newnode.in 14.7 KB
Newer Older
1 2 3 4
#!/usr/bin/perl -w

#
# EMULAB-COPYRIGHT
5
# Copyright (c) 2003-2008 University of Utah and the Flux Group.
6 7 8 9 10 11 12
# All rights reserved.
#

#
# newnode - a script for moving nodes from the new_ tables into production.
#

13
use lib '@prefix@/lib';
14 15

use libdb;
16
use libtestbed;
17
use libxmlrpc;
18
use NodeType;
Robert Ricci's avatar
Robert Ricci committed
19
use English;
20
use OSinfo;
21
use Getopt::Std;
22 23 24

use strict;

25 26 27 28 29
my $TB         = "@prefix@";
my $ELABINELAB = @ELABINELAB@;
my $RPCSERVER  = "@OUTERBOSS_NODENAME@";
my $RPCPORT    = "@OUTERBOSS_XMLRPCPORT@";
my $RPCCERT    = "@OUTERBOSS_SSLCERTNAME@";
30 31

my $switchmac = "$TB/sbin/switchmac";
32 33 34 35
my $os_load = "$TB/bin/os_load";
my $os_select = "$TB/bin/os_select";
my $newnode_reboot = "$TB/sbin/newnode_reboot";
my $named_setup = "$TB/sbin/named_setup";
36 37 38
my $nalloc = "$TB/bin/nalloc";
my $nfree = "$TB/bin/nfree";
my $dhcpd_makeconf = "$TB/sbin/dhcpd_makeconf";
39
my $exports_setup = "$TB/sbin/exports_setup";
40 41 42

my $dhcpd_rc = "/usr/local/etc/rc.d/2.dhcpd.sh";

43
my $sudo = "/usr/local/bin/sudo -S";
44 45

#
46
# MFS to boot the nodes into initially
47
#
48 49
my $MFS_INITIAL = TB_OSID_FREEBSD_MFS();
my $OS_OPSNODE  = "OPSNODE-BSD";
50 51

#
52
# Initial event system state to put the nodes into
53
#
54 55
my $STATE_INITIAL = TBDB_NODESTATE_SHUTDOWN;
my $STATE_OPSNODE = TBDB_NODESTATE_ISUP;
56 57 58 59 60 61

#
# Number of vnodes to create for each physical node
#
my $NUM_VNODES = 50;

62 63 64 65 66
#
# Default number of vnodes that can be hosted on any pnode
#
my $DEFAULT_PCVM_COUNT = 20;

67 68 69 70 71 72 73 74
#
# Experiments we might put nodes into
#
my $PID_HWDOWN = NODEDEAD_PID();
my $EID_HWDOWN = NODEDEAD_EID();
my $PID_OPSNODE = TBOPSPID();
my $EID_OPSNODE = "opsnodes";

75 76
# Turn off line buffering on output
$| = 1;
77

78 79 80 81
if (!TBAdmin()) {
    die "Sorry, only testbed administrators can run this script!\n";
}

82 83 84
#
# Handle command-line options
#
85 86 87
sub usage() {
    die "Usage: $0 [-f] <node_id> ...\n";
}
88 89 90 91 92 93 94 95 96 97
my $force_unconnected_interfaces = 0;
my %options = ();
if (! getopts("f", \%options)) {
    usage();
}

if ($options{f}) {
    $force_unconnected_interfaces = 1;
}

98 99 100 101 102 103
#
# We seem to have to do this so that ssh gets proper root permissions to read
# the key file. Argh.
#
$UID = $EUID = 0;

104
if (@ARGV < 1) {
105
    usage();
106 107
}

108 109 110
#
# The user has to be able to run sudo, so they can restart dhcpd.
#
111
if (system "$sudo /bin/pwd < /dev/null") {
112 113 114
    die "You must be able to sudo to root to use this script\n";
}

115 116 117 118 119 120 121 122 123 124
#
# For vnodes - figure out the jail IP base
#
my $IPBASE;
if (TBDB_JAILIPBASE =~ /^(\d+).(\d+).(\d+).(\d+)/) {
    $IPBASE = "$1.$2";
} else {
    die "Problem with JAILIPBASE\n"
}

125 126 127 128 129
my @node_ids = @ARGV;

