vnode_setup.in 14.9 KB
Newer Older
1
#!/usr/bin/perl -wT
Leigh B. Stoller's avatar
Leigh B. Stoller committed
2 3 4

#
# EMULAB-COPYRIGHT
5
# Copyright (c) 2000-2004, 2007 University of Utah and the Flux Group.
Leigh B. Stoller's avatar
Leigh B. Stoller committed
6 7
# All rights reserved.
#
8 9 10 11 12 13 14 15 16 17 18 19 20
use English;
use Getopt::Std;

#
# Set up the vnode state on a virtual (multiplexed) node.
#
# XXX - This script should only be run from os_setup!
#
# The output is all jumbled together since the updates are issued in parallel.
# Might be a pain when debugging. 
# 
sub usage()
{
21
    print STDOUT "Usage: vnode_setup [-f] [-k] [-j] [-p] [-n <numbatch>] [-w <wait_time>] <pid> <eid> [node ...]\n";
22 23
    exit(-1);
}
24
my  $optlist = "fdkjpn:w:";
25 26 27 28 29 30 31 32 33 34

#
# We don't want to run this script unless its the real version.
# That is, it must be setuid root. 
#
if ($EUID != 0) {
    die("*** $0:\n".
	"    Must be root! Maybe its a development version?\n");
}

35 36 37 38 39 40 41
#
# Configure variables
#
my $TB		= "@prefix@";
my $TESTMODE    = @TESTMODE@;
my $TBOPS       = "@TBOPSEMAIL@";
my $TBLOGS      = "@TBLOGSEMAIL@";
42
my $CLIENT_BIN  = "@CLIENT_BINDIR@";
43 44 45 46 47 48

my $ssh		= "$TB/bin/sshtb -n";
my $debug       = 0;
my $force       = 0;
my $failed      = 0;
my $killmode    = 0;
49 50
my $jailonly    = 0;
my $plabonly    = 0;
Kirk Webb's avatar
Kirk Webb committed
51
my $numbatch    = 10;
Kirk Webb's avatar
Kirk Webb committed
52
my $childwait   = 120;
53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76
my $dbuid;

#
# Load the Testbed support stuff. 
#
use lib "@prefix@/lib";
use libdb;
use libtestbed;

# un-taint path
$ENV{'PATH'} = '/bin:/usr/bin:/usr/local/bin';
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

# Turn off line buffering on output
$| = 1; 

#
# Parse command arguments. Once we return from getopts, all that should be
# left are the required arguments.
#
%options = ();
if (! getopts($optlist, \%options)) {
    usage();
}
77
if (@ARGV < 2) {
78 79 80 81 82 83 84 85 86 87 88
    usage();
}
if (defined($options{"f"})) {
    $force = 1;
}
if (defined($options{"d"})) {
    $debug = 1;
}
if (defined($options{"k"})) {
    $killmode = 1;
}
89 90 91 92 93 94
if (defined($options{"j"})) {
    $jailonly = 1;
}
if (defined($options{"p"})) {
    $plabonly = 1;
}
Kirk Webb's avatar
Kirk Webb committed
95 96 97 98 99 100 101 102
if (defined($options{"n"})) {
    if ($options{"n"} =~ /^(\d+)$/) {
        $numbatch = $1;
    }
    else {
        die ("*** Bad data in numbatch: $options{'n'}");
    }
}
Kirk Webb's avatar
Kirk Webb committed
103 104 105 106 107 108 109 110
if (defined($options{"w"})) {
    if ($options{"w"} =~ /^(\d+)$/) {
        $childwait = $1;
    }
    else {
        die ("*** Bad data in wait_time: $options{'w'}");
    }
}
Kirk Webb's avatar
Kirk Webb committed
111

112 113
my $pid   = shift(@ARGV);
my $eid   = shift(@ARGV);
114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130

#
# Untaint the arguments.
#
if ($pid =~ /^([-\@\w]+)$/) {
    $pid = $1;
}
else {
    die("*** Bad data in pid: $pid\n");
}	
if ($eid =~ /^([-\@\w]+)$/) {
    $eid = $1;
}
else {
    die("*** Bad data in eid: $eid\n");
}

131 132 133 134
if ($plabonly && $jailonly) {
    die("*** '-j' and '-p' are mutually exclusive.");
}

