improve reliability:
- don't invoke intermediate shells on exec(), allowing to actually kill children - when we are going down, only kill remaining children, not the whole pgrp - don't use a SIGCHLD handler, but call waitpid explicitly at the right places - exit(1) in case of an error others: - a little cleanup - new option -t <timeout> ok pval@
This commit is contained in:
sturm
parent
ff1520f3fd
commit
91eb345927
@ -55,7 +55,7 @@ our $MAKEFLAGS = "BATCH=Yes BIN_PACKAGES=Yes BULK=Yes TRUST_PACKAGES=Yes";
|
||||
our $PORTSDIR = $ENV{'PORTSDIR'} || "/usr/ports";
|
||||
our $TMPDIR = $ENV{'PKG_TMPDIR'} || '/var/tmp';
|
||||
our $TIMEOUT = 10;
|
||||
our $SSH = "/usr/bin/ssh -n -o ConnectTimeout=$TIMEOUT";
|
||||
our @SSH = ("/usr/bin/ssh", "-n", "-o ConnectTimeout=$TIMEOUT");
|
||||
|
||||
# -A <Arch>: specify architecture of build hosts
|
||||
# -b: build dependency file
|
||||
@ -65,8 +65,9 @@ our $SSH = "/usr/bin/ssh -n -o ConnectTimeout=$TIMEOUT";
|
||||
# -L <Logdir>: use <Logdir> instead of $PORTSDIR/logs/$ARCH
|
||||
# -S <SUBDIRLIST>: use <SUBDIRLIST> instead of all ports
|
||||
# -T <Dependency File>: use <Dependency File> instead of a temporary one
|
||||
our ($opt_A, $opt_b, $opt_d, $opt_e, $opt_F, $opt_L, $opt_S, $opt_T);
|
||||
getopts('A:bdeF:L:S:T:');
|
||||
# -t <Timeout>: use this timeout instead of the default
|
||||
our ($opt_A, $opt_b, $opt_d, $opt_e, $opt_F, $opt_L, $opt_S, $opt_T, $opt_t);
|
||||
getopts('A:bdeF:L:S:T:t:');
|
||||
|
||||
$ARCH = $opt_A if defined $opt_A;
|
||||
|
||||
@ -81,6 +82,8 @@ unless (defined $opt_T) {
|
||||
UNLINK => 0 );
|
||||
}
|
||||
|
||||
$TIMEOUT = $opt_t if defined $opt_t;
|
||||
|
||||
our @dead_children = ();
|
||||
|
||||
sub child_handler()
|
||||
@ -95,6 +98,19 @@ sub child_handler()
|
||||
}
|
||||
}
|
||||
|
||||
sub term_handler()
|
||||
{
|
||||
local $SIG{CHLD} = "IGNORE";
|
||||
local $SIG{INT} = "IGNORE";
|
||||
local $SIG{TERM} = "IGNORE";
|
||||
|
||||
foreach my $h (keys %{$CHECK_HOSTS}, keys %{$CHILD}) {
|
||||
kill INT => $h;
|
||||
}
|
||||
|
||||
clean_up(1);
|
||||
}
|
||||
|
||||
sub reap_children()
|
||||
{
|
||||
while (my $c = pop @dead_children) {
|
||||
@ -104,7 +120,9 @@ sub reap_children()
|
||||
|
||||
sub mark_as_down($)
|
||||
{
|
||||
push(@DOWN_HOSTS, shift);
|
||||
my $host = shift;
|
||||
print "*** lost $host\n";
|
||||
push(@DOWN_HOSTS, $host);
|
||||
}
|
||||
|
||||
sub mark_as_free($)
|
||||
@ -122,25 +140,26 @@ sub check_host($)
|
||||
# parent
|
||||
my $begin = time();
|
||||
$CHECK_HOSTS->{$pid} = undef;
|
||||
child_handler();
|
||||
while (not defined $CHECK_HOSTS->{$pid}) {
|
||||
# give ssh a chance to timeout by itself
|