#
# Now, loop through the nodes given, and add each one
#
Robert Ricci's avatar
Robert Ricci committed
130
my (@succeeded_nodes, @reboot_IPs);
131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146
NODE: foreach my $node_id (@node_ids) {
    my $query_result;

    #
    # Check to make sure said node does not already exist!
    #
    $query_result = DBQueryFatal("SELECT node_id FROM nodes WHERE " .
	"node_id='$node_id'");
    if ($query_result->num_rows()) {
	warn "Node $node_id failed: a node with that name already exists!\n";
	next NODE;
    }

    #
    # Grab information about the node from the new_nodes table
    #
147
    $query_result = DBQueryFatal("SELECT new_node_id, type, IP, temporary_IP, ".
148
	"floor, building, loc_x, loc_y, contact, phone, room, role " .
149
	"FROM new_nodes WHERE node_id='$node_id'");
150 151 152 153
    if (!$query_result->num_rows()) {
	warn "Node $node_id failed: No pending node with that name exists!\n";
	next NODE;
    }
154
    my ($new_node_id, $type, $IP, $tempIP, $floor, $building, $loc_x, $loc_y,
155
	$contact, $phone, $room, $role) = $query_result->fetchrow();
156

157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177
    #
    # Make sure they've given it an IP address that looks valid
    #
    if ($IP !~ /^(\d+).(\d+).(\d+).(\d+)/) {
	warn "Node $node_id has an invalid IP address ($IP) - skipping\n";
	next NODE;
    }

    #
    # Make sure some other node doesn't already have this IP address on its
    # control interface
    #
    $query_result = DBQueryFatal("SELECT node_id FROM interfaces WHERE " .
	"IP='$IP' AND role='" . TBDB_IFACEROLE_CONTROL() . "'");
    if ($query_result->num_rows()) {
	my ($existing_node_id) = $query_result->fetchrow();
	warn "Node $node_id failed: $existing_node_id is already using IP " .
		"address $IP\n";
	next NODE;
    }

178 179 180 181
    #
    # Make sure that the new node is of a valid type, and grab a few other
    # things to fill in as initial values
    #
182 183 184 185 186
    my $typeinfo = NodeType->Lookup($type);
    if (!defined($typeinfo)) {
	warn "No such nodetype $type is defined in the DB!\n";
	next NODE;
    }
187
    my ($control_iface,$virtnode_capacity,$adminmfs,$adminmfs_osid);
188 189 190 191 192 193 194 195 196

    if ($typeinfo->control_interface(\$control_iface) || !$control_iface) {
	warn "No control interface for $type is defined in the DB!\n";
	next NODE;
    }
    if ($typeinfo->virtnode_capacity(\$virtnode_capacity)) {
	warn "Failure looking up virtnode_capacity for $type in the DB!\n";
	next NODE;
    }
197
    if ($typeinfo->adminmfs_osid(\$adminmfs_osid)) {
198
	warn "Failure looking up adminmfs_osid for $type in the DB!\n";
199 200
	next NODE;
    }
201 202
    $virtnode_capacity = 0
	if (!defined($virtnode_capacity));
203 204 205 206 207 208 209 210

    # Find object for the adminfs.
    if (defined($adminmfs_osid)) {
	$adminmfs = OSinfo->Lookup($adminmfs_osid);
    }
    else {
	$adminmfs = OSinfo->Lookup(TBOPSPID(), $MFS_INITIAL);
    }
211 212 213 214

    #
    # Grab the node's MACs from the new_interfaces table
    #
215
    $query_result = DBQueryFatal("SELECT card, MAC, interface_type, " .
216
	"switch_id, switch_card, switch_port, cable, len, role " .
217
	"FROM new_interfaces WHERE new_node_id='$new_node_id'");
218 219 220 221 222 223
    if (!$query_result->num_rows()) {
	warn "Node $node_id failed: Must have at least one interface!\n";
	next NODE;
    }

    my %interfaces;
224
    while (my ($card, $MAC, $iface_type, $switch_id, $switch_card,
225
	    $switch_port, $cable, $len, $ifrole) = $query_result->fetchrow()) {
226 227 228
	#
	# Get some more information about this interface type
	#
229
	my $iface_query = DBQueryFatal("SELECT max_speed, full_duplex, " .
230
	    "connector FROM interface_types WHERE type='$iface_type'");
231
	if (!$iface_query->num_rows()) {
232
	    warn "Node $node_id failed: Interface $card is of unknown type " .
233
	    	"$iface_type\n";
234 235 236
	    next NODE;
	}

