This patch causes mon to run alerts asychronously.
diff -ur 0.99.1/mon work/mon
--- 0.99.1/mon Sat Aug 18 15:37:53 2001
+++ work/mon Fri Sep 7 10:20:26 2001
@@ -2772,10 +2775,25 @@
my ($summary, $tmnow, $buf);
$tmnow = time;
+
+ #
+ # Since this only waits while a monitor is running, alerts which are
+ # run while no monitors are active will become zombies. They'll be
+ # reaped when a monitor is next started up. This should be harmless
+ # with any normal configuration. It could be fixed by either
+ # removing this test (and so always checking for dead kids, even
+ # when no monitors are active), or by integrating non-monitors into
+ # %runningpid and testing that rather than %running here.
+ #
return if (keys %running == 0);
while ((my $p = waitpid (-1, &WNOHANG)) >0)
{
+ if (!exists $runningpid{$p}) {
+ debug (1, "reaped non-monitor, pid $p status $?\n");
+ next;
+ }
+
my ($group, $service) = split (/\//, $runningpid{$p});
my $sref = \%{$watch{$group}->{$service}};
@@ -4531,14 +4549,39 @@
syslog ("alert", "calling $alerttype $a for" .
" $args{group}/$args{service} ($alert,$args{args}) $summary");
- my $pid = open (ALERT, "|-");
+ #
+ # As a simple way to avoid blocking while writing data to the alert,
+ # I fork an extra kid to do the job. First I fork one kid. This
+ # one forks another which execs the alert, then it writes the data
+ # to it. The server process continues without waiting.
+ #
+ # XXX This could be done more efficiently by forking a single kid to
+ # exec the alert, then using the main select loop to write data to
+ # it.
+ #
+ my $pid = fork;
if (!defined $pid) {
syslog ('err', "could not fork: $!");
return undef;
}
#
- # child, the actual alert
+ # first child, fork off the alert and feed data to it
+ #
+ if ($pid == 0) {
+ $pid = open (ALERT, "|-");
+ if (!defined $pid) {
+ syslog ('err', "could not fork: $!");
+ exit;
+ }
+ if ($pid != 0) {
+ print ALERT $args{"output"};
+ exit;
+ }
+ }
+
+ #
+ # second child, the actual alert
#
if ($pid == 0) {
#
@@ -4599,13 +4642,6 @@
}
exit;
}
-
- #
- # this will block if the alert is sucking gas
- #
- print ALERT $args{"output"};
- close (ALERT);
- waitpid $pid, 0;
#
# test alerts don't count
--
Roderick Schertler
[EMAIL PROTECTED]
