--- loncom/loncnew 2003/08/19 09:31:46 1.19 +++ loncom/loncnew 2003/10/27 10:09:21 1.30 @@ -2,7 +2,7 @@ # The LearningOnline Network with CAPA # lonc maintains the connections to remote computers # -# $Id: loncnew,v 1.19 2003/08/19 09:31:46 foxr Exp $ +# $Id: loncnew,v 1.30 2003/10/27 10:09:21 foxr Exp $ # # Copyright Michigan State University Board of Trustees # @@ -45,6 +45,55 @@ # Change log: # $Log: loncnew,v $ +# Revision 1.30 2003/10/27 10:09:21 foxr +# Tighten up a few compares to eq and flip a few debug levels around... nothing +# critical +# +# Revision 1.29 2003/10/21 14:24:42 foxr +# Fix little typo that may explain growth of connections +# +# Revision 1.28 2003/10/14 15:36:21 albertel +# - making it easier to run loncnew, +# /etc/init.d/loncontrol startnew +# /etc/init.d/loncontrol restartnew +# will now start loncnew in place of lonc +# +# Revision 1.27 2003/10/07 11:23:03 foxr +# Installed and tested code to process reinit in parent server. +# +# Revision 1.26 2003/09/30 11:11:17 foxr +# Add book-keeping hashes to support the re-init procedure. +# +# Revision 1.25 2003/09/23 11:22:14 foxr +# Tested ability to receive sigusr2 This is now logged and must be +# properly implemented as a re-read of hosts and re-init of appropriate +# children. +# +# Revision 1.24 2003/09/16 09:46:42 foxr +# Added skeletal infrastructure to support SIGUSR2 update hosts request. +# +# Revision 1.23 2003/09/15 09:24:49 foxr +# Add use strict and fix all the fallout from that. +# +# Revision 1.22 2003/09/02 10:34:47 foxr +# - Fix errors in host dead detection logic (too many cases where the +# retries left were not getting incremented or just not checked). +# - Added some additional status to the ps axuww display: +# o Remaining retries on a host. +# o >>> DEAD <<< indicator if I've given up on a host. +# - Tested the SIGHUP will reset the retries remaining count (thanks to +# the above status stuff, and get allow the loncnew to re-try again +# on the host (thanks to the log). +# +# Revision 1.21 2003/08/26 09:19:51 foxr +# How embarrassing... put in the SocketTimeout function in loncnew and forgot +# to actually hook it into the LondTransaction. Added this to MakeLondConnection +# where it belongs... hopefully transactions (not just connection attempts) will +# timeout more speedily than the socket errors will catch it. +# +# Revision 1.20 2003/08/25 18:48:11 albertel +# - fixing a forgotten ; +# # Revision 1.19 2003/08/19 09:31:46 foxr # Get socket directory from configuration rather than the old hard coded test # way that I forgot to un-hard code. @@ -81,7 +130,7 @@ # Revision 1.10 2003/06/24 02:46:04 foxr # Put a limit on the number of times we'll retry a connection. # Start getting the signal stuff put in as well...note that need to get signals -# going or else 6the client will permanently give up on dead servers. +# going or else the client will permanently give up on dead servers. # # Revision 1.9 2003/06/13 02:38:43 foxr # Add logging in 'expected format' @@ -95,7 +144,7 @@ # complete coding to support deferred transactions. # # - +use strict; use lib "/home/httpd/lib/perl/"; use lib "/home/foxr/newloncapa/types"; use Event qw(:DEFAULT ); @@ -119,12 +168,6 @@ use LONCAPA::HashIterator; # # Disable all signals we might receive from outside for now. # -#$SIG{QUIT} = IGNORE; -#$SIG{HUP} = IGNORE; -#$SIG{USR1} = IGNORE; -#$SIG{INT} = IGNORE; -#$SIG{CHLD} = IGNORE; -#$SIG{__DIE__} = IGNORE; # Read the httpd configuration file to get perl variables @@ -137,20 +180,22 @@ my %perlvar = %{$perlvarref}; # parent and shared variables. my %ChildHash; # by pid -> host. +my %HostToPid; # By host -> pid. +my %HostHash; # by loncapaname -> IP. my $MaxConnectionCount = 10; # Will get from config later. my $ClientConnection = 0; # Uniquifier for client events. my $DebugLevel = 0; -my $NextDebugLevel= 10; # So Sigint can toggle this. +my $NextDebugLevel= 2; # So Sigint can toggle this. my $IdleTimeout= 3600; # Wait an hour before pruning connections. # # The