--- loncom/loncnew 2007/03/28 00:23:46 1.80 +++ loncom/loncnew 2007/05/01 01:04:23 1.85 @@ -2,7 +2,7 @@ # The LearningOnline Network with CAPA # lonc maintains the connections to remote computers # -# $Id: loncnew,v 1.80 2007/03/28 00:23:46 albertel Exp $ +# $Id: loncnew,v 1.85 2007/05/01 01:04:23 albertel Exp $ # # Copyright Michigan State University Board of Trustees # @@ -60,7 +60,6 @@ use LONCAPA::Stack; use LONCAPA::LondConnection; use LONCAPA::LondTransaction; use LONCAPA::Configuration; -use LONCAPA::HashIterator; use Fcntl qw(:flock); @@ -73,7 +72,8 @@ my %perlvar = %{$perlvarref}; # # parent and shared variables. -my %ChildHash; # by pid -> host. +my %ChildPid; # by pid -> host. +my %ChildHost; # by host. my %listening_to; # Socket->host table for who the parent # is listening to. my %parent_dispatchers; # host-> listener watcher events. @@ -94,6 +94,8 @@ my $executable = $0; # Get the full # The variables below are only used by the child processes. # my $RemoteHost; # Name of host child is talking to. +my $RemoteHostId; # default lonid of host child is talking to. +my @all_host_ids; my $UnixSocketDir= $perlvar{'lonSockDir'}; my $IdleConnections = Stack->new(); # Set of idle connections my %ActiveConnections; # Connections to the remote lond. @@ -112,7 +114,6 @@ my $LondConnecting = 0; # True wh -my $hosts_tab = 0; # True if we are using a static hosts.tab my $I_am_child = 0; # True if this is the child process. # @@ -333,9 +334,7 @@ sub child_exit { open(LOCK,">$lock_file"); print LOCK "Contents not important"; close(LOCK); - if ($hosts_tab) { - unlink(&GetLoncSocketPath()); - } + unlink(&GetLoncSocketPath()); if ($message) { die($message); @@ -503,6 +502,9 @@ the data and Event->w->fd is the socket sub ClientWritable { my $Event = shift; my $Watcher = $Event->w; + if (!defined($Watcher)) { + &child_exit(-1,'No watcher for event in ClientWritable'); + } my $Data = $Watcher->data; my $Socket = $Watcher->fd; @@ -566,6 +568,7 @@ sub ClientWritable { } } else { $Watcher->cancel(); # A delayed request...just cancel. + return; } } @@ -1150,10 +1153,12 @@ sub LondWritable { } } + =pod =cut + sub QueueDelayed { Debug(3,"QueueDelayed called"); @@ -1162,16 +1167,7 @@ sub QueueDelayed { Debug(4, "Delayed path: ".$path); opendir(DIRHANDLE, $path); - my @all_host_ids; - my $host_iterator = &LondConnection::GetHostIterator(); - while (!$host_iterator->end()) { - my ($host_id,$host_name) = @{$host_iterator->get()}[0,3]; - if ($host_name eq $RemoteHost) { - push(@all_host_ids, $host_id); - } - $host_iterator->next(); - } - my $host_id_re = '(?:'.join('|',@all_host_ids).')'; + my $host_id_re = '(?:'.join('|',map {quotemeta($_)} (@all_host_ids)).')'; my @alldelayed = grep(/\.$host_id_re$/, readdir(DIRHANDLE)); closedir(DIRHANDLE); foreach my $dfname (sort(@alldelayed)) { @@ -1206,7 +1202,8 @@ sub MakeLondConnection { .GetServerPort()); my $Connection = LondConnection->new(&GetServerHost(), - &GetServerPort()); + &GetServerPort(), + &GetHostId()); if($Connection eq undef) { # Needs to be more robust later. Log("CRITICAL","Failed to make a connection with lond."); @@ -1214,6 +1211,7 @@ sub MakeLondConnection { return 0; # Failure. } else { + $LondConnecting = 1; # Connection in progress. # The connection needs to have writability # monitored in order to send the init sequence # that starts the whole authentication/key @@ -1244,7 +1242,6 @@ sub MakeLondConnection { } Log("SUCESS", "Created connection ".$ConnectionCount ." to host ".GetServerHost()); - $LondConnecting = 1; # Connection in progress. return 1; # Return success. } @@ -1496,6 +1493,18 @@ sub GetServerHost { =pod +=head2 GetServerId + +Returns the hostid whose lond we talk with. + +=cut + +sub GetHostId { + return $RemoteHostId; # Setup by the fork. +} + +=pod + =head2 GetServerPort Returns the lond port number. @@ -1749,7 +1758,7 @@ sub ChildProcess { # Create a new child for host passed in: sub CreateChild { - my ($host, $socket) = @_; + my ($host, $hostid) = @_; my $sigset = POSIX::SigSet->new(SIGINT); sigprocmask(SIG_BLOCK, $sigset); @@ -1758,18 +1767,15 @@ sub CreateChild { my $pid = fork; if($pid) { # Parent $RemoteHost = "Parent"; - $ChildHash{$pid} = $host; + $ChildPid{$pid} = $host; sigprocmask(SIG_UNBLOCK, $sigset); - + undef(@all_host_ids); } else { # child. + $RemoteHostId = $hostid; ShowStatus("Connected to ".