135
#
136 137
# Verify permission to muck with this experiment. Note that this script
# is run as root from the plab monitor daemon.
138
#
139
if ($UID && !TBAdmin($UID) &&
140 141 142
    !TBExptAccessCheck($UID, $pid, $eid, TB_EXPT_DESTROY)) {
    die("*** $0:\n".
	"    You do not have permission to mess with $pid/$eid!\n");
143 144 145
}

#
146
# Get the list of nodes in this experiment.
147
#
148 149
my @nodes = ExpNodes($pid, $eid);
if (! @nodes) {
150 151 152
    warn("*** $0:\n".
	 "    No allocated nodes in experiment $pid/$eid!\n");
    exit(0);
153
}
154 155

# Nodes on the command line. Operate only on this set.
156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180
if (@ARGV) {
    my %fulllist = ();

    # Temporary hash list for searching.
    foreach my $node ( @nodes ) {
	$fulllist{$node} = 1;
    }
    @nodes = ();

    foreach my $node ( @ARGV ) {
	if ($node =~ /^([-\@\w]+)$/) {
	    $node = $1;

	    if (!defined($fulllist{$node})) {
		die("*** $0:\n".
		     "   Node $node is not allocated to $pid/$eid!\n");
	    }
	}
	else {
	    die("Bad node name: $node.");
	}
	push(@nodes, $node);
    }
}
my $exptstate = ExpState($pid, $eid);
181 182 183 184

# Just the vnodes mam.
foreach my $node (@nodes) {
    my $pnode;
185
    my $jailed;
186
    my $plab;
187
    my $allocstate;
188
    my $mode = ($killmode ? "teardown" : "setup");
189
    my $remote = TBIsNodeRemote($node); 
190

191
    if (! TBIsNodeVirtual($node, \$jailed, \$plab)) {
192 193
	next;
    }
194
    if (($plabonly || $jailonly) and 
195
        !(($plabonly && $plab) || ($jailonly && ($jailed || $remote)))) {
196 197
        next;
    }
198 199 200 201
    if (! TBPhysNodeID($node, \$pnode)) {
	die("*** $0:\n".
	    "    No physical node for $node!\n");
    }
202

203 204
    TBGetNodeAllocState($node, \$allocstate);