variables below are only used by the child processes. # my $RemoteHost; # Name of host child is talking to. -my $UnixSocketDir= $perlvar{'lonSockDir'} +my $UnixSocketDir= $perlvar{'lonSockDir'}; my $IdleConnections = Stack->new(); # Set of idle connections my %ActiveConnections; # Connections to the remote lond. my %ActiveTransactions; # LondTransactions in flight. @@ -160,8 +205,8 @@ my $ConnectionCount = 0; my $IdleSeconds = 0; # Number of seconds idle. my $Status = ""; # Current status string. my $RecentLogEntry = ""; -my $ConnectionRetries=5; # Number of connection retries allowed. -my $ConnectionRetriesLeft=5; # Number of connection retries remaining. +my $ConnectionRetries=2; # Number of connection retries allowed. +my $ConnectionRetriesLeft=2; # Number of connection retries remaining. # # The hash below gives the HTML format for log messages @@ -259,7 +304,7 @@ sub GetPeername { my $peerip; if($AdrFamily == AF_INET) { ($peerport, $peerip) = sockaddr_in($peer); - my $peername = gethostbyaddr($iaddr, $AdrFamily); + my $peername = gethostbyaddr($peerip, $AdrFamily); return $peername; } elsif ($AdrFamily == AF_UNIX) { my $peerfile; @@ -280,7 +325,7 @@ sub Debug { my $level = shift; my $message = shift; if ($level <= $DebugLevel) { - Log("INFO", "-Debug- $message host = $RemotHost"); + Log("INFO", "-Debug- $message host = $RemoteHost"); } } @@ -320,7 +365,9 @@ sub ShowStatus { sub SocketTimeout { my $Socket = shift; - KillSocket($Socket); + KillSocket($Socket); # A transaction timeout also counts as + # a connection failure: + $ConnectionRetriesLeft--; } =pod @@ -334,8 +381,12 @@ Invoked each timer tick. sub Tick { my $client; - ShowStatus(GetServerHost()." Connection count: ".$ConnectionCount); - + if($ConnectionRetriesLeft > 0) { + ShowStatus(GetServerHost()." Connection count: ".$ConnectionCount + ." Retries remaining: ".$ConnectionRetriesLeft); + } else { + ShowStatus(GetServerHost()." >> DEAD <<"); + } # Is it time to prune connection count: @@ -343,7 +394,7 @@ sub Tick { ($WorkQueue->Count() == 0)) { # Idle connections and nothing to do? $IdleSeconds++; if($IdleSeconds > $IdleTimeout) { # Prune a connection... - $Socket = $IdleConnections->pop(); + my $Socket = $IdleConnections->pop(); KillSocket($Socket); } } else { @@ -352,7 +403,7 @@ sub Tick { # # For each inflight transaction, tick down its timeout counter. # - foreach $item (keys %ActiveTransactions) { + foreach my $item (keys %ActiveTransactions) { my $Socket = $ActiveTransactions{$item}->getServer(); $Socket->Tick(); } @@ -365,12 +416,18 @@ sub Tick { if ($ConnectionRetriesLeft > 0) { my $Connections = ($Requests <= $MaxConnectionCount) ? $Requests : $MaxConnectionCount; - Debug(1,"Work but no connections, start ".$Connections." of them"); - for ($i =0; $i < $Connections; $i++) { - MakeLondConnection(); + Debug(5,"Work but no connections, start ".$Connections." of them"); + my $successCount = 0; + for (my $i =0; $i < $Connections; $i++) { + $successCount += MakeLondConnection(); + } + if($successCount == 0) { # All connections failed: + Debug(5,"Work in queue failed to make any connectiouns\n"); + EmptyQueue(); # Fail pending transactions with con_lost. } } else { - Debug(1,"Work in queue, but gave up on connections..flushing\n"); + ShowStatus(GetServerHost()." >>> DEAD!!! <<<"); + Debug(5,"Work in queue, but gave up on connections..flushing\n"); EmptyQueue(); # Connections can't be established. } @@ -417,19 +474,19 @@ sub ServerToIdle { my $Socket = shift; # Get the socket. delete($ActiveTransactions{$Socket}); # Server has no transaction - &Debug(6, "Server to idle"); + &Debug(5, "Server to idle"); # If there's work to do, start the transaction: - $reqdata = $WorkQueue->dequeue(); # This is a LondTransaction - unless($reqdata eq undef) { - Debug(9, "Queue gave request data: ".$reqdata->getRequest()); + my $reqdata = $WorkQueue->dequeue(); # This is a LondTransaction + if ($reqdata ne undef) { + Debug(5, "Queue gave request data: ".