$RemoteHost); $SIG{INT} = 'DEFAULT'; sigprocmask(SIG_UNBLOCK, $sigset); - if(defined $socket) { - &ChildProcess($socket); - } else { - ChildProcess; # Does not return. - } + &ChildProcess(); # Does not return. } } @@ -1797,67 +1803,50 @@ sub parent_client_connection { my ($event) = @_; my $watcher = $event->w; my $socket = $watcher->fd; - if ($hosts_tab) { - - # Lookup the host associated with this socket: - - my $host = $listening_to{$socket}; - - # Start the child: - - - - &Debug(9,"Creating child for $host (parent_client_connection)"); - &CreateChild($host, $socket); - - # Clean up the listen since now the child takes over until it exits. - - $watcher->cancel(); # Nolonger listening to this event - delete($listening_to{$socket}); - delete($parent_dispatchers{$host}); - $socket->close(); - - } else { - my $connection = $socket->accept(); # Accept the client connection. - Event->io(cb => \&get_remote_hostname, - poll => 'r', - data => "", - fd => $connection); - } + my $connection = $socket->accept(); # Accept the client connection. + Event->io(cb => \&get_remote_hostname, + poll => 'r', + data => "", + fd => $connection); } } sub get_remote_hostname { - my ($event) = @_; - my $watcher = $event->w; - my $socket = $watcher->fd; - - my $thisread; - my $rv = $socket->recv($thisread, POSIX::BUFSIZ, 0); - Debug(8, "rcv: data length = ".length($thisread)." read =".$thisread); - if (!defined($rv) || length($thisread) == 0) { - # Likely eof on socket. - Debug(5,"Client Socket closed on lonc for p_c_c"); - close($socket); - $watcher->cancel(); - return; - } + my ($event) = @_; + my $watcher = $event->w; + my $socket = $watcher->fd; - my $data = $watcher->data().$thisread; - $watcher->data($data); - if($data =~ /\n$/) { # Request entirely read. - chomp($data); - } else { - return; - } - - &Debug(5,"Creating child for $data (parent_client_connection)"); - &CreateChild($data); - - # Clean up the listen since now the child takes over until it exits. - $watcher->cancel(); # Nolonger listening to this event - $socket->send("done\n"); - $socket->close(); + my $thisread; + my $rv = $socket->recv($thisread, POSIX::BUFSIZ, 0); + Debug(8, "rcv: data length = ".length($thisread)." read =".$thisread); + if (!defined($rv) || length($thisread) == 0) { + # Likely eof on socket. + Debug(5,"Client Socket closed on lonc for p_c_c"); + close($socket); + $watcher->cancel(); + return; + } + + my $data = $watcher->data().$thisread; + $watcher->data($data); + if($data =~ /\n$/) { # Request entirely read. + chomp($data); + } else { + return; + } + + &Debug(5,"Creating child for $data (parent_client_connection)"); + (my $hostname,my $lonid,@all_host_ids) = split(':',$data); + $ChildHost{$hostname}++; + if ($ChildHost{$hostname} == 1) { + &CreateChild($hostname,$lonid); + } else { + &Log('WARNING',"Request for a second child on $hostname"); + } + # Clean up the listen since now the child takes over until it exits. + $watcher->cancel(); # Nolonger listening to this event + $socket->send("done\n"); + $socket->close(); } # parent_listen: @@ -1908,7 +1897,7 @@ sub parent_listen { sub parent_clean_up { my ($loncapa_host) = @_; - Debug(5, "parent_clean_up: $loncapa_host"); + Debug(-1, "parent_clean_up: $loncapa_host"); my $socket_file = &GetLoncSocketPath($loncapa_host); unlink($socket_file); # No problem if it doesn't exist yet [startup e.g.] @@ -1917,14 +1906,12 @@ sub parent_clean_up { } -# listen_on_all_unix_sockets: -# This sub initiates a listen on all unix domain lonc client sockets. -# This will be called in the case where we are trimming idle processes. -# When idle processes are trimmed, loncnew starts up with no children, -# and only spawns off children when a connection request occurs on the -# client unix socket. The spawned child continues to run until it has -# been idle a while at which point it eventually exits and once more -# the parent picks up the listen. + +# This sub initiates a listen on the common unix domain lonc client socket. +# loncnew starts up with no children, and only spawns off children when a +# connection request occurs on the common client unix socket. The spawned +# child continues to run until it has been idle a while at which point it +# eventually exits and once more the parent picks up the listen. # # Parameters: # NONE @@ -1933,18 +1920,6 @@ sub parent_clean_up { # Returns: # NONE # -sub listen_on_all_unix_sockets { - Debug(5, "listen_on_all_unix_sockets"); - my $host_iterator = &LondConnection::GetHostIterator(); - while (!$host_iterator->end()) { - my $host_entry_ref = $host_iterator->get(); - my $host_name = $host_entry_ref->[3]; - Debug(9, "Listen for $host_name"); - &parent_listen($host_name); - $host_iterator->next(); - } -} - sub listen_on_common_socket { Debug(5, "listen_on_common_socket"); &parent_listen(); @@ -1969,16 +1944,13 @@ sub server_died { } # need the host to restart: - my $host = $ChildHash{$pid}; + my $host = $ChildPid{$pid}; if($host) { # It's for real... &Debug(9, "Caught sigchild for $host"); - delete($ChildHash{$pid}); - if ($hosts_tab) { - &parent_listen($host); - } else { - &parent_clean_up($host); - } - + delete($ChildPid{$pid}); + delete($ChildHost{$host}); + &parent_clean_up($host); + } else { &Debug(5, "Caught sigchild for pid not in hosts hash: $pid"); } @@ -2036,11 +2008,7 @@ Log("CRITICAL", "--------------- Startin LondConnection::ReadConfig; # Read standard config files. $RemoteHost = "[parent]"; -if ($hosts_tab) { - &listen_on_all_unix_sockets(); -} else { - &listen_on_common_socket(); -} +&listen_on_common_socket(); $RemoteHost = "Parent Server"; @@ -2065,15 +2033,9 @@ $parent_handlers{INT} = Event->signal(cb $parent_handlers{TERM} = Event->signal(cb => \&Terminate, desc => "Parent TERM handler", signal => "TERM"); -if ($hosts_tab) { - $parent_handlers{HUP} = Event->signal(cb => \&Restart, - desc => "Parent HUP handler.", - signal => "HUP"); -} else { - $parent_handlers{HUP} = Event->signal(cb => \&KillThemAll, - desc => "Parent HUP handler.", - signal => "HUP"); -} +$parent_handlers{HUP} = Event->signal(cb => \&KillThemAll, + desc => "Parent HUP handler.", + signal => "HUP"); $parent_handlers{USR1} = Event->signal(cb => \&CheckKids, desc => "Parent USR1 handler", signal => "USR1"); @@ -2110,7 +2072,7 @@ sub CheckKids { foreach my $host (keys %parent_dispatchers) { print $fh "LONC Parent process listening for $host\n"; } - foreach my $pid (keys %ChildHash) { + foreach my $pid (keys %ChildPid) { Debug(2, "Sending USR1 -> $pid"); kill 'USR1' => $pid; # Tell Child to report status. } @@ -2154,11 +2116,7 @@ sub UpdateKids { # The down side is transactions that are in flight will get timed out # (lost unless they are critical). - if ($hosts_tab) { - &Restart(); - } else { - &KillThemAll(); - } + &KillThemAll(); } @@ -2191,13 +2149,19 @@ SIGHUP. Responds to sigint and sigterm. sub KillThemAll { Debug(2, "Kill them all!!"); - local($SIG{CHLD}) = 'IGNORE'; # Our children >will< die. - foreach my $pid (keys %ChildHash) { - my $serving = $ChildHash{$pid}; + + #local($SIG{CHLD}) = 'IGNORE'; + # Our children >will< die. + # but we need to catch their death and cleanup after them in case this is + # a restart set of kills + my @allpids = keys(%ChildPid); + foreach my $pid (@allpids) { + my $serving = $ChildPid{$pid}; ShowStatus("Nicely Killing lonc for $serving pid = $pid"); Log("CRITICAL", "Nicely Killing lonc for $serving pid = $pid"); kill 'QUIT' => $pid; } + ShowStatus("Finished killing child processes off."); } @@ -2209,12 +2173,12 @@ sub really_kill_them_all_dammit { Debug(2, "Kill them all Dammit"); local($SIG{CHLD} = 'IGNORE'); # In case some purist reenabled them. - foreach my $pid (keys %ChildHash) { - my $serving = $ChildHash{$pid}; + foreach my $pid (keys %ChildPid) { + my $serving = $ChildPid{$pid}; &ShowStatus("Nastily killing lonc for $serving pid = $pid"); Log("CRITICAL", "Nastily killing lonc for $serving pid = $pid"); kill 'KILL' => $pid; - delete($ChildHash{$pid}); + delete($ChildPid{$pid}); my $execdir = $perlvar{'lonDaemons'}; unlink("$execdir/logs/lonc.pid"); } @@ -2242,6 +2206,14 @@ sub Terminate { exit 0; } + +sub my_hostname { + use Sys::Hostname; + my $name = &hostname(); + &Debug(9,"Name is $name"); + return $name; +} + =pod =head1 Theory