205 206 207
    #
    # On remote nodes, or when forcemode is on, always do the deed.
    # Otherwise, look at experiment state.
208 209
    #
    if (!$force) {
210 211 212 213
	if ($exptstate eq EXPTSTATE_SWAPPING) {
	    #
	    # When swapping, local vnodes go down with the physnode.
	    #
214
	    if (! $remote) {
215 216 217 218
		print "$node will $mode with local node $pnode.\n";
		next;
	    }
	    elsif ($allocstate eq TBDB_ALLOCSTATE_DOWN) {
219 220 221 222 223 224 225
		if (!$plab) {
		    print "$node failed to boot; skipping $mode.\n";
		    next;
		}
		# Plab nodes need to be cleaned up.
		print "$node failed to boot; changing to cleanup.\n";
		$mode = "cleanup";
226
	    }
227
	}
228 229
	elsif ($exptstate eq EXPTSTATE_ACTIVATING ||
	       $exptstate eq EXPTSTATE_MODIFY_RESWAP) {
230
	    #
231 232 233 234 235 236 237 238 239 240 241
	    # The allocstate determines if the vnode actually needs to
	    # be setup or torndown. Note that a failed experiment will
	    # cause a bunch of vnodes to be torndown, while in the
	    # ACTIVATING state. See os_setup and assign_wrapper; the
	    # idea is to avoid doing setup/teardown up vnodes on
	    # machines that are rebooting anyway, or that failed.
	    # Complicated by modify which could add/subtract a vnode on
	    # an existing machine, but not reboot the machine. Note that
	    # free (now unused) vnodes will land in RES_TEARDOWN. It is
	    # assumed that these booted okay, and need to be torndown,
	    # even though they are not RES_READY.
242
	    #
243
	    if (! $remote) {
244 245
		if ($killmode) {
		    if ($allocstate eq TBDB_ALLOCSTATE_DOWN) {
246 247 248 249
			print "$node failed to boot; skipping $mode.\n";
			next;
		    }
		    elsif ($allocstate eq TBDB_ALLOCSTATE_RES_INIT_CLEAN()) {
250 251 252
			print "$node never booted; skipping $mode.\n";
			next;
		    }
253 254 255 256
		    elsif ($allocstate eq TBDB_ALLOCSTATE_RES_READY()) {
			print "$node will $mode with local node $pnode.\n";
			next;
		    }
257 258
		}
		elsif ($allocstate eq TBDB_ALLOCSTATE_RES_READY()) {
259
		    print "$node is already setting up on local node $pnode\n";
260 261 262
		    next;
		}
	    }
263 264
	    else {
		if ($killmode) {
265 266 267
		    if ($allocstate eq TBDB_ALLOCSTATE_DEAD) {
			# plab only. See below.
			print "$node failed to initialize; skipping $mode.\n";
268 269
			next;
		    }
270 271 272 273 274 275 276 277 278
		    elsif ($allocstate eq TBDB_ALLOCSTATE_DOWN) {
			if (!$plab) {
			    print "$node failed to boot; skipping $mode.\n";
			    next;
			}
			# Plab nodes need to be cleaned up.
			print "$node failed to boot; changing to cleanup.\n";
			$mode = "cleanup";
		    }
279 280 281 282
		    elsif ($allocstate eq TBDB_ALLOCSTATE_RES_INIT_CLEAN()) {
			print "$node never booted; skipping $mode.\n";
			next;
		    }
283
		}
284 285
		elsif ($allocstate eq TBDB_ALLOCSTATE_RES_READY()) {
		    print "$node is already set up on remote node $pnode\n";
286 287
		    next;
		}
288
		elsif ($allocstate eq TBDB_ALLOCSTATE_RES_INIT_DIRTY()) {
289
		    print "$node needs a reboot on remote node $pnode\n";
290 291
		    $mode = "reboot";
		}
292 293 294 295 296 297 298
		elsif ($allocstate eq TBDB_ALLOCSTATE_RES_RECONFIG()) {
		    print "$node needs a reconfig on remote node $pnode\n";
		    # We do not actually reconfig virtual nodes; just
		    # reboot them. Might reconfig someday, in which case
		    # this would move up into os_setup.
		    $mode = "reboot";
		}
299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321
		elsif ($plab && $allocstate eq TBDB_ALLOCSTATE_RES_INIT_CLEAN()) {
		    # This is a special case.  If we reuse one of the plab 
		    # nodes, but use a different vname in the topo, there will
		    # be an entry in the database and a slice will be reserved
		    # because it's never torn down.  However, we need to 
		    # skip plabnode alloc and go straight to vnodesetup.
		    #
		    # BUT, we only can do this if the sliver entry is already
		    # in the database!
		    # Also note that this could eventually cause problems
		    # if there is a mistaken sliver entry in the DB.

		    $res = DBQueryFatal(
			"select slicename,node_id from plab_slice_nodes" . 
			" where pid='$pid' and eid='$eid' " . 
                        "   and node_id='$node'");
		    if ($res->numrows == 1) {
			# node exists; change mode to resetup
			$mode = "resetup";
			print "Doing a resetup on '$node'\n";
		    }
		    
		}
322 323
	    }
	}
324 325 326 327 328 329
    }

    #
    # When setting up a vnode, force its event state into SHUTDOWN since
    # no telling what its initial state is. 
    # 
Kirk Webb's avatar
Kirk Webb committed
330 331
    # XXX: Don't we always want to set this?
    #
332
    if ($mode eq "teardown" || $mode eq "reboot") {
333 334
	TBSetNodeEventState($node, TBDB_NODESTATE_SHUTDOWN);
    }
335 336

    #
337 338 339 340 341 342
    # Put this into the list of calls we have to make in the next loop
    #
    push @vnodes, [$node, $pnode, $mode, $jailed, $plab];

}

343

344 345
my $children = 0;
my %child_vnodes = ();
Kirk Webb's avatar
Kirk Webb committed
346 347
print "vnode_setup running at parallelization: $numbatch ". 
    "wait_time: $childwait\n";
348 349
while (1) {

Kirk Webb's avatar
Kirk Webb committed
350
    # Space out the invocation of child processes a little.
Kirk Webb's avatar
Kirk Webb committed
351
    sleep(1);
Kirk Webb's avatar
Kirk Webb committed
352

353 354 355 356 357 358 359
    #
    # We're done when we've hit the last vnode, and we've outlived all of our
    # children
    #
    if ((!@vnodes) && ($children == 0)) {
	last;
    }
360

