force)) { $force=0; }
1045 1046 1047 1048 1049 1050 1051 1052

    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.

1053
    if ($command eq $TBREBOOT && $retry >=4) {
1054 1055
	announce("Node $params has tried rebooting $retry times and has \n".
		 "still not been successful. Please look into it soon.\n".
1056 1057 1058
	"" );#	 "In the meantime, $params will be powered off.\n");
	# Just return...
	return 0;
1059
	# change my command to poweroff.
1060
	#$command = $TBPOWEROFF;
1061 1062 1063 1064 1065 1066
    }

    foreach ($command) {
	/^$TBREBOOT$/ && do {
	    # For reboot, the params is a comma-separated list of nodes
	    my @nodes = split(",",$params);
1067 1068 1069 1070 1071 1072 1073 1074

	    # 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");
1075 1076
		    if (($nodes{$node}{state} ne TBDB_NODESTATE_ISUP) &&
			(!$nodes{$node}{noretry}) ) {
1077
			# This node shouldn't be rebooted now...
1078 1079
			# XXX Send feedback here somehow!

1080 1081 1082 1083 1084 1085 1086 1087 1088 1089
			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; }
	    }

1090 1091 1092 1093 1094
	    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.

1095
	    my $cmd = "$nodereboot -r $nodelist";
1096
	    my $redir = " 2>&1 >> $rebootlog &";
1097 1098 1099
	    debug("$cmd $redir\n");
	    system("date $redir");
	    system($cmd.$redir) and
1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121
	      notify("$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};
1122
	    info("Sending power $func for nodes: $nodelist\n");
1123 1124 1125 1126
	    # Permissions were checked in order to send the message,
	    # so we don't need to do any fancy stuff here.

	    my $cmd = "$power $func $nodelist &";
1127 1128
	    debug("$cmd\n") and
	    system($cmd) and
1129 1130 1131 1132 1133 1134 1135 1136 1137 1138
	      notify("$params/$command: ".
		     "Command '$cmd' failed, error $?: $!\n");

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

1139 1140 1141 1142
#
# Check if we need to generate an ISUP
#
sub checkGenISUP($) {
Mac Newbold's avatar
Mac Newbold committed
1143 1144 1145 1146 1147 1148 1149 1150 1151 1152
    my ($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
1153

Mac Newbold's avatar
Mac Newbold committed
1154 1155 1156 1157 1158 1159
    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
1160

Mac Newbold's avatar
Mac Newbold committed
1161 1162
    # If os will send ISUP on its own, do nothing here.
    if ($can{"isup"}) {
Mac Newbold's avatar
Mac Newbold committed
1163
	debug("$node: Will send own ISUP\n");
Mac Newbold's avatar
Mac Newbold committed
1164 1165
	return 0;
    }
Mac Newbold's avatar
Mac Newbold committed
1166

Mac Newbold's avatar
Mac Newbold committed
1167 1168 1169 1170 1171 1172
    # 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;
1173
    }
Mac Newbold's avatar
Mac Newbold committed
1174

Mac Newbold's avatar
Mac Newbold committed
1175
    # If os doesn't support ping or isup, stated sends ISUP just after
Mac Newbold's avatar
Mac Newbold committed
1176 1177 1178 1179 1180 1181 1182
    # 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
1183 1184
}

1185 1186 1187 1188
# 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
1189
    my ($node) = @_;
1190

Mac Newbold's avatar
Mac Newbold committed
1191
    # XXX: I don't want to do this every time, for performance reaons,
Mac Newbold's avatar
Mac Newbold committed
1192
    # but we need to make sure that we don't get into an inconsistent
Mac Newbold's avatar
Mac Newbold committed
1193 1194 1195 1196 1197 1198
    # 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'");
1199