$reqdata->getRequest()); &StartRequest($Socket, $reqdata); } else { # There's no work waiting, so push the server to idle list. - &Debug(8, "No new work requests, server connection going idle"); + &Debug(5, "No new work requests, server connection going idle"); $IdleConnections->push($Socket); } } @@ -475,7 +532,7 @@ sub ClientWritable { # request. &Debug(9,"Send result is ".$result." Defined: ".defined($result)); - if(defined($result)) { + if($result ne undef) { &Debug(9, "send result was defined"); if($result == length($Data)) { # Entire string sent. &Debug(9, "ClientWritable data all written"); @@ -546,7 +603,7 @@ The transaction that is being completed. =cut sub CompleteTransaction { - &Debug(6,"Complete transaction"); + &Debug(5,"Complete transaction"); my $Socket = shift; my $Transaction = shift; @@ -555,7 +612,7 @@ sub CompleteTransaction { StartClientReply($Transaction, $data); } else { # Delete deferred transaction file. Log("SUCCESS", "A delayed transaction was completed"); - LogPerm("S:$Client:".$Transaction->getRequest()); + LogPerm("S:$Transaction->getClient() :".$Transaction->getRequest()); unlink $Transaction->getFile(); } } @@ -585,7 +642,6 @@ sub StartClientReply { &Debug(8," Reply was: ".$data); my $Serial = $ActiveClients{$Client}; my $desc = sprintf("Connection to lonc client %d", - $Serial); Event->io(fd => $Client, poll => "w", @@ -623,6 +679,9 @@ sub FailTransaction { Debug(1," Replying con_lost to ".$transaction->getRequest()); StartClientReply($transaction, "con_lost\n"); } + if($ConnectionRetriesLeft <= 0) { + Log("CRITICAL", "Host marked dead: ".GetServerHost()); + } } @@ -634,6 +693,7 @@ sub FailTransaction { =cut sub EmptyQueue { + $ConnectionRetriesLeft--; # Counts as connection failure too. while($WorkQueue->Count()) { my $request = $WorkQueue->dequeue(); # This is a transaction FailTransaction($request); @@ -648,7 +708,7 @@ Close all connections open on lond prior =cut sub CloseAllLondConnections { - foreach $Socket (keys %ActiveConnections) { + foreach my $Socket (keys %ActiveConnections) { KillSocket($Socket); } } @@ -700,7 +760,7 @@ sub KillSocket { # work queue, the work all gets failed with con_lost. # if($ConnectionCount == 0) { - EmptyQueue; + EmptyQueue(); } } @@ -769,7 +829,7 @@ sub LondReadable { my $Socket = $Watcher->data; my $client = undef; - &Debug(6,"LondReadable called state = ".$State); + &Debug(6,"LondReadable called state = ".$Socket->GetState()); my $State = $Socket->GetState(); # All action depends on the state. @@ -790,6 +850,7 @@ sub LondReadable { } $Watcher->cancel(); KillSocket($Socket); + $ConnectionRetriesLeft--; # Counts as connection failure return; } SocketDump(6,$Socket); @@ -823,9 +884,13 @@ sub LondReadable { } elsif ($State eq "Idle") { # If necessary, complete a transaction and then go into the # idle queue. + # Note that a trasition to idle indicates a live lond + # on the other end so reset the connection retries. + # + $ConnectionRetriesLeft = $ConnectionRetries; # success resets the count $Watcher->cancel(); if(exists($ActiveTransactions{$Socket})) { - Debug(8,"Completing transaction!!"); + Debug(5,"Completing transaction!!"); CompleteTransaction($Socket, $ActiveTransactions{$Socket}); } else { @@ -1035,8 +1100,7 @@ sub QueueDelayed { Debug(4, "Delayed path: ".$path); opendir(DIRHANDLE, $path); - @alldelayed = grep /\.$RemoteHost$/, readdir DIRHANDLE; - Debug(4, "Got ".$alldelayed." delayed files"); + my @alldelayed = grep /\.$RemoteHost$/, readdir DIRHANDLE; closedir(DIRHANDLE); my $dfname; my $reqfile; @@ -1073,26 +1137,27 @@ sub MakeLondConnection { my $Connection = LondConnection->new(&GetServerHost(), &GetServerPort()); - if($Connection == undef) { # Needs to be more robust later. + if($Connection eq undef) { # Needs to be more robust later. Log("CRITICAL","Failed to make a connection with lond."); $ConnectionRetriesLeft--; return 0; # Failure. } else { - $ConnectionRetriesLeft = $ConnectionRetries; # success resets the count + # The connection needs to have writability # monitored in order to send the init sequence # that starts the whole authentication/key # exchange underway. # my $Socket = $Connection->GetSocket(); - if($Socket == undef) { + if($Socket eq undef) { die "did not get a socket from the connection"; } else { &Debug(9,"MakeLondConnection got socket: ".