237
	my ($max_speed, $full_duplex, $connector) = $iface_query->fetchrow();
238

239 240 241 242
	#
	# Do a sanity check - make sure that we have a switch recorded for all
	# experimental interfaces
	#
243
	unless ($force_unconnected_interfaces || ($role eq "ctrlnode")) {
244
	    my $iface = "eth$card";
245 246
	    if (($iface ne $control_iface) && ($connector !~ /wireless/i)
		&& !$switch_id) {
247 248 249 250 251 252
		warn "Node $node_id failed: Don't know which switch " .
		    "card $iface is connected to\n";
		next NODE;
	    }
	}

253 254 255
	#
	# Stash it away...
	#
256
	$interfaces{$card} = [$MAC, $iface_type, $max_speed, $full_duplex,
257
		$switch_id, $switch_card, $switch_port, $cable, $len, $ifrole];
258 259 260 261 262 263

    }

    #
    # Make up a priority (just used for sorting)
    #
264 265
    $node_id =~ /^(.*\D)(\d+)$/;
    my ($prefix,$nodenum) = ($1, $2);
266
    my $priority;
267 268
    if (defined $nodenum) {
	$priority = $nodenum;
269 270 271 272
    } else {
	$priority = 1;
    }

273 274 275 276 277 278 279 280 281 282 283 284 285
    #
    # Figure out which project and experiment this node needs to go into
    #
    my ($pid, $eid);
    if ($role eq "ctrlnode") {
	$pid = $PID_OPSNODE;
	$eid = $EID_OPSNODE;
    } else {
	$pid = $PID_HWDOWN;
	$eid = $EID_HWDOWN;
    }

    #
286
    # Figure out what osid/opmode this node should run
287
    #
288 289
    my ($osid,$opmode);
    
290
    if ($role eq "ctrlnode") {
291
	my $osinfo = OSinfo->Lookup(TBOPSPID(), $OS_OPSNODE);
292

293
	$osid   = $osinfo->osid();
Leigh B. Stoller's avatar
Leigh B. Stoller committed
294
	$opmode = $osinfo->op_mode();
295 296 297
    }
    else {
	$osid   = $adminmfs->osid();
Leigh B. Stoller's avatar
Leigh B. Stoller committed
298
	$opmode = $adminmfs->op_mode();
299 300 301 302 303 304 305 306 307 308 309 310
    }

    #
    # Figure out what state to start the node in
    #
    my $state;
    if ($role eq "ctrlnode") {
	$state = $STATE_OPSNODE;
    } else {
	$state = $STATE_INITIAL;
    }

311 312 313 314 315 316 317 318 319 320 321 322 323
    #
    # When operating inside an inner elab, we need an extra bit of information
    # from the outer boss; which the IP of the outer control network. This is
    # strictly for bootinfo which has some really gross hacks in it.
    #
    my $outer_control_ip;
    
    if ($ELABINELAB) {
        libxmlrpc::Config({"server"  => $RPCSERVER,
			   "verbose" => 0,
			   "cert"    => $RPCCERT,
			   "portnum" => $RPCPORT});

324 325 326 327 328 329 330 331
	#
	# XXX force MAC to a string since simple XML parser may otherwise
	# think it is an int (i.e., all digits) or even a double (i.e.,
	# all digits with an 'e' in the middle).  Don't laugh, I spent a
	# lot of time figuring this out when it really happened!
	#
	my $macstr = RPC::XML::string->new($interfaces{0}[0]);

332
	my $rval = libxmlrpc::CallMethod("elabinelab", "newnode_info",
333
					 {"mac" => $macstr});
334 335 336 337 338 339 340 341 342 343 344 345 346

	if (!defined($rval)) {
	    print "XMPRPC server did not return control iface info!\n";
	    next NODE;
	}
	else {
	    #
	    # Grab what we need by searching the interface marked as
	    # the outer control network.
	    #
	    my %iface_array = %{ $rval->{"interfaces"} };
		
	    foreach my $mac (keys(%iface_array)) {
347 348
		my $aref   = $iface_array{$mac};
		my $ifrole = $aref->{"role"};
349

