-Nonstop xCAT, xcatd starting replaces old xCAT instance with nearly no interruption
git-svn-id: https://svn.code.sf.net/p/xcat/code/xcat-core/trunk@6680 8638fb3e-16cb-4fca-ae20-7b5d299a9bcd
This commit is contained in:
jbjohnso
parent
2cf271a1f9
commit
e032aa9375
@ -67,7 +67,7 @@ require xCAT::NotifHandler;
|
||||
|
||||
my $dbworkerpid; #The process id of the database worker
|
||||
my $dbworkersocket;
|
||||
my $dbsockpath = "/tmp/xcat/dbworker.sock";
|
||||
my $dbsockpath = "/tmp/xcat/dbworker.sock.".$$;
|
||||
my $exitdbthread;
|
||||
my $dbobjsforhandle;
|
||||
my $intendedpid;
|
||||
|
||||
@ -191,8 +191,23 @@ sub daemonize {
|
||||
my %cmd_handlers;
|
||||
sub do_installm_service {
|
||||
#This function servers as a handler for messages from installing nodes
|
||||
|
||||
my $socket;
|
||||
my $installpidfile;
|
||||
my $retry=1;
|
||||
$SIG{USR2} = sub {
|
||||
xCAT::MsgUtils->message("S","xcatd install monitor $$ quiescing");
|
||||
unlink("/tmp/xcat/installservice.pid"); close($socket); $quit=1; };
|
||||
if (open($installpidfile,"<","/tmp/xcat/installservice.pid")) {
|
||||
my $pid = <$installpidfile>;
|
||||
if ($pid) {
|
||||
$retry=100; #grace period for old instance to get out of the way, 5 seconds
|
||||
kill 12,$pid;
|
||||
yield(); # let peer have a shot at closure
|
||||
}
|
||||
close($installpidfile);
|
||||
}
|
||||
while (not $socket and $retry) {
|
||||
$retry--;
|
||||
if ($inet6support) {
|
||||
$socket = IO::Socket::INET6->new(LocalPort=>$sport,
|
||||
Proto => 'tcp',
|
||||
@ -204,6 +219,11 @@ if ($inet6support) {
|
||||
ReuseAddr => 1,
|
||||
Listen => 8192);
|
||||
}
|
||||
sleep 0.05; #up to 50 ms outage possible
|
||||
}
|
||||
open($installpidfile,">","/tmp/xcat/installservice.pid"); #if here, everyone else has unlinked installservicepid or doesn't care
|
||||
print $installpidfile $$;
|
||||
close($installpidfile);
|
||||
|
||||
unless ($socket) {
|
||||
xCAT::MsgUtils->message("S","xcatd unable to open install monitor services on $sport");
|
||||
@ -337,6 +357,13 @@ if ($inet6support) {
|
||||
}
|
||||
}
|
||||
}
|
||||
if (open($installpidfile,"<","/tmp/xcat/installservice.pid")) {
|
||||
my $pid = <$installpidfile>;
|
||||
if ($pid == $$) { #if our pid, unlink the file, otherwise, we managed to see the pid after someone else created it
|
||||
unlink("/tmp/xcat/installservice.pid");
|
||||
}
|
||||
close($installpidfile);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@ -349,8 +376,23 @@ sub do_udp_service { #This function opens up a UDP port
|
||||
#Explicitly, to handle whatever operations nodes periodically send during discover state
|
||||
#Could be used for heartbeating and such as desired
|
||||
$dispatch_requests=0;
|
||||
my $socket;
|
||||
my $udppidfile;
|
||||
my $retry=1;
|
||||
my $socket;
|
||||
$SIG{USR2} = sub {
|
||||
xCAT::MsgUtils->message("S","xcatd udp service $$ quiescing");
|
||||
unlink("/tmp/xcat/udpservice.pid"); close($socket); $quit=1; };
|
||||
if (open($udppidfile,"<","/tmp/xcat/udpservice.pid")) {
|
||||
my $pid = <$udppidfile>;
|
||||
if ($pid) {
|
||||
$retry=100; #grace period for old instance to get out of the way, 5 seconds
|
||||
kill 12,$pid;
|
||||
yield(); # let peer have a shot at closure
|
||||
}
|
||||
close($udppidfile);
|
||||
}
|
||||
my $select = new IO::Select;
|
||||
while (not $socket and $retry) {
|
||||
if ($inet6support) {
|
||||
$socket = IO::Socket::INET6->new(LocalPort => $port,
|
||||
Proto => 'udp',
|
||||
@ -360,6 +402,11 @@ if ($inet6support) {
|
||||
Proto => 'udp',
|
||||
Domain => AF_INET);
|
||||
}
|
||||
sleep 0.05;
|
||||
}
|
||||
open($udppidfile,">","/tmp/xcat/udpservice.pid"); #if here, everyone else has unlinked udpservicepid or doesn't care
|
||||
print $udppidfile $$;
|
||||
close($udppidfile);
|
||||
|
||||
openlog("xCAT UDP",'','local4');
|
||||
unless ($socket) {
|
||||
@ -435,6 +482,13 @@ if ($inet6support) {
|
||||
exit 1;
|
||||
}
|
||||
}
|
||||
if (open($udppidfile,"<","/tmp/xcat/udpservice.pid")) {
|
||||
my $pid = <$udppidfile>;
|
||||
if ($pid == $$) { #if our pid, unlink the file, otherwise, we managed to see the pid after someone else created it
|
||||
unlink("/tmp/xcat/udpservice.pid");
|
||||
}
|
||||