$Socket); } - - $event = Event->io(fd => $Socket, + $Connection->SetTimeoutCallback(\&SocketTimeout); + + my $event = Event->io(fd => $Socket, poll => 'w', cb => \&LondWritable, data => $Connection, @@ -1148,7 +1213,7 @@ sub StartRequest { $ActiveTransactions{$Lond} = $Request; $Lond->InitiateTransaction($Request->getRequest()); - $event = Event->io(fd => $Socket, + my $event = Event->io(fd => $Socket, poll => "w", cb => \&LondWritable, data => $Lond, @@ -1187,12 +1252,17 @@ sub QueueTransaction { my $LondSocket = $IdleConnections->pop(); if(!defined $LondSocket) { # Need to queue request. - Debug(8,"Must queue..."); + Debug(5,"Must queue..."); $WorkQueue->enqueue($requestData); if($ConnectionCount < $MaxConnectionCount) { - Debug(4,"Starting additional lond connection"); - if(MakeLondConnection() == 0) { - EmptyQueue(); # Fail transactions, can't make connection. + if($ConnectionRetriesLeft > 0) { + Debug(5,"Starting additional lond connection"); + if(MakeLondConnection() == 0) { + EmptyQueue(); # Fail transactions, can't make connection. + } + } else { + ShowStatus(GetServerHost()." >>> DEAD !!!! <<<"); + EmptyQueue(); # It's worse than that ... he's dead Jim. } } } else { # Can start the request: @@ -1224,7 +1294,7 @@ sub ClientRequest { my $rv = $socket->recv($thisread, POSIX::BUFSIZ, 0); Debug(8, "rcv: data length = ".length($thisread) ." read =".$thisread); - unless (defined $rv && length($thisread)) { + unless (defined $rv && length($thisread)) { # Likely eof on socket. Debug(5,"Client Socket closed on lonc for ".$RemoteHost); close($socket); @@ -1358,6 +1428,8 @@ sub SetupLoncListener { Child USR1 signal handler to report the most recent status into the status file. +We also use this to reset the retries count in order to allow the +client to retry connections with a previously dead server. =cut sub ChildStatus { my $event = shift; @@ -1368,6 +1440,7 @@ sub ChildStatus { my $fh = IO::File->new(">>$docdir/lon-status/loncstatus.txt"); print $fh $$."\t".$RemoteHost."\t".$Status."\t". $RecentLogEntry."\n"; + $ConnectionRetriesLeft = $ConnectionRetries; } =pod @@ -1463,15 +1536,16 @@ sub CreateChild { my $host = shift; $RemoteHost = $host; Log("CRITICAL", "Forking server for ".$host); - $pid = fork; + my $pid = fork; if($pid) { # Parent $RemoteHost = "Parent"; - $ChildHash{$pid} = $RemoteHost; + $ChildHash{$pid} = $host; + $HostToPid{$host}= $pid; sigprocmask(SIG_UNBLOCK, $sigset); } else { # child. ShowStatus("Connected to ".$RemoteHost); - $SIG{INT} = DEFAULT; + $SIG{INT} = 'DEFAULT'; sigprocmask(SIG_UNBLOCK, $sigset); ChildProcess; # Does not return. } @@ -1508,7 +1582,7 @@ if ($childpid != 0) { # ShowStatus("Parent writing pid file:"); -$execdir = $perlvar{'lonDaemons'}; +my $execdir = $perlvar{'lonDaemons'}; open (PIDSAVE, ">$execdir/logs/lonc.pid"); print PIDSAVE "$$\n"; close(PIDSAVE); @@ -1527,8 +1601,9 @@ Log("CRITICAL", "--------------- Startin my $HostIterator = LondConnection::GetHostIterator; while (! $HostIterator->end()) { - $hostentryref = $HostIterator->get(); + my $hostentryref = $HostIterator->get(); CreateChild($hostentryref->[0]); + $HostHash{$hostentryref->[0]} = $hostentryref->[4]; $HostIterator->next(); } $RemoteHost = "Parent Server"; @@ -1545,11 +1620,13 @@ $SIG{INT} = \&Terminate; $SIG{TERM} = \&Terminate; $SIG{HUP} = \&Restart; $SIG{USR1} = \&CheckKids; +$SIG{USR2} = \&UpdateKids; # LonManage update request. while(1) { - $deadchild = wait(); + my $deadchild = wait(); if(exists $ChildHash{$deadchild}) { # need to restart. - $deadhost = $ChildHash{$deadchild}; + my $deadhost = $ChildHash{$deadchild}; + delete($HostToPid{$deadhost}); delete($ChildHash{$deadchild}); Log("WARNING","Lost child pid= ".$deadchild. "Connected to host ".