> > I suspect we need a general timeout mechanism for plugin dispatches.
> > Right now, we only timeout input.
>
> I've raised this issue before - search for thread "smtp connections
> never timeout". Ask acknowledged this as a "pretty serious bug", but I
> don't think there has ever been any followup.
>
> I don't think we want to add timeout login to each plugin - there
> should be a timeout within dispatch().
I agree.
But do we really want to use signals for it? If so, it would be
pretty easy to put a simple timeout handler in TcpServer::dispatch().
consider this proof of concept.
I haven't tried this:
# TcpServer::dispatch
sub dispatch {
my $self = shift;
my ($cmd) = lc shift;
my $result;
local $SIG{ALRM} = sub {
die "timeout in plugin '$cmd'\n";
};
alarm $self->config("plugin_timeout") || 120;
$result = $self->SUPER::dispatch(@_);
if ($@ ~= /^timeout /) {
$self->fault("timeout dispatching plugin '$cmd'");
} else {
die $@ if $@; # rethrow
}
alarm 0;
return $result;
}
The eval is necessary to prevent the timeout from killing the entire
program, unless of course that's what we want. (But I don't think it
is.)
The actual plugin dispatch is also wrapped in an eval, and that will
catch timeouts in the plugin itself. If for some reason there's a
timeout not inside that inner eval, we'll catch it here.
-R