close($udppidfile);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@ -487,15 +541,19 @@ unless ($foreground) {
|
||||
|
||||
$dbmaster=xCAT::Table::init_dbworker;
|
||||
my $CHILDPID=0; #Global for reapers
|
||||
my %immediatechildren;
|
||||
sub generic_reaper {
|
||||
while (waitpid(-1,WNOHANG) > 0) {
|
||||
while (($CHILDPID=waitpid(-1,WNOHANG)) > 0) {
|
||||
yield;
|
||||
}
|
||||
$SIG{CHLD} = \&generic_reaper;
|
||||
}
|
||||
|
||||
sub ssl_reaper {
|
||||
while (waitpid(-1,WNOHANG) > 0) {
|
||||
while (($CHILDPID=waitpid(-1,WNOHANG)) > 0) {
|
||||
if ($immediatechildren{$CHILDPID}) {
|
||||
delete $immediatechildren{$CHILDPID};
|
||||
}
|
||||
$sslclients--;
|
||||
}
|
||||
$SIG{CHLD} = \&ssl_reaper;
|
||||
@ -569,21 +627,51 @@ unless ($pid_MON) {
|
||||
xexit(0);
|
||||
}
|
||||
$$progname="xcatd: SSL listener";
|
||||
|
||||
#setup signal in NotifHandler so that the cache can be updated
|
||||
xCAT::NotifHandler::setup($$, $dbmaster);
|
||||
|
||||
#start the monitoring process
|
||||
xCAT_monitoring::monitorctrl::start($$);
|
||||
|
||||
my $peername;
|
||||
my $ssltimeout;
|
||||
my $retry=1;
|
||||
openlog("xCAT SSL","","local4");
|
||||
my $listener;
|
||||
if ($inet6support) {
|
||||
$listener = IO::Socket::INET6->new(
|
||||
LocalPort => $port,
|
||||
Listen => 8192,
|
||||
Reuse => 1,
|
||||
);
|
||||
} else {
|
||||
$listener = IO::Socket::INET->new(
|
||||
LocalPort => $port,
|
||||
Listen => 8192,
|
||||
Reuse => 1,
|
||||
);
|
||||
my $mainpidfile;
|
||||
$SIG{USR2} = sub {
|
||||
xCAT::MsgUtils->message("S","xcatd main service $$ quiescing");
|
||||
unlink("/tmp/xcat/mainservice.pid"); close($listener); $quit=1; $listener=0; };
|
||||
if (open($mainpidfile,"<","/tmp/xcat/mainservice.pid")) {
|
||||
my $pid = <$mainpidfile>;
|
||||
if ($pid) {
|
||||
$retry=100; #grace period for old instance to get out of the way, 5 seconds
|
||||
kill 12,$pid;
|
||||
yield(); # let peer have a shot at closure
|
||||
}
|
||||
close($mainpidfile);
|
||||
}
|
||||
while (not $listener and $retry) {
|
||||
$retry--;
|
||||
if ($inet6support) {
|
||||
$listener = IO::Socket::INET6->new(
|
||||
LocalPort => $port,
|
||||
Listen => 8192,
|
||||
Reuse => 1,
|
||||
);
|
||||
} else {
|
||||
$listener = IO::Socket::INET->new(
|
||||
LocalPort => $port,
|
||||
Listen => 8192,
|
||||
Reuse => 1,
|
||||
);
|
||||
}
|
||||
sleep(0.05);
|
||||
}
|
||||
open($mainpidfile,">","/tmp/xcat/mainservice.pid"); #if here, everyone else has unlinked mainservicepid or doesn't care
|
||||
print $mainpidfile $$;
|
||||
close($mainpidfile);
|
||||
|
||||
unless ($listener) {
|
||||
kill 2, $pid_UDP;
|
||||
@ -597,15 +685,6 @@ unless ($listener) {
|
||||
die "ERROR:Unable to start xCAT service on port $port.";
|
||||
}
|
||||
closelog();
|
||||
|
||||
#setup signal in NotifHandler so that the cache can be updated
|
||||
xCAT::NotifHandler::setup($$, $dbmaster);
|
||||
|
||||
#start the monitoring process
|
||||
xCAT_monitoring::monitorctrl::start($$);
|
||||
|
||||
my $peername;
|
||||
my $ssltimeout;
|
||||
until ($quit) {
|
||||
$SIG{CHLD} = \&ssl_reaper; #set here to ensure that signal handler is not corrupted during loop
|
||||
next unless my $cnnection=$listener->accept;
|
||||
@ -614,6 +693,9 @@ until ($quit) {
|
||||
sleep 0.1; #Keep processor utilization down
|
||||
}
|
||||
my $child = xCAT::Utils->xfork(); #Yes we fork, IO::Socket::SSL is not threadsafe..
|
||||
if ($child) {
|
||||
$immediatechildren{$child}=1;
|
||||
}
|
||||
|
||||
unless (defined $child) {
|
||||
xCAT::MsgUtils->message("S","xCATd cannot fork");
|
||||
@ -691,7 +773,15 @@ if ($inet6support) {
|
||||
$sslclients++; #THROTTLE
|
||||
$cnnection->close();
|
||||
}
|
||||
$listener->close;
|
||||
if (open($mainpidfile,"<","/tmp/xcat/mainservice.pid")) {
|
||||
my $pid = <$mainpidfile>;
|
||||
if ($pid == $$) { #if our pid, unlink the file, otherwise, we managed to see the pid after someone else created it
|
||||
unlink("/tmp/xcat/mainservice.pid");
|
||||
}
|
||||
close($mainpidfile);
|
||||
}
|
||||
if ($listener) { $listener->close; }
|
||||
while (keys %immediatechildren) { yield(); }
|
||||
xCAT::Table::shut_dbworker;
|
||||
if ($dbmaster) {
|
||||
kill 2, $dbmaster;
|
||||
|
||||
Loading…
Reference in New Issue
Block a user