350
		if ($ifrole eq TBDB_IFACEROLE_OUTER_CONTROL()) {
351 352 353 354 355 356 357 358 359 360 361
		    $outer_control_ip = $aref->{"IP"};
		    last;
		}
	    }
	}
	if (!defined($outer_control_ip)) {
	    print "Could not determine outer control interface IP!\n";
	    next NODE;
	}
    }

362 363 364
    #
    # Okay, time to actually add the node!
    #
365 366 367 368 369
    my $uuid = NewUUID();
    if (!defined($uuid)) {
	print "Could not generate a UUID!\n";
	next NODE;
    }
370 371
    
    DBQueryFatal("INSERT INTO nodes SET node_id='$node_id', type='$type', " .
372
	"phys_nodeid='$node_id', role='$role', priority=$priority, " .
373
    	"eventstate='$state', op_mode='$opmode', uuid='$uuid', " .
374
	"def_boot_osid='$osid', " .
375
	"inception=now(), ".
376 377
	"state_timestamp=unix_timestamp(NOW()), " .
	"op_mode_timestamp=unix_timestamp(NOW())");
378

379 380
    DBQueryFatal("insert into node_hostkeys (node_id) ".
		 "values ('$node_id')");
381 382 383 384 385
    DBQueryFatal("insert into node_status ".
		 "(node_id, status, status_timestamp) ".
		 "values ('$node_id', 'down', now()) ");
    DBQueryFatal("insert into node_activity ".
		 "(node_id) values ('$node_id')");
386 387
    DBQueryFatal("insert into node_utilization ".
		 "(node_id) values ('$node_id')");
388

389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418
    #
    # Copy data into the location_info table, if they provided any
    #
    if ($floor || $building || $loc_x || $loc_y || $contact
	|| $phone || $room) {

	#
	# Handle columns that could have come to us NULL (and thus are undef)
	#
	if (defined $floor) {
	    $floor = "'$floor'";
	} else {
	    $floor = "NULL";
	}
	if (defined $building) {
	    $building = "'$building'";
	} else {
	    $building = "NULL";
	}
	if (defined $room) {
	    $room = "'$room'";
	} else {
	    $room = "NULL";
	}

	DBQueryFatal("INSERT INTO location_info SET node_id='$node_id', " .
	    "floor=$floor, building=$building, loc_x=$loc_x, loc_y=$loc_y, ".
	    "contact='$contact', phone='$phone', room=$room;");
    }

419 420 421
    #
    # Add some vnodes
    #
422
    if (defined $nodenum && ($role eq "testnode")) {
423 424 425 426
	my $vtype = $type;
	if (!($vtype =~ s/pc/pcvm/)) {
	    $vtype = "$vtype-vm";
	}
427
	DBQueryFatal("INSERT INTO node_auxtypes set node_id='$node_id', " .
428
	    "type='pcvm', count=$virtnode_capacity");
429
	DBQueryFatal("INSERT INTO node_auxtypes set node_id='$node_id', " .
430
	    "type='$vtype', count=$virtnode_capacity");
431 432
    }

433
    while (my ($card, $aref) = each %interfaces) {
434
	my ($MAC, $iface_type, $max_speed, $duplex, $switch_id, $switch_card,
435
	    $switch_port, $cable, $len, $ifrole) = @$aref;
436
	my $iface = "eth$card";
437 438 439 440 441
	my $iface_IP;
	my $wire_type;
	my $iface_role;

	# Role from new_interfaces overrides anything we decide here.
442 443
	if (defined($ifrole)) {
	    $iface_role = $ifrole;
444 445
	}
	elsif ($iface eq $control_iface) {
446
	    $iface_role = TBDB_IFACEROLE_CONTROL();
447
	}
448 449 450 451 452 453 454 455 456 457 458 459 460
	else {
	    $iface_role = TBDB_IFACEROLE_EXPERIMENT();
	}

	# Okay, now set up some stuff for the inserts.
	if ($iface_role eq TBDB_IFACEROLE_CONTROL()) {
	    $iface_IP  = $IP;
	    $wire_type = "Control";
	}
	elsif ($iface_role eq TBDB_IFACEROLE_OUTER_CONTROL()) {
	    $iface_IP  = $outer_control_ip;
	    $wire_type = "OuterControl";
	}
461 462
	else {
	    # Experimental interfaces start with no speed set. 
Kirk Webb's avatar
Kirk Webb committed
463
	    $max_speed = 0;
464 465
	    $iface_IP  = "";
	    $wire_type = "Node";
466
	}
467

