From 20775a437705cec3c4a6687e655f88d4e61dbd0b Mon Sep 17 00:00:00 2001
From: jbjohnso <jbjohnso@8638fb3e-16cb-4fca-ae20-7b5d299a9bcd>
Date: Fri, 3 Apr 2009 14:40:21 +0000
Subject: [PATCH] -Have blade and ipmi plugins only decrease known child count
 when the child exiting belongs to it explicitly -Have blade and ipmi plugin
 children wait longer for parent acknowledgement -Ignore failure to deliver
 acknowledgements to children that give up and exit before that can occur

git-svn-id: https://svn.code.sf.net/p/xcat/code/xcat-core/trunk@3075 8638fb3e-16cb-4fca-ae20-7b5d299a9bcd
---
 xCAT-server/lib/xcat/plugins/blade.pm | 6 +++---
 xCAT-server/lib/xcat/plugins/ipmi.pm  | 6 +++---
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/xCAT-server/lib/xcat/plugins/blade.pm b/xCAT-server/lib/xcat/plugins/blade.pm
index f424265db..52cc45b77 100644
--- a/xCAT-server/lib/xcat/plugins/blade.pm
+++ b/xCAT-server/lib/xcat/plugins/blade.pm
@@ -264,7 +264,7 @@ sub waitforack {
     my $select = new IO::Select;
     $select->add($sock);
     my $str;
-    if ($select->can_read(10)) { # Continue after 10 seconds, even if not acked...
+    if ($select->can_read(60)) { # Continue after 60 seconds, even if not acked...
         if ($str = <$sock>) {
         } else {
            $select->remove($sock); #Block until parent acks data
@@ -2178,7 +2178,7 @@ sub process_request {
 
 
   my $children = 0;
-  $SIG{CHLD} = sub { my $cpid; while ($cpid = waitpid(-1, WNOHANG) > 0) { delete $mm_comm_pids{$cpid}; $children--; } };
+  $SIG{CHLD} = sub { my $cpid; while ($cpid = waitpid(-1, WNOHANG) > 0) { if ($mm_comm_pids{$cpid}) { delete $mm_comm_pids{$cpid}; $children--; } } };
   my $inputs = new IO::Select;;
   foreach my $info (@$moreinfo) {
     $info=~/^\[(.*)\]\[(.*)\]\[(.*)\]/;
@@ -2688,7 +2688,7 @@ sub forward_data {
       while ($data !~ /ENDOFFREEZE6sK4ci/) {
         $data .= <$rfh>;
       }
-      print $rfh "ACK\n";
+      eval { print $rfh "ACK\n"; }; #Ignore ack loss due to child giving up and exiting, we don't actually explicitly care about the acks
       my $responses=thaw($data);
       foreach (@$responses) {
         $callback->($_);
diff --git a/xCAT-server/lib/xcat/plugins/ipmi.pm b/xCAT-server/lib/xcat/plugins/ipmi.pm
index d991b1a52..94113b006 100644
--- a/xCAT-server/lib/xcat/plugins/ipmi.pm
+++ b/xCAT-server/lib/xcat/plugins/ipmi.pm
@@ -361,7 +361,7 @@ sub waitforack {
     my $select = new IO::Select;
     $select->add($sock);
     my $str;
-    if ($select->can_read(10)) { # Continue after 10 seconds, even if not acked...
+    if ($select->can_read(60)) { # Continue after 60 seconds, even if not acked...
         if ($str = <$sock>) {
         } else {
            $select->remove($sock); #Block until parent acks data
@@ -6200,7 +6200,7 @@ sub process_request {
   }
 
     my $children = 0;
-    $SIG{CHLD} = sub {my $kpid; do { $kpid = waitpid(-1, WNOHANG); if ($kpid > 0) { delete $bmc_comm_pids{$kpid}; $children--; } } while $kpid > 0; };
+    $SIG{CHLD} = sub {my $kpid; do { $kpid = waitpid(-1, WNOHANG); if ($bmc_comm_pids{$kpid}) { delete $bmc_comm_pids{$kpid}; $children--; } } while $kpid > 0; };
     my $sub_fds = new IO::Select;
     foreach (@donargs) {
       while ($children > $ipmimaxp) { 
@@ -6291,7 +6291,7 @@ sub forward_data { #unserialize data from pipe, chunk at a time, use magic to de
       while ($data !~ /ENDOFFREEZE6sK4ci/) {
         $data .= <$rfh>;
       }
-      print $rfh "ACK\n";
+      eval { print $rfh "ACK\n"; };  # Ignore ack loss to child that has given up and exited
       my $responses=thaw($data);
       foreach (@$responses) {
         #save the nodes that has errors and the ones that has no-op for use by the node status monitoring