||||
if ($begin + $TIMEOUT + 2 > time()) {
|
||||
sleep(1);
|
||||
} else {
|
||||
# ssh did not terminate in time, kill it
|
||||
kill('TERM', $pid);
|
||||
kill INT => $pid;
|
||||
return -1;
|
||||
}
|
||||
child_handler();
|
||||
}
|
||||
return $CHECK_HOSTS->{$pid};
|
||||
} else {
|
||||
# child
|
||||
$SIG{CHLD} = "DEFAULT";
|
||||
$SIG{INT} = "DEFAULT";
|
||||
$SIG{TERM} = "DEFAULT";
|
||||
|
||||
exec("$SSH $host exit 0 > /dev/null 2>&1");
|
||||
die("exec(): $!");
|
||||
exec @SSH, $host, "exit 0";
|
||||
die "exec(): $!";
|
||||
}
|
||||
}
|
||||
|
||||
@ -167,7 +186,6 @@ sub check_hosts()
|
||||
my $host = $FREE_HOSTS[$i];
|
||||
my $retval = check_host($host);
|
||||
if ($retval != 0) {
|
||||
print "*** lost $host\n";
|
||||
mark_as_down($host);
|
||||
splice(@FREE_HOSTS, $i, 1);
|
||||
$i--;
|
||||
@ -181,7 +199,6 @@ sub check_hosts()
|
||||
|
||||
if ($retval != 0) {
|
||||
my $port = $CHILD->{$pid}[PORT];
|
||||
print "*** lost $host\n";
|
||||
mark_as_down($host);
|
||||
delete $childpid{$port};
|
||||
delete $CHILD->{$pid};
|
||||
@ -192,6 +209,8 @@ sub check_hosts()
|
||||
sub update_after_child($)
|
||||
{
|
||||
my $pid = shift;
|
||||
return unless defined $CHILD->{$pid};
|
||||
|
||||
my $host = $CHILD->{$pid}[HOST];
|
||||
my $port = $CHILD->{$pid}[PORT];
|
||||
my $retval = $CHILD->{$pid}[RETVAL];
|
||||
@ -208,8 +227,6 @@ sub update_after_child($)
|
||||
|
||||
remove_port($port);
|
||||
} elsif ($retval == 255) {
|
||||
print "<== host $host is down\n";
|
||||
|
||||
delete $childpid{$port};
|
||||
mark_as_down($host);
|
||||
|
||||
@ -226,11 +243,14 @@ sub update_after_child($)
|
||||
|
||||
sub find_free_host()
|
||||
{
|
||||
child_handler();
|
||||
reap_children();
|
||||
check_hosts();
|
||||
|
||||
while (@FREE_HOSTS == 0) {
|
||||
sleep(1);
|
||||
|
||||
child_handler();
|
||||
reap_children();
|
||||
check_hosts();
|
||||
}
|
||||
@ -345,7 +365,6 @@ sub build_package($$$$)
|
||||
return;
|
||||
} else {
|
||||
# child
|
||||
$SIG{CHLD} = "DEFAULT";
|
||||
$SIG{INT} = "DEFAULT";
|
||||
$SIG{TERM} = "DEFAULT";
|
||||
$0 = "dpb [slave] - $port";
|
||||
@ -357,14 +376,18 @@ sub build_package($$$$)
|
||||
if (defined $opt_d) {
|
||||
sleep(1);
|
||||
} else {
|
||||
my $arg = "$SSH $host 'cd $PORTSDIR/$port && ";
|
||||
my $arg = "cd $PORTSDIR/$port && ";
|
||||
$arg .= "FLAVOR=\"$flavor\" " if defined $flavor;
|
||||
$arg .= "$MAKE $MAKEFLAGS package' > $FIFO{$host} 2>&1";
|
||||
$arg .= "$MAKE $MAKEFLAGS package";
|
||||
open STDOUT, '>', "$FIFO{$host}" or
|
||||
die "Cannot redirect STDOUT: $!";
|
||||
open STDERR, ">&STDOUT" or
|
||||
die "Cannot redirect STDERR: $!";