361 362 363
    #
    # There are more free slots
    #
Kirk Webb's avatar
Kirk Webb committed
364
    if (($children < $numbatch) && @vnodes) {
365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388
	#
	# Look for a vnode that is not on a pnode we're already working on
	# 
	# XXX - do this!
	
	my ($vnode, $pnode, $mode, $jailed, $plab) = @{pop @vnodes};
    
	print STDOUT "Doing $mode of vnode $vnode on $pnode ...\n";

	#
	# Run an ssh command in a child process, protected by an alarm to
	# ensure that the ssh is not hung up forever if the machine is in some
	# funky state.
	# 
	my $syspid = fork();

	if ($syspid) {
	    #
	    # Just keep track of it, we'll wait for it finish down below
	    #
	    $child_vnodes{$syspid} =
		[$vnode, $pnode, $mode, $jailed, $plab, time()];
	    $children++;
	} else {
389 390
	    TBdbfork();	# So we get the event system fork too ...
	    
391
            my $exval = 0;
392 393 394
	    # Must change our real UID to root so that ssh will work.
	    $UID = 0;

395
	    if ($mode eq "setup" && ($plab || !$jailed)) {
Kirk Webb's avatar
Kirk Webb committed
396 397 398 399
                # Make sure vnode is in the proper state before trying to
                # bring it up.
                # XXX: do this for all vnodes (see above)?
                TBSetNodeEventState($vnode, TBDB_NODESTATE_SHUTDOWN);
400 401 402 403 404 405 406 407 408 409 410 411 412 413 414
		if ($plab) {
		    if (TBForkCmd("$TB/sbin/plabnode ". 
				  ($force ? "-f" : "").
				  " alloc $pid $eid $vnode", 1)) {
			print STDERR "*** $0:\n" .
			    "    Plab node allocation failed\n";
			# Should check DB state instead.
			exit(99);
		    }
		}
		else {
		    if (TBForkCmd("$ssh -host $pnode $CLIENT_BIN/vnodesetup ".
				  " -i $vnode", 1)) {
			exit(99);
		    }
415
		}
416 417
                # Make sure the system knows we now have state on the node!
                TBSetNodeAllocState($vnode, TBDB_ALLOCSTATE_RES_INIT_DIRTY);
418
	    }
419 420 421 422 423
	    elsif (!($plab && ($mode eq "cleanup" || $mode eq "teardown"))) {
		# Cleanup is used only on plab nodes.
		# Don't try to teardown plab vnodes; it's just asking for
		# trouble as the shutdown may hang or take too long.  It's
		# best to simply try and free the vserver below.
424 425
		my $args = (($mode eq "teardown") ? "-k " :
			    ($mode eq "reboot" ? "-r " : " "));
426
		$args   .= ($jailed   ? "-j " : ($plab ? "-p " : "-i "));
427
		$args   .= "$vnode ";
428
		
429
		$exval = TBForkCmd("$ssh -host $pnode ".
430 431 432 433 434
                                 "$CLIENT_BIN/vnodesetup $args",1);
            }

            # Free the plab node lease if necessary.
            if ($plab && ($mode eq "teardown" || $mode eq "cleanup")) {
435
                TBSetNodeEventState($vnode, TBDB_NODESTATE_SHUTDOWN);
436 437 438 439 440
                exec("$TB/sbin/plabnode free $pid $eid $vnode");
                die("*** $0:\n".
                    "    exec failed!\n");
            }
	    exit($exval);
441 442 443 444 445 446 447 448 449 450 451
	}
    } else {
	#
	# We have too many of the little rugrats, wait for one to die
	#

	#
	# Set up a timer - we want to kill processes after they hit 120 seconds
	# old (not much of a life, is it?), so we find the first one marked for
	# death.
	#
452
	my $oldest;
453
	my $oldestpid = 0;
Kirk Webb's avatar
Kirk Webb committed
454
        my $oldestvnode = "";
455 456
	while (my ($pid, $aref) = each %child_vnodes) {
	    my ($vnode, $pnode, $mode, $jailed, $plab, $birthtime) = @$aref;
457
	    if ((!$oldestpid) || ($birthtime < $oldest)) {
458 459
		$oldest = $birthtime;
		$oldestpid = $pid;
Kirk Webb's avatar
Kirk Webb committed
460
                $oldestvnode = $vnode;
461 462 463 464 465 466 467 468 469 470 471
	    }
	}

