From 182e950bf52066dc74037b0be66c1818c01cfe2c Mon Sep 17 00:00:00 2001
From: "Leigh B. Stoller" <stoller@flux.utah.edu>
Date: Mon, 24 Jun 2002 20:27:29 +0000
Subject: [PATCH] Hmm, forgot to commit the node_update changes that switch to
a client initiated model for updating remote nodes.
---
tbsetup/node_update.in | 61 +++++++++++++++++++++++++++++++++++++++---
1 file changed, 58 insertions(+), 3 deletions(-)
diff --git a/tbsetup/node_update.in b/tbsetup/node_update.in
index d39e38b021..7a6556f2ac 100644
--- a/tbsetup/node_update.in
+++ b/tbsetup/node_update.in
@@ -88,6 +88,7 @@ my $logname;
my %pids = ();
my $failed = 0;
my $dbuid;
+my @remotenodes = ();
#
# We don't want to run this script unless its the real version.
@@ -195,6 +196,24 @@ if (! @nodes) {
fatal("No Nodes in the experiment");
}
+#
+# For widearea nodes, just update flag them in the DB. The nodes will
+# notice this and suck the data over when they can.
+#
+foreach my $node ( @nodes ) {
+ if (TBIsNodeRemote($node) && !TBIsNodeVirtual($node)) {
+ print STDOUT "Marking widearea node $node for update.\n";
+
+ #
+ # No point in incrementing the flag past 2!
+ #
+ DBQueryFatal("update nodes set update_accounts=update_accounts+1 ".
+ "where node_id='$node' and update_accounts<2");
+
+ push(@remotenodes, $node);
+ }
+}
+
#
# We want some overlap, but not too much since we could burn up
# a lot processes on wedged nodes. Issue a small number in parallel,
@@ -203,6 +222,13 @@ if (! @nodes) {
#
my $maxpids = 0;
foreach my $node ( @nodes ) {
+ #
+ # Remote nodes handled just above.
+ #
+ if (TBIsNodeRemote($node) || TBIsNodeVirtual($node)) {
+ next;
+ }
+
while ($maxpids >= $maxchildren) {
my $thispid = waitpid(-1, 0);
my $thisnode = $pids{$thispid};
@@ -212,7 +238,7 @@ foreach my $node ( @nodes ) {
print STDERR "Update of node $thisnode failed!\n";
}
else {
- print STDOUT "$thisnode updated ...\n";
+ print STDOUT "$thisnode updated.\n";
}
delete($pids{$thispid});
@@ -236,7 +262,36 @@ foreach my $thispid ( keys(%pids) ) {
print STDERR "Update of node $node failed!\n";
}
else {
- print STDOUT "$node updated ...\n";
+ print STDOUT "$node updated.\n";
+ }
+}
+
+#
+# Wait a couple of minutes for remote nodes to do the update. We know
+# this cause tmcd decrements the update_accounts flag whenever a node
+# picks up new accounts.
+#
+if (@remotenodes) {
+ print STDOUT "Waiting a while for widearea nodes to update ...\n";
+ sleep(90);
+ my $foo = pop(@remotenodes);
+ my $str = "node_id='$foo' ";
+
+ foreach my $node ( @remotenodes ) {
+ $str .= "or node_id='$node' ";
+ }
+ my $query_result =
+ DBQueryFatal("select node_id,update_accounts from nodes ".
+ "where ($str)");
+
+ while (my ($node,$notdone) = $query_result->fetchrow_array) {
+ if ($notdone) {
+ $failed++;
+ print STDERR "Update of node $node has not completed!\n";
+ }
+ else {
+ print STDOUT "$node updated.\n";
+ }
}
}
@@ -255,7 +310,7 @@ sub UpdateNode {
my($node) = @_;
my($syspid, $mypid);
- print STDOUT "Updating $node ...\n";
+ print STDOUT "Updating $node.\n";
#
# We need to know if its a remote or local node, so we know how
--
GitLab