|
||||
|
||||
start_logger($host);
|
||||
|
||||
exec($arg);
|
||||
die("exec(): $!");
|
||||
exec @SSH, $host, $arg;
|
||||
die "exec(): $!";
|
||||
}
|
||||
exit 0;
|
||||
}
|
||||
@ -401,11 +424,11 @@ sub start_logger()
|
||||
|
||||
unless (-p $FIFO{$host}) {
|
||||
system("mkfifo $FIFO{$host}") and
|
||||
die("Cannot create $FIFO{$host}: $!");
|
||||
die "Cannot create $FIFO{$host}: $!";
|
||||
}
|
||||
|
||||
my $pid = fork();
|
||||
die("fork: $!") unless defined $pid;
|
||||
die "fork: $!" unless defined $pid;
|
||||
|
||||
if ($pid > 0) {
|
||||
# parent
|
||||
@ -413,42 +436,37 @@ sub start_logger()
|
||||
return;
|
||||
} else {
|
||||
# child
|
||||
$SIG{CHLD} = "DEFAULT";
|
||||
# dies on its own on EOF
|
||||
$SIG{INT} = "DEFAULT";
|
||||
$SIG{TERM} = "DEFAULT";
|
||||
|
||||
exec("$LOGGER < $FIFO{$host} > /dev/null 2>&1");
|
||||
die("Failed to start logger: $!");
|
||||
die "Failed to start logger: $!";
|
||||
}
|
||||
}
|
||||
|
||||
sub clean_up()
|
||||
sub clean_up($)
|
||||
{
|
||||
$SIG{INT} = "IGNORE";
|
||||
$SIG{TERM} = "IGNORE";
|
||||
|
||||
kill('TERM', -$$);
|
||||
|
||||
# only remove self generated dependency file
|
||||
unlink($opt_T) if ref $opt_T;
|
||||
foreach my $h (keys %FIFO) {
|
||||
unlink($FIFO{$h});
|
||||
}
|
||||
|
||||
exit(0);
|
||||
exit(shift);
|
||||
}
|
||||
|
||||
# MAIN
|
||||
# collect dependency data
|
||||
$SIG{INT} = \&clean_up;
|
||||
$SIG{TERM} = \&clean_up;
|
||||
$SIG{INT} = \&term_handler;
|
||||
$SIG{TERM} = \&term_handler;
|
||||
$0 = "dpb [master]";
|
||||
|
||||
# collect dependency data
|
||||
if (defined $opt_b) {
|
||||
my $arg = "cd $PORTSDIR && $MAKE ";
|
||||
|
||||
if (defined $opt_S) {
|
||||
die("SUBDIRLIST $opt_S not found!") unless (-f $opt_S);
|
||||
die "SUBDIRLIST $opt_S not found!" unless (-f $opt_S);
|
||||
$arg .= "SUBDIRLIST=$opt_S ";
|
||||
}
|
||||
|
||||
@ -463,7 +481,6 @@ if (defined $opt_b) {
|
||||
parse_dependency_file();
|
||||
parse_hosts_file();
|
||||
|
||||
$SIG{CHLD} = \&child_handler;
|
||||
check_hosts();
|
||||
|
||||
my @keys_prereqs = (keys %prereqs_of);
|
||||
@ -509,9 +526,11 @@ do {
|
||||
}
|
||||
}
|
||||
|
||||
reap_children();
|
||||
check_hosts();
|
||||
|
||||
child_handler();
|
||||
reap_children();
|
||||
|
||||
# create new key set, taking currently building ports into account
|
||||
@keys_childpid = (keys %childpid);
|
||||
@keys_prereqs = ();
|
||||
@ -523,5 +542,6 @@ do {
|
||||
|
||||
} while ($#keys_prereqs >= 0 or $#keys_childpid >= 0);
|
||||
|
||||
clean_up();
|
||||
print "==> done, cleaning up\n";
|
||||
clean_up(0);
|
||||
|
||||
|
||||
Loading…
Reference in New Issue
Block a user