Merge branch 'master' of git-public.flux.utah.edu:/flux/git/emulab-devel

tbsetup/mapper.in

 #!/usr/bin/perl -w
 #
 # EMULAB-COPYRIGHT
-# Copyright (c) 2000-2009 University of Utah and the Flux Group.
+# Copyright (c) 2000-2010 University of Utah and the Flux Group.
 # All rights reserved.
 #
 use strict;
@@ -284,6 +284,8 @@ sub AssignLoop()
 
     TBDebugTimeStamp("mapper loop started");
     while (1) {
+	my $gotlock = 0;
+	
 	chat("Assign run $currentrun\n");
 
 	my $prefix = ($debug || $regression ? "$pid-$eid" : "$pid-$eid-$$");
@@ -297,6 +299,28 @@ sub AssignLoop()
 	    chat("Trying assign on an empty testbed.\n");
 	}
 
+	#
+	# Serialize with the pool daemon if using shared nodes.
+	# XXX When using shared nodes, only one can proceed at a
+	# time through assignment. This is okay for now since few
+	# experiments are using shared nodes. Eventually needs to be
+	# a barrier. 
+	#
+	if ((!($impotent || $regression)) && $vtop->sharednodecount()) {
+	    while (1) {
+		my $lock_result =
+		    DBQueryWarn("select get_lock('pool_daemon', 10)");
+		fatal("DB Error tring to get pool_daemon lock")
+		    if (!defined($lock_result));
+
+		($gotlock) = $lock_result->fetchrow_array();
+		last
+		    if ($gotlock);
+		
+		chat("Waiting for pool daemon lock ...\n");
+	    }
+	}
+
 	#
 	# RunAssign returns  0 if successful.
 	#           returns -1 if failure, but assign says to stop trying.
@@ -305,6 +329,11 @@ sub AssignLoop()
 	#
 	my $retval = RunAssign($precheck, $prefix);
 
+	if ($gotlock) {
+	    DBQueryWarn("select release_lock('pool_daemon')")
+		or fatal("Could not release the pool lock");
+	}
+
 	# Success!
 	last
 	    if ($retval == 0);

tbsetup/named_setup.in

@@ -153,6 +153,13 @@ while (my ($node_id,$pid,$eid,$vname,$erole) = $db_result->fetchrow_array()) {
         ($erole eq "boss" || $erole eq "boss+router")) {
         push @{$cnames{$node_id}}, "www.$eid.$pid";
     }
+    #
+    # Special case for inner elab ops; add CNAME for event-server.
+    #
+    if (defined($erole) &&
+        ($erole eq "ops" || $erole eq "ops+fs")) {
+        push @{$cnames{$node_id}}, "event-server.$eid.$pid";
+    }
 }
 
 

tbsetup/pool_daemon.in

@@ -19,10 +19,11 @@ sub usage()
 	"Use the -d option to prevent daemonization\n";
     exit(-1);
 }
-my $optlist = "dn";
+my $optlist = "dne";
 my $debug    = 0;
 my $impotent = 0;
-my $startup  = 0;
+my $killme   = 0;
+my $nofree   = 1;
 
 #
 # This should run as root.
@@ -88,10 +89,13 @@ if (@ARGV != 0) {
     usage();
 }
 if (defined($options{"d"})) {
-    $debug = $options{"d"};
+    $debug = 1;
 }
 if (defined($options{"n"})) {
-    $impotent = $options{"n"};
+    $impotent = 1;
+}
+if (defined($options{"e"})) {
+    $nofree = 1;
 }
 
 if (!$impotent && CheckDaemonRunning("pool_daemon")) {
@@ -164,21 +168,26 @@ if (!defined($image)) {
     exit(0);
 }
 
-while (1) {
+#
+# And handler for TERM since we really do not want this to be
+# interrupted. Just set a flag that will cause it to exit at
+# the next loop.
+#
+sub sigterm()
+{
+    print "Got a TERM signal; arranging to exit soon\n";
+    $killme = 1;
+}
+$SIG{TERM} = \&sigterm;
+
+while (!$killme) {
     my $disabled;
     
-    # Use a long period; we do not want the pool to change too fast.
-    if (!$startup) {
-	$startup++;
-    }
-    else {
-	sleep(120);
-    }
     print "Pool Daemon running at ".`date`;
 
     if (! TBGetSiteVar("web/nologins", \$disabled) || $disabled) {
 	print "  Skipping this loop cause of nologins\n";
-	next;
+	goto loop;
     }
 