468 469
	DBQueryFatal("INSERT INTO interfaces SET node_id='$node_id', " .
	    "card=$card, port=1, mac='$MAC', IP='$iface_IP', " .
470
	    "interface_type='$iface_type', iface='$iface', uuid=UUID(), " .
Kirk Webb's avatar
Kirk Webb committed
471
	    "current_speed='$max_speed', duplex=$duplex, role='$iface_role'");
472

473 474 475
	DBQueryFatal("INSERT INTO interface_state SET node_id='$node_id', " .
	    "card=$card, port=1, iface='$iface'");

476 477 478 479
	if (!$switch_id) {
	    print "No switch found for ${node_id}:$iface - skipping\n";
	    next;
	}
480 481 482 483 484 485 486 487

	my $cable_len = "";
	if ($cable) {
	    $cable_len .= ", cable=$cable";
	}
	if ($len) {
	    $cable_len .= ", len=$len";
	}
488 489 490
	DBQueryFatal("INSERT INTO wires SET type='$wire_type', " .
	    "node_id1='$node_id', card1=$card, port1=1, " .
	    "node_id2='$switch_id', card2='$switch_card', " .
491
	    "port2='$switch_port' $cable_len");
492 493 494
    }

    #
495 496
    # Put it into hwdown for now - I would put them in reloading, but I'm
    # afriad the reload_daemon might do the wrong thing to them
497
    #
498
    system "$nalloc $pid $eid $node_id";
499 500 501 502

    #
    # Remove the node from the new_ tables
    #
503
    DBQueryFatal("DELETE FROM new_nodes WHERE new_node_id=$new_node_id");
504
    DBQueryFatal("DELETE FROM new_interfaces WHERE new_node_id=$new_node_id");
505

506
    print "$node_id succesfully added!\n";
507 508

    push @succeeded_nodes, $node_id;
Robert Ricci's avatar
Robert Ricci committed
509 510 511 512 513 514 515

    #
    # Only reboot regular nodes, not ops nodes
    #
    if ($role ne "ctrlnode") {
	push @reboot_IPs, $tempIP;
    }
516 517 518 519 520 521 522 523 524 525 526 527
}

#
# No point in restarting dhcpd, etc. if there are no nodes that succeeded
#
if (!@succeeded_nodes) {
    die "No nodes succeeded, exiting early\n";
}

#
# Re-generate dhcpd.conf
#
528
print "Re-generating dhcpd.conf\n";
529
if (system("$dhcpd_makeconf -i -r")) {
530 531
    die "Unable to open regenerate dhcpd config file\n";
}
532 533 534 535 536 537 538 539 540 541 542 543

print "Restarting dhcpd: $sudo $dhcpd_rc stop\n";
my $sudo_rv = system "$sudo $dhcpd_rc stop";
if ($sudo_rv) {
    die "Error stopping dhcpd - return value was $sudo_rv\n";
}
sleep 2;
print "Restarting dhcpd: $sudo $dhcpd_rc start\n";
$sudo_rv = system "$sudo $dhcpd_rc start";
if ($sudo_rv) {
    die "Error starting dhcpd - return value was $sudo_rv\n";
}
544

545 546 547 548 549
print "Setting up nameserver\n";
my $named_rv = system "$named_setup";
if ($named_rv) {
    die "Error running named_setup - return value was $named_rv\n";
}
550

551 552 553 554 555 556 557 558 559
#
# Before we boot nodes into the MFS, we have to make sure they can mount
# NFS filesystems
#
print "Running exports_setup\n";
my $exports_rv = system "$exports_setup";
if ($exports_rv) {
    warn "WARNING - exports_setup returned $exports_rv";
}
560

561 562 563 564
#
# Start rebooting nodes
#
print "Rebooting nodes...\n";
Robert Ricci's avatar
Robert Ricci committed
565
foreach my $IP (@reboot_IPs) {
566 567
    print "Rebooting $IP\n";
    system "$newnode_reboot $IP\n";
568 569
}

570 571 572 573 574
print "\n\n";
print "Finished - when you are satisifed that the nodes are working\n";
print "correctly, use nfree on boss to free them from the emulab-ops/hwdown\n";
print "experiment.\n";

575
#
576 577 578
# TODO -
#    disable interfaces
#    console setup
579
#