	#
	# Sanity check
	#
	if (!$oldest) {
	    die "*** $0\n".
	    	"Uh oh, I have no children left, something is wrong!\n";
	}

472
	#
Kirk Webb's avatar
Kirk Webb committed
473
	# If the oldest has already expired, just kill it off right now, and go
474 475 476
	# back around the loop
	#
	my $now = time();
Kirk Webb's avatar
Kirk Webb committed
477
	my $waittime = ($oldest + $childwait) - time();
478

479
	#
Kirk Webb's avatar
Kirk Webb committed
480
	# Kill off the oldest if it gets too old while we're waiting.
481
	#
Kirk Webb's avatar
Kirk Webb committed
482 483
        my $childpid = -1;
        my $exitstatus = -1;
484

Kirk Webb's avatar
Kirk Webb committed
485 486 487 488 489 490 491 492 493
        eval {
            local $SIG{ALRM} = sub { die "alarm clock" };

            if ($waittime <= 0) {
                print "*** $0: timeout waiting for vnode: $oldestvnode\n";
                kill("TERM",$oldestpid);
            } else {
                alarm($waittime);
            }
494

Kirk Webb's avatar
Kirk Webb committed
495 496 497 498 499 500 501
            $childpid = wait();
            alarm 0;
            $exitstatus = $?;
        };
        if ($@) {
            die unless $@ =~ /alarm clock/;
            next;
502 503
        }

504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521
	#
	# Another sanity check
	#
	if ($childpid < 0) {
	    die "*** $0\n".
	    	"wait() returned <0, something is wrong!\n";
	}

	#
	# Look up to see what vnode, etc. this was associated with - if we
	# don't know about this child, ignore it
	#
	my $aref = $child_vnodes{$childpid};
	next unless @$aref;
	my ($vnode, $pnode, $mode, $jailed, $plab, $birthtime) = @$aref;
	$children--;
	delete $child_vnodes{$childpid};

522
	#
523
	# Look for setup failure, reported back through ssh.
524
	# 
525
	if ($exitstatus) {
526 527
	    print STDERR "vnode $vnode $mode on $pnode returned $?.\n"
		if $debug;
528

529
	    if ($exitstatus == 256) {
530
		print STDERR "$vnode is not running sshd.\n" if $debug;
531 532
	    }
	    elsif ($exitstatus == 15) {
533
		print STDERR "$vnode is wedged.\n" if $debug;
534
	    }
535 536 537
	    elsif ($exitstatus >> 8 == 99) {
		print STDERR "$vnode did not allocate properly.\n" if $debug;
	    }
538

539 540
	    push @failed_nodes, [$vnode, $pnode, $mode, $exitstatus];
	    
541 542
	    warn("*** $0:\n".
		 "    Virtual node $vnode $mode failure!\n");
543
	}
544

545 546 547 548 549 550 551 552 553 554
	if ($plab) {
	    #
	    # If the node was in the setup process, then mark its allocstate
	    # as down so os_setup knows not to bother waiting for it. DEAD
	    # is a temp state, different then DOWN. It indicates the node
	    # was never instantiated (important distinction for plab). 
	    #
	    if ($exitstatus &&
		(($mode eq "setup") || ($mode eq "reboot"))) {
		TBSetNodeAllocState($vnode, TBDB_ALLOCSTATE_DEAD());
555 556
	    }
	}
557 558 559
    }
}

560 561 562 563 564 565
#
# In force node, do not bother with this other stuff. 
#
exit(scalar(@failed_nodes))
    if ($force);

566 567 568 569 570 571
#
# Send mail to testbed-ops about failed nodes
#
if (@failed_nodes) { 
    my $failed_lines = join("\n",map { join("\t",@{$_}) } @failed_nodes);
    SENDMAIL($TBOPS, "Virtual Node failure for $pid/$eid",
572 573 574
	     "The following virtual nodes failed: \n" .
	     "vnode\t\tpnode\tmode\texit status\n" .
	     $failed_lines);
575 576
}

577 578 579 580 581 582
if ($killmode) {
    print STDOUT "Vnode teardown finished.\n";
}
else {
    print STDOUT "Vnode setup initiated on all nodes ...\n";
}
583
exit(0);