     Node->FlushAll();
@@ -196,6 +205,24 @@ while (1) {
     my $minpoolsize  = TBGetSiteVar("general/minpoolsize");
     my $poolnodetype = TBGetSiteVar("general/poolnodetype");
 
+    #
+    # Serialize this part with the mapper.
+    #
+    if (!$impotent) {
+	while (1) {
+	    my $lock_result =
+		DBQueryWarn("select get_lock('pool_daemon', 5)");
+	    fatal("DB Error tring to get pool_daemon lock")
+		if (!defined($lock_result));
+
+	    my ($gotlock) = $lock_result->fetchrow_array();
+	    last
+		if ($gotlock);
+		
+	    print "Waiting for pool daemon lock ...\n";
+	}
+    }
+	
     #
     # Look to see how each of the nodes is packed. This is
     # advisory; we will not know for sure until tables locked
@@ -220,7 +247,7 @@ while (1) {
 	next
 	    if ($vnodecount < 0);
 
-	if ($vnodecount == 0) {
+	if ($vnodecount == 0 && !$nofree) {
 	    print "$node no longer has virtual nodes on it.\n";
 	    # Free the node unless we would go below the minpoolsize.
 	    if (scalar(@nodelist) - scalar(keys(%tofree)) > $minpoolsize) {
@@ -231,6 +258,7 @@ while (1) {
 	}
 	# Count up loaded vs. unloaded nodes.
 	my $factor = $maxsharecount / $vnodecount;
+	print "$node load factor is $factor\n";
 	if ($factor < 0.5) {
 	    $unloaded++;
 	}
@@ -248,14 +276,14 @@ while (1) {
 	    my $key = (keys(%tofree))[0];
 	    delete($tofree{$key});
 	}
-	elsif (scalar(@nodelist) < $maxpoolsize) {
+ 	elsif (scalar(@nodelist) < $maxpoolsize) {
 	    $newcount++;
 	}
     }
     if (! (keys(%tofree) || $newcount)) {
 	exit(0)
 	    if ($impotent);
-	next;
+	goto loop;
     }
 
     #
@@ -268,7 +296,7 @@ while (1) {
     
     if (!open(NS, ">$tmpfile")) {
 	notify("Could not create $tmpfile");
-	next;
+	goto loop;
     }
     print NS "# Auto generated by the pool daemon\n\n";
     print NS "source tb_compat.tcl\n";
@@ -311,7 +339,7 @@ while (1) {
     chmod(0775, $tmpfile);
 
     exit(0)
-	if ($impotent);
+	if ($impotent || $killme);
 
     # Must do this each time before fork.
     tblog_new_session();
@@ -335,7 +363,7 @@ while (1) {
 
 	    if ($error_data->{'cause'} eq "temp") {
 		print "Temporary resource shortage; try again later\n";
-		next;
+		goto loop;
 	    }
 	    fatal("swapmod failed");
 	}
@@ -348,6 +376,12 @@ while (1) {
 	exec("$SWAPEXP -q -w -n -s modify $pid $eid $tmpfile");
 	die("Could not exec $SWAPEXP\n");
     }
+  loop:
+    DBQueryWarn("select release_lock('pool_daemon')")
+	or fatal("Could not release the pool lock");
+    
+    # Use a long period; we do not want the pool to change too fast.
+    sleep(120);
 }
 
 #

tbsetup/tbswap.in

@@ -199,6 +199,7 @@ if (!defined($experiment)) {
 }
 my $special = ($pid eq "testbed" || $pid eq "tbres" ||
 	       $pid eq "emulab-ops" || $pid eq "utahstud");
+my $newsetup = ($pid eq "testbed");
 
 #
 # Print starting message.
@@ -1285,8 +1286,10 @@ sub doSwapin($) {
     #
     print "Resetting OS and rebooting.\n";
     TBDebugTimeStamp("launching os_setup");
-    if (!($os_setup_pid = fork())) { 
-	exec("os_setup $pid $eid") or return 1;
+    if (!($os_setup_pid = fork())) {
+	my $oscmd = ($newsetup ? "os_setup_new" : "os_setup");
+	
+	exec("$oscmd $pid $eid") or return 1;
     } elsif ($os_setup_pid == -1) {
 	tberror "Fork failed.";
 	return 1;