--- loncom/Attic/lonc 2001/06/12 15:32:38 1.16 +++ loncom/Attic/lonc 2001/08/30 20:02:28 1.17 @@ -41,7 +41,7 @@ sub catchexception { die($@); } -$childmaxattempts=10; +$childmaxattempts=5; # -------------------------------- Set signal handlers to record abnormal exits @@ -116,7 +116,11 @@ sub REAPER { # ta sub HUNTSMAN { # signal handler for SIGINT local($SIG{CHLD}) = 'IGNORE'; # we're going to kill our children - kill 'INT' => keys %children; + map { + $wasserver=$children{$_}; + &logthis('Closing '.$wasserver.': '.&subreply('exit',$wasserver)); + kill ('INT',$_); + } keys %children; my $execdir=$perlvar{'lonDaemons'}; unlink("$execdir/logs/lonc.pid"); &logthis("CRITICAL: Shutting down"); @@ -125,7 +129,11 @@ sub HUNTSMAN { # si sub HUPSMAN { # signal handler for SIGHUP local($SIG{CHLD}) = 'IGNORE'; # we're going to kill our children - kill 'INT' => keys %children; + map { + $wasserver=$children{$_}; + &logthis('Closing '.$wasserver.': '.&subreply('exit',$wasserver)); + kill ('INT',$_); + } keys %children; &logthis("CRITICAL: Restarting"); unlink("$execdir/logs/lonc.pid"); my $execdir=$perlvar{'lonDaemons'}; @@ -222,7 +230,7 @@ while (1) { # See who died and start new one foreach $thisserver (keys %hostip) { if (!$childpid{$thisserver}) { - if ($childatt{$thisserver}<=$childmaxattempts) { + if ($childatt{$thisserver}<$childmaxattempts) { $childatt{$thisserver}++; &logthis( "INFO: Trying to reconnect for $thisserver " @@ -287,6 +295,13 @@ print $remotesock "$answer"; $answer=<$remotesock>; chomp($answer); &logthis("Init reply for $conserver: >$answer<"); +if ($answer ne 'ok') { + my $st=120+int(rand(240)); + &logthis( +"WARNING: Init failed $conserver ($st secs)"); + sleep($st); + exit; +} sleep 5; print $remotesock "pong\n"; $answer=<$remotesock>;