$deadhost); @@ -1579,7 +1656,7 @@ sub CheckKids { my $now=time; my $local=localtime($now); print $fh "LONC status $local - parent $$ \n\n"; - foreach $pid (keys %ChildHash) { + foreach my $pid (keys %ChildHash) { Debug(2, "Sending USR1 -> $pid"); kill 'USR1' => $pid; # Tell Child to report status. sleep 1; # Wait so file doesn't intermix. @@ -1588,6 +1665,114 @@ sub CheckKids { =pod +=head1 UpdateKids + +parent's SIGUSR2 handler. This handler: + +=item + +Rereads the hosts file. + +=item + +Kills off (via sigint) children for hosts that have disappeared. + +=item + +QUITs children for hosts that already exist (this just forces a status display +and resets the connection retry count for that host. + +=item + +Starts new children for hosts that have been added to the hosts.tab file since +the start of the master program and maintains them. + +=cut + +sub UpdateKids { + + Log("INFO", "Updating connections via SIGUSR2"); + + # Just in case we need to kill our own lonc, we wait a few seconds to + # give it a chance to receive and relay lond's response to the + # re-init command. + # + + sleep(2); # Wait a couple of seconds. + + my %hosts; # Indexed by loncapa hostname, value=ip. + + # Need to re-read the host table: + + + LondConnection::ReadConfig(); + my $I = LondConnection::GetHostIterator; + while (! $I->end()) { + my $item = $I->get(); + $hosts{$item->[0]} = $item->[4]; + $I->next(); + } + + # The logic below is written for clarity not for efficiency. + # Since I anticipate that this function is only rarely called, that's + # appropriate. There are certainly ways to combine the loops below, + # and anyone wishing to obscure the logic is welcome to go for it. + # Note that we don't re-direct sigchild. Instead we do what's needed + # to the data structures that keep track of children to ensure that + # when sigchild is honored, no new child is born. + # + + # For each existing child; if it's host doesn't exist, kill the child. + + foreach my $child (keys %ChildHash) { + my $oldhost = $ChildHash{$child}; + if (!(exists $hosts{$oldhost})) { + Log("CRITICAL", "Killing child for $oldhost host no longer exists"); + delete $ChildHash{$child}; + delete $HostToPid{$oldhost}; + kill 'QUIT' => $child; + } + } + # For each remaining existing child; if it's host's ip has changed, + # Restart the child on the new IP. + + foreach my $child (keys %ChildHash) { + my $oldhost = $ChildHash{$child}; + my $oldip = $HostHash{$oldhost}; + if ($hosts{$oldhost} ne $oldip) { + + # kill the old child. + + Log("CRITICAL", "Killing child for $oldhost host ip has changed..."); + delete $ChildHash{$child}; + delete $HostToPid{$oldhost}; + kill 'QUIT' => $child; + + # Do the book-keeping needed to start a new child on the + # new ip. + + $HostHash{$oldhost} = $hosts{$oldhost}; + CreateChild($oldhost); + } + } + # Finally, for each new host, not in the host hash, create a + # enter the host and create a new child. + # Force a status display of any existing process. + + foreach my $host (keys %hosts) { + if(!(exists $HostHash{$host})) { + Log("INFO", "New host $host discovered in hosts.tab..."); + $HostHash{$host} = $hosts{$host}; + CreateChild($host); + } else { + kill 'HUP' => $HostToPid{$host}; # status display. + } + } +} + + +=pod + =head1 Restart Signal handler for HUP... all children are killed and @@ -1597,11 +1782,11 @@ the config file. =cut sub Restart { - KillThemAll; # First kill all the children. + &KillThemAll; # First kill all the children. Log("CRITICAL", "Restarting"); my $execdir = $perlvar{'lonDaemons'}; unlink("$execdir/logs/lonc.pid"); - exec("$execdir/lonc"); + exec("$execdir/loncnew"); } =pod @@ -1616,7 +1801,7 @@ SIGHUP. Responds to sigint and sigterm. sub KillThemAll { Debug(2, "Kill them all!!"); local($SIG{CHLD}) = 'IGNORE'; # Our children >will< die. - foreach $pid (keys %ChildHash) { + foreach my $pid (keys %ChildHash) { my $serving = $ChildHash{$pid}; Debug(2, "Killing lonc for $serving pid = $pid"); ShowStatus("Killing lonc